CN118823024A - A method for identifying machining defects of air-cooled radiators - Google Patents

Abstract

The present invention relates to the field of image processing technology, and provides a method for identifying processing defects of air-cooled radiators, comprising: obtaining an original image and performing preprocessing to obtain a preprocessed image, wherein the original image includes several continuous frames of multiple surface processing processes; performing superpixel segmentation on the preprocessed image to obtain multiple superpixel regions, and obtaining their regional image features; dividing the multiple superpixel regions into noise regions and surface regions; obtaining the noise influence coefficient of each superpixel region by analyzing the degree of interference of each superpixel region by the noise region; obtaining the adaptive filtering strength coefficient of each superpixel region based on the noise influence coefficient of each superpixel region, and performing Gaussian filtering and smoothing on the original image based on the adaptive filtering strength coefficient of each superpixel region to obtain a smooth surface image; and identifying defective regions based on the smooth surface image. The method effectively improves the accuracy and reliability of the final defect recognition result.

Description

A method for identifying machining defects of air-cooled radiators

Technical Field

The present invention relates to the technical field of image processing, and in particular to a method for identifying processing defects of an air-cooled radiator.

Background Art

Air-cooled radiators are an important device for heat dissipation, especially in artificial intelligence and automobile engines. Air-cooled radiators reduce heat through air flow to keep the device within a safe operating temperature range. However, during the manufacturing and processing process, air-cooled radiators may have various defects, which may affect their heat dissipation performance and lifespan, and even cause instability or damage to the device. Therefore, the identification of processing defects of air-cooled radiators is the key to ensuring the quality, safety and reliability of radiator products.

In the existing air-cooled radiator processing defect recognition method, computer vision technology is used to automatically analyze and identify processing defects in the radiator surface image. However, since air-cooled radiators usually undergo different surface treatments, such as spraying, anodizing, etc., these treatments will change the optical properties of the radiator surface, resulting in uneven light reflection or absorption in the optical image, causing noise points of different brightness to appear in the image, thereby interfering with the recognition of the imaging system, making it difficult for existing defect recognition methods to accurately identify.

Summary of the invention

In order to solve the technical problem that the existing defect recognition method is difficult to accurately identify, the purpose of the present invention is to provide a method for identifying processing defects of air-cooled radiators. The technical solution adopted is as follows:

A method for identifying machining defects of an air-cooled radiator, comprising:

Acquire an original image and perform preprocessing to obtain a preprocessed image, wherein the original image includes a plurality of continuous frames of a plurality of surface treatment processes;

Performing superpixel segmentation on the preprocessed image to obtain a plurality of superpixel regions, and acquiring regional image features of the plurality of superpixel regions;

Dividing the plurality of super-pixel regions into noise regions and surface regions according to regional image features of the plurality of super-pixel regions;

The noise influence coefficient of each super-pixel region is obtained by analyzing the interference degree of each super-pixel region by the noise region;

Obtaining an adaptive filtering strength coefficient of each super-pixel region based on the noise influence coefficient of each super-pixel region, and performing Gaussian filtering and smoothing processing on the original image based on the adaptive filtering strength coefficient of each super-pixel region to obtain a smooth surface image;

Defective regions are identified based on the smooth surface image.

Preferably, the step of dividing the plurality of super-pixel regions into noise regions and surface regions according to regional image features of the plurality of super-pixel regions comprises:

According to the regional image features of the plurality of super-pixel regions, a surface characteristic stability factor and an information loss factor of each super-pixel region are obtained;

The noise coefficient of each super-pixel region is calculated based on the surface characteristic stability factor and the information loss factor of each super-pixel region;

The multiple super-pixel regions are divided into a noise region and a surface region according to the noise coefficient of each super-pixel region.

Preferably, the regional image features include regional pixel data; the regional pixel data includes the number of regional edge pixels, the total number of pixels in the region, and the pixel values and coordinates of the pixels in the region;

The surface characteristic stability factor of each superpixel region is obtained based on the regional image features of the multiple superpixel regions, and the surface characteristic stability factor of each superpixel region is calculated based on the number of regional edge pixels in each superpixel region, the total number of pixels in the region, and the difference between the same frame images of two adjacent surface processing processes.

