WO2024021662A1 - 检测电芯表面缺陷的方法和装置 - Google Patents

检测电芯表面缺陷的方法和装置 Download PDF

Info

Publication number
WO2024021662A1
WO2024021662A1 PCT/CN2023/085076 CN2023085076W WO2024021662A1 WO 2024021662 A1 WO2024021662 A1 WO 2024021662A1 CN 2023085076 W CN2023085076 W CN 2023085076W WO 2024021662 A1 WO2024021662 A1 WO 2024021662A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
cell surface
detected
surface area
defect
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/CN2023/085076
Other languages
English (en)
French (fr)
Inventor
孙鸿基
陈飞
江冠南
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Contemporary Amperex Technology Co Ltd
Original Assignee
Contemporary Amperex Technology Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Contemporary Amperex Technology Co Ltd filed Critical Contemporary Amperex Technology Co Ltd
Priority to EP23744030.0A priority Critical patent/EP4336443B1/en
Priority to ES23744030T priority patent/ES3035473T3/es
Priority to US18/454,184 priority patent/US12406349B2/en
Publication of WO2024021662A1 publication Critical patent/WO2024021662A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/11—Region-based segmentation
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/90—Determination of colour characteristics

Definitions

  • This application relates to the field of battery technology and the field of machine vision inspection technology, and in particular to a method and device for detecting cell surface defects, a method and device for training a defect detection neural network model, electronic equipment, computer-readable storage media and computer program products .
  • Electric vehicles have become an important part of the sustainable development of the automobile industry due to their advantages in energy conservation and environmental protection.
  • battery technology is an important factor related to their development.
  • rechargeable batteries referring to batteries that can be activated by charging to activate active materials and continue to be used after the battery is discharged, also known as secondary batteries
  • a battery box and a series and/or parallel connection located in the battery box.
  • a battery cell is the smallest unit in a battery that provides an energy source.
  • the battery core is a key component for electrochemical reactions in a battery cell. Its main structure includes an anode pole, a cathode pole, and a separator that separates the anode pole and cathode pole.
  • This application aims to solve at least one of the technical problems existing in the prior art.
  • one purpose of this application is to propose a method and device for detecting surface defects of battery cells, a method and device for training a defect detection neural network model, electronic equipment, computer-readable storage media, and computer program products to improve the performance of battery cells.
  • the production yield rate is to propose a method and device for detecting surface defects of battery cells, a method and device for training a defect detection neural network model, electronic equipment, computer-readable storage media, and computer program products to improve the performance of battery cells.
  • An embodiment of the first aspect of the present application provides a method for detecting surface defects of a battery core.
  • the method includes: using an image acquisition unit to obtain an initial image including the surface of the battery core; preprocessing the initial image to obtain at least one image to be detected of the surface of the battery core; and inputting at least one image to be detected into a defect detection neural network model, and Obtain the detection results output by the defect detection neural network model.
  • the detection results are used to indicate whether there are defects on the surface of the battery core.
  • the defect characteristics in the image can be highlighted to facilitate subsequent defect detection neural network.
  • Network model detection can extract more features from images, thereby improving the efficiency and accuracy of defect detection.
  • At least one image to be detected includes a plurality of images to be detected, the initial image includes a cell surface area where the cell surface is located, and the initial image is preprocessed to obtain at least one to-be-detected image of the cell surface.
  • the image includes: extracting the battery core surface area from the initial image; and segmenting the battery core surface area to obtain multiple images to be detected. Since some defects on the surface of the battery core are relatively small, for example, they only account for a small part of the entire battery core surface. By extracting the surface area of the battery core and segmenting it to obtain small-sized images to be detected, the characteristics of these defects can be It is more prominent in the inspection image, thereby improving the accuracy of defect detection.
  • the initial image has a first exposure
  • extracting the cell surface area from the initial image includes: directly extracting the cell surface area from the initial image; and adjusting the exposure of the cell surface area to obtain A cell surface area having a second degree of exposure, wherein the first degree of exposure is higher than the second degree of exposure.
  • the above method uses an initial image with a higher exposure to extract the surface area of the battery core, and then adjusts the exposure of the surface area of the battery core. This can make the characteristics of certain defects more prominent in the low-exposure image, thereby improving the performance of the battery core. Accuracy of surface area extraction and subsequent defect detection.
  • the first exposure is configured such that the pixel value range of the cell surface area in the initial image is different from the pixel value range of other areas in the initial image except the cell surface area.
  • extracting the cell surface area from the initial image further includes: extracting the cell surface area of the preset color channel from the cell surface area with the second exposure, and segmenting the cell surface area.
  • obtaining multiple images to be detected includes: segmenting the cell surface area of a preset color channel to obtain multiple images to be detected. Since the characteristics of certain defects on the cell surface are more prominent in the image of the preset color channel, the characteristics of these defects can be highlighted by extracting the cell surface area of the preset color channel, thereby facilitating the subsequent detection of the defect detection neural network model. , to improve the accuracy of defect detection.
  • the default color channel is the G channel. Since blue glue usually exists on the surface of the battery core, extracting the blue glue from the cell surface area of the G channel can avoid extracting the blue glue, thereby reducing the impact of blue glue on defect feature extraction. This facilitates the subsequent defect detection neural network model to detect defects in the cell surface area to further improve the accuracy of defect detection.
  • directly extracting the cell surface area from the initial image includes: extracting the cell surface area from the initial image based at least on a pixel value range of the cell surface area. Since there are certain differences in the pixel values of various areas on the initial image, the above technical solution for extracting the battery core surface area based on the pixel values of each area on the initial image can quickly and effectively extract the battery core surface area.
  • the multiple images to be detected are of equal size.
  • the above method can easily meet the input image size requirements of the defect detection neural network model and promote the detection of the defect detection neural network model.
  • the detection results include the defect confidence corresponding to each image to be detected, and inputting at least one image to be detected into the defect detection neural network model, and obtaining the detection result output by the defect detection neural network model includes: adding at least An image to be detected is input into a defect detection neural network model to obtain the defect confidence of each image to be detected in at least one image to be detected; and based on the defect confidence of each image to be detected in at least one image to be detected, determine the battery core Whether there are defects on the surface.
  • the above defect detection neural network model can extract more features from images, thereby improving the efficiency and accuracy of defect detection.
  • determining whether there is a defect on the cell surface based on the defect confidence of each of the at least one to-be-detected image includes: determining whether the defect confidence of each of the at least one to-be-detected image is less than equal to the defect threshold; and in response to determining that the defect confidence of one or more of the images to be detected in the at least one image to be detected is greater than the defect threshold, determining that a defect exists on the surface of the cell.
  • the above method determines whether there are defects on the surface of the battery core based on the defect confidence output by the defect detection neural network model and the preset defect threshold, thereby improving the efficiency and accuracy of defect detection.
  • the defect is a crush defect.
  • An embodiment of the second aspect of the present application provides a method for training a defect detection neural network model.
  • the method includes: obtaining a sample image including the surface of the battery core, the sample image including preset defects on the surface of the battery core; preprocessing the sample image to obtain at least one sample to be detected image of the battery core surface; converting the at least one sample to be detected image Input the defect detection neural network model and obtain the detection results output by the defect detection neural network model.
  • the detection results are used to indicate whether there are defects on the cell surface; and based on the detection results and preset defects, adjust the parameters of the defect detection neural network model.
  • the above-mentioned embodiments of the present application can highlight the defect characteristics in the image to promote subsequent detection of the defect detection neural network model; on the other hand, they can improve the efficiency and accuracy of the defect detection neural network model.
  • At least one image of the sample to be detected includes a plurality of images of the sample to be detected, the sample image includes a cell surface area where the cell surface is located, and the sample image is preprocessed to obtain at least one of the cell surface
  • the image of the sample to be detected includes: extracting the surface area of the battery core from the sample image; and segmenting the surface area of the battery core to obtain multiple images of the sample to be detected. Since some defects on the surface of the battery core are relatively small, for example, they only account for a small part of the entire battery core surface. By extracting the surface area of the battery core and segmenting it to obtain small-sized images to be detected, the characteristics of these defects can be It is more prominent in the detection image, thereby improving the training efficiency and accuracy of the defect detection neural network model.
  • the sample image has a first exposure
  • extracting the cell surface area from the sample image includes: directly extracting the cell surface area from the sample image; and adjusting the exposure of the cell surface area to obtain A cell surface area having a second degree of exposure, wherein the first degree of exposure is higher than the second degree of exposure.
  • the above-mentioned method uses an initial image with a higher exposure to extract the cell surface area, and then adjusts the exposure of the cell surface area. This can make certain defect features more prominent in low-exposure images, thereby improving defect detection nerves. The training efficiency and accuracy of the network model.
  • extracting the cell surface area from the sample image further includes: extracting the cell surface area of the preset color channel from the cell surface area with the second exposure, and segmenting the cell surface area.
  • obtaining multiple samples to be detected images includes: segmenting the cell surface area of the preset color channel to obtain multiple samples to be detected images. Since the characteristics of certain defects on the surface of some cells are more prominent in the image of the preset color channel, the characteristics of these defects can be highlighted by extracting the cell surface area of the preset color channel, thereby facilitating the subsequent development of the defect detection neural network model. Detection to improve the training efficiency and accuracy of the defect detection neural network model.
  • the embodiment of the third aspect of the present application provides a device for detecting crush defects on the surface of a battery core, including: an acquisition unit configured to use an image acquisition unit to acquire an initial image including the surface of the battery core; a preprocessing unit, The processing unit is configured to preprocess the initial image to obtain at least one to-be-detected image of the cell surface; and the detection unit is configured to input the at least one to-be-detected image into a defect detection neural network model and obtain the defect detection The detection results output by the neural network model are used to indicate whether there are defects on the surface of the battery core.
  • This embodiment can achieve the same technical effect as the corresponding method mentioned above.
  • the embodiment of the fourth aspect of the present application provides a device for training a defect detection neural network model, including: an image acquisition unit, the image acquisition unit is configured to acquire a sample image including a battery core surface, and the sample image includes a preset image of the battery core surface.
  • an image processing unit the preprocessing module is configured to preprocess the sample image to obtain at least one sample to be detected image of the cell surface
  • a defect detection unit the detection unit is configured to input at least one sample to be detected image into the defect Detect the neural network model and obtain the detection output by the defect detection neural network model
  • the detection result is used to indicate whether there is a defect on the surface of the battery core
  • the adjustment unit is configured to adjust the parameters of the defect detection neural network model based on the detection result and the preset defect.
  • An embodiment of the fifth aspect of the present application provides an electronic device, including at least one processor and a memory communicatively connected to the at least one processor, wherein the memory stores instructions that can be executed by at least one processor, and the instructions are processed by at least one processor.
  • the processor is executed, so that at least one processor can execute the method for detecting cell surface defects according to the present application and/or the method for training a defect detection neural network model according to the present application.
  • An embodiment of the sixth aspect of the present application provides a computer-readable storage medium storing computer instructions, wherein the computer instructions are configured to cause the computer to execute the method for detecting cell surface defects according to the present application and/or the training according to the present application. Defect detection neural network model approach.
  • An embodiment of the seventh aspect of the present application provides a computer program product, including a computer program, wherein when executed by a processor, the computer program implements the method for detecting cell surface defects according to the present application and/or the training defects according to the present application. Methods for detecting neural network models.
  • Figure 1 is a schematic structural diagram of a vehicle according to some embodiments of the present application.
  • Figure 2 is a schematic diagram of the exploded structure of a battery according to some embodiments of the present application.
  • Figure 3 is a schematic diagram of the exploded structure of a battery cell according to some embodiments of the present application.
