JPH0749945A - Discriminating device - Google Patents

Abstract

PURPOSE:To improve the efficiency of an arithmetic processing for executing the discrimination and a discrimination rate of a discrimination object by executing a discriminating operation for discriminating the discrimination object by using a specified sensor signal and a weight coefficient stored in a memory. CONSTITUTION:This device is provided with an arithmetic part 13 for executing a discriminating operation, based on a sensor signal and a weight coefficient, a learning part 14 for executing learning related to the weighting coefficient, a memory 15 for storing the weighting coefficient, etc. Also, a control part 16 generates and outputs a sensor output control signal to a sensor signal input part 12 in order to fetch only the sensor signal outputted from a sensor 10, and also, generates and outputs a discriminating operation control signal to the arithmetic part 13, as well in order to execute an operation by using the weighting coefficient corresponding to the sensor signal outputted from the sensor 10. In such a way, by using only the sensor signal whose degree of contribution to discrimination of a discrimination object is high, in the sensor signals outputted from a sensor group 11, and its weighting coefficient, the discriminating operation is executed and the discrimination is executed, therefore, the discriminating operation does not become redundant, the discrimination can be executed efficiently, and the discrimination rate can be improved.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an identification device which is suitable for use in a manufacturing process, for example, and which can reduce the amount of calculation processing when identifying an identification object and can perform identification efficiently.

[0002]

2. Description of the Related Art FIG. 29 is a block diagram showing the structure of a conventional identification device. In the figure, reference numeral 1 denotes a sensor for dividing the wavelength range of visible light into a certain range and receiving and detecting n spectra respectively. The spectroscope 2 made up of a group is an arithmetic unit that calculates a plurality of output signals output from the spectroscope 1 and outputs the identification result of the identification object.

Next, the operation will be described. In this identification device, an object A, an object B, and an object C that are objects of identification are
The spectroscopic spectrum of 1 is detected by the spectroscope 1. As a result, the output signal of the object A from the spectroscope 1 has a spectrum as shown in FIG. 30 (a), and the object B has the spectrum shown in FIG. 30 (b). If the object C has the spectrum shown in FIG. 7C, the object A, the object B, and the object C are calculated for the components of wavelength p, wavelength q, and wavelength r, respectively. The object A, the object B, and the object C are identified and distinguished. Further, when there are a large number of spectral spectra of the object A, the object B, and the object C in addition to the wavelength p, the wavelength q, and the wavelength r, calculation is performed for all these wavelength components, and the object A and the object B and the object C will be identified and distinguished.

[0004]

Since the conventional identification device is constructed as described above, the spectrum output from the spectroscope 1 without considering the characteristics of the object to be identified. As a result of performing the arithmetic processing for identification on all the spectra, there are problems that the time required for the arithmetic processing becomes long and the accuracy rate of the identification result decreases.

The inventions of claims 1 to 7 have been made to solve the above-mentioned problems, and are capable of efficiently performing the arithmetic processing for identification and improving the identification rate. Aim to get.

It is an object of the present invention to obtain an identification device having an improved man / machine interface.

It is an object of the invention of claim 13 to obtain an identification device which can identify an object to be identified based on a gloss characteristic.

It is an object of the present invention to provide an identifying device for identifying an object to be identified based on gloss characteristics and color characteristics.

It is an object of the invention of claim 15 to obtain an identification device capable of simultaneously inputting a plurality of color features.

[0010]

The identification device according to the invention of claim 1 performs learning for identifying an object to be identified,
A calculation unit executes a discrimination calculation for discriminating the discrimination target object by using a sensor signal group outputted from a specific sensor group having a high degree of contribution to discrimination and a weighting coefficient stored in a memory. It is provided with a control unit for controlling.

According to a second aspect of the present invention, an identification device performs learning for identifying an object to be identified based on color-separated three-primary color signals, and a specific three-primary-color signal and a memory having a high degree of contribution to the identification are stored in a memory. A control unit that controls the calculation unit to execute the identification calculation by using the stored weighting factor is provided.

According to the third aspect of the present invention, the discrimination device performs learning for discriminating the discrimination target object with respect to the physical characteristics of the surface of the discrimination target object based on the image Fourier-transformed by the lens, and performs the discrimination. A control unit that controls a calculation unit to execute a discrimination calculation for discriminating the discrimination target object by using a specific sensor signal having a high degree of contribution and a weighting coefficient stored in the memory Is.

According to a fourth aspect of the present invention, there is provided an identification device for identifying the identification target object with respect to a pattern of the identification target object based on a sensor signal obtained from a planar arrangement sensor including photoelectric elements arranged on a plane. A control unit that performs learning and controls such that the calculation unit executes a discrimination calculation for discriminating the discrimination target using the sensor signal that contributes to discrimination of the pattern and the weighting coefficient stored in the memory. It is equipped with and.

According to a fifth aspect of the present invention, there is provided an identification device for identifying the identification target object by using geometrical characteristics of the identification target object based on a sensor signal obtained from a sensor including a photoelectric element group arranged on a plane. And a control unit that controls the arithmetic unit to perform an identification operation for identifying the identification target object by using the geometrical features that contribute to the identification and the weighting factors stored in the memory. It is equipped with.

According to a sixth aspect of the present invention, in the identification device, the controller outputs the sensor output control signal based on the partial differential value of the output layer of the memory, which is composed of the weighting factors forming the multilayer neural network, with respect to the input layer. It was created.

According to a seventh aspect of the present invention, an identifying device identifies an input signal used for learning when obtaining a partial differential value with respect to an input layer of an output layer of a memory, which is composed of weighting factors constituting a multilayer neural network. The intermediate value of the input signal is selected for each object and the intermediate value is generated to obtain the partial differential value.

According to another aspect of the present invention, there is provided an identification device which performs an inverse operation from a given identification state and a weighting coefficient stored in a memory and which is used for identification, and outputs a desired sensor signal and a desirable identification. The learning unit performs learning by calculating a weighting coefficient from the error with the signal indicating the state and modifying the weighting coefficient of the memory, and instructing the computing unit to correct the signal indicating the desired identification state and the coefficient. A signal is input from an input device and the generated predicted sensor signal is displayed on a display device.

In the identification device according to the ninth aspect of the present invention, the three-dimensional model synthesis operation unit synthesizes the three-dimensional model of the identification target using the average image data created from the image data of the plurality of learning samples related to the identification target. It is something that is done.

According to a tenth aspect of the present invention, the three-dimensional model display device displays the three-dimensional model of the identification object and the original image synthesized by the calculation unit, and the user can display the three-dimensional model of the identification object. The difference between the model and the original image is input from the difference information input device.

According to an eleventh aspect of the present invention, there is provided an identification device which produces a three-dimensional model of an identification object and an original image which are combined by a difference image calculation unit, and an image of a difference between the three-dimensional model and the original image of the identification object. It is designed to be displayed on a display device.

According to the twelfth aspect of the present invention, in the identification device, the rank based on the rating scale is input from the rank scale input device to the plurality of combined three-dimensional models to be identified displayed on the three-dimensional model display device. It was done like this.

According to a thirteenth aspect of the present invention, an identification device extracts a plurality of pieces of gloss characteristic information from a signal output from an image pickup unit that picks up an object to be identified, and extracts the gloss characteristic information and the weighting coefficient stored in the memory. Is calculated and the discrimination result is calculated and output, and a weighting factor is calculated from an error between the output and a signal indicating a desired discrimination state, and the weighting factor is output to the memory for learning, and discrimination is performed from the signal and the weighting factor. The object is identified.

In the identification apparatus according to the fourteenth aspect of the present invention, the gloss feature extraction unit extracts a plurality of gloss feature information from the sensor signal, and a plurality of specific wavelengths are detected as color features by the color feature detection unit. The characteristic information, the color characteristic information, and the weighting coefficient stored in the memory are calculated, the discrimination result is output from the arithmetic unit, and the weighting factor is calculated from the error between the output of the arithmetic unit and the signal indicating the desired discrimination state to perform learning. Further, a sensor output control signal for taking in the specific color characteristic information is generated by the gloss characteristic information and the weighting coefficient stored in the memory, and the color characteristic detection unit is operated based on the sensor output control signal. The identification calculation is executed by the calculation unit using the detected color feature information and the weighting coefficient stored in the memory.

According to a fifteenth aspect of the present invention, in the identification device, the light of the identification object imaged by the imaging unit is partially shielded by a mask, and the light passing through the mask is reflected like a plurality of half mirrors. Based on the output of the photoelectric filter group that detects the light transmitted through each of the interference filters and the interference filter that transmits only the light of a specific wavelength to the reflected light,
The color feature information processing section for identifying the color feature of the identification object is provided.

[0025]

According to the identification device of the present invention, by performing learning, a sensor output control signal for taking in a specific sensor signal group having a high degree of contribution to the identification of the sensor signal is generated, and the sensor signal input unit. The sensor signal taken in by is specified as a sensor signal that contributes to the identification, and an identification calculation for identifying the identification target object is performed using the identified sensor signal and the weighting coefficient stored in the memory.
The sensor signal amount taken in by the sensor signal input unit is suppressed, and the efficiency of the arithmetic processing for performing identification and the identification rate of the identification target can be improved.