Preferably, obtaining the information missing factor of each superpixel region according to the regional image features of the plurality of superpixel regions comprises:

Obtain pixel mean data according to the pixel values of the pixel points in the region to form a pixel mean data set, and obtain the frequency of occurrence of each pixel mean data in the pixel mean data set;

The information missing factor of each superpixel region is calculated based on the pixel mean and the frequency of occurrence of all superpixel regions.

Preferably, the calculation formula of the information missing factor of the superpixel area is:

In the formula, For the The first The serial number of the superpixel region, For the The frequency of the pixel mean of a superpixel region appearing in the pixel mean data set, Represents the The average frequency of pixels in the superpixel region, For the All the first The average of the pixel mean frequencies in the superpixel region, Represents the The prominence coefficient of the mean pixel value in the superpixel region is Represents the The total number of all superpixel regions closely adjacent to the superpixel region, Represents a collection The The serial number of the superpixel region, Represents the The pixel mean of the superpixel region is Represents the The pixel mean of the superpixel region is Represents the The information loss factor of the superpixel region.

Preferably, the noise coefficient of the superpixel area is calculated as follows:

In the formula, is the noise factor, is the information missing factor, is the surface property stabilization factor.

Preferably, the plurality of pixel regions are divided into noise regions and surface regions according to the noise coefficient of each pixel region, a threshold is preset, and it is determined whether the noise coefficient of each pixel region is greater than the preset threshold, if so, it is a noise region, otherwise, it is a surface region.

Preferably, the noise influence coefficient of each superpixel area is obtained by analyzing the degree of interference of each superpixel area by the noise area, and the noise influence coefficient of each superpixel area is calculated according to the surface characteristic stability factor of each superpixel area and the surface characteristic stability factors of a preset number of noise areas around it, and the Euclidean distance between the two.

Preferably, the calculation formula of the noise influence coefficient of the super pixel area is:

In the formula, For the The surface property stability factor of the superpixel area, For the The total number of the four nearest noise regions around the superpixel region, for The first in the collection The number of the noise region, For the The superpixel region is The Euclidean distance of the noise region, For the The surface characteristic stability factor of the noise region, For the The noise influence coefficient of the super pixel area.

Preferably, the adaptive filtering strength coefficient of each superpixel area is obtained based on the noise influence coefficient of each superpixel area, the maximum standard deviation of the Gaussian filter is calculated, and the product of the maximum standard deviation of the Gaussian filter and the noise influence coefficient of the superpixel area is used as the adaptive filtering strength coefficient of the superpixel area.

The present invention has the following beneficial effects: the processing defect identification method of the air-cooled radiator proposed in the present invention obtains the original image of several continuous frames including multiple surface treatment processes, divides multiple super-pixel areas into noise areas and surface areas according to the regional image characteristics of the multiple super-pixel areas, and then designs an adaptive filtering intensity coefficient according to the interference degree of each super-pixel area affected by noise to ensure the accuracy and reliability of subsequent image processing. It can avoid the problem that the optical properties of the surface of the air-cooled radiator are changed due to different surface treatments such as spraying, anodizing, etc., resulting in uneven light reflection or absorption in the optical image, causing noise points of different brightness in the image, thereby interfering with the recognition of the imaging system, and effectively improves the accuracy and reliability of the final defect recognition result.

BRIEF DESCRIPTION OF THE DRAWINGS

In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

FIG1 is a schematic flow chart of a method for identifying machining defects of an air-cooled heat sink provided by the present invention;

FIG2 is a schematic diagram of an image of the surface of an air-cooled heat sink obtained in the present invention;

FIG. 3 is a schematic diagram of a sub-flow chart of step S3 in the method for identifying machining defects of an air-cooled heat sink provided by the present invention.

DETAILED DESCRIPTION

In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the processing defect identification method of an air-cooled radiator proposed by the present invention, its specific implementation method, structure, features and effects, in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

The specific scheme of the method for identifying machining defects of an air-cooled radiator provided by the present invention is described in detail below with reference to the accompanying drawings.

Please refer to FIG1 , which shows a flow chart of a method for identifying machining defects of an air-cooled heat sink provided by an embodiment of the present invention, including:

Step S1, acquiring an original image and preprocessing it to obtain a preprocessed image, wherein the original image includes a plurality of continuous frames of a plurality of surface treatment processes;

Step S2, performing superpixel segmentation on the preprocessed image to obtain a plurality of superpixel regions, and obtaining regional image features of the plurality of superpixel regions;

Step S3, dividing the multiple super-pixel regions into noise regions and surface regions according to regional image features of the multiple super-pixel regions;

Step S4, obtaining a noise influence coefficient of each super-pixel region by analyzing the interference degree of each super-pixel region caused by the noise region;

Step S5, obtaining an adaptive filtering strength coefficient of each super-pixel region based on the noise influence coefficient of each super-pixel region, and performing Gaussian filtering and smoothing processing on the original image based on the adaptive filtering strength coefficient of each super-pixel region to obtain a smooth surface image;

Step S6: identifying defective areas based on the smooth surface image.