  • Figure 4 is a schematic flowchart of a method for detecting cell surface defects according to some embodiments of the present application.
  • Figure 5 is a schematic flowchart of preprocessing initial images according to some embodiments of the present application.
  • Figure 6 is a schematic diagram of the detection results for each image to be detected in multiple images to be detected according to some embodiments of the present application.
  • Figure 7 is a schematic flowchart of a method for training a defect detection neural network model according to some embodiments of the present application.
  • Figure 8 is a structural block diagram of a method for detecting cell surface defects according to some embodiments of the present application.
  • Figure 9 is a structural block diagram of a method for training a defect detection neural network model according to some embodiments of the present application.
  • Figure 10 is a schematic flowchart of a method for detecting surface defects of a battery core according to other embodiments of the present application.
  • an embodiment means that a particular feature, structure or characteristic described in connection with the embodiment can be included in at least one embodiment of the present application.
  • the appearances of this phrase in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments. Those skilled in the art understand, both explicitly and implicitly, that the embodiments described herein may be combined with other embodiments.
  • multiple refers to more than two (including two).
  • multiple groups refers to two or more groups (including two groups), and “multiple pieces” refers to It is more than two pieces (including two pieces).
  • Power batteries are not only used in energy storage power systems such as hydropower, thermal power, wind power and solar power stations, but are also widely used in electric vehicles such as electric bicycles, electric motorcycles and electric cars, as well as in many fields such as military equipment and aerospace. . As the application fields of power batteries continue to expand, their market demand is also constantly expanding.
  • batteries may produce various defects in various process stages.
  • the battery core needs to be hot-pressed and shaped.
  • crushing defects may appear on the surface of the battery core, which will not only affect the appearance of the battery core, but also lead to quality problems of the battery core, such as poor self-discharge performance of the battery core.
  • multiple light sources and multiple cameras are usually used to collect images of the surface of the product to be tested under light sources in a certain wavelength range based on the spectral reflection characteristics of the product to be tested. , and then extract the surface area from the collected image and input it into a traditional machine learning model (for example, support vector machine SVM) to obtain the detection results output by the machine learning model, or by collecting light sources at different angles
  • a traditional machine learning model for example, support vector machine SVM
  • the features extracted from images by traditional machine learning technology are limited and the accuracy of detection is very dependent on the imaging quality of the image acquisition unit.
  • the light source used when collecting images and the stability of the lens will all have an impact on test results.
  • the method of image fusion will affect the prominence of certain defects (for example, crush defects), thereby affecting the degree of identifiable defect features, and will also increase the time and time for image acquisition. Reduce the speed of the algorithm. It can be seen that the detection efficiency and detection accuracy of defect detection methods in related technologies are low, which can easily lead to over-inspection or missed inspection of products, which can easily have an adverse impact on the industry. How to improve the production yield of battery cells has become an urgent technical problem to be solved in this field.
  • the applicant has conducted in-depth research and provided a method and device for detecting cell surface defects, a method and device for training a defect detection neural network model, electronic equipment, computer-readable storage media, and computer program products , can quickly and accurately detect defects on the surface of the battery core, thereby improving the production yield of the battery core.
  • the embodiments of the present application are applied in various production processes of the battery core, for example, after the hot pressing and shaping stage of the battery core.
  • the embodiments of this application utilize computer vision technology and deep learning technology.
  • an image acquisition unit is used to obtain an initial image including the surface of the battery core; then, the initial image is preprocessed to obtain at least one image to be detected of the surface of the battery core; Finally, at least one image to be detected is input into the defect detection neural network model, and the detection result output by the defect detection neural network model is obtained.
  • the embodiments of the present application can highlight the defect features in the image to facilitate the subsequent detection of the defect detection neural network model. On the other hand, more features can be extracted from the image, thereby improving the efficiency of defect detection. and accuracy.
  • the embodiments of the present application are applied in various production processes of cell components of various power batteries or energy storage batteries, for example, after the hot pressing and shaping process.
  • an electric device 1000 according to an embodiment of the present application is used as an example.
  • FIG. 1 is a schematic structural diagram of a vehicle 1000 provided by some embodiments of the present application.
  • the vehicle 1000 may be a fuel vehicle, a gas vehicle or a new energy vehicle, and the new energy vehicle may be a pure electric vehicle, a hybrid vehicle or an extended-range vehicle, etc.
  • the battery 100 is disposed inside the vehicle 1000 , and the battery 100 may be disposed at the bottom, head, or tail of the vehicle 1000 .
  • the battery 100 may be used to power the vehicle 1000 , for example, the battery 100 may serve as an operating power source for the vehicle 1000 .
  • the vehicle 1000 may also include a controller 200 and a motor 300 .
  • the controller 200 is used to control the battery 100 to provide power to the motor 300 , for example, for starting, navigating, and driving the vehicle 1000 to meet operating power requirements.
  • the battery 100 can not only be used as an operating power source for the vehicle 1000 , but also can be used as a driving power source for the vehicle 1000 , replacing or partially replacing fuel or natural gas to provide driving power for the vehicle 1000 .
  • FIG. 2 is an exploded view of the battery 100 provided by some embodiments of the present application.
  • the battery 100 includes a case 10 and battery cells 20 , and the battery cells 20 are accommodated in the case 10 .
  • the box 10 is used to provide an accommodation space for the battery cells 20 .
  • the box 10 may include a first part 11 and a second part 12, the first part 11 and the second part 12 are mutually covered, and the first part 11 and the second part 12 jointly define a space for accommodating battery cells.
  • the battery cell 20 may be a secondary battery or a primary battery; it may also be a lithium-sulfur battery, a sodium-ion battery or a magnesium-ion battery, but is not limited thereto.
  • the battery cell 20 may be in the shape of a cylinder, a flat body, a rectangular parallelepiped or other shapes.
  • FIG. 3 is an exploded structural diagram of a battery cell 20 provided in some embodiments of the present application.
  • the battery cell 20 refers to the smallest unit that constitutes the battery.
  • the battery cell 20 includes an end cover 21 , a casing 22 , a battery cell assembly 23 (hereinafter also referred to as a battery cell), and other functional components.
  • the end cap 21 refers to a component that covers the opening of the case 22 to isolate the internal environment of the battery cell 20 from the external environment.
  • the shape of the end cap 21 can be adapted to the shape of the housing 22 to fit the housing 22 .
  • the housing 22 is a component used to cooperate with the end cover 21 to form an internal environment of the battery cell 20 , wherein the formed internal environment can be used to accommodate the battery core assembly 23 , electrolyte and other components.
  • the battery cell assembly 23 is a component in the battery cell 100 where electrochemical reactions occur.
  • One or more battery core assemblies 23 may be contained within the housing 22 .
  • the cell assembly 23 is mainly formed by winding or stacking positive electrode sheets and negative electrode sheets, and a separator is usually provided between the positive electrode sheets and the negative electrode sheets.
  • Figure 4 is a schematic flowchart of a method 400 for detecting cell surface defects in some embodiments of the present application
  • Figure 5 is a schematic flowchart of preprocessing an initial image in some embodiments of the present application
  • Figure 6 is a schematic flowchart of a method for preprocessing an initial image in some embodiments of the present application.
  • a schematic diagram of the detection results of each image to be detected in multiple images to be detected. The method 400 is described in detail below with reference to FIGS. 4 to 6 .
  • a method 400 for detecting surface defects of a battery core may include the following steps S401 to S403.
  • step S401 an initial image 500 including the cell surface is acquired using an image acquisition unit (not shown).
  • step S402 the initial image 500 is preprocessed to obtain at least one to-be-detected image 512 of the cell surface.
  • step S403 at least one to-be-detected image 512 is input into the defect detection neural network model, and the detection result output by the defect detection neural network model is obtained.
  • the detection result is used to indicate whether there is a defect 511 on the cell surface.
  • the specific type of image acquisition unit is not limited.
  • the image acquisition unit can use a CCD industrial camera, or a CMOS industrial camera, etc.
  • the image acquisition unit may use a line array camera or an area array camera.
  • the acquired initial image 500 includes at least the cell surface area 510 .
  • the initial image 500 may also include a clamp area 530 for fixing the battery core, a background area 520 other than the battery core surface area 510 and the clamp area 530, and the like.
  • the defects 511 on the cell surface may be, for example, crush defects, scratch defects, bump defects, etc., and the application is not limited thereto.
  • the at least one to-be-detected image 512 acquired after preprocessing in step S402 may include one or more to-be-detected images.
  • preprocessing operations such as exposure adjustment and color channel extraction can only be performed on the initial image 500 without segmenting it.
  • the proportion of defects for example, crush defects
  • the initial image 500 can be segmented to obtain multiple images p1 to be detected. ,...,pn.
  • the characteristics of defects in the image can include gray value, color, shape, contour, texture, etc.
  • the defect detection neural network model can be a deep learning model, such as ResNet18, DNN, etc. Deep learning models can extract multiple features from images, such as gray value, color, shape, contour, texture, etc., thereby improving the accuracy of defect detection.
  • the detection results output by the defect detection neural network model may include the defect confidence corresponding to each image to be detected.
  • the server may include one or more general purpose computers, dedicated server computers (such as PC (Personal Computer) servers, UNIX servers, midrange servers), blade servers, mainframe computers, server clusters or any other suitable arrangement and/or combination.
  • the server may be a distributed system server, a server combined with a blockchain, a cloud server, an intelligent cloud computing server with artificial intelligence technology, or an intelligent cloud host, etc.
  • Client devices may include various types of computer devices, such as portable handheld devices, general purpose computers (such as personal computers and laptop computers), workstation computers, and the like.
  • method 400 may also be executed at an industrial computer equipped with chips in the battery cell production line.
  • the chip can be a system on chip (System on Chip, SoC), etc.
  • the method 400 of the embodiment of the present application preprocesses the initial image 500 and uses a defect detection neural network model to detect the defects 511 in the preprocessed initial image 500.
  • the defect characteristics in the image can be highlighted to facilitate subsequent defects.
  • Detection using neural network models can extract more features from images, thereby improving the efficiency and accuracy of defect detection.
  • At least one image to be detected 512 includes a plurality of images to be detected p1,..., pn, and the initial image 500 includes the cell surface area 510 where the cell surface is located.
  • preprocessing the initial image 500 to obtain at least one to-be-detected image 512 of the battery core surface may include: extracting the battery core surface area 510 from the initial image 500; and segmenting the battery core surface area 510 to obtain Acquire multiple images to be detected 512.
  • the battery core surface area 510 is first extracted from the initial image 500, and then the extracted battery core surface area 510 is segmented to obtain multiple images p1,...,pn to be detected.
  • the sizes of the multiple segmented images to be detected may be the same (as shown in Figure 5), or they may be different from each other, or part of them may be the same size and another part may be of different sizes.
  • the initial image 500 has a first exposure. Extracting the cell surface area 510 from the initial image 500 may include: directly extracting the cell surface area 510 from the initial image 500; and adjusting the exposure of the cell surface area 510 to obtain a cell surface with a second exposure. Region 510a, where the first exposure is higher than the second exposure.
  • the first exposure may be a high exposure.
  • the image with the first exposure can be obtained by one or more methods of enlarging the aperture of the industrial camera, extending the exposure time of the industrial camera, increasing the brightness in the environment, and the like. Since there is usually blue glue on the surface of the cell, an image with high exposure can significantly distinguish the cell surface area 510 from other areas (e.g., the clamp area 530, the background area 520) in the initial image 500 (e.g., both The range of pixel values varies greatly).
  • the cell surface area 510 with the first exposure is adjusted to the cell surface area 510a with the second exposure, so as to highlight the characteristics of the defect 511 in the cell surface area 510a, as shown in FIG. 5 .
  • This can improve the accuracy of ROI extraction in the region of interest and the accuracy of subsequent defect detection.
  • the first exposure is configured such that the pixel value range of the cell surface area 510 in the initial image 500 is different from the pixel value ranges of other areas in the initial image 500 except the cell surface area 510 .
  • other areas in the initial image 500 except the cell surface area 510 may include, for example, a clamp area 530 and a background area 520 .
  • the exposure of the acquired initial image 500 can be adjusted by at least one of enlarging the aperture of the industrial camera, extending the exposure time of the industrial camera, increasing the brightness in the environment, etc., so that the cell surface area 510 is consistent with the initial image 500
  • the pixel value ranges of other areas 520 and 530 are quite different.
  • the above embodiment makes it easy to distinguish the cell surface area 510 from other areas 520 and 530 in the initial image 500 with the first exposure, thereby facilitating the extraction of the cell surface area 510 from the initial image 500 with the first exposure.