According to the second aspect of the present invention, the discrimination device learns the weighting factors used for discrimination with respect to the three primary color signals, and uses the specific three primary color signals having a high degree of contribution to the discrimination and the weighting factors stored in the memory. Since the identification calculation for identifying the identification target is performed, the efficiency of the calculation process for identifying the identification target and the identification rate of the identification target can be improved.

In the identification device according to the third aspect of the present invention, an image according to the physical characteristics of the surface of the identification object Fourier-transformed by the lens is detected by the plane arrangement sensor arranged on a plane, and the plane arrangement sensor is arranged. The weighting coefficient used for the discrimination calculation is learned for the sensor signal output from the sensor signal, and the discrimination calculation is performed by the weighting coefficient stored in the memory and the specific sensor signal having a high degree of contribution to the discrimination. Therefore, it is possible to improve the efficiency of the calculation process and the identification rate of the identification object.

According to a fourth aspect of the present invention, in the identification device, the pattern of the object to be identified is detected by a planar arrangement sensor including photoelectric elements arranged on a plane, and a sensor signal output from a sensor having a high degree of contribution to the identification. Is performed by learning the weighting coefficient used for identification, and the identification operation is performed by the specific sensor signal having a high degree of contribution to the identification and the weighting coefficient stored in the memory. It is possible to improve the efficiency and the identification rate.

In the identification apparatus according to the invention of claim 5, the identification object is detected by a sensor composed of a photoelectric element group arranged on a plane, and the geometrical feature of the identification object is extracted from the sensor signal to identify the object. Since the weighting coefficient used for the identification of the contributing geometric features is learned and the identification operation is performed by the geometrical features having a high degree of contributing to the identification and the corresponding weighting factors stored in the memory,
It is possible to improve the efficiency of the arithmetic processing for identifying the geometrical feature of the identification object and the identification rate.

In the discrimination device according to the invention of claim 6, the sensor signal having a high degree of contribution to discrimination is judged based on a partial differential value with respect to the input layer of the output layer of the memory composed of the weighting factors constituting the multilayer neural network. Based on this, a specific sensor signal that contributes to identification is captured, and an identification calculation for identifying the identification object is performed using this sensor signal and the weighting coefficient stored in the memory. The sensor signal amount taken in by the input unit is suppressed, and it is possible to improve the efficiency of the arithmetic processing for performing identification and the identification rate.

In the identification apparatus according to the invention of claim 7, the weighting coefficient used for the identification operation for identifying the identification object is learned, and the sensor signal having a high degree of contribution to the identification has a multilayer neural network. Partial differential value for the input layer of the output layer of the memory composed of the weighting factor, especially the input signal used for learning is divided for each identification object, the intermediate value of the input signal is selected, and the intermediate value is used. Since the discrimination calculation for discriminating the discrimination target object is carried out by using the sensor signal and the weighting coefficient stored in the memory, which are taken in based on the obtained partial differential value, the sensor signal taken in by the sensor signal input section. The amount is suppressed, and it becomes possible to improve the efficiency of the arithmetic processing for performing identification and the identification rate.

According to an eighth aspect of the present invention, in the identification device, the output of the arithmetic unit for performing the inverse operation from the given identification state and the weighting coefficient stored in the memory to generate the predicted sensor signal and the signal indicating the desired identification state. The weighting coefficient used for discrimination is calculated from the error between the values of and and learning is performed, and the error is corrected by inputting a signal indicating a desired discrimination state and a signal instructing the correction of the weighting coefficient from the input device. It In this case, by displaying the generated predicted sensor signal on the display device, it becomes easy to compare the display with a learning sample and input a signal for instructing the correction, thereby improving the man / machine interface. It will be possible.

According to the ninth aspect of the present invention, the identification device synthesizes the three-dimensional model of the identification target by using the average image data created from the image data of a plurality of learning samples related to the identification target, and displays it on the display device. , Improve the man / machine interface by providing a visually easy-to-understand display.

The identification device according to the invention of claim 10 is 3
By displaying the three-dimensional model of the identification object and the original image on the three-dimensional model display device, it becomes possible to easily understand the difference between the three-dimensional model and the original image. The difference between the 3D model and the original image can be input, and the man / machine interface can be improved.

In the identification apparatus according to the invention of claim 11, the three-dimensional model of the identification object and the original image, and the difference image between the three-dimensional model of the identification object and the original image are displayed on the difference image display device. Therefore, the user can easily understand the difference between the three-dimensional model of the identification object and the original image, and the man / machine interface can be improved.

The identification device according to the invention of claim 12 is 3
By enabling the rank scale input device to input the rank based on the rating scale for the plurality of combined three-dimensional models to be identified displayed on the three-dimensional model display device, the weight coefficient can be modified. Therefore, it is possible to improve the man / machine interface.

According to a thirteenth aspect of the present invention, in the identification device, an identification result obtained by calculation from a plurality of gloss characteristic information extracted from a signal output from the image pickup section and a weighting coefficient stored in the memory, and a desirable identification state. Learning is performed by calculating a weighting coefficient from the error of, and the identification object is identified as a result of the identification operation using the gloss information and the weighting coefficient. Therefore, the identification object is efficiently evaluated based on its gloss feature. It is possible to identify well.

In the discriminating apparatus according to the fourteenth aspect of the invention, the weighting coefficient is learned from the difference between the discrimination result obtained by the gloss characteristic information, the color characteristic information and the weighting coefficient stored in the memory and the desired discrimination state. Further, the specific color feature information is fetched based on the gloss feature information and the weighting coefficient, and an identification calculation is performed using the color feature information and the weighting factor to obtain the gloss feature information and the color feature information. It is possible to efficiently identify the identification object based on the weighting factor.

In the identification apparatus according to the fifteenth aspect of the present invention, the light of the identification target imaged by the imaging unit is partially shielded by the mask, and the light passing through the mask is reflected by the half mirrors and specified by the interference filter. Only the light of the wavelength is transmitted, the color feature information of the identification object is detected by the photoelectric element group, and the color of the identification object can be identified based on the simultaneously detected color feature information.

[0040]

【Example】

Example 1. An embodiment of the invention of claim 1 will be described below with reference to the drawings. FIG. 1 is a block diagram showing the configuration of the identification device of this embodiment. In the figure, 11 is a plurality of sensors 1.
A sensor group consisting of 0, 12 is a sensor signal input section for taking in the sensor signal output from the sensor group 11, 13 is an arithmetic section for performing an identification operation based on the sensor signal and the weight coefficient, and 14 is a learning for the weight coefficient. Learning section to perform, 1
Reference numeral 5 is a memory for storing the weighting coefficient, and 16 is a control unit for controlling the sensor signal input unit 12 and the arithmetic unit 13.

Next, the operation will be described. FIG. 2 is a flowchart showing the operation of the identification device of FIG. Step ST1 is a learning process, step ST2 is an analysis process, step ST3 is a re-learning process, and step ST4 is a learning process by judging whether or not the error between the output of the operation unit 13 and the teacher signal is in the process of converging to a threshold value. Convergence determination processing for determining whether or not has ended, step ST5 is a sensor signal selection processing including analysis processing, re-learning processing, and convergence determination processing, step ST
Reference numeral 6 is an identification process for performing an identification calculation or the like for identifying the identification object.

First, the learning process of step ST1 will be described. The sensor output when the two types of objects A and B to be identified are measured by a plurality of sensors of the sensor group 1 is
It is assumed that the sensor p and the sensor q are characteristic as shown in FIG. 3A, and the sensor q and the sensor r are characteristic as shown in FIG. 3B. Description will be made assuming that learning is performed using a back propagation model of a network (neural network).

First, the weighting coefficient for discriminating between the object A and the object B is learned. The memory 15 is assumed to be initialized by an appropriate method. A large number of teacher samples of the object A and the object B are prepared, each of which is detected by the sensor group 1, and the detected signal is sent to the calculation unit 13 through the sensor signal input unit 2.

The computing unit 13 uses the signal sent from the sensor signal input unit 2 and the weighting coefficient stored in the memory 15 to perform a discrimination operation such as a sum of products calculation based on the characteristics of the neurons of the neural network. . On the other hand, the learning unit 14 corrects the value of the weighting coefficient stored in the memory 15 from the error between the output of the calculation unit 13 and the teacher signal given from the outside. This correction operation is performed until the error becomes smaller than a predetermined threshold value, and when the error becomes smaller than the threshold value, the learning ends.

Next, the analysis processing in step ST2 will be described. In this analysis processing, the control unit 16 causes the memory 15 to operate.
The weighting coefficient stored in is checked using, for example, a sensitivity analysis method, and the degree of contribution such as how much each sensor signal contributes to the identification calculation is checked. Then, a sensor output control signal is sent to the sensor signal input unit 12 so as to input a sensor signal according to the degree of contribution. In the present embodiment, it is analyzed that, as a result of the analysis, the contributions of the sensor p and the sensor r are high, and the contributions of other sensors are low. Then, this is divided into two, "contribute" and "do not contribute".

Further, in the re-learning process of step ST3,
The sensor output control signal is sent to the sensor signal input unit 12 so as to input only the "contributing sensor signal" from the control unit 16, that is, the sensor signals output from the sensor p and the sensor r, and the sensor signal is narrowed down and reinitialized. The value of the weighting coefficient stored in the memory 15 is stored in the step ST.
The correction is made in the same manner as the learning process of 2. At this time, the memory 1
The previous weighting factor in 5 is also saved.