The method for identifying processing defects of an air-cooled radiator proposed in the present invention obtains original images of several continuous frames including multiple surface treatment processes, divides multiple super-pixel areas into noise areas and surface areas according to regional image features of multiple super-pixel areas, and then designs an adaptive filtering intensity coefficient according to the degree of interference of each super-pixel area affected by noise to ensure the accuracy and reliability of subsequent image processing. It can avoid the problem that the optical properties of the surface of the air-cooled radiator are changed due to different surface treatments such as spraying, anodizing, etc., which leads to uneven light reflection or absorption in the optical image, and noise points of different brightness appear in the image, thereby interfering with the recognition of the imaging system, and effectively improves the accuracy and reliability of the final defect recognition result.

The main purpose of the present invention is to solve the problem that the optical characteristics of traditional defect detection methods change due to surface treatment, which makes it difficult to accurately identify defects. The present invention proposes a more accurate and reliable defect identification method.

The specific scenario targeted by the present invention is: air-cooled radiator processing defects refer to various problems that occur during the manufacturing process, such as: poor welding quality (cracks, pores, etc. in the weld), material defects (metal fatigue cracks on the surface of the air-cooled radiator, etc.). The present invention targets traditional defect detection methods that rely on analyzing specific features in the image to identify defects, but these methods are interfered with due to changes in optical properties caused by surface treatment, making accurate identification of defects difficult. An adaptive image denoising method based on local area optical features is proposed to ensure accuracy in subsequent defect identification. It can be understood that the present invention further processes the image on the basis of traditional defect detection methods to ensure the accuracy and reliability of the final defect identification.

The following is a detailed description of each step:

In step S1, an image of the surface of the air-cooled radiator may be captured by an industrial camera to obtain a multi-frame image set. Specifically, multiple high-resolution industrial cameras may be installed 2 meters around the processing streamline of the air-cooled radiator (ensuring that the camera matrix can accurately cover the surface of the air-cooled radiator at different angles), and the image of the surface of the air-cooled radiator after each processing is captured every 1 second (as shown in FIG2 ), and a multi-frame image set A of the radiator surface within a period of time (one hour) is obtained; it should be noted that when collecting the radiator surface image, only the process of surface processing of the radiator (such as anodizing, electroplating, frosting, etc.) is collected and obtained.

The preprocessing includes increasing contrast, grayscale processing, etc., in order to facilitate subsequent image processing. Specifically, it is to perform image processing on multiple frames of image A, including increasing contrast, grayscale processing, etc.; and adjust the calculated focus size ratio value f(x, y) (acquired based on local image gradient), and dynamically extract image features according to the focus size ratio value f(x, y). This operation helps to dynamically select the appropriate size to extract features in different image areas (known technology); to ensure the reliability of subsequent image analysis results.

In step S2, superpixel segmentation refers to the process of subdividing a digital image into multiple image sub-regions (a collection of pixels). A superpixel region is a small region composed of a series of pixels with adjacent positions and similar characteristics such as color, brightness, and texture. These small regions retain effective information for further image segmentation and generally do not destroy the boundary information of objects in the image.

In the existing air-cooled heat sink processing defect recognition method, a superpixel region segmentation algorithm is usually used to obtain multiple superpixel regions with different similar features in the heat sink surface image, so as to set thresholds based on the superpixel features to screen the defective regions; however, in actual operation, the surface treatment of the heat sink during processing will affect the change of the optical properties of the heat sink surface, resulting in a large number of brightness change areas, namely noise areas, in the image, and the simple segmentation method based on a single threshold cannot accurately distinguish between the noise area and the real surface defect area. Therefore, it is necessary to further perform feature analysis on the abnormal defect area to accurately distinguish between noise and real defects on the heat sink surface.