  • the color channels may be R channel (red channel), G channel (green channel), and B channel (blue channel). Since the characteristics of certain defects 511 (for example, crush defects) are more prominent in images of specific color channels, the characteristics of these defects can be highlighted by extracting the cell surface area of the preset color channel, as shown in Figure 5 The color channel is shown in cell surface area 510b.
  • R channel red channel
  • G channel green channel
  • B channel blue channel
  • the above-mentioned preset color channel is the G channel.
  • directly extracting the cell surface area 510 from the initial image 500 may include: extracting the cell surface area 510 from the initial image 500 based at least on a pixel value range of the cell surface area.
  • the initial image 500 may be an RGB image, and the application is not limited thereto.
  • Pixel values can be the brightness values of individual color channels of a color image.
  • the pixel value of each pixel in an RGB image can include a brightness value for each color channel, that is, red (R), green (G), or blue (B), generally ranging from 0 to 255, with the brightest being 255.
  • the darkest value is 0.
  • the pixel value range of the cell surface area 510 can be determined based on experience or battery production process specification requirements. For example, the pixel value range can be obtained based on the pixel value range of the cell surface imaged under the same lighting conditions.
  • the above technical solution of extracting the battery core surface area based on the pixel values of each area on the initial image can quickly and effectively extract the battery core surface area 510.
  • the sizes of the multiple segmented images p1,...,pn to be detected are equal.
  • the above method can easily meet the input image size requirements of the defect detection neural network model and promote the detection of the defect detection neural network model.
  • step S403, inputting at least one image to be detected 512 into the defect detection neural network model, and obtaining the detection result output by the defect detection neural network model may include: inputting at least one image to be detected 512 into the defect detection neural network model.
  • the network model is used to obtain the defect confidence of each image to be detected in at least one image to be detected; and based on the defect confidence of each image to be detected in the at least one image to be detected, determine whether there is a defect on the surface of the cell.
  • the defect confidence level may be a probability value used to evaluate the possibility of defects in the image to be detected.
  • Figure 6 shows the detection results t1,...tn output by the neural network model for corresponding images to be detected among multiple images to be detected, in which the crush defect 611 of the image to be detected corresponding to the detection result tn is the most significant. It can be seen that the defect confidence (i.e., crush score) of the detection result tn output by the neural network model is the highest, that is, 0.996668, which indicates that the image to be detected corresponding to the detection result tn has the highest probability of having a crush defect.
  • the above defect detection neural network model can extract more features of the image, thereby improving the efficiency and accuracy of defect detection.
  • determining whether there is a defect on the cell surface based on the defect confidence of each of the at least one to-be-detected images 512 includes: determining the defect confidence of each of the at least one to-be-detected images. whether the degree is less than or equal to the defect threshold; and in response to determining that the defect confidence of one or more images to be detected in the at least one to-be-detected image 512 is greater than the defect threshold, it is determined that there is a defect on the surface of the cell.
  • the defect threshold can be determined based on experience or battery production process specification requirements.
  • the above-mentioned method of determining whether there are defects on the surface of the battery core based on the defect confidence output by the defect detection neural network model and the preset defect threshold can improve the efficiency and accuracy of defect detection.
  • the method 400 may further include: in response to determining that the defect confidence of each to-be-detected image in the at least one to-be-detected image 512 is less than or equal to the defect threshold, determining that there is no defect on the surface of the cell and issuing a message to continue the cell. Instructions to flow to the next process.
  • the above-mentioned defect 511 is a crush defect.
  • the defects 511 on the cell surface may also be scratch defects, bump defects, etc., for example.
  • a method 700 for training a defect detection neural network model may include the following steps S701 to S704.
  • step S701 a sample image including the surface of the battery core is acquired, and the sample image includes preset defects on the surface of the battery core.
  • step S702 the sample image is preprocessed to obtain at least one image of the sample to be detected on the surface of the battery core.
  • step S703 at least one sample image to be detected is input into the defect detection neural network model, and the detection results output by the defect detection neural network model are obtained.
  • the detection results are used to indicate whether there are defects on the surface of the battery core.
  • step S704 the parameters of the defect detection neural network model are adjusted based on the detection results and preset defects.
  • the above sample image acquisition method is the same as the initial image acquisition method in method 400, and will not be described in detail here.
  • the characteristics of the above-mentioned preset defects, the sample image to be detected, the defect detection neural network model, etc. are the same as the characteristics of the defects, the image to be detected, the defect detection neural network model, etc. in the method 400, and will not be described in detail here.
  • the parameters of the defect detection neural network model can be continuously adjusted through the gradient descent method based on the detection results and preset defects, so as to minimize the loss function of the defect detection neural network model.
  • Servers include one or more general purpose computers, dedicated server computers (e.g., PC (Personal Computer) servers, UNIX servers, midrange servers), blade servers, mainframe computers, server clusters, or any other suitable arrangement and/or combination .
  • the server may be a distributed system server, a server combined with a blockchain, a cloud server, an intelligent cloud computing server with artificial intelligence technology, or an intelligent cloud host, etc.
  • Client devices may include various types of computer devices, such as portable handheld devices, general purpose computers (such as personal computers and laptop computers), workstation computers, and the like.
  • the method 700 of the embodiment of the present application can, on the one hand, highlight the defect features in the image to promote subsequent detection of the defect detection neural network model, and on the other hand, it can improve the efficiency and accuracy of the defect detection neural network model.
  • At least one sample image to be detected includes a plurality of sample images to be detected, the sample image includes a cell surface area where the cell surface is located, and step S702 is to preprocess the sample image to obtain the cell surface.
  • the at least one sample to-be-detected image of the core surface includes: extracting the battery core surface area from the sample image; and segmenting the battery core surface area to obtain multiple sample-to-be-detected images.
  • the sizes of the multiple segmented sample images to be detected may be the same or different from each other, or some parts may be the same size and the other parts may be different in size.
  • some defects on the cell surface are relatively small, for example, they only account for a small part of the entire cell surface, by extracting the cell surface area and segmenting it to obtain small-size images to be detected, the characteristics of these defects can be It is more prominent in the detection image, thereby improving the training efficiency and accuracy of the defect detection neural network model.
  • the sample image has a first exposure
  • extracting the battery core surface area from the sample image includes: directly extracting the battery core surface area from the sample image; and adjusting the exposure of the battery core surface area, To obtain a cell surface area with a second exposure degree, wherein the first exposure degree is higher than the second exposure degree.
  • the characteristics of the above first exposure are the same as the characteristics of the first exposure in method 400 and will not be described in detail here. This can improve the training efficiency and accuracy of the defect detection neural network model.
  • extracting the cell surface area from the sample image further includes: extracting the cell surface area of the preset color channel from the cell surface area with the second exposure, and comparing the cell surface area to divide, Obtaining multiple samples to be detected images includes: segmenting the cell surface area of a preset color channel to obtain multiple samples to be detected images.
  • the characteristics of the above-mentioned preset color channel are the same as the characteristics of the preset color channel in method 400, and will not be described in detail here.
  • directly extracting the battery core surface area from the sample image may include: extracting the battery core surface area from the initial image based on at least a pixel value range of the battery core surface area.
  • the above characteristics of the pixel value and the pixel value range are the same as the characteristics of the pixel value and the pixel value range in the method 400 respectively, and will not be described in detail here.
  • the embodiment of the present application also provides a device 800 for detecting crush defects on the surface of a battery core, including: an acquisition unit 801 , a preprocessing unit 802 and a detection unit 803 .
  • the acquisition unit 801 is configured to acquire an initial image including the cell surface using the image acquisition unit.
  • the preprocessing unit 802 is configured to preprocess the initial image to obtain at least one to-be-detected image of the cell surface.
  • the detection unit 803 is configured to input at least one image to be detected into the defect detection neural network model, and obtain the detection result output by the defect detection neural network model. The detection result is used to indicate whether there is a defect on the surface of the battery core.
  • each module of the device 800 shown in FIG. 8 may correspond to each step in the method 400 described with reference to FIG. 4 . Accordingly, the operations, features and advantages described above with respect to method 400 apply equally to apparatus, 800 and the modules it includes. For the sake of brevity, certain operations, features, and advantages are not described again here.
  • the above-mentioned virtual device in the embodiment of the present application can, on the one hand, highlight the defect features in the image to facilitate subsequent detection of the defect detection neural network model, and on the other hand, it can extract more features from the image, thereby improving the efficiency and effectiveness of defect detection. Accuracy.
  • this embodiment of the present application also provides a device 900 for training a defect detection neural network model, including: an image acquisition unit 901 , an image processing unit 902 , a defect detection unit 903 and an adjustment unit 904 .
  • the image acquisition unit 901 is configured to acquire a sample image including a cell surface, where the sample image includes preset defects on the cell surface.
  • the preprocessing module 902 is configured to preprocess the sample image to obtain at least one sample to be detected image of the cell surface.
  • the defect detection unit 903 is configured to input at least one sample image to be detected into the defect detection neural network model, and obtain the detection results output by the defect detection neural network model. The detection results are used to indicate whether there are defects on the surface of the battery core.
  • the adjustment unit 904 is configured to adjust the parameters of the defect detection neural network model based on the detection results and the preset defects.
  • each module of the device 900 shown in FIG. 9 may correspond to each step in the method 700 described with reference to FIG. 8 . Accordingly, the operations, features, and advantages described above with respect to method 700 apply equally to apparatus 900 and the modules it includes. For the sake of brevity, certain operations, features, and advantages are not described again here.
  • the above-mentioned virtual device in the embodiment of the present application can, on the one hand, highlight the defect characteristics in the image to promote subsequent detection of the defect detection neural network model, and on the other hand, can improve the efficiency and accuracy of the defect detection neural network model.
  • An embodiment of the present application further provides an electronic device, including at least one processor and a memory communicatively connected to the at least one processor, wherein the memory stores instructions that can be executed by at least one processor, and the instructions are executed by at least one processor, To enable at least one processor to execute the aforementioned method 400 and/or method 700.
  • Embodiments of the present application also provide a computer-readable storage medium storing computer instructions, wherein the computer instructions are configured to cause the computer to execute the aforementioned method 400 and/or method 700.
  • An embodiment of the present application also provides a computer program product, including a computer program, wherein the computer program implements the aforementioned method 400 and/or method 700 when executed by a processor.
  • a method 1000 for detecting surface defects of a battery core includes the following steps S1001 to S1011.
  • step S1001 an image acquisition unit is used to acquire an initial image including the surface of the battery core, and the initial image has a first exposure.
  • step S1002 the cell surface area is directly extracted from the initial image.
  • step S1003 the exposure of the battery core surface area is adjusted to obtain the battery core surface area with a second exposure, where the first exposure is higher than the second exposure.
  • step S1004 the cell surface area of the preset color channel is extracted from the cell surface area with the second exposure.
  • step S1005 the cell surface area of the preset color channel is segmented to obtain multiple images to be detected.
  • step S1006 multiple images to be detected are input into the defect detection neural network model to obtain the defect confidence of each image to be detected in at least one image to be detected.
  • step S1007 it is determined whether the defect confidence of each to-be-detected image in the plurality of to-be-detected images is less than or equal to the defect threshold.
  • step S1008 in response to determining that the defect confidence of any one of the plurality of images to be detected is greater than the defect threshold, it is determined that a defect exists on the surface of the battery core.
  • step S1009 in response to determining that the defect confidence of all the images to be detected among the multiple images to be detected is less than or equal to the defect threshold, it is determined that there are no defects on the surface of the battery core and an instruction is issued to continue flowing the battery core to the next process.
  • the method 1000 of the embodiment of the present application can, on the one hand, highlight the features of prominent defects in the image to facilitate subsequent detection of the defect detection neural network model; on the other hand, it can extract more features of the image, thereby facilitating more efficient identification of defects. To improve the efficiency and accuracy of defect detection.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Computational Linguistics (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Health & Medical Sciences (AREA)
  • Data Mining & Analysis (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Image Analysis (AREA)