When it is determined in the learning end determination process of step ST4 that the error between the output of the arithmetic unit 13 and the teacher signal given from the outside is converged to a predetermined threshold value, the analysis process of step ST2 is performed. When it is determined that the error exceeds the threshold value and does not converge any more, the weighting coefficient of the memory 15 is set to the result of the previous re-learning process. Return step ST3
The relearning process of is ended.

After the learning is completed, in the actual identification processing in step ST6, the sensor p sent to the calculation unit 13 is sent.
And the sensor signal of the sensor r is subjected to an identification calculation in which a predetermined weighting coefficient stored in the memory 15 is added to perform the identification.

The objects to be identified are (a), (b),
As shown in (c), sensor p, sensor q, and sensor q, respectively.
And the sensor r, and the three target objects having characteristics that are characteristic for each sensor signal related to the sensor p and the sensor r, the weighting coefficient is learned as described above, and the weighting coefficient after learning is completed. When the degree of contribution to the identification is examined, it is found that the degree of contribution of the sensor p, the sensor q, and the sensor r is high, and the degree of contribution of other sensors is low.

Therefore, the control unit 16 controls the sensor signal input unit 1
2, a sensor output control signal is generated and output so as to capture only the sensor signals output from the sensors p, q, and r, and is also output to the calculation unit 13 from the sensors p, q, and r. The identification calculation control signal is generated and output so as to perform the calculation using the weighting factor corresponding to the sensor signal.

As described above, according to the present embodiment, among the sensor signals output from the sensor group 11, only the sensor signal having a large degree of contribution to the identification of the identification object and the weighting coefficient are used for the identification. The calculation is performed and identified,
The identification calculation is not redundant, the identification can be performed efficiently, and the identification rate is improved.

In the above embodiments, there are two and three objects, and there are two and three sensors that output sensor signals showing characteristic characteristics. However, it goes without saying that the present invention can also be applied to the case where there are an arbitrary number of identification objects and sensors showing characteristic characteristics.

In the above embodiment, the case where the learning unit 14 performs learning using the back propagation model of the neural network has been described, but it is also applied to the case where learning is performed using another learning model such as multivariate analysis. It goes without saying that you can do it.

Furthermore, in the above-mentioned embodiment, the case where the control unit 16 performs the sensitivity analysis has been described, but it goes without saying that the present invention can be applied to the case where the degree of the contribution to the discrimination is analyzed using another method.

Furthermore, in the above-mentioned embodiment, the case where the degree of contribution is divided into "contribution" and "non-contribution" has been described. However, so-called multi-value threshold processing for dividing this into multiple stages is performed. It goes without saying that it is also applicable to the case.

Example 2. An embodiment of the invention of claim 2 will be described below with reference to the drawings. FIG. 4 is a block diagram showing the configuration of the identifying apparatus of this embodiment. In the identifying apparatus of this embodiment, the color and further the object to be identified are identified based on the three primary color signals of R, G and B. 4, parts that are the same as or equivalent to those in FIG. 1 are given the same reference numerals and explanations thereof are omitted.

In the figure, reference numeral 21 is a color image pickup device which is a sensor for outputting R, G, B three primary color signals, and 22 is three amplifiers 22 corresponding to R, G, B three primary color signals.
It is a three-primary-color signal input unit composed of a, 22b, and 22c.

Next, the operation will be described. Assuming that there are 10 types of objects to be identified from A1 to A10, 1
The three primary color signals ai = (r output from the color imaging device 21 for each of the 0 types of objects A1 to A10
i, gi, bi) (where i is 1, 2, 3 ... 10)
Are represented in a three-dimensional vector space as shown in FIG. In this embodiment, the learning process of step ST1 and the analysis process of step ST2 are performed on such an identification object according to the algorithm shown in FIG.

In the analysis process of step ST2, as is apparent from the part shown in FIG. 5, it is determined that the analysis result has a greater contribution degree to the R signal and the B signal than the G signal. Then, the degree of contribution, that is, the degree of contribution is expressed as a normalized number from "0" to "1". Further, in the re-learning process of step ST3, the amplifiers 22a, 22b, 22c of the three primary color signal input unit 22 are provided with
The control unit 16 outputs a sensor output control signal so as to amplify each primary color signal in proportion to the contribution, and in this state, the re-learning process of step ST3 is performed.

In the actual identification processing of step ST6, the primary color signal amplified in proportion to the contribution and the weighting coefficient stored in the memory 15 are calculated for identification.

In the embodiment described above, the case where the number of objects is 10 has been described, but the number of objects is 10.
Of course, the number is not limited to one, and any number may be used.

As described above, according to the present embodiment, each of the R, G, and B three primary color signals output from the color image pickup device 21 is amplified in accordance with the degree of contribution to the identification of the identification object. The color can be identified based on the primary color signal and the weighting coefficient obtained as a result of learning.

Example 3. An embodiment of the invention of claim 3 will be described below with reference to the drawings. FIG. 6 is a block diagram showing the configuration of the identifying apparatus of this embodiment. In the identifying apparatus of this embodiment, the shape is identified from the image Fourier-transformed by the lens. 6, parts that are the same as or equivalent to those in FIG. 1 are assigned the same reference numerals and explanations thereof are omitted.

In FIG. 6, reference numeral 30 is a lens for Fourier-transforming the image of the object, and 31 is a plane arrangement sensor consisting of a photoelectric element group arranged in a plane.

Next, the operation will be described with reference to the flowchart of FIG. In this embodiment, two types of objects A, which are identification objects, an object B, a lens 30, and a planar arrangement sensor 31 are arranged at a constant distance, and when coherent light is applied to the identification object, a plane is obtained. As shown in FIG. 7A or FIG. 7B, a pattern indicating the Fourier transform of the surface of the identification object may be observed on the placement sensor 31 according to the physical characteristics of the surface of the identification object. Known (for example, "Image processing by electronic computer" By R
openfield pp66-74, Kyoritsu Shuppan), the learning process of step ST1 and the analysis process of step ST2 are performed according to the flowchart of FIG.

In this embodiment, as a result of the analysis processing in step ST2, it is determined that the contributions of the photoelectric elements arranged near the center and the photoelectric elements arranged near the four corners are high. In the re-learning process of the next step ST3, the sensor output control signal is output to the sensor signal input unit 12 so as to capture only the signal output from the photoelectric element having a high contribution from the control unit 16, and the re-learning is performed in this state. Do.

Even in the actual identification processing in step ST6, only the signal output from the photoelectric element having a high contribution sent to the arithmetic unit 13 and the weighting coefficient stored in the memory 15 corresponding to this signal are calculated. To identify them.

In the embodiment described above, the case where the identification objects are the two kinds of the objects A and B has been described, but it goes without saying that the number of the identification objects may be any number. Absent.

As described above, according to this embodiment, it is possible to efficiently identify the identification object based on the surface texture.

Example 4. An embodiment of the invention of claim 4 will be described below with reference to the drawings. FIG. 8 is a block diagram showing the configuration of this embodiment. In the identification apparatus of this embodiment, the pattern of the identification object is identified based on the sensor signal detected by the photoelectric element group arranged in a plane. In FIG. 8, parts that are the same as or equivalent to those in FIG. 6 are given the same reference numerals and description thereof is omitted.

Next, the operation will be described with reference to the flowchart of FIG. In this embodiment, the patterns of the two types of the objects A and B to be identified are arranged in the planar arrangement sensor 3
When the image is picked up by 1 and input through the sensor signal input unit 2, the patterns shown in FIGS. 9A and 9B are obtained. The learning process of step ST1 shown in FIG. 2 is performed for this pattern, and the analysis process of step ST2 is further performed.

In this embodiment, as a result of the analysis processing in step ST2, it is determined that the photoelectric elements arranged in the areas of the first quadrant and the second quadrant have a high degree of contribution. In the re-learning process of the next step ST3, only the sensor signal of the area (in this case, the area of the first quadrant and the second quadrant) to which the photoelectric element having a high contribution belongs is fetched from the control unit 16.
A sensor output control signal is output to the sensor signal input unit 2, and relearning is performed in this state. Even in the actual identification processing in step ST6, only the sensor signal obtained from the area to which the photoelectric element having a high contribution output to the calculation unit 13 belongs and the weighting coefficient stored in the memory 15 corresponding to this sensor signal are used. Calculate and identify.

As described above, according to the present embodiment, the pattern of the identification object can be efficiently identified based on the sensor signal detected by the photoelectric element group arranged in a plane.

In the embodiment described above, the case where the objects are the two kinds of objects A and B has been described, but it goes without saying that the number of objects may be any number.

Example 5. An embodiment of the invention of claim 5 will be described below with reference to the drawings. FIG. 10 is a block diagram showing the configuration of the present embodiment. In the identification device of this embodiment, based on the geometrical characteristics of the pattern of the identification object from the sensor signal obtained by the photoelectric element group arranged in a plane. Identification. 10, parts that are the same as or correspond to those in FIG. 8 are assigned the same reference numerals and explanations thereof are omitted.

In the figure, reference numeral 32 is a geometric feature extraction unit for extracting a geometric feature from the pattern to be identified which is picked up by the planar arrangement sensor 31.