The present invention is based on the traditional method to obtain several super-pixel areas in the radiator surface image, analyzes the regional image features of the local area of the air-cooled radiator surface at different times, obtains the interference degree of the current area affected by noise, and designs the adaptive image denoising strength to ensure the accuracy and reliability of the subsequent acquisition of the air-cooled radiator surface image. This is specifically reflected in steps S3-S5.

In step S3, in general, the surface treatment of some areas of the air-cooled heat sink is uneven during the processing, which will cause the optical properties of these areas (such as reflectivity or diffuse reflectivity) to be different from those of the surrounding areas. This difference may appear as a brightness change area in the image, that is, a noise area. Preferably, as shown in FIG3, step S3 includes:

Step S31: obtaining a surface characteristic stability factor and an information loss factor of each superpixel region according to regional image features of the plurality of superpixel regions;

Step S32: Calculate the noise coefficient of each super-pixel region based on the surface characteristic stability factor and the information loss factor of each super-pixel region;

Step S33: Divide the multiple super-pixel regions into noise regions and surface regions according to the noise coefficient of each super-pixel region.

The regional image features include regional pixel data; the regional pixel data includes the number of regional edge pixels, the total number of pixels in the region, and the pixel values and coordinates of the pixels in the region. That is, based on the continuous frames in the multi-frame image set A, the canny edge detection algorithm is used to obtain the edge curve of each superpixel region, count the number of edge pixels in each superpixel region and the total number of pixels in the region; obtain the pixel values of all pixels in each superpixel region; at the same time, obtain the number and coordinates of all edge pixels in each superpixel region in the target frame, and obtain the number and coordinates of all edge pixels in the superpixel region at the same position in the next frame.

In step S31, in this step, it can be divided into two steps: obtaining the surface characteristic stability factor of the super-pixel area and obtaining the information missing factor of the super-pixel area, which can be performed simultaneously or successively.

In obtaining the surface characteristic stability factor of the super-pixel region, it can be understood that the reflection and diffuse reflection of light under different surface characteristics will produce different brightness changes, so that the noise area shows changes in position and shape, that is, fluidity, under different image acquisition conditions or processing steps; for the real defect area, due to the stability of its heat sink surface structure, the characteristics of its local area are relatively stable at different times. Preferably, the surface characteristic stability factor of each super-pixel region is obtained based on the regional image characteristics of multiple super-pixel regions, and the surface characteristic stability factor of each super-pixel region is calculated based on the number of regional edge pixels in each super-pixel region, the total number of pixels in the region, and the difference between the same frame images of two adjacent surface processing processes.

In some specific implementations, the surface characteristic stability factor of the superpixel region is obtained as:

The number of edge points whose coordinates overlap in two adjacent frames is calculated and divided by the number of all edge points in the superpixel area in the target frame to obtain the structural stability factor, denoted as D. The larger the value, the more stable and unchanging edge curves exist in the same area in two adjacent frames.

Construct a specific formula for the surface characteristic stability factor of the target superpixel area:

In the formula, Represents the first Superpixel area, represents the set of all surface treatments that the target heat sink has undergone. Represents The first in the collection Surface treatment process, Represents The same as the Surface treatment process adjacent to the surface treatment process, Represents the radiator The superpixel region is The first frame of the surface image of the surface treatment process (here it can be selected based on experience), Represents the radiator The superpixel region is The first frame of surface image in the surface processing process, Represents the radiator The superpixel region is The first frame surface image and the second frame surface image of the surface processing process The structural stability factor corresponding to the first frame surface image of the surface treatment process, Represents the The number of edge pixels in a superpixel area, Represents the The number of pixels in a superpixel region, Represents the The surface property stability factor of the superpixel area, is the mean square error calculation symbol, based on the radiator The superpixel region is and The pixel value difference at the same position in the first frame surface image of each surface processing flow is used to calculate the mean square error. is the normalization function. and It suffices to select the same frame of surface images in two adjacent surface treatment processes, and based on experience, the first frame is preferred.

It indicates the The ratio of the difference in the pixel value set of a super-pixel region in different frame images to the stable structure coefficient of the region. When the super-pixel region is a defective region, its surface characteristics are usually manifested as obvious edge curve features between images at different processing stages due to the stability of the radiator structure, and the edge curve is stable and will not change with the surface processing. Therefore, when the target super-pixel region is a defective region after uneven surface treatment, even if the pixels of this region have huge differences between different frames, its internal edge curve is relatively stable and will not have large differences. The correction logic is performed, and the structural stability of the target area is corrected by combining the number of edge curves in the area, that is, the edge pixel content, that is, The larger the value, the The greater the possibility that a superpixel region is a defective region, The larger the value of .