Abstract

一种检测电芯表面缺陷的方法和装置。方法包括利用图像采集单元获取包括电芯表面的初始图像（S401）；对初始图像进行预处理，以获得电芯表面的至少一个待检测图像（S402）；以及将至少一个待检测图像输入缺陷检测神经网络模型，并获取缺陷检测神经网络模型所输出的检测结果，检测结果用于指示电芯表面是否存在缺陷（S403）。一方面可以突出图像中缺陷的特征，以便于后续缺陷检测神经网络模型的检测，另一方面可以从图像中提取出的更多特征，从而提高缺陷检测的效率和精度。

Description

检测电芯表面缺陷的方法和装置
交叉引用
本申请引用于2022年7月29日递交的名称为“检测电芯表面缺陷的方法和装置”的第202210907837.9号中国专利申请，其通过引用被全部并入本申请。
技术领域
本申请涉及电池技术领域和机器视觉检测技术领域，尤其涉及一种检测电芯表面缺陷的方法和装置、训练缺陷检测神经网络模型的方法和装置、电子设备、计算机可读存储介质以及计算机程序产品。
背景技术
节能减排是汽车产业可持续发展的关键，电动车辆由于其节能环保的优势成为汽车产业可持续发展的重要组成部分。对于电动车辆而言，电池技术又是关乎其发展的一项重要因素。
在相关技术中，充电电池(指在电池放电后可通过充电的方式使活性物质激活而继续使用的电池，又称二次电池)包括电池箱以及位于电池箱内的通过串联和/或并联方式组合的多个电池单体。电池单体是电池中提供能量来源的最小单元。电芯是电池单体中发生电化学反应的关键部件，其主要结构包括阳极极片、阴极极片、以及将阳极极片和阴极极片间隔的隔膜。
如何提高电芯的生产良品率，是本领域亟待解决的技术问题。
发明内容
本申请旨在至少解决现有技术中存在的技术问题之一。为此，本申请的一个目的在于提出一种检测电芯表面缺陷的方法和装置、训练缺陷检测神经网络模型的方法和装置、电子设备、计算机可读存储介质以及计算机程序产品，以提高电芯的生产良品率。
本申请第一方面的实施例提供一种检测电芯表面缺陷的方法。方法包括：利用图像采集单元获取包括电芯表面的初始图像；对初始图像进行预处理，以获得电芯表面的至少一个待检测图像；以及将至少一个待检测图像输入缺陷检测神经网络模型，并获取缺陷检测神经网络模型所输出的检测结果，检测结果用于指示电芯表面是否存在缺陷。
本申请实施例的技术方案中，通过对初始图像进行预处理并利用缺陷检测神经网络模型检测预处理后的初始图像中的缺陷，一方面可以突出图像中的缺陷特征，以便于后续缺陷检测神经网络模型的检测，另一方面可以从图像中提取出更多特征，从而提高缺陷检测的效率和精度。
在一些实施例中，至少一个待检测图像包括多个待检测图像，初始图像包括电芯表面所处的电芯表面区，并且对初始图像进行预处理，以获得电芯表面的至少一个待检测图像包括：从初始图像中提取出电芯表面区；以及对电芯表面区进行分割，以获取多个待检测图像。由于电芯表面的某些缺陷比较小，例如仅占整个电芯表面的一小部分，通过提取电芯表面区并对其进行分割以获得小尺寸的待检测图像，使得这些缺陷的特征在待检测图像中更加突出，从而提高缺陷检测的准确度。
在一些实施例中，初始图像具有第一曝光度，并且从初始图像中提取出电芯表面区包括：从初始图像中直接提取电芯表面区；以及调整电芯表面区的曝光度，以获得具有第二曝光度的电芯表面区，其中，第一曝光度高于第二曝光度。上述利用曝光度较高的初始图像提取电芯表面区，然后再调整电芯表面区的曝光度，由此可以使得某些缺陷的特征在低曝光度的图像中更加突出，从而可以提高电芯表面区提取的准确性和后续缺陷检测的准确性。
在一些实施例中，第一曝光度被配置为使得初始图像中的电芯表面区的像素值范围不同于初始图像中除电芯表面区外的其他区域的像素值范围。上述实施方式使得在第一曝光度的初始图像中电芯表面区易与其他区域区分开，从而便于在具有第一曝光度的初始图像中提取出电芯表面区。
在一些实施例中，从初始图像中提取出电芯表面区还包括：从具有第二曝光度的电芯表面区提取出预设颜色通道的电芯表面区，并且对电芯表面区进行分割，以获取多个待检测图像包括：对预设颜色通道的电芯表面区进行分割，以获取多个待检测图像。由于电芯表面的某些缺陷的特征在预设颜色通道的图像中更加突出，因此通过提取预设颜色通道的电芯表面区可以突出这些缺陷的特征，从而便于后续缺陷检测神经网络模型的检测，以提高缺陷检测的准确性。
在一些实施例中，预设颜色通道为G通道。由于电芯表面通常存在蓝胶，通过提取G通道的电芯表面区可以避免提取出蓝胶，从而降低蓝胶对于缺陷特征提取的影响。由此便于后续缺陷检测神经网络模型对电芯表面区中的缺陷进行检测，以进一步提高缺陷检测的准确性。
在一些实施例中，从初始图像中直接提取电芯表面区包括：至少基于电芯表面区的像素值范围，从初始图像中提取电芯表面区。由于初始图像上的各个区的像素值存在一定差异，上述基于初始图像的各个区上的像素值提取电芯表面区的技术方案，可以快速且有效地提取出电芯表面区。
在一些实施例中，多个待检测图像的大小相等。上述通过将电芯表面区分割成大小相同的多个待检测图像，从而便于满足缺陷检测神经网络模型的输入图像尺寸要求，并且促进缺陷检测神经网络模型的检测。
在一些实施例中，检测结果包括每一个待检测图像相应的缺陷置信度，并且将至少一个待检测图像输入缺陷检测神经网络模型，并获取缺陷检测神经网络模型所输出的检测结果包括：将至少一个待检测图像输入缺陷检测神经网络模型，以获得至少一个待检测图像中每一待检测图像的缺陷置信度；以及基于至少一个待检测图像中每一待检测图像的缺陷置信度，确定电芯表面是否存在缺陷。上述缺陷检测神经网络模型可以从图像中提取出更多特征，从而提高缺陷检测的效率和精度。
在一些实施例中，基于至少一个待检测图像中每一待检测图像的缺陷置信度，确定电芯表面是否存在缺陷包括：确定至少一个待检测图像中每一待检测图像的缺陷置信度是否小于等于缺陷阈值；以及响应于确定至少一个待检测图像中一个或多个待检测图像的缺陷置信度大于缺陷阈值，确定电芯表面存在缺陷。上述基于缺陷检测神经网络模型所输出的缺陷置信度和预设的缺陷阈值判断电芯表面是否存在缺陷，从而可以提高缺陷检测的效率和精度。
在一些实施例中，缺陷为压伤缺陷。通过利用上述方法检测压伤缺陷，从而可以提高压伤缺陷检测的效率和精度。
本申请第二方面的实施例提供一种训练缺陷检测神经网络模型的方法。方法包括：获取包括电芯表面的样本图像，样本图像包括电芯表面的预设缺陷；对样本图像进行预处理，以获得电芯表面的至少一个样本待检测图像；将至少一个样本待检测图像输入缺陷检测神经网络模型，并获取缺陷检测神经网络模型所输出的检测结果，检测结果用于指示电芯表面是否存在缺陷；以及基于检测结果和预设缺陷，调整缺陷检测神经网络模型的参数。
本申请上述实施例一方面可以突出图像中的缺陷特征，以促进后续缺陷检测神经网络模型的检测，另一方面可以提高缺陷检测神经网络模型的效率和精度。
在一些实施例中，至少一个样本待检测图像包括多个样本待检测图像，样本图像包括电芯表面所处的电芯表面区，并且对样本图像进行预处理，以获得电芯表面的至少一个样本待检测图像包括：从样本图像中提取出电芯表面区；以及对电芯表面区进行分割，以获取多个样本待检测图像。由于电芯表面的某些缺陷比较小，例如仅占整个电芯表面的一小部分，通过提取电芯表面区并对其进行分割以获得小尺寸的待检测图像，使得这些缺陷的特征在待检测图像中更加突出，从而提高缺陷检测神经网络模型的训练效率以及准确性。
在一些实施例中，样本图像具有第一曝光度，并且从样本图像中提取出电芯表面区包括：从样本图像中直接提取电芯表面区；以及调整电芯表面区的曝光度，以获得具有第二曝光度的电芯表面区，其中，第一曝光度高于第二曝光度。上述利用曝光度较高的初始图像提取电芯表面区，然后再调整电芯表面区的曝光度，由此可以使得某些缺陷的特征在低曝光度的图像中更加突出，从而提高缺陷检测神经网络模型的训练效率以及准确性。
在一些实施例中，从样本图像中提取出电芯表面区还包括：从具有第二曝光度的电芯表面区提取出预设颜色通道的电芯表面区，并且对电芯表面区进行分割，以获取多个样本待检测图像包括：对预设颜色通道的电芯表面区进行分割，以获取多个样本待检测图像。由于一些电芯表面的某些缺陷的特征在预设颜色通道的图像中更加突出，因此通过提取预设颜色通道的电芯表面区可以突出这些缺陷的特征，从而便于后续缺陷检测神经网络模型的检测，以提高缺陷检测神经网络模型的训练效率以及准确性。
本申请第三方面的实施例提供一种检测电芯表面的压伤缺陷的装置，包括：获取单元，获取单元被配置为利用图像采集单元获取包括电芯表面的初始图像；预处理单元，预处理单元被配置为对初始图像进行预处理，以获得电芯表面的至少一个待检测图像；以及检测单元，检测单元被配置为将至少一个待检测图像输入缺陷检测神经网络模型，并获取缺陷检测神经网络模型所输出的检测结果，检测结果用于指示电芯表面是否存在缺陷。该实施例方案可以获得与前述相应的方法相同的技术效果。
本申请第四方面的实施例提供一种训练缺陷检测神经网络模型的装置，包括：图像获取单元，图像获取单元被配置为获取包括电芯表面的样本图像，样本图像包括电芯表面的预设缺陷；图像处理单元，预处理模块被配置为对样本图像进行预处理，以获得电芯表面的至少一个样本待检测图像；缺陷检测单元，检测单元被配置为将至少一个样本待检测图像输入缺陷检测神经网络模型，并获取缺陷检测神经网络模型所输出的检测 结果，检测结果用于指示电芯表面是否存在缺陷；以及调整单元，调整单元被配置为基于检测结果和预设缺陷，调整缺陷检测神经网络模型的参数。该实施例方案可以获得与前述相应的方法相同的技术效果。