Next, the operation will be described with reference to the flowchart of FIG. In the present embodiment, two types of images of the object A and the object B of the identification object are picked up by the planar arrangement sensor 31, and the geometric features are extracted by the geometric feature extraction unit 32. The two-dimensional pattern as shown in b) and the n feature quantities an and bn shown in FIG. 12 are obtained. It is assumed that the object A and the object B have the same area and that there is a significant difference in the peripheral length and the moment.

In this embodiment, the learning process of step ST1 and the analysis process of step ST2 shown in FIG. 2 are performed on the n feature quantities an and bn shown in FIG. Then, as a result of the analysis processing in step ST2, it is determined that the degree of contribution is high with respect to the feature amount having a significant difference such as the perimeter or the moment. In the re-learning process of the next step ST3, a control signal is sent from the control unit 16 to the sensor signal input unit 12 so as to capture only the feature amount having a significant difference such as a perimeter and a moment having a high contribution, Re-learning is performed in this state.

Even in the actual identification processing in step ST6, the calculation is performed using only the feature amount having a high contribution output to the calculation unit 13 and the corresponding weighting coefficient stored in the memory 15 to obtain the target object. Identify.

As described above, according to this embodiment, the object to be identified can be efficiently identified based on its geometric feature.

In the above-described embodiment, the case where the objects are the two kinds of objects A and B has been described. However, the number of objects may be any number, and the feature amount may be the surroundings. It goes without saying that it is not limited to length or moment.

Example 6. Hereinafter, claim 6 and claim 7
An embodiment of the invention will be described with reference to the drawings. Note that the configuration of the identification apparatus of the present embodiment is referred to the block diagram shown in FIG. 1, and the algorithm for identification is referred to the flowchart shown in FIG. In the identification device of this embodiment, the memory 15 is composed of weighting factors constituting a multi-layered neural network, and the control unit (arithmetic control unit) 16 outputs a specific sensor signal based on the partial differential value of the output layer with respect to the input layer. A control signal to be taken in will be generated.

First, the process until the control section 16 generates a control signal for taking in a specific sensor signal based on the partial differential value of the output layer with respect to the input layer will be described.
In FIG. 14, reference numeral 41 denotes an L-th layer unit which transfers input / output data, and its conceptual diagram is shown in FIG.
In the figure, 42 to 46 are weighting coefficients, which are coefficients having values from "0" to "1" defined between the unit 41 and other units. Also, "f" is s
This is the igoid function. FIG. 15 shows the sensor signals detected by the sensor group 11 for the five learning samples of the object A, which are five types of signals a1 to a5.

Each unit constituting the neural network is shown in FIG.
Input / output data is transferred as shown in FIG. 3, and the sensor signal input unit 12 is provided in each input unit of the input layer (first layer).
If a signal indicating a desired identification state is supplied from the control unit 16 to each output unit of the output layer (Nth layer), then the mth layer of the Lth layer is supplied.
The output of the unit is expressed by the following equation (1).

[0085]

[Equation 1]

Here, iLm is the total sum of the inputs of the L-th layer m-th unit, OLm is the output of the L-th layer m-th unit, and θLm is the L-th L-th unit.
Threshold value of m-th layer unit, f is input / output function of unit, W
Lmk is a coupling load between the m-th unit of the L-th layer and the k-th unit of the (L-1) th layer.

When the above equation (1) is differentiated by i1j, the following equation (2) is obtained.

[0088]

[Equation 2]

Here, N = 3, that is, a three-layer neural network is considered, and in Expression (2), L is 3, and f is sig.
Using the oid function, f '(iLm) is expressed by the following equation (3).

[0090]

[Equation 3]

Therefore, the following equation (4) is obtained.

[0092]

[Equation 4]

In the learned neural network, the weighting factor is already known. Therefore, if an appropriate input value is given to the first layer (input layer), the equation (4) can be calculated. The input value can be appropriately given, but as an example, there is a method using an intermediate value signal.

In the method using the intermediate value signal, the signal of the same identification object (FIG. 15) is selected from the input signals used for learning.
(5 signals of a1 to a5, etc.) are extracted.
Then, an intermediate value of the above signals is selected for each sensor, and this is collected to generate a new intermediate value signal a0, which is used as an input value for obtaining the partial differential value.

Step ST shown in the flowchart of FIG.
In the state where the learning process of 1 is performed, an appropriate weighting coefficient is learned and stored in the memory 15. The control unit 16 uses the equation (4) in the analysis process in step ST2.
Is used to obtain the partial differential value for each input unit of all output units. As a result, as shown in FIG. 16, it becomes clear that the partial differential values are different depending on the input unit. In this case, the input unit having a large partial differential value greatly contributes to the discrimination, and the input unit having a small partial differential value contributes little to the discrimination.
Therefore, this partial differential value is subjected to threshold processing such as binarization or multi-value conversion, and the sensor signal corresponding to the input unit is taken in the input unit, that is, the sensor signal input unit 12 based on the value obtained as a result. The control signal is generated and output to the sensor signal input unit 12.

In the above embodiment, the case where 3 is used as L and the sigmoid function is used as f has been described, but it is also applicable to the case where an arbitrary number of layers is used as L and another appropriate function is used as f. Needless to say.

As described above, according to this embodiment, the sensor signal having a high degree of contribution to the discrimination is based on the partial differential value with respect to the input layer of the output layer of the memory, which is composed of the weighting factors constituting the multilayer neural network. Therefore, it is possible to improve the efficiency of the arithmetic processing for performing identification and the identification rate of the identification object.

Example 7. An embodiment of the invention of claim 8 will be described below with reference to the drawings. FIG. 17 is a block diagram showing the configuration of the present embodiment. In FIG. 17, parts that are the same as or corresponding to those in FIG.

In the figure, reference numeral 48 is an input device for inputting an evaluation scale or evaluation quantity for one learning sample, and 49 is a display device.

Next, the operation will be described with reference to the flowchart of FIG. In this embodiment, it is assumed that the learning unit 14 performs learning using a factorization method, which is a typical method, in the learning process of step ST2 shown in FIG. First, the learning unit 14 collects the inspector's evaluation scale and evaluation amount for the object A and the object B, and causes the arithmetic unit 13 to perform a factor analysis on the inspector's evaluation scale and evaluation amount. The matrix R is obtained and stored in the memory 15.

In the analysis process of the next step ST2, the N sensor groups 11 corresponding to the N factors obtained by the factor analysis, and the function fn (n
N) is determined experimentally. First, n appropriate sensors are prepared as sensor candidates of the sensor group 11, and the weighting coefficient of the function fn stored in the memory 15 is initialized appropriately. Next, the evaluation scale and the evaluation amount for one learning sample are input from the input device 48, and the calculation unit 13 calculates the predicted sensor output value using the function fn and the factor load matrix R. At the same time, the sensor output is displayed on the display device 49 by a visual display method that is easy for humans to understand. Then, the display and the learning sample are compared with each other, and if there is no difference, the learning process is ended. If there is a difference, the difference value is input from the input device 48, and the calculation unit 13 inputs the function fn.
Calculate the correction amount of the weighting factor of.

Further, in the re-learning process of step ST3,
The learning unit 14 corrects the weighting coefficient stored in the memory 15 based on the calculated correction amount. Then, this operation is repeated until the difference is eliminated. If the difference does not decrease, the sensor candidate of the sensor group 11 is replaced with another sensor and the same operation as described above is performed.

Using the factor load matrix R, the sensor group 11, and the function fn determined in this way, the computing unit 13 performs the identification process of step ST6.

As described above, in this embodiment, the predicted sensor signal generated by performing the inverse calculation from the given identification state and the weighting coefficient stored in the memory is displayed on the display device in a visually easy-to-understand manner. Since it is displayed in the format, it is possible to easily visually judge the degree of difference between the display and the learning sample by comparing them.

In the embodiment described above, the learning unit 1
Although item 4 describes the case where learning is performed using the factor analysis method, it goes without saying that it can be applied to the case where learning is performed using another learning model such as a neural network.

Example 8. An embodiment of the invention of claim 9 will be described below with reference to the drawings. FIG. 19 is an explanatory diagram for explaining a visual display method that is easy for humans to understand performed by the identification device according to the present embodiment. In the figure, 51 is an identification target object and 52 is a display device (three-dimensional model display device). ) 49 is a three-dimensional model of the identification object visually displayed.
For the identification device of this embodiment, refer to the block diagram shown in FIG.

Next, the operation will be described. First, when learning is performed in the learning process of step ST1 in FIG. 2, image data of all learning samples to be identified are input. Then, average image data of these image data is created. Further, a reflectance distribution model, a transmittance distribution model, etc. at each point on the surface of the identification object shown in FIG. 21 are created from this average image data. Next, in the analysis processing of step ST2, for example, when the evaluation scale for one learning sample is input as “the surface glows well” and the evaluation quantity is “I think so much”, the calculation unit (three-dimensional model synthesis calculation unit) 13 Function fn and factor loading matrix R
Is used to calculate the predicted sensor output value shown in FIG. This sensor output is multiplied by a reflectance distribution model, a transmittance distribution model, and the like created in advance with appropriate weights, and the three-dimensional model to be identified is synthesized and displayed on the display device 49.

As described above, in this embodiment, the three-dimensional model of the identification object is synthesized by using the average image data created from the image data of the plurality of identification objects, and the display format is easy to understand visually. Can be displayed.