In obtaining the information missing factor of the superpixel area, in the surface image of the air-cooled radiator, there is a significant contrast difference between the real defect area and the noise area and the normal area. The noise area has different brightness characteristics compared with the normal area and the real defect due to the structural characteristics, resulting in the masking and loss of local area image information. Therefore, all noise areas in the surface image can be obtained by analyzing the brightness difference characteristics between different areas and surrounding areas. Preferably, the information missing factor of each superpixel area is obtained according to the regional image characteristics of multiple superpixel areas, including: obtaining pixel mean data according to the pixel values of the pixel points in the area, forming a pixel mean data set, and obtaining the frequency of each pixel mean data in the pixel mean data set; the information missing factor of each superpixel area is calculated according to the pixel mean and the frequency of occurrence of all superpixel areas.

In some specific implementations, the information missing factor for obtaining the superpixel region is:

Obtain the pixel mean data of each superpixel area in the surface image to form a data set B. At the same time, obtain the frequency of each mean data in the set;

Construct the information missing factor formula for the specific target superpixel area:

In the formula, For the The first The serial number of the superpixel region, For the The frequency of the pixel mean of a superpixel region appearing in the pixel mean data set, Represents the The average frequency of pixels in the superpixel region, For the All the first The average of the pixel mean frequencies in the superpixel region, Represents the The prominence coefficient of the mean pixel value in the superpixel region is Represents the The total number of all superpixel regions closely adjacent to the superpixel region, Represents a collection The The serial number of the superpixel region, Represents the The pixel mean of the superpixel region is Represents the The pixel mean of the superpixel region is For the The information loss factor of the superpixel region.

In summary, It indicates the prominence of the mean pixel value of the target superpixel region in the overall surface image. Since the area of the normal region in the surface image must be much larger than the area of the region, and the mean pixel value of the normal region is usually similar. Therefore, the pixel value of the region with a greater prominence can effectively represent the normal pixel value of the target radiator surface; when evaluating the missing information of the target region, the information missing can be evaluated based on the brightness difference of the remaining closely adjacent regions. At the same time, a larger weight is set for the pixel value close to the normal surface area to ensure that the analysis result can better highlight the missing information of the original image caused by the abnormal brightness of the target region. The larger the value, The larger the value of .

In step S32, preferably, the noise coefficient of the super pixel area is calculated as:

In the formula, is the noise factor, is the information missing factor, is the surface property stabilization factor.

is the noise coefficient of the area. The larger its value is, the larger the information missing coefficient of the superpixel area is. At the same time, the smaller the stability of its surface characteristics is, the greater the possibility that the area is an abnormal area and a noise area.

In step S33, preferably, the plurality of pixel regions are divided into noise regions and surface regions according to the noise coefficient of each pixel region, a threshold is preset, and it is determined whether the noise coefficient of each pixel region is greater than the preset threshold, if so, it is a noise region, otherwise, it is a surface region. The present invention sets the threshold based on experience , the noise coefficient of any superpixel area Compare with the threshold, the area greater than the threshold is considered as the noise area, and the area less than the threshold is considered as the surface area (including the normal area and the defect area). At this point, the normal area, defect area and noise area in the radiator surface image are accurately distinguished.

In step S4, it can be understood that when filtering and smoothing any super-pixel region in the radiator surface image, the denser the noise region around it is, the closer it may be to the real defect boundary, or even overlap it. Therefore, for any long pixel region, before performing subsequent filtering and smoothing, it is necessary to comprehensively consider its regional characteristics and the density of the surrounding noise region to obtain the noise influence coefficient of its region. Preferably, the noise influence coefficient of each super-pixel region obtained by analyzing the degree of interference of each super-pixel region by the noise region is calculated according to the surface characteristic stability factor of each super-pixel region and the surface characteristic stability factors of a preset number of the noise regions around it, and the Euclidean distance between the two.

In some specific implementations, the noise influence coefficient of the superpixel region is obtained as:

Get the four nearest noise regions around any superpixel region, and the Euclidean distance between the centroids of any superpixel region;

Construct a specific noise impact coefficient formula:

In the formula, For the The surface property stability factor of the superpixel area, For the The total number of the four nearest noise regions around the superpixel region, for The first in the collection The number of the noise region, For the The superpixel region is The Euclidean distance of the noise region, For the The surface characteristic stability factor of the noise region, For the The noise influence coefficient of the super pixel area. It can be understood that The value of is 4.