本申请第五方面的实施例提供一种电子设备，包括至少一个处理器以及与至少一个处理器通信连接的存储器，其中，存储器存储有能够被至少一个处理器执行的指令，指令被至少一个处理器执行，以使至少一个处理器能够执行根据本申请的检测电芯表面缺陷的方法和/或根据本申请的训练缺陷检测神经网络模型的方法。
本申请第六方面的实施例提供一种存储有计算机指令的计算机可读存储介质，其中，计算机指令配置为使计算机执行根据本申请的检测电芯表面缺陷的方法和/或根据本申请的训练缺陷检测神经网络模型的方法。
本申请第七方面的实施例提供一种计算机程序产品，包括计算机程序，其中，计算机程序在被处理器执行时实现根据本申请的检测电芯表面缺陷的方法和/或根据本申请的训练缺陷检测神经网络模型的方法。
上述说明仅是本申请技术方案的概述，为了能够更清楚了解本申请的技术手段，而可依照说明书的内容予以实施，并且为了让本申请的上述和其它目的、特征和优点能够更明显易懂，以下特举本申请的具体实施方式。
附图说明
在附图中，除非另外规定，否则贯穿多个附图相同的附图标记表示相同或相似的部件或元素。这些附图不一定是按照比例绘制的。应该理解，这些附图仅描绘了根据本申请公开的一些实施方式，而不应将其视为是对本申请范围的限制。
图1为本申请一些实施例的车辆的结构示意图；
图2为本申请一些实施例的电池的分解结构示意图；
图3为本申请一些实施例的电池单体的分解结构示意图；
图4为本申请一些实施例的检测电芯表面缺陷的方法的流程示意图；
图5为本申请一些实施例的对初始图像进行预处理的流程示意图；
图6为本申请一些实施例的针对多个待检测图像中每个待检测图像的检测结果的示意图；
图7为本申请一些实施例的训练缺陷检测神经网络模型的方法的流程示意图；
图8为本申请一些实施例的检测电芯表面缺陷的方法的结构框图；
图9为本申请一些实施例的训练缺陷检测神经网络模型的方法的结构框图；以及
图10为本申请另一些实施例的检测电芯表面缺陷的方法的流程示意图。
附图标记说明：
车辆1000；
电池100，控制器200，马达300；
箱体10，第一部分11，第二部分12；
电池单体20，端盖21，壳体22，电芯组件23；
初始图像500，电芯表面区510，缺陷511，夹具区530，背景区520，具有第二曝光度的电芯表面区510a，预设颜色通道的电芯表面区510b，待检测图像512、p1…pn，检测结果t1…tn，压伤缺陷611。
具体实施方式
下面将结合附图对本申请技术方案的实施例进行详细的描述。以下实施例仅用于更加清楚地说明本申请的技术方案，因此只作为示例，而不能以此来限制本申请的保护范围。
除非另有定义，本文所使用的所有的技术和科学术语与属于本申请的技术领域的技术人员通常理解的含义相同；本文中所使用的术语只是为了描述具体的实施例的目的，不是旨在于限制本申请；本申请的说明书和权利要求书及上述附图说明中的术语“包括”和“具有”以及它们的任何变形，意图在于覆盖不排他的包含。
在本申请实施例的描述中，技术术语“第一”“第二”等仅用于区别不同对象，而不能理解为指示或暗示相对重要性或者隐含指明所指示的技术特征的数量、特定顺序或主次关系。在本申请实施例的描述中，“多个”的含义是两个以上，除非另有明确具体的限定。
在本文中提及“实施例”意味着，结合实施例描述的特定特征、结构或特性可以包含在本申请的至少一个实施例中。在说明书中的各个位置出现该短语并不一定均是指相同的实施例，也不是与其它实施例互斥的独立的或备选的实施例。本领域技术人员显式地和隐式地理解的是，本文所描述的实施例可以与其它实施例相结合。
在本申请实施例的描述中，术语“和/或”仅仅是一种描述关联对象的关联关系，表示可以存在三种关系，例如A和/或B，可以表示：单独存在A，同时存在A和B，单独存在B这三种情况。另外，本文中字符“/”，一般表示前后关联对象是一种“或”的关系。
在本申请实施例的描述中，术语“多个”指的是两个以上(包括两个)，同理，“多组”指的是两组以上(包括两组)，“多片”指的是两片以上(包括两片)。
在本申请实施例的描述中，技术术语“中心”“纵向”“横向”“长度”“宽度”“厚度”“上”“下”“前”“后”“左”“右”“竖直”“水平”“顶”“底”“内”“外”“顺时针”“逆时针”“轴向”“径向”“周向”等指示的方位或位置关系为基于附图所示的方位或位置关系，仅是为了便于描述本申请实施例和简化描述，而不是指示或暗示所指的装置或元件必须具有特定的方位、以特定的方位构造和操作，因此不能理解为对本申请实施例的限制。
在本申请实施例的描述中，除非另有明确的规定和限定，技术术语“安装”“相连”“连接”“固定”等术语应做广义理解，例如，可以是固定连接，也可以是可拆卸连接，或成一体；也可以是机械连接，也可以是电连接；可以是直接相连，也可以通过中间媒介间接相连，可以是两个元件内部的连通或两个元件的相互作用关系。对于本领域的普通技术人员而言，可以根据具体情况理解上述术语在本申请实施例中的具体含义。
目前，从市场形势的发展来看，动力电池的应用越加广泛。动力电池不仅被应用于水力、火力、风力和太阳能电站等储能电源系统，而且还被广泛应用于电动自行车、电动摩托车、电动汽车等电动交通工具，以及军事装备和航空航天等多个领域。随着动力电池应用领域的不断扩大，其市场的需求量也在不断地扩增。
然而，在电池的生产过程中，电池在各个工序段可能产生各种缺陷。例如，为了确保电芯中正极片、隔膜和负极片之间的平整度以使得电芯的厚度和高度满足特定要求，需要对电芯进行热压整形。此时，如果电芯表面受力不均匀，在电芯表面可能出现压伤缺陷，这不仅会影响电芯外观，而且会导致电芯的质量问题，例如电芯自放电性能差等。
相关技术中，在对产品表面的压伤缺陷进行检测时，通常通过多个光源以及多个相机，根据待测产品的光谱反射特性、在一定的波长范围的光源下采集待测产品表面的图像，然后从所采集的图像中提取出表面区域并将其输入到传统的机器学习模型(例如，支持向量机SVM)中，以获得机器学习模型所输出的检测结果，或者通过采集在不同角度光源的条件下拍摄的多个图像，并对多个图像进行融合，然后对融合后的图像进行缺陷识别。
本申请的申请人注意到，传统机器学习技术从图像中提取的特征有限并且检测的准确性非常依赖于图像采集单元的成像质量，例如，采集图像时所使用的光源以及镜头的稳定性均会对检测结果产生影响。此外，图像融合的方式将影响某些缺陷(例如，压伤缺陷)的突出性，从而影响缺陷特征的可识别程度，并且还会增加采集图像的时间和 降低算法的速度。由此可见，相关技术中的缺陷检测方法的检测效率和检测精度较低，容易导致产品的过检或漏检，从而容易给产业带来不利的影响。如何提高电芯的生产良品率，成为本领域亟待解决的技术问题。
基于发现的上述技术问题，申请人经过深入研究，提供了一种检测电芯表面缺陷的方法和装置、训练缺陷检测神经网络模型的方法和装置、电子设备、计算机可读存储介质以及计算机程序产品，可以快速而准确地检测出电芯表面的缺陷，从而可以提高电芯的生产良品率。
本申请实施例方案应用于电芯的各个生产工序中，例如实施在电芯的热压整形阶段后。本申请实施例方案利用计算机视觉技术和深度学习技术，首先，利用图像采集单元获取包括电芯表面的初始图像；然后，对初始图像进行预处理，以获得电芯表面的至少一个待检测图像；最后，将至少一个待检测图像输入缺陷检测神经网络模型，并获取缺陷检测神经网络模型所输出的检测结果。本申请实施例方案相比相关技术，一方面可以突出图像中的缺陷特征，以便于后续缺陷检测神经网络模型的检测，另一方面可以从图像中提取出更多特征，从而提高缺陷检测的效率和精度。
本申请实施例方案应用于各种动力电池或储能电池的电芯组件的各个生产工序中，例如，热压整形的工艺之后。
以下实施例为了方便说明，以本申请一实施例的一种用电装置为车辆1000为例进行说明。
请参照图1，图1为本申请一些实施例提供的车辆1000的结构示意图。车辆1000可以为燃油汽车、燃气汽车或新能源汽车，新能源汽车可以是纯电动汽车、混合动力汽车或增程式汽车等。车辆1000的内部设置有电池100，电池100可以设置在车辆1000的底部或头部或尾部。电池100可以用于车辆1000的供电，例如，电池100可以作为车辆1000的操作电源。车辆1000还可以包括控制器200和马达300，控制器200用来控制电池100为马达300供电，例如，用于车辆1000的启动、导航和行驶时的工作用电需求。
在本申请一些实施例中，电池100不仅可以作为车辆1000的操作电源，还可以作为车辆1000的驱动电源，代替或部分地代替燃油或天然气为车辆1000提供驱动动力。
请参照图2，图2为本申请一些实施例提供的电池100的爆炸图。电池100包括箱体10和电池单体20，电池单体20容纳于箱体10内。其中，箱体10用于为电池单体20提供容纳空间。在一些实施例中，箱体10可以包括第一部分11和第二部分12，第一部分11与第二部分12相互盖合，第一部分11和第二部分12共同限定出用于容纳电池单 体20的容纳空间。电池单体20可以为二次电池或一次电池；还可以是锂硫电池、钠离子电池或镁离子电池，但不局限于此。电池单体20可呈圆柱体、扁平体、长方体或其它形状等。
请参照图3，图3为本申请一些实施例提供的电池单体20的分解结构示意图。电池单体20是指组成电池的最小单元。如图3，电池单体20包括有端盖21、壳体22、电芯组件23(在下文中也称为电芯)以及其他的功能性部件。
端盖21是指盖合于壳体22的开口处以将电池单体20的内部环境隔绝于外部环境的部件。不限地，端盖21的形状可以与壳体22的形状相适应以配合壳体22。