Example 9. An embodiment of the invention of claim 10 will be described below with reference to the drawings. FIG. 22 is an explanatory diagram showing a visual display method that is easy for humans to understand on the display device of the identification device according to the present embodiment. 22, parts that are the same as or correspond to those in FIG. 19 are assigned the same reference numerals and explanations thereof are omitted. In the figure, 53 is an original image of the identification object. The block diagram shown in FIG. 17 is referred to as the identification device of this embodiment.

Next, the operation will be described. In the analysis process of the present embodiment, the three-dimensional model 52 to be identified is displayed and the original image 53 is also displayed. 3D model for identification 5
2 is created and displayed by the method described in the sixth embodiment, for example. Then, the original image 53 and the three-dimensional model 52 are compared, and if there is a difference, the difference is input from the input device (difference information input device) 48 using the given rating scale and rating.
For example, the rating scale of "roughness" is instructed such as "there are many (three-dimensional model) is much". As a result, the calculation unit (three-dimensional model synthesis calculation unit) 13 corrects the weighting coefficient.

As described above, according to this embodiment, the display device (three-dimensional model display device) 49 is provided with the identification object 3
By displaying the three-dimensional model and the original image, the user can easily understand the difference between the three-dimensional model and the original image, and further input information about the difference between the three-dimensional model and the original image. The weighting factor can be modified based on this input.

In the above embodiment, the input device 48 has been described on the assumption that a keyboard capable of inputting characters is used. However, the present invention can be applied to an input device of a type in which a selection menu displayed on the display device is selected with a mouse. Needless to say.

Example 10. An embodiment of the invention of claim 11 will be described below with reference to the drawings. FIG. 23 is an explanatory diagram showing a visual display method that is easy for humans to understand on the display device of the identification device according to the present embodiment. In FIG. 23, the same or corresponding parts as those in FIG. 22 are designated by the same reference numerals and the description thereof will be omitted. In the figure, 54 is the original image 53 of the identification object.
And an image of the difference between the three-dimensional model 52. The block diagram shown in FIG. 17 is referred to as the identification device of this embodiment.

Next, the operation will be described. In the analysis processing of the present embodiment, a plurality of images 54 of the difference between the three-dimensional model 52 of the identification target and the original image 53 of the identification target are displayed, and the history of changes is also displayed in this case.
By displaying the image of the difference, the difference between the two can be clearly recognized, and the input device 4 using the rating scale and the rating amount.
Even when the difference is input from 8, the input operation may be performed so that the difference image disappears.

Example 11. An embodiment of the invention of claim 12 will be described below with reference to the drawings. FIG. 24 is an explanatory diagram showing a visual display method that is easy for humans to understand on the display device of the identification device according to the present embodiment. In FIG. 24, parts that are the same as or equivalent to those in FIG. In the figure, 52a, 52b and 52c are a plurality of three-dimensional models of the identification object. The block diagram shown in FIG. 17 is referred to as the identification device of this embodiment.

Next, the operation will be described. In the analysis processing of the present embodiment, together with the original image 53 of the identification target, a plurality of three-dimensional models of the identification target with different rating scales of a certain rating scale are combined, and at the same time a display device (three-dimensional model display device) 49.
To display. Then, a person ranks the three-dimensional model based on another rating scale, and inputs it from the input device (rank scale input device) 48. The rank input by a person from the input device 48 is treated as a rank scale, and the calculation unit 13
At, the correlation with the rating scale is calculated, and the correction amount of the weighting coefficient stored in the memory 15 is calculated. Then, the learning unit 14 corrects the weight coefficient of the memory 15 according to the calculated correction amount.

As described above, according to the present embodiment, the rank based on the rating scale is input from the input device 48 with respect to the three-dimensional model of the plurality of combined identification objects displayed on the display device 49. Therefore, the value of the weighting coefficient can be easily corrected visually.

Example 12. An embodiment of the invention of claim 13 will be described below with reference to the drawings. FIG. 25 is a block diagram showing the configuration of the control device according to the present embodiment. The same or corresponding parts as in FIG. In the figure, reference numeral 61 denotes an image pickup unit for picking up an image of an identification object, and 6
Reference numeral 2 denotes a gloss feature input unit that extracts and outputs a gloss feature from a signal output as a result of the image pickup unit 61 picking up an image of the identification target.

Next, the operation will be described based on the flowchart shown in FIG. First, the learning process of step ST11 will be described. In this learning process, a learning process using a back propagation model of a neural network (neural network) is performed. First, the imaging unit 61 captures images of N types of objects An (n N) to be identified. The gloss feature input unit extracts a plurality of gloss features from the signal output from the image pickup unit 61. The plurality of gloss features are the degree of transparency based on an analysis of human senses,
It has multiple gloss characteristics including the magnitude of brightness change, brightness unevenness, and the sharpness of the reflected light distribution curve.

The learning section 14 carries out learning using a back propagation model of a neural network (neural network), initializes the memory 15 by an appropriate method,
A large number of learning samples of the object An (n N) are prepared, each of them is imaged by the image pickup unit 61, and the plurality of gloss features are extracted from the signal output from the image pickup unit 61 by the gloss feature input unit 62 and calculated. It is output to the unit 13. Computing unit 1
At 3, the calculation is performed using the expected value generation circuit 13 signal sent from the gloss feature input unit 62 and the weighting coefficient stored in the memory 15.

The learning section 14 corrects the value of the weighting coefficient stored in the memory 15 from the error between the signal output from the arithmetic section 13 and the teacher signal given from the outside. This correction operation is performed until the error becomes smaller than a predetermined threshold value, and when the error becomes less than the threshold value, the learning ends.

Then, in the actual identification processing of step ST12, the gloss feature sent to the operation unit 13 is subjected to an operation in which a predetermined weighting coefficient stored in the memory 12 is added to perform the identification. To do.

As described above, according to this embodiment, the identification result calculated by the plurality of gloss feature information extracted from the signal output from the image pickup unit 61 and the weighting coefficient stored in the memory 15 are obtained. Learning is performed by calculating a weighting coefficient from an error with a desired identification state, and the identification calculation is performed using the gloss information and the weighting coefficient to perform identification, and thus the identification target is identified based on the gloss feature. can do.

In the above embodiment, the case where the learning unit 14 performs learning using the back propagation model of the neural network has been described, but it is also applied to the case where learning is performed using another learning model such as multivariate analysis. It goes without saying that you can do it.

Example 13 An embodiment of the invention of claim 14 will be described below with reference to the drawings. FIG. 27 is a block diagram showing the configuration of the present embodiment, and the same or corresponding parts as in FIG. 1 will be assigned the same reference numerals and explanations thereof will be omitted. In the figure, reference numeral 71 denotes a gloss feature input section for extracting P gloss features Gp (pP) which are the plurality of gloss features described in the twelfth embodiment from the signals output from the sensor group 11, and 72 designates Q identified features. It is a color feature detection unit that extracts the wavelength Wq (qQ).

Next, the operation will be described based on the flowchart shown in FIG. First, the learning process of step ST1 will be described. In this learning process, a learning process using a back propagation model of a neural network (neural network) is performed. N types of objects An (n N) with high gloss as identification objects and M with low gloss
Assuming a type of object Bm (m M), the gloss feature input unit 71 extracts a plurality of gloss features from the signals output from the sensor group 11.

The memory 15 is initialized by an appropriate method, a large number of learning samples of the objects An (n N) and Bm (m M) are prepared, these are detected by the sensor group 11, and output from the sensor group 11. A plurality of gloss features and color features are respectively extracted from the signal by the gloss feature input unit 71 and the color feature detection unit 72, and signals corresponding to these gloss features and color features are output to the calculation unit 13.

In the calculation section 13, the signals sent from the gloss feature input section 71 and the color feature detection section 72 and the memory 1 are sent.
Calculation is performed using the weighting coefficient stored in 5.

A learning signal is externally given to the learning section 14, and the value of the weighting coefficient stored in the memory 15 is corrected from the error between the output of the calculating section 13 and the learning signal. This correction operation is performed until the error becomes smaller than a predetermined threshold value, and when the error becomes less than the threshold value, the learning ends.

In the analysis process of step ST2, the control unit 1
6 is a weighting coefficient stored in the memory 15, for example, by using a method such as sensitivity analysis, when each identification object has a gloss feature such as high or low gloss, how much each color feature is used for the identification calculation. When the degree of contribution is examined, for example, the gloss feature A of the high gloss object A1
When 1 g is (g11, g21, ... gP1), the color features having a high degree of contribution are W1 and W3, and the gloss feature of the low-brightness object B1 is (g12, g22 ,.
P2), the color features with a high degree of contribution are W2 and W.
It becomes clear that it is 4.

Further, in the re-learning process of step ST3,
A control signal is sent to the color feature input unit 72 so as to input only the color feature signal contributing from the control unit 16 in accordance with the gloss feature to be identified, and the weighting factor stored in the reinitialized memory 15 is sent. The value of is corrected in the same manner as the learning process of step ST1. At this time, the previous state of the memory 15 is also saved, and if the error becomes smaller than a predetermined threshold value, the process returns to the analysis processing of step ST2, and if the error does not become smaller than the predetermined threshold value, the memory 15 is set to the previous state. Then, the re-learning process of step ST3 is completed. And
This re-learning process is performed for each gloss information, and the weighting coefficient is stored in the memory 15.