In summary, the noise impact of the target area is quantified by analyzing the surface characteristic stability factor of the target area itself and combining the aggregation degree and noise intensity of the four nearest noise areas around it, that is, The larger the value is, the more it indicates that the target area itself has a certain degree of unstable surface characteristics, or even if it has relatively stable surface characteristics, there is a relatively dense noise area with high noise intensity around it. In this case, it is easier to blur the real edge curve of the target area. When performing subsequent image smoothing processing, this part of the area needs to be set with a larger smoothing effect.

In step S5, according to the calculated local area noise influence degree, the original image is subjected to Gaussian filtering and smoothing processing to reduce the influence of noise points and increase the reliability of the abnormal area after smoothing, so as to obtain a more accurate surface image, so as to facilitate the subsequent further analysis of surface defects. Preferably, based on the noise influence coefficient of each super-pixel area, the adaptive filtering strength coefficient of each super-pixel area is obtained, the maximum standard deviation of the Gaussian filter is calculated, and the product of the maximum standard deviation of the Gaussian filter and the noise influence coefficient of the super-pixel area is used as the adaptive filtering strength coefficient of the super-pixel area.

Specifically, a Gaussian filter with an area of L is set. In the Gaussian filter, the sum of the weights of all elements is 1. Therefore, in theory, the maximum filtering strength of the Gaussian filter should be infinitely close to "the weight of the central element is 1, and the weights of the other elements are 0", that is, the maximum standard deviation that the Gaussian filter theoretically obeys is:

L represents the area of Gaussian filtering, Represents the square difference between the center weight and the weight mean when the center element weight is 1. represents the sum of the squared differences between the remaining L-1 elements and the weight mean when their weights are 0, so It represents the variance of all element weights in the Gaussian filter when the weight of the central element of the Gaussian filter is 1. The maximum standard deviation obtained by taking the square root of the variance is Z.

It should be noted that the maximum standard deviation of the Gaussian filter is calculated in order to adaptively set the Gaussian filter strength required for each signal area based on it. It is only used as reference data and has no practical significance.

Representative The noise impact coefficient of the area, is the Gaussian filter standard deviation after adjustment in the i-th region;

The maximum standard deviation of the Gaussian filter with area L is multiplied by the noise influence coefficient of the i-th region to obtain the adjusted Gaussian filter standard deviation of the i-th region. .

According to Gaussian filtering, a set of The element weight set of each position in the filter has a one-to-one correspondence with the pixel value of each position on the radiator surface image. The data is weighted averaged to obtain the filtered surface image. The above is a well-known technology and will not be described in detail here. Finally, the smoothed radiator surface image is obtained, that is, the smoothed surface image.

In step S6, based on the obtained smooth surface image, edge detection is used to extract possible defective area features from the image, and the extracted features are identified. For example, template matching technology can be used to detect and identify defective areas to obtain more accurate defective areas. Finally, the defective areas are visualized and a report is generated for further analysis and processing, so as to better evaluate and process the surface processing quality of the air-cooled radiator.

It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

Claims (10)

1. The method for identifying the processing defects of the air-cooled radiator is characterized by comprising the following steps of:

Acquiring an original image and preprocessing the original image to obtain a preprocessed image, wherein the original image comprises a plurality of continuous frames of a plurality of surface processing flows;

performing super-pixel segmentation on the preprocessed image to obtain a plurality of super-pixel areas, and obtaining area image features of the plurality of super-pixel areas;

dividing the super pixel areas into noise areas and surface areas according to the area image characteristics of the super pixel areas;

Obtaining noise influence coefficients of each super pixel region by analyzing the interference degree of each super pixel region by the noise region;

obtaining self-adaptive filtering intensity coefficients of each super-pixel region based on the noise influence coefficients of each super-pixel region, and performing Gaussian filtering smoothing processing on an original image based on the self-adaptive filtering intensity coefficients of each super-pixel region to obtain a smooth surface image;

And identifying a defective area based on the smooth surface image.