壳体22是用于配合端盖21以形成电池单体20的内部环境的组件，其中，形成的内部环境可以用于容纳电芯组件23、电解液以及其他部件。
电芯组件23是电池单体100中发生电化学反应的部件。壳体22内可以包含一个或更多个电芯组件23。电芯组件23主要由正极片和负极片卷绕或层叠放置形成，并且通常在正极片与负极片之间设有隔膜。
图4为本申请一些实施例的检测电芯表面缺陷的方法400的流程示意图；图5为本申请一些实施例的对初始图像进行预处理的流程示意图；图6为本申请一些实施例的针对多个待检测图像中每个待检测图像的检测结果的示意图。下面结合图4至图6详细描述方法400。
如图4所示，本申请一些实施例提供的检测电芯(例如，如图3所示的电芯组件23)表面缺陷的方法400，可包括以下步骤S401至步骤S403。
在步骤S401，利用图像采集单元(未示出)获取包括电芯表面的初始图像500。
在步骤S402，对初始图像500进行预处理，以获得电芯表面的至少一个待检测图像512。
在步骤S403，将至少一个待检测图像512输入缺陷检测神经网络模型，并获取缺陷检测神经网络模型所输出的检测结果，检测结果用于指示电芯表面是否存在缺陷511。
图像采集单元的具体类型不限。例如，图像采集单元可以采用CCD工业相机，或者CMOS工业相机等。又例如，图像采集单元可以采用线阵相机或面阵相机等。
如图5所示，所获取的初始图像500至少包括电芯表面区510。此外，初始图像500还可以包括用于固定电芯的夹具区530以及除电芯表面区510和夹具区530之外的背景区520等。电芯表面的缺陷511例如可以是压伤缺陷、划伤缺陷、凸点缺陷等，本申请不限于此。
在步骤S402中预处理后所获取的至少一个待检测图像512可以包括一个或多个待检测图像。具体地，例如，在电芯表面的缺陷占比较大时，可以仅对初始图像500进行曝光度调整、颜色通道提取等预处理操作，而不对其进行分割。又例如，在电芯表面的缺陷(例如，压伤缺陷)的占比较小时，为了提高缺陷的占比以便缺陷的特征更加突出，可以对初始图像500进行分割，以获得多个待检测图像p1,…,pn。在本文中，图像中缺陷的特征可以包括灰度值、颜色、形状、轮廓以及纹理等。
缺陷检测神经网络模型可以是深度学习模型，例如，ResNet18、DNN等。深度学习模型可以从图像中提取多个特征，例如，灰度值、颜色、形状、轮廓以及纹理等中的多者，由此可以提高缺陷检测的精度。其中，缺陷检测神经网络模型所输出的检测结果可包括每一个待检测图像相应的缺陷置信度。
上述方法400可以在服务器、或客户端设备处执行。其中，服务器可以包括一个或多个通用计算机、专用服务器计算机(例如PC(个人计算机)服务器、UNIX服务器、中端服务器)、刀片式服务器、大型计算机、服务器群集或任何其他适当的布置和/或组合。在一些实施方式中，服务器可以为分布式系统的服务器、或者是结合了区块链的服务器、云服务器、带人工智能技术的智能云计算服务器或智能云主机等。客户端设备可以包括各种类型的计算机设备，例如便携式手持设备、通用计算机(诸如个人计算机和膝上型计算机)、工作站计算机等。替代地，方法400也可以在电芯生产线中装备有芯片的工控机处执行。其中，芯片可以是系统级芯片(System on Chip，SoC)等。
本申请实施例的方法400，通过对初始图像500进行预处理并利用缺陷检测神经网络模型检测预处理后的初始图像500中的缺陷511，一方面可以突出图像中的缺陷特征，以便于后续缺陷检测神经网络模型的检测，另一方面可以从图像中提取出更多特征，从而提高缺陷检测的效率和精度。
根据本申请的一些实施例，至少一个待检测图像512包括多个待检测图像p1,…,pn，初始图像500包括电芯表面所处的电芯表面区510。步骤S402、对初始图像500进行预处理，以获得电芯表面的至少一个待检测图像512可包括：从初始图像500中提取出电芯表面区510；以及对电芯表面区510进行分割，以获取多个待检测图像512。
也就是说，如图5所示，首先从初始图像500中提取出电芯表面区510，然后对提取出的电芯表面区510进行分割，以获得多个待检测图像p1,…,pn。在一些示例中，所分割出的多个待检测图像的大小可以相同(如图5所示)，也可以互不同，或者一部分大小相同而另一部分大小互不同。
由于电芯表面的某些缺陷比较小，例如仅占整个电芯表面的一小部分，通过提取电芯表面区并对其进行分割以获得小尺寸的待检测图像，使得这些缺陷的特征在待检测图像中更加突出，从而提高缺陷检测的准确度。
根据本申请的一些实施例，初始图像500具有第一曝光度。从初始图像500中提取出电芯表面区510可包括：从初始图像500中直接提取电芯表面区510；以及调整电芯表面区510的曝光度，以获得具有第二曝光度的电芯表面区510a，其中，第一曝光度高于第二曝光度。
第一曝光度可以是高曝光度。例如，可以通过放大工业相机的光圈、或者延长工业相机的曝光时间、增加环境中的亮度等中的一种或多种方式获取具有第一曝光度的图像。由于电芯表面通常有蓝胶，具有高曝光度的图像可以将电芯表面区510和初始图像500中的其他区域(例如，夹具区530、背景区520)显著区分开(例如，两者的像素值范围差别较大)。但是，在具有高曝光度的电芯表面区510中，某些缺陷511(例如，压伤缺陷)与电芯表面区510的其他部分难以区分开(例如，两者的像素值范围差别较小)，如图5所示。因此，将具有第一曝光度的电芯表面区510调整为具有第二曝光度的电芯表面区510a，以便于在电芯表面区510a中突出缺陷511的特征，如图5所示。
由此可以提高感兴趣区域ROI提取的准确性和后续缺陷检测的准确性。
根据本申请的一些实施例，第一曝光度被配置为使得初始图像500中的电芯表面区510的像素值范围不同于初始图像500中除电芯表面区510外的其他区域的像素值范围。
如图5所示，初始图像500中除电芯表面区510外的其他区域例如可包括夹具区530和背景区520。可以通过放大工业相机的光圈、或者延长工业相机的曝光时间、增加环境中的亮度等中的至少一种方式调整所获取的初始图像500的曝光度，以使得电芯表面区510与初始图像500中的其他区域520、530的像素值范围差别较大。
上述实施方式使得在第一曝光度的初始图像500中电芯表面区510易与其他区域520、530区分开，从而便于在具有第一曝光度的初始图像500中提取出电芯表面区510。
根据本申请的一些实施例，从初始图像500中提取出电芯表面区510还可包括：从具有第二曝光度的电芯表面区510a提取出预设颜色通道的电芯表面区510b，并且对电芯表面区510进行分割，以获取多个待检测图像p1,…,pn可包括对预设颜色通道的电芯表面区510b进行分割，以获取多个待检测图像p1,…,pn。
例如，在初始图像500是RGB图像时，颜色通道可以是R通道(红色通道)、G通道(绿色通道)、B通道(蓝色通道)。由于某些缺陷511(例如，压伤缺陷)的特征在特定的颜色通道的图像中更加突出，因此通过提取预设颜色通道的电芯表面区可以突出这些缺陷的特征，如图5中预设颜色通道的电芯表面区510b所示。
由此便于后续缺陷检测神经网络模型对电芯表面区中的缺陷进行检测，以进一步提高缺陷检测的准确性。
根据本申请的一些实施例，上述预设颜色通道为G通道。
由于电芯表面通常存在蓝胶，通过提取G通道的电芯表面区可以避免提取出蓝胶，从而降低蓝胶对于缺陷特征提取的影响。
由此便于后续缺陷检测神经网络模型对电芯表面区中的缺陷进行检测，以进一步提高缺陷检测的准确性。
根据本申请的一些实施例，从初始图像500中直接提取电芯表面区510可包括：至少基于电芯表面区的像素值范围，从初始图像500中提取电芯表面区510。
例如，初始图像500可以是RGB图像，本申请不限于此。像素值可以是彩色图像的各个颜色通道的亮度值。例如，RGB图像中的每个像素的像素值可以包括每个颜色通道，即红(R)、绿(G)或蓝(B)的亮度值，范围一般从0到255，最亮为255，最暗为0。电芯表面区510的像素值范围可以根据经验或者电池生产工艺规格要求来确定。例如，该像素值范围可以根据电芯表面在相同光照条件下成像的像素值范围来获取。
由于初始图像500上的各个区的像素值存在一定差异，上述基于初始图像的各个区上的像素值提取电芯表面区的技术方案，可以快速且有效地提取出电芯表面区510。
根据本申请的一些实施例，如图5所示，分割出的多个待检测图像p1,…,pn的大小相等。
上述通过将电芯表面区分割成大小相同的多个待检测图像，从而便于满足缺陷检测神经网络模型的输入图像尺寸要求，并且促进缺陷检测神经网络模型的检测。
根据本申请的一些实施例，步骤S403、将至少一个待检测图像512输入缺陷检测神经网络模型，并获取缺陷检测神经网络模型所输出的检测结果可包括：将至少一个待检测图像输入缺陷检测神经网络模型，以获得至少一个待检测图像中每一待检测图像的缺陷置信度；以及基于至少一个待检测图像中每一待检测图像的缺陷置信度，确定电芯表面是否存在缺陷。
其中，缺陷置信度可以是一个概率值，用于评价待检测图像中存在缺陷的可能性。图6中示出了神经网络模型针对多个待检测图像中相应的待检测图像所输出的检测结果t1,…tn，其中检测结果tn所对应的待检测图像的压伤缺陷611最显著。可以看到，神经网络模型所输出的检测结果tn的缺陷置信度(即，压伤得分)最高，即，0.996668，其指示检测结果tn所对应的待检测图像存在压伤缺陷的概率最大。
上述缺陷检测神经网络模型可以提取图像的更多特征，从而提高缺陷检测的效率和精度。