In the actual identification processing of step ST6, according to the signal corresponding to the gloss feature sent to the calculation unit 13,
A signal corresponding to a predetermined color feature is input, a predetermined weighting coefficient stored in the memory 15 is read out, and a calculation is performed to identify it.

As described above, according to the present embodiment, it is possible to efficiently identify the object to be identified by the gloss characteristic information, the color characteristic information and the weighting coefficient stored in the memory.

In the embodiment described above, the learning unit 1
Although 4 has been described as performing learning using a back propagation model of a neural network (neural network), it goes without saying that it can also be applied to learning using other learning models such as multivariate analysis.

Further, although the case where the control unit 16 performs the sensitivity analysis has been described, it is needless to say that the present invention can be applied to the case where the degree of the contribution to the discrimination is analyzed using another method.

Further, although the degree of contribution is classified into "contribution" and "non-contribution", it can be applied to the case where multi-valued threshold processing for classifying into multiple stages is applied. There is no end.

Example 14. An embodiment of the invention of claim 15 will be described below with reference to the drawings. FIG. 28 is an explanatory diagram showing the configuration of the color feature detection unit of the identification device of this embodiment. In the figure, 81, 83, 85a, 85b and 85c are lenses, 82 is a mask for blocking a part of the optical path, 84a, 84b,
Reference numeral 84c is a half mirror, 86a, 86b and 86c are interference filters that transmit only light of different specific wavelengths, and 87a, 87b and 87c are photoelectric elements. Reference numeral 90 denotes an image pickup unit, and the lenses 81 and 83 and the mask 82 are provided in the image pickup unit 90. Reference numeral 91 is a color feature information processing unit for identifying the color feature of the identification object based on the color feature information.

Next, the operation will be described. Light emitted from the identification object arranged in front of the lens 81 is condensed by the lens 82 and guided to the mask 82. At that time, unnecessary light components such as regular reflection components are blocked by the mask 82. The light passing through the mask 82 is guided to the half mirrors 84a, 84b, 84c via the lens 83, the light reflected by the half mirror 84a is condensed by the lens 85a, and only the component of the specific wavelength is collected by the interference filter 86a. It is extracted and collected on the photoelectric element 87a. The same applies to the half mirrors 84b and 84c. Then, the signals output from the photoelectric elements 87a, 87b, 87c become signals corresponding to the color features of the identification target, and are output to the color feature information processing unit 91 and processed to identify the color features of the identification target. It

As described above, according to the present embodiment, it is possible to identify the color feature of the identification object by extracting only the component of the specific wavelength of the light from the identification object.

[0140]

As described above, according to the first aspect of the invention, the sensor output control for fetching the sensor signal group according to the weighting coefficient for the sensor signal stored in the memory as a result of learning and contributing to the identification. A signal is output to the sensor signal input unit, and the identification target is identified using the specific sensor signal captured by the sensor signal input unit based on the sensor output control signal and the weighting coefficient stored in the memory. With this configuration, the efficiency of the arithmetic processing for performing identification is improved, and the identification rate is improved.

According to the invention of claim 2, as a result of the learning, the three primary color signals which are taken in by the three primary color signal input section and contribute to the discrimination and the weighting factors stored in the memory are stored according to the weighting factors stored in the memory. Since the object to be identified is identified by using and, the efficiency of the arithmetic processing for performing the identification based on the three primary color signals is improved, and the identification rate is improved.

According to the invention of claim 3, among the sensor signals output from the plane-arranged sensor arranged in a plane for receiving the image Fourier-transformed by the lens, the weight for the sensor signal contributing to the discrimination. According to the coefficient, the sensor signal input unit is configured to identify the object to be identified by using the specific sensor signal captured and the weighting coefficient stored in the memory. Therefore, the identification based on the sensor signal is performed. As a result, the efficiency of the arithmetic processing for performing the operation is improved and the identification rate is improved.

According to the fourth aspect of the invention, it contributes to the discrimination among the sensor signals output from the planar arrangement sensor including the photoelectric element groups arranged in a plane for detecting the information about the pattern of the identification object. According to the weighting coefficient for the sensor signal, the sensor signal input unit is configured to identify the identification target by using the specific sensor signal captured by the sensor signal and the weighting coefficient stored in the memory. There is an effect that the efficiency of the arithmetic processing for performing the identification based on is improved and the identification rate is improved.

According to the fifth aspect of the present invention, the sensor signal output from the sensor composed of the photoelectric element group arranged on the plane is taken in, the geometric feature of the identification object is extracted, and the identification is performed based on the specific geometric feature. Since it is configured to perform the above, there is an effect that the efficiency of the arithmetic processing for performing the identification based on the geometric feature is improved and the identification rate is improved.

According to the sixth aspect of the present invention, a partial differential value of the output layer of the multilayer neural network with respect to the input layer in the memory composed of the weighting factors constituting the multilayer neural network is calculated, and this partial differential value is obtained. Since the specific sensor signal that contributes to the identification is fetched based on the above, there is an effect that the arithmetic processing for the identification can be efficiently performed and the identification rate of the identification object can be improved.

According to the invention of claim 7, the input signal used for learning is divided for each identification object, an intermediate value of the input signal is selected, a partial differential value is obtained using the intermediate value, and the partial differential value is obtained. Since the specific sensor signal is fetched based on the above, there is an effect that the identification target can be efficiently identified and the identification rate can be improved.

According to the eighth aspect of the present invention, the predicted sensor signal generated by performing the inverse operation from the given identification state and the weighting coefficient stored in the memory is displayed on the display device. By comparing this display with the learning sample, the difference between the two can be clearly determined, and an identification device with an improved man / machine interface can be obtained.

According to the invention of claim 9, the three-dimensional model of the object to be identified is synthesized using the average image data created from the image data of the plurality of objects to be identified, and is displayed on the display device. There is an effect that an identification device having an improved man / machine interface capable of displaying a three-dimensional model can be obtained.

According to the tenth aspect of the invention, the three-dimensional model display device displays the three-dimensional model of the identification object and the original image synthesized by the arithmetic unit, and the difference information input device is further provided. Therefore, it is possible to perform display in a visual display format that is easy to understand, and further, the user can input the difference between the three-dimensional model and the original image of the identification object from the difference information input device. There is an effect that an identification device with an improved machine interface can be obtained.

According to the eleventh aspect of the invention, the difference image between the three-dimensional model and the original image of the identification object and the difference image between the three-dimensional model and the original image of the identification object are displayed on the difference image display device. Since it is configured, there is an effect that an identification device having an improved man / machine interface can be obtained.

According to the twelfth aspect of the invention, the plurality of combined identification objects 3 displayed on the three-dimensional model display device are displayed.
Since the user is able to input the rank based on the rating scale of the dimensional model from the rank scale input device, there is provided an identification device having an improved man / machine interface, which can modify the weighting coefficient according to the rank. There is an effect to be obtained.

According to the thirteenth aspect of the invention, the plurality of gloss characteristic information extracted from the signal output from the image pickup section and the weighting coefficient stored in the memory are calculated and the discrimination result is output. There is an effect that an identification device that can efficiently identify the glossy feature of the identification object can be obtained.

According to the fourteenth aspect of the invention, the sensor output control signal for taking in the specific color characteristic information is generated by the gloss characteristic information and the learned weighting coefficient stored in the memory, and this sensor output control is performed. The color feature information detected by the color feature detection unit based on the signal and the weighting coefficient stored in the memory are used to cause the calculation unit to perform the identification calculation. There is an effect that a discriminating apparatus for discriminating the discrimination target efficiently based on the gloss characteristic and the color characteristic is obtained by learning the coefficient.

According to the fifteenth aspect of the present invention, the light of the identification target imaged by the imaging section is partially shielded by the mask,
Since the light passing through the mask is reflected by the half mirrors respectively, only the light having a specific wavelength is transmitted by the interference filter, and the photoelectric feature group is configured to detect the color feature information of the identification target. There is an effect that the identification target can be efficiently identified based on the information.

[Brief description of drawings]

FIG. 1 is a block diagram showing an identification device according to an embodiment of the present invention.

FIG. 2 is a flowchart showing an operation of the identification device according to the embodiment of the invention of claim 1;

FIG. 3 is an explanatory diagram showing sensor signals output from a sensor group of an identification device according to an embodiment of the present invention.

FIG. 4 is a block diagram showing an identification device according to an embodiment of the invention of claim 2;

FIG. 5 is an explanatory diagram showing three primary color signals of R, G, B output from a color image pickup device of an identification device according to an embodiment of the present invention.