2. The method of claim 1, wherein the step of dividing the super pixel regions into a noise region and a surface region according to region image characteristics of the super pixel regions comprises:

obtaining a surface characteristic stability factor and an information deletion factor of each super pixel region according to the region image characteristics of the plurality of super pixel regions;

calculating to obtain the noise coefficient of each super pixel region based on the surface characteristic stability factor and the information deletion factor of each super pixel region;

And dividing the super pixel areas into a noise area and a surface area according to the noise coefficient of each super pixel area.

3. The method of claim 2, wherein the area image features include area pixel data; the regional pixel data comprises the number of regional edge pixel points, the total number of the regional pixel points, the pixel values and coordinates of the regional pixel points;

The surface characteristic stability factors of the super pixel areas are obtained according to the area image characteristics of the plurality of super pixel areas, namely, the surface characteristic stability factors of the super pixel areas are obtained according to the number of the area edge pixel points in the super pixel areas, the total number of the pixel points in the areas and the difference of the same frame of image of two adjacent surface treatment processes.

4. The method for identifying a processing defect of an air-cooled radiator according to claim 3, wherein obtaining an information deletion factor of each super pixel region from region image features of a plurality of the super pixel regions comprises:

Obtaining pixel mean value data according to pixel values of pixel points in the region, forming a pixel mean value data set, and obtaining the occurrence frequency of each pixel mean value data in the pixel mean value data set;

And calculating the information deletion factor of each super pixel region according to the pixel mean value and the occurrence frequency of all the super pixel regions.

5. The method for identifying machining defects of an air-cooled radiator according to claim 4, wherein the information missing factor of the super pixel area is calculated by the formula:

In the method, in the process of the invention, Is the firstThe first super pixel regionThe sequence number of the individual super-pixel regions,Is the firstThe frequency with which the pixel mean of the super pixel regions occurs in the pixel mean data set,Represents the firstThe pixel mean frequency of the super pixel regions,To divide byAll of the first super pixel regionsThe average of the pixel average frequency of the super pixel regions,Represents the firstThe saliency coefficient of the pixel mean of the super pixel region,Represents the firstThe total number of all super pixel regions immediately adjacent around each super pixel region,Represents a collectionThe first of (3)The sequence number of the individual super-pixel regions,Represents the firstThe pixel mean value of the individual super-pixel regions,Represents the firstThe pixel mean value of the individual super-pixel regions,Is the firstInformation deletion factor of each super pixel region.

6. The method for identifying machining defects of an air-cooled radiator according to claim 4, wherein a calculation formula of a noise coefficient of the super pixel area is:

In the method, in the process of the invention, For the noise figure to be a noise figure,As the information-missing factor(s),Is a surface property stabilization factor.

7. The method for recognizing processing defects of an air-cooled radiator according to claim 6, wherein the dividing of the plurality of pixel areas into noise areas and surface areas according to the noise coefficients of the pixel areas is performed by presetting a threshold value, judging whether the noise coefficient of each pixel area is greater than the preset threshold value, if so, the pixel area is the noise area, and if not, the pixel area is the surface area.

8. The method for recognizing processing defects of an air-cooled radiator according to claim 7, wherein the noise influence coefficients of the super-pixel regions are obtained by analyzing the interference degree of the super-pixel regions by the noise regions, and the noise influence coefficients of the super-pixel regions are obtained by calculating the surface characteristic stability factors of the super-pixel regions and the surface characteristic stability factors of the surrounding preset number of the noise regions, and the euclidean distance between the surface characteristic stability factors and the surrounding preset number of the noise regions.

9. The method for identifying machining defects of an air-cooled radiator according to claim 8, wherein a calculation formula of a noise influence coefficient of the super pixel area is:

In the method, in the process of the invention, Is the firstThe surface characteristic stabilization factor of each super pixel region,For the total number of four noise regions closest to the circumference of the ith super pixel region,Is thatThe first in the collectionThe sequence number of the individual noise region,Is the firstSuper pixel region and the firstThe euclidean distance of the individual noise regions,Is the firstThe surface property stability factor of the individual noise regions,Is the firstNoise influence coefficients of the individual super pixel areas.

10. The method for identifying a processing defect of an air-cooled radiator according to any one of claims 1 to 9, wherein obtaining an adaptive filter intensity coefficient of each super-pixel region based on the noise influence coefficient of each super-pixel region is to calculate a maximum standard deviation of gaussian filtering, and a product of the maximum standard deviation of gaussian filtering and the noise influence coefficient of the super-pixel region is used as the adaptive filter intensity coefficient of the super-pixel region.