根据本申请的一些实施例，基于至少一个待检测图像512中每一待检测图像的缺陷置信度，确定电芯表面是否存在缺陷包括：确定至少一个待检测图像中每一待检测图像的缺陷置信度是否小于等于缺陷阈值；以及响应于确定至少一个待检测图像512中一个或多个待检测图像的缺陷置信度大于缺陷阈值，确定电芯表面存在缺陷。
也就是说，如果确定至少一个待检测图像512中任一个待检测图像的缺陷置信度大于缺陷阈值，即可确定电芯表面存在缺陷。上述缺陷阈值可以根据经验或者电池生产工艺规格要求来确定。
上述基于缺陷检测神经网络模型所输出的缺陷置信度和预设的缺陷阈值判断电芯表面是否存在缺陷，可以提高缺陷检测的效率和精度。
根据本申请一些实施例，方法400还可包括：响应于确定至少一个待检测图像512中每一待检测图像的缺陷置信度小于等于缺陷阈值，确定电芯表面不存在缺陷并发出使电芯继续流转到下一工序的指令。
根据本申请的一些实施例，上述缺陷511为压伤缺陷。
通过利用上述方法400检测压伤缺陷，从而可以提高压伤缺陷检测的效率和精度。
替代地，电芯表面的缺陷511例如也可以是划伤缺陷、凸点缺陷等。
如图7所示，本申请一些实施例提供的训练缺陷检测神经网络模型的方法700，可包括以下步骤S701至步骤S704。
在步骤S701，获取包括电芯表面的样本图像，样本图像包括电芯表面的预设缺陷。
在步骤S702，对样本图像进行预处理，以获得电芯表面的至少一个样本待检测图像。
在步骤S703，将至少一个样本待检测图像输入缺陷检测神经网络模型，并获取缺陷检测神经网络模型所输出的检测结果，检测结果用于指示电芯表面是否存在缺陷。
在步骤S704，基于检测结果和预设缺陷，调整缺陷检测神经网络模型的参数。
上述样本图像的获取方式与方法400中的初始图像的获取方式相同，在此不再详述。此外上述预设缺陷、样本待检测图像、缺陷检测神经网络模型等的特征分别与方法400中的缺陷、待检测图像、缺陷检测神经网络模型等的特征相同，在此不再详述。
在一些示例中，可以基于检测结果和预设缺陷，通过梯度下降法来不断调整缺陷检测神经网络模型的参数，以使得缺陷检测神经网络模型的损失函数最小化。
上述方法700可以在服务器、或客户端设备处执行。其中，服务器包括一个或多个通用计算机、专用服务器计算机(例如PC(个人计算机)服务器、UNIX服务器、中端服务器)、刀片式服务器、大型计算机、服务器群集或任何其他适当的布置和/或组合。在一些实施方式中，服务器可以为分布式系统的服务器、或者是结合了区块链的服务器、云服务器、带人工智能技术的智能云计算服务器或智能云主机等。客户端设备可以包括各种类型的计算机设备，例如便携式手持设备、通用计算机(诸如个人计算机和膝上型计算机)、工作站计算机等。
本申请实施例的方法700一方面可以突出图像中的缺陷特征，以促进后续缺陷检测神经网络模型的检测，另一方面可以提高缺陷检测神经网络模型的效率和精度。
根据本申请的一些实施例，至少一个样本待检测图像包括多个样本待检测图像，样本图像包括电芯表面所处的电芯表面区，并且步骤S702、对样本图像进行预处理，以获得电芯表面的至少一个样本待检测图像包括：从样本图像中提取出电芯表面区；以及对电芯表面区进行分割，以获取多个样本待检测图像。
所分割出的多个样本待检测图像的大小可以相同，也可以互不同，或者一部分大小相同而另一部分大小互不同。
由于电芯面的某些缺陷比较小，例如仅占整个电芯表面的一小部分，通过提取电芯表面区并对其进行分割以获得小尺寸的待检测图像，使得这些缺陷的特征在待检测图像中更加突出，从而提高缺陷检测神经网络模型的训练效率以及准确性。
根据本申请的一些实施例，样本图像具有第一曝光度，并且从样本图像中提取出电芯表面区包括：从样本图像中直接提取电芯表面区；以及调整电芯表面区的曝光度，以获得具有第二曝光度的电芯表面区，其中，第一曝光度高于第二曝光度。上述第一曝光度的特征与方法400中的第一曝光度的特征相同，在此不再详述。由此可以提高提高缺陷检测神经网络模型的训练效率以及准确性。
根据本申请的一些实施例，从样本图像中提取出电芯表面区还包括：从具有第二曝光度的电芯表面区提取出预设颜色通道的电芯表面区，并且对电芯表面区进行分割， 以获取多个样本待检测图像包括：对预设颜色通道的电芯表面区进行分割，以获取多个样本待检测图像。
上述预设颜色通道的特征与方法400中的预设颜色通道的特征相同，在此不再详述。
由此，便于后续缺陷检测神经网络模型对电芯表面区中的缺陷进行检测，以进一步提高缺陷检测神经网络模型的训练效率以及准确性。
根据本申请的一些实施例，从样本图像中直接提取电芯表面区可包括：至少基于电芯表面区的像素值范围，从初始图像中提取电芯表面区。
上述像素值和像素值范围的特征分别与方法400中的像素值和像素值范围的特征相同，在此不再详述。
如图8所示，本申请实施例还提供一种检测电芯表面的压伤缺陷的装置800，包括：获取单元801、预处理单元802以及检测单元803。获取单元801被配置为利用图像采集单元获取包括电芯表面的初始图像。预处理单元802被配置为对初始图像进行预处理，以获得电芯表面的至少一个待检测图像。检测单元803被配置为将至少一个待检测图像输入缺陷检测神经网络模型，并获取缺陷检测神经网络模型所输出的检测结果，检测结果用于指示电芯表面是否存在缺陷。
应当理解，图8中所示装置800的各个模块可以与参考图4描述的方法400中的各个步骤相对应。由此，上面针对方法400描述的操作、特征和优点同样适用于装置、800及其包括的模块。为了简洁起见，某些操作、特征和优点在此不再赘述。
本申请实施例的上述虚拟装置，一方面可以突出图像中的缺陷特征，以便于后续缺陷检测神经网络模型的检测，另一方面可以从图像中提取出更多特征，从而提高缺陷检测的效率和精度。
如图9所示，本申请实施例还提供一种训练缺陷检测神经网络模型的装置900，包括：图像获取单元901、图像处理单元902、缺陷检测单元903以及调整单元904。图像获取单元901被配置为获取包括电芯表面的样本图像，样本图像包括电芯表面的预设缺陷。预处理模块902被配置为对样本图像进行预处理，以获得电芯表面的至少一个样本待检测图像。缺陷检测单元903被配置为将至少一个样本待检测图像输入缺陷检测神经网络模型，并获取缺陷检测神经网络模型所输出的检测结果，检测结果用于指示电芯表面是否存在缺陷。调整单元904被配置为基于检测结果和预设缺陷，调整缺陷检测神经网络模型的参数。
应当理解，图9中所示装置900的各个模块可以与参考图8描述的方法700中的各个步骤相对应。由此，上面针对方法700描述的操作、特征和优点同样适用于装置900及其包括的模块。为了简洁起见，某些操作、特征和优点在此不再赘述。
本申请实施例的上述虚拟装置，一方面可以突出图像中的缺陷特征，以促进后续缺陷检测神经网络模型的检测，另一方面可以提高缺陷检测神经网络模型的效率和精度。
本申请实施例还提供一种电子设备，包括至少一个处理器以及与至少一个处理器通信连接的存储器，其中，存储器存储有可被至少一个处理器执行的指令，指令被至少一个处理器执行，以使至少一个处理器能够执行前述方法400和/或方法700。
本申请实施例还提供一种存储有计算机指令的计算机可读存储介质，其中，计算机指令配置为使计算机执行前述方法400和/或方法700。
本申请实施例还提供一种计算机程序产品，包括计算机程序，其中，计算机程序在被处理器执行时实现前述方法400和/或方法700。
如图10所示，本申请一些实施例提供的检测电芯表面缺陷的方法1000，包括以下步骤S1001至步骤S1011。
在步骤S1001，利用图像采集单元获取包括电芯表面的初始图像，初始图像具有第一曝光度。
在步骤S1002，从初始图像中直接提取电芯表面区。
在步骤S1003，调整电芯表面区的曝光度，以获得具有第二曝光度的电芯表面区，其中，第一曝光度高于第二曝光度。
在步骤S1004，从具有第二曝光度的电芯表面区提取出预设颜色通道的电芯表面区。
在步骤S1005，对预设颜色通道的电芯表面区进行分割，以获取多个待检测图像。
在步骤S1006，将多个待检测图像输入缺陷检测神经网络模型，以获得至少一个待检测图像中每一待检测图像的缺陷置信度。
在步骤S1007，确定多个待检测图像中每一待检测图像的缺陷置信度是否小于等于缺陷阈值。
在步骤S1008，响应于确定多个待检测图像中任一个待检测图像的缺陷置信度大于缺陷阈值，确定电芯表面存在缺陷。
在步骤S1009，响应于确定多个待检测图像中所有的待检测图像的缺陷置信度小于等于缺陷阈值，确定电芯表面不存在缺陷并发出将电芯继续流转到下一工序的指令。
上述方法1000中的各个步骤与方法400中的相应的步骤的特征相同。为了简洁起见，在此不再赘述。
本申请实施例的方法1000一方面可以突出在图像中突出缺陷的特征，以便于后续缺陷检测神经网络模型的检测，另一方面可以提取图像的更多特征，从而便于更高效地识别出缺陷，以提高缺陷检测的效率和精度。
最后应说明的是：以上各实施例仅用以说明本申请的技术方案，而非对其限制；尽管参照前述各实施例对本申请进行了详细的说明，本领域的普通技术人员应当理解：其依然可以对前述各实施例所记载的技术方案进行修改，或者对其中部分或者全部技术特征进行等同替换；而这些修改或者替换，并不使相应技术方案的本质脱离本申请各实施例技术方案的范围，其均应涵盖在本申请的权利要求和说明书的范围当中。尤其是，只要不存在结构冲突，各个实施例中所提到的各项技术特征均可以任意方式组合起来。本申请并不局限于文中公开的特定实施例，而是包括落入权利要求的范围内的所有技术方案。