FIG. 6 is a block diagram showing an identification device according to an embodiment of the invention of claim 3;

FIG. 7 is an explanatory diagram showing an image Fourier-transformed by a lens in the identification device according to the embodiment of the invention of claim 3;

FIG. 8 is a block diagram showing an identification device according to an embodiment of the invention of claim 4;

FIG. 9 is an explanatory diagram showing patterns of two types of identification objects in the identification device according to the embodiment of the invention of claim 4;

FIG. 10 is a block diagram showing an identification device according to an embodiment of the invention of claim 5;

FIG. 11 is an explanatory diagram showing a two-dimensional pattern obtained as a result of imaging an object to be identified by a photoelectric element group in the identifying device according to the embodiment of the invention of claim 5;

FIG. 12 is an explanatory diagram showing a plurality of types of feature amounts extracted by a geometric feature extraction unit in the identification device according to the embodiment of the invention of claim 5;

FIG. 13 is an explanatory diagram showing a multi-layered neural network of a memory in the identification device according to the embodiment of the invention of claim 6;

FIG. 14 is an explanatory diagram showing a unit of the L-th layer of the multilayer neural network of the memory in the identification device according to the embodiment of the invention of claim 6;

FIG. 15 is an explanatory diagram of signals detected by the sensor group and five intermediate value signals of the five learning samples of the object A in the identification apparatus according to the embodiments of the sixth and seventh inventions.

FIG. 16 is an explanatory diagram showing a partial differential value for each input unit of all output units of the multilayer neural network of the memory in the identification device according to the embodiment of the invention of claim 6;

FIG. 17 is a block diagram showing an identification device according to an embodiment of the invention of claim 8;

FIG. 18 is an explanatory diagram showing a factor loading matrix in the identification device according to the embodiment of the invention of claim 8;

FIG. 19 is an explanatory diagram for explaining a visual display method that is easy for humans to understand in a display device of an identification device according to an exemplary embodiment of the present invention;

FIG. 20 is an explanatory diagram showing predicted sensor output values in the identification device according to the embodiment of the invention of claim 9;

FIG. 21 is an explanatory diagram for explaining a reflectance distribution model and a transmittance distribution model of each point on the surface of the identification target in the identification device according to the embodiment of the invention of claim 9;

FIG. 22 is an explanatory diagram showing a visual display method that is easy for a human to understand in a display device of an identification device according to an exemplary embodiment of the present invention;

FIG. 23 is an explanatory diagram showing a visual display method that is easy for a human to understand in a display device of an identification device according to an embodiment of the invention in claim 11;

FIG. 24 is an explanatory diagram showing a visual display method that is easy for a human to understand on a display device of an identification device according to an embodiment of the invention of claim 12;

FIG. 25 is a block diagram showing an identification device according to an embodiment of the invention of claim 13;

FIG. 26 is a flow chart for explaining the operation of the identification device according to an embodiment of the invention of claim 13;

FIG. 27 is a block diagram showing an identification device according to an embodiment of the invention of claim 14;

FIG. 28 is an explanatory diagram showing a configuration of a color feature input unit of an identification device according to an embodiment of the invention of claim 15;

FIG. 29 is a block diagram showing a configuration of a conventional identification device.

FIG. 30 is an explanatory diagram for explaining the identifying operation of the conventional identifying device.

[Explanation of symbols]

 Reference Signs List 11 sensor group 12 sensor signal input unit 13 calculation unit (three-dimensional model synthesis calculation unit) 14 learning unit 15 memory 16 control unit (calculation control unit) 21 color image pickup device 22 3 primary color signal input unit 22a, 22b, 22c amplifier 30 Lens 31 Planar arrangement sensor 32 Geometric feature extraction unit 48 Input device (difference information input device, rank scale input device) 49 Display device (3D model display device) 52 3D model 53 Original image 54 Difference image 61,90 Imaging unit 62,71 Gloss characteristic input section 72 Color characteristic detection section 82 Masks 84a, 84b, 84c Half mirrors 86a, 86b, 86c Interference filter 87a, 87b, 87c Photoelectric element group 91 Color characteristic information processing section

─────────────────────────────────────────────────── ───

[Procedure amendment]

[Submission date] January 13, 1994

[Procedure Amendment 1]

[Document name to be amended] Statement

[Name of item to be corrected] 0024

[Correction method] Change

[Correction content]

According to a fifteenth aspect of the present invention, in the identification device, the light of the identification object imaged by the imaging unit is partially shielded by a mask, and the light passing through the mask is made into a plurality of half mirrors .
More reflected, based on the output of the photoelectric filter group that detects the light transmitted through each of the interference filter and the interference filter that transmits only the light of a specific wavelength to the reflected light respectively,
The color feature information processing section for identifying the color feature of the identification object is provided.

[Procedure Amendment 2]

[Document name to be amended] Statement

[Correction target item name] 0086

[Correction method] Change

[Correction content]

Here, i ^L _m is the sum of the inputs of the L-th layer m-th unit, O ^L _m is the output of the L-th layer m-th unit, and θ ^L _m is the L-th L-th unit.
Threshold value of m-th layer unit, f is input / output function of unit, W
^L _mk is a coupling load between the m-th unit in the L-th layer and the k-th unit in the (L-1) th layer.

[Procedure 3]

[Document name to be amended] Statement

[Correction target item name] 0089

[Correction method] Change

[Correction content]

Here, N = 3, that is, a three-layer neural network is considered, and in Expression (2), L is 3, and f is sig.
Using the oid function, f ′ (i ^L _m ) is expressed by the following equation (3).

[Procedure amendment 4]

[Document name to be amended] Statement

[Correction target item name] 0096

[Correction method] Change

[Correction content]

In the above embodiment, the case where 3 is used as L and the sigmoid function is used as f has been described, but an arbitrary number of layers is used as L, and a threshold function is used as f.
The present invention can be applied when using other suitable functions like.

[Procedure Amendment 5]

[Document name to be amended] Statement

[Correction target item name] 0101

[Correction method] Change

[Correction content]

In the analysis process of the next step ST2, the N sensor groups 11 corresponding to the N factors obtained by the factor analysis and the function fn (n∈n) between the factors and the sensor groups 11 are processed.
N) is determined experimentally. First, n appropriate sensors are prepared as sensor candidates of the sensor group 11, and the weighting coefficient of the function fn stored in the memory 15 is initialized appropriately. Next, the evaluation scale and the evaluation amount for one learning sample are input from the input device 48, and the calculation unit 13 calculates the predicted sensor output value using the function fn and the factor load matrix R. At the same time, the sensor output is displayed on the display device 49 by a visual display method that is easy for humans to understand. Then, the display and the learning sample are compared with each other, and if there is no difference, the learning process is ended. If there is a difference, the difference value is input from the input device 48, and the calculation unit 13 inputs the function fn.
Calculate the correction amount of the weighting factor of.

[Procedure correction 6]

[Document name to be amended] Statement

[Name of item to be corrected] 0119

[Correction method] Change

[Correction content]

Next, the operation will be described based on the flowchart shown in FIG. First, the learning process of step ST11 will be described. In this learning process, a learning process using a back propagation model of a neural network (neural network) is performed. First, the imaging unit 61 images N types of objects An (nεN) to be identified. The gloss feature input unit extracts a plurality of gloss features from the signal output from the image pickup unit 61. The plurality of gloss features are the degree of transparency based on an analysis of human senses,
It has multiple gloss characteristics including the magnitude of brightness change, brightness unevenness, and the sharpness of the reflected light distribution curve.

[Procedure Amendment 7]

[Document name to be amended] Statement

[Correction target item name] 0120

[Correction method] Change

[Correction content]

The learning section 14 carries out learning using a back propagation model of a neural network (neural network), initializes the memory 15 by an appropriate method,
A large number of learning samples of the object An (nεN) are prepared, each of which is imaged by the image pickup unit 61, and the plurality of gloss features are extracted from the signal output from the image pickup unit 61 by the gloss feature input unit 62. Output to the calculation unit 13. Computing unit 1
At 3, the calculation is performed using the expected value generation circuit 13 signal sent from the gloss feature input unit 62 and the weighting coefficient stored in the memory 15.

[Procedure Amendment 8]

[Document name to be amended] Statement

[Correction target item name] 0124

[Correction method] Change

[Correction content]

In the above embodiment, the case where the learning unit 14 carries out learning using the back propagation model of the neural network has been described, but learning is carried out using other learning models such as multivariate analysis. It goes without saying that it can be applied to cases.

[Procedure Amendment 9]

[Document name to be amended] Statement

[Name of item to be corrected] 0125

[Correction method] Change

[Correction content]

Example 13 An embodiment of the invention of claim 14 will be described below with reference to the drawings. FIG. 27 is a block diagram showing the configuration of the present embodiment, and the same or corresponding parts as in FIG. 1 will be assigned the same reference numerals and explanations thereof will be omitted. In the figure, reference numeral 71 denotes a gloss feature input unit for extracting P gloss features Gp (pεP) , which are the plurality of gloss features described in the twelfth embodiment, from signals output from the sensor group 11, and 72 denotes Q.
It is a color feature detection unit that extracts the specified wavelengths Wq (qεQ) .

[Procedure Amendment 10]

[Document name to be amended] Statement

[Name of item to be corrected] 0126

[Correction method] Change

[Correction content]

Next, the operation will be described based on the flowchart shown in FIG. First, the learning process of step ST1 will be described. In this learning process, a learning process using a back propagation model of a neural network (neural network) is performed. N types of objects An (nεN) having high gloss and M having low gloss as identification objects
Assuming a type of object Bm (mεM) , the gloss feature input unit 71 extracts a plurality of gloss features from the signal output from the sensor group 11.

[Procedure Amendment 11]

[Document name to be amended] Statement

[Name of item to be corrected] 0127

[Correction method] Change

[Correction content]

The memory 15 is initialized by an appropriate method, a large number of learning samples of the objects An (nεN) and Bm (mεM) are prepared, detected by the sensor group 11, and output from the sensor group 11. The gloss feature input section 71 and the color feature detection section 72 respectively extract a plurality of gloss features and color features from the generated signal, and output signals corresponding to these gloss features and color features to the computing section 13.

[Procedure Amendment 12]

[Document name to be corrected] Drawing

[Correction target item name] Figure 9

[Correction method] Change

[Correction content]

[Figure 9]

[Procedure Amendment 13]

[Document name to be corrected] Drawing

[Correction target item name] Fig. 28

[Correction method] Change

[Correction content]

FIG. 28

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁶ Identification code Internal reference number FI Technical display location G06T 1/00 G06F 15/64 310

Claims (15)

[Claims] 1. A sensor for detecting an identification object and outputting a sensor signal, a sensor signal input section for taking in a sensor signal output from the sensor, a memory for storing a weighting coefficient used for identification, and the sensor signal input. From the error between the calculation unit that performs the discrimination calculation using the sensor signal obtained by the section and the weighting coefficient stored in the memory and outputs the discrimination result, and the output of the calculation unit and the signal indicating the desired discrimination state. A learning unit that calculates a weighting factor used for identification and corrects the weighting factor stored in the memory, and capture of a specific sensor signal highly contributing to identification based on the weighting factor stored in the memory. A sensor output control signal for commanding is output to the sensor signal input section, and the sensor signal input section outputs the sensor signal acquired based on the sensor output control signal. An identification device, comprising: a control unit that causes the operation unit to perform an identification operation for identifying the identification object using the weighting coefficient stored in the memory. 2. The sensor is a color image pickup device for detecting the three color-separated color signals of the object to be identified and outputting as a sensor signal, and the sensor signal input section has amplifiers corresponding to the three color-primary signals, respectively. The identification device according to claim 1, wherein the identification device is a three-primary-color signal input unit that takes in the three-primary-color signals. 3. The identification device according to claim 1, wherein the sensor is a plane arrangement sensor which is arranged on a plane to receive the image Fourier-transformed by the lens and which outputs the image as a sensor signal. 4. The identification according to claim 1, wherein the sensor is a planar arrangement sensor including photoelectric element groups arranged on a plane for detecting information on a pattern of an identification object and outputting a sensor signal. apparatus. 5. A plane arrangement sensor comprising photoelectric elements arranged on a plane, a geometric feature extraction unit for taking in a sensor signal output from the plane arrangement sensor and extracting a geometric feature of an identification target, A memory for storing a weighting factor to be used, an arithmetic unit for performing an identification operation using the geometrical feature obtained by the geometrical feature extraction unit and the weighting factor stored in the memory, and outputting an identification result for the geometrical feature; , A learning unit that calculates a weighting coefficient used for identification from the error between the output of the operation unit and a signal indicating a desired identification state and corrects the weighting coefficient stored in the memory, and a weighting coefficient stored in the memory. A sensor output control signal for instructing the acquisition of a specific sensor signal having a high degree of contribution to identification is generated for the geometric feature extraction unit, and the sensor output control signal is set to the sensor output control signal. An identification device comprising: a control unit that causes an operation unit to perform an identification operation using the geometric feature extracted by the geometric feature extraction unit and the weighting coefficient stored in the memory. 6. A sensor, a sensor signal input section for taking in a sensor signal output from the sensor, a memory having a weighting factor forming a multilayer neural network, and a sensor obtained by the sensor signal input section. A weighting coefficient of the memory is calculated by calculating a weighting coefficient from an error between an output of the operation unit and an output of the identification unit and a signal indicating a desired identification state, and performing an identification operation based on the signal and the weighting coefficient of the memory. And a learning unit for correcting the partial differential value with respect to the input layer of the output layer of the multi-layered neural network in the memory, and based on the partial differential value, a sensor output control signal for instructing acquisition of a specific sensor signal is described above. An identification device, comprising: a control unit that outputs to a sensor signal input unit and causes the operation unit to perform an identification operation. 7. The control unit divides an input signal used for learning for each identification object, selects an intermediate value of each input signal, further generates an intermediate value thereof to obtain a partial differential value, and the partial differential value is obtained. 7. The identification device according to claim 6, wherein the identification device is an operation control unit that generates a sensor output control signal that takes in a specific sensor signal based on a value and causes the operation unit to perform an identification operation. 8. A sensor, a sensor signal input section for taking in a sensor signal output from the sensor, a memory for storing a weighting coefficient used for identification, a sensor signal obtained by the sensor signal input section and the memory. An arithmetic unit that performs an identification operation using the stored weighting factors and outputs an identification result, and that performs an inverse operation from the given identification state and the weighting factors stored in the memory to generate a predicted sensor signal; , A weighting coefficient used for the above discrimination is calculated from the error between the output of the calculation unit and a signal indicating a desired discrimination state, and the weighting coefficient of the memory is corrected, or the weighting coefficient of the memory is calculated based on the signal instructing the correction. A learning unit for correction, a control unit for causing the calculation unit to perform an identification calculation and the inverse calculation, a signal indicating a desired identification state for the calculation unit, and a signal for instructing the correction. An identification device comprising: an input device for inputting and a display device for displaying the generated predicted sensor signal. 9. The calculation unit calculates the sensor signal obtained by the sensor signal input unit and the weighting coefficient stored in the memory and outputs a discrimination result, and the discrimination state provided and the memory stored in the memory. Three-dimensional model synthesis for synthesizing a three-dimensional model to be identified using average image data created from image data of a plurality of learning samples related to the object to be identified 9. The identification device according to claim 8, wherein the identification device is a computing unit, and the display device is a three-dimensional model display device that displays the generated predicted sensor signal and the three-dimensional model. 10. An arithmetic unit calculates the sensor signal obtained by the sensor signal input unit and a weighting coefficient stored in the memory and used for identification, and outputs an identification result, and a given identification state and the memory. A three-dimensional model synthesizing calculation unit that generates a sensor signal that is inversely calculated from the weighting coefficient stored in the And the display device is a three-dimensional model display device for displaying the three-dimensional model and the original image of the synthesized identification object, and the input device is a signal indicating a desired identification state for the three-dimensional model synthesis calculation unit and It is a difference information input device for inputting a signal for instructing the correction or for a user inputting information on a difference between the three-dimensional model of the identification object and the original image. The identification device according to claim 8. 11. A computing unit computes the sensor signal obtained by the sensor signal input unit and a weighting coefficient stored in the memory to output a discrimination result, and also stores a given discrimination state and the memory stored in the memory. The predicted object signal is generated by performing an inverse operation from the weighting coefficient, and further, based on the image data of a plurality of learning samples related to the object to be identified, 3
The display device is a difference image calculation unit that synthesizes a three-dimensional model and generates an image of a difference between the three-dimensional model and the original image, and the display device is the synthesized three-dimensional model of the identification object and the original image, and the three-dimensional model. 9. The identification device according to claim 8, wherein the identification device is a difference image display device that displays an image of the difference between the original image and the original image. 12. The arithmetic unit synthesizes a plurality of three-dimensional models of the identification object having different evaluation and quantification of a predetermined evaluation scale together with the original image of the identification object, and the input rank is used as the rank scale to obtain the rating scale. Calculate the correlation and obtain the correction amount of the weight count 3
The input device is a rank scale input device that inputs a rank based on another rating scale for the three-dimensional model as a rank scale, and the display device displays the three-dimensional model together with the original image. The identification device according to claim 8, which is a three-dimensional model display device for displaying. 13. An image pickup section for picking up an image of an object to be identified, a gloss feature input section for extracting a plurality of pieces of gloss feature information from a signal output from the image pickup section, a memory for storing a weighting factor, and the gloss feature input. Calculating the weighting factor from the difference between the output of the calculating unit and the output of the calculating unit and the signal indicating the desired identification state. And a learning unit that corrects the weighting coefficient of the memory. 14. A sensor, a gloss feature input section for extracting a plurality of gloss feature information from a sensor signal output from the sensor, and a color for detecting a plurality of specific wavelengths from the sensor signal output from the sensor group. A feature detection unit, a memory for storing a weighting factor, a gloss feature information obtained by the gloss feature input unit, a color feature information obtained by the color feature detection unit, and a weighting factor stored in the memory. Then, the calculation unit that outputs the identification result, the learning unit that calculates the weighting coefficient from the error between the output of the calculation unit and the signal indicating the desired identification state and corrects the weighting coefficient of the memory, and the gloss feature input unit are extracted. A control signal for fetching the specific color feature information based on the gloss feature information and the weighting coefficient stored in the memory is generated to the color feature detection unit, and based on the control signal. And a control unit that causes the arithmetic unit to perform an identification calculation using the color characteristic information detected by the color characteristic detection unit and the weighting coefficient stored in the memory. 15. The sensor is an image pickup unit for picking up an image of an identification target, and the color feature detection unit includes a mask for partially blocking the light of the identification target, and a plurality of halves for reflecting the light passing through the mask. A mirror, an interference filter that transmits only light of a specific wavelength with respect to the light reflected by each of the plurality of half mirrors, a photoelectric element group that detects light transmitted through each of the interference filters, and the photoelectric element group 15. A color feature information processing unit for detecting the color feature information of the identification object based on the output of FIG.
The identification device described.