Claims (20)

  1. 一种检测电芯表面缺陷的方法，包括：
    利用图像采集单元获取包括所述电芯表面的初始图像；
    对所述初始图像进行预处理，以获得所述电芯表面的至少一个待检测图像；以及
    将所述至少一个待检测图像输入缺陷检测神经网络模型，并获取所述缺陷检测神经网络模型所输出的检测结果，所述检测结果用于指示所述电芯表面是否存在缺陷。
  2. 根据权利要求1所述的方法，其中，所述至少一个待检测图像包括多个待检测图像，所述初始图像包括所述电芯表面所处的电芯表面区，并且
    对所述初始图像进行预处理，以获得所述电芯表面的至少一个待检测图像包括：
    从所述初始图像中提取出所述电芯表面区；以及
    对所述电芯表面区进行分割，以获取所述多个待检测图像。
  3. 根据权利要求2所述的方法，其中，所述初始图像具有第一曝光度，并且
    从所述初始图像中提取出所述电芯表面区包括：
    从所述初始图像中直接提取所述电芯表面区；以及
    调整所述电芯表面区的曝光度，以获得具有第二曝光度的电芯表面区，其中，所述第一曝光度高于所述第二曝光度。
  4. 根据权利要求3所述的方法，其中，所述第一曝光度被配置为使得所述初始图像中的电芯表面区的像素值范围不同于所述初始图像中除所述电芯表面区外的其他区域的像素值范围。
  5. 根据权利要求3或4所述的方法，其中，从所述初始图像中提取出所述电芯表面区还包括：
    从具有所述第二曝光度的电芯表面区提取出预设颜色通道的电芯表面区，并且
    对所述电芯表面区进行分割，以获取所述多个待检测图像包括：
    对所述预设颜色通道的电芯表面区进行分割，以获取所述多个待检测图像。
  6. 根据权利要求5所述的方法，其中，所述预设颜色通道为G通道。
  7. 根据权利要求3至6中任一项所述的方法，其中，从所述初始图像中直接提取所述电芯表面区包括：
    至少基于所述电芯表面区的像素值范围，从所述初始图像中提取所述电芯表面区。
  8. 根据权利要求2至7中任一项所述的方法，其中，所述多个待检测图像的大小相等。
  9. 根据权利要求1至8中任一项所述的方法，其中，所述检测结果包括每一个所述待检测图像相应的缺陷置信度，并且将所述至少一个待检测图像输入缺陷检测神经网络模型，并获取所述缺陷检测神经网络模型所输出的检测结果包括：
    将所述至少一个待检测图像输入所述缺陷检测神经网络模型，以获得所述至少一个待检测图像中每一待检测图像的缺陷置信度；以及
    基于所述至少一个待检测图像中每一待检测图像的缺陷置信度，确定所述电芯表面是否存在缺陷。
  10. 根据权利要求9所述的方法，其中，基于所述至少一个待检测图像中每一待检测图像的缺陷置信度，确定所述电芯表面是否存在缺陷包括：
    确定所述至少一个待检测图像中每一待检测图像的缺陷置信度是否小于等于缺陷阈值；以及
    响应于确定所述至少一个待检测图像中一个或多个待检测图像的缺陷置信度大于缺陷阈值，确定所述电芯表面存在缺陷。
  11. 根据权利要求1至10中任一项所述的方法，其中，所述缺陷为压伤缺陷。
  12. 一种训练缺陷检测神经网络模型的方法，包括：
    获取包括电芯表面的样本图像，所述样本图像包括所述电芯表面的预设缺陷；
    对所述样本图像进行预处理，以获得所述电芯表面的至少一个样本待检测图像；
    将所述至少一个样本待检测图像输入所述缺陷检测神经网络模型，并获取所述缺陷检测神经网络模型所输出的检测结果，所述检测结果用于指示所述电芯表面是否存在缺陷；以及
    基于所述检测结果和所述预设缺陷，调整所述缺陷检测神经网络模型的参数。
  13. 根据权利要求12所述的方法，其中，所述至少一个样本待检测图像包括多个样本待检测图像，所述样本图像包括所述电芯表面所处的电芯表面区，并且
    对所述样本图像进行预处理，以获得所述电芯表面的至少一个样本待检测图像包括：
    从所述样本图像中提取出所述电芯表面区；以及
    对所述电芯表面区进行分割，以获取所述多个样本待检测图像。
  14. 根据权利要求13所述的方法，其中，所述样本图像具有第一曝光度，并且
    从所述样本图像中提取出所述电芯表面区包括：
    从所述样本图像中直接提取所述电芯表面区；以及
    调整所述电芯表面区的曝光度，以获得具有第二曝光度的电芯表面区，其中，所述第一曝光度高于所述第二曝光度。
  15. 根据权利要求14所述的方法，其中，从所述样本图像中提取出所述电芯表面区还包括：
    从具有所述第二曝光度的电芯表面区提取出预设颜色通道的电芯表面区，并且
    对所述电芯表面区进行分割，以获取所述多个样本待检测图像包括：
    对所述预设颜色通道的电芯表面区进行分割，以获取所述多个样本待检测图像。
  16. 一种检测电芯表面的压伤缺陷的装置，包括：
    获取单元，所述获取单元被配置为利用图像采集单元获取包括所述电芯表面的初始图像；
    预处理单元，所述预处理单元被配置为对所述初始图像进行预处理，以获得所述电芯表面的至少一个待检测图像；以及
    检测单元，所述检测单元被配置为将所述至少一个待检测图像输入缺陷检测神经网络模型，并获取所述缺陷检测神经网络模型所输出的检测结果，所述检测结果用于指示所述电芯表面是否存在缺陷。
  17. 一种训练缺陷检测神经网络模型的装置，包括：
    图像获取单元，所述图像获取单元被配置为获取包括电芯表面的样本图像，所述样本图像包括所述电芯表面的预设缺陷；
    图像处理单元，所述预处理模块被配置为对所述样本图像进行预处理，以获得所述电芯表面的至少一个样本待检测图像；
    缺陷检测单元，所述检测单元被配置为将所述至少一个样本待检测图像输入所述缺陷检测神经网络模型，并获取所述缺陷检测神经网络模型所输出的检测结果，所述检测结果用于指示所述电芯表面是否存在缺陷；以及
    调整单元，所述调整单元被配置为基于所述检测结果和所述预设缺陷，调整所述缺陷检测神经网络模型的参数。
  18. 一种电子设备，包括至少一个处理器以及与所述至少一个处理器通信连接的存储器，其中，
    所述存储器存储有能够被所述至少一个处理器执行的指令，所述指令被所述至少一个处理器执行，以使所述至少一个处理器能够执行根据权利要求1至11中任一项所述的方法和/或根据权利要求12至15中任一项所述的方法。
  19. 一种存储有计算机指令的计算机可读存储介质，其中，所述计算机指令配置为使计算机执行根据权利要求1至11中任一项所述的方法和/或根据权利要求12至15中任一项所述的方法。
  20. 一种计算机程序产品，包括计算机程序，其中，所述计算机程序在被处理器执行时实现根据权利要求1至11中任一项所述的方法和/或根据权利要求12至15中任一项所述的方法。
PCT/CN2023/085076 2022-07-29 2023-03-30 检测电芯表面缺陷的方法和装置 Ceased WO2024021662A1 (zh)

Priority Applications (3)

Application Number Priority Date Filing Date Title
EP23744030.0A EP4336443B1 (en) 2022-07-29 2023-03-30 Method and apparatus for detecting defect of battery cell surface
ES23744030T ES3035473T3 (en) 2022-07-29 2023-03-30 Method and apparatus for detecting defect of battery cell surface
US18/454,184 US12406349B2 (en) 2022-07-29 2023-08-23 Method and apparatus for detecting defect on surface of cell

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202210907837.9A CN115829912A (zh) 2022-07-29 2022-07-29 检测电芯表面缺陷的方法和装置
CN202210907837.9 2022-07-29

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US18/454,184 Continuation US12406349B2 (en) 2022-07-29 2023-08-23 Method and apparatus for detecting defect on surface of cell

Publications (1)

Publication Number Publication Date
WO2024021662A1 true WO2024021662A1 (zh) 2024-02-01

Family

ID=85522990

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2023/085076 Ceased WO2024021662A1 (zh) 2022-07-29 2023-03-30 检测电芯表面缺陷的方法和装置

Country Status (3)

Country Link
CN (1) CN115829912A (zh)
HU (1) HUE071907T2 (zh)
WO (1) WO2024021662A1 (zh)

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117723491A (zh) * 2024-02-07 2024-03-19 宁德时代新能源科技股份有限公司 电芯防爆阀的检测系统及检测方法
CN117929280A (zh) * 2024-03-21 2024-04-26 宁德时代新能源科技股份有限公司 电池蓝膜包覆系统及电池蓝膜包覆的检测方法
CN118570554A (zh) * 2024-07-17 2024-08-30 敬业钢铁有限公司 基于图像处理的钢铁缺陷预测方法及系统
CN119780093A (zh) * 2024-11-05 2025-04-08 比亚迪股份有限公司 金属异物检测系统、方法、装置、设备及存储介质
CN120122578A (zh) * 2025-02-24 2025-06-10 青岛贝斯兰半导体科技有限公司 一种隔膜泵的生产过程控制方法、设备及介质
WO2025255946A1 (zh) * 2024-06-13 2025-12-18 宁德时代新能源科技股份有限公司 电芯绝缘膜检测系统及其方法

Families Citing this family (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115829912A (zh) * 2022-07-29 2023-03-21 宁德时代新能源科技股份有限公司 检测电芯表面缺陷的方法和装置
ES3035473T3 (en) 2022-07-29 2025-09-03 Contemporary Amperex Technology Hong Kong Ltd Method and apparatus for detecting defect of battery cell surface
CN116067619B (zh) * 2023-03-07 2023-08-11 宁德时代新能源科技股份有限公司 电池卷绕检测装置、卷绕装置、检测方法、设备和介质
CN116630262B (zh) * 2023-05-17 2025-02-07 江苏正力新能电池技术股份有限公司 基于神经网络的电芯检测方法和系统
CN116309564B (zh) * 2023-05-17 2023-08-11 厦门微图软件科技有限公司 基于人工智能图像识别的电芯外观缺陷检测方法及系统
CN116543267B (zh) * 2023-07-04 2023-10-13 宁德时代新能源科技股份有限公司 图像集处理方法、图像分割方法、装置、设备和存储介质
CN117611591B (zh) * 2024-01-24 2024-05-14 俐玛精密测量技术(苏州)有限公司 电芯缺陷的工业ct检测方法、装置、电子设备及存储介质
CN118037740A (zh) * 2024-04-15 2024-05-14 宁德时代新能源科技股份有限公司 极片检测方法、装置、设备、存储介质和计算机程序产品

Citations (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109472284A (zh) * 2018-09-18 2019-03-15 浙江大学 一种基于无偏嵌入零样本学习的电芯缺陷分类方法
CN109598721A (zh) * 2018-12-10 2019-04-09 广州市易鸿智能装备有限公司 电池极片的缺陷检测方法、装置、检测设备和存储介质
CN111060514A (zh) * 2019-12-02 2020-04-24 精锐视觉智能科技(上海)有限公司 缺陷检测方法、装置及终端设备
CN111598877A (zh) * 2020-05-18 2020-08-28 河北工业大学 一种基于生成对抗网络的锂电池表面缺陷检测方法
CN111640091A (zh) * 2020-05-14 2020-09-08 阿丘机器人科技(苏州)有限公司 产品缺陷的检测方法及计算机存储介质
CN112241699A (zh) * 2020-10-13 2021-01-19 无锡先导智能装备股份有限公司 物体缺陷类别识别方法、装置、计算机设备和存储介质
CN112634254A (zh) * 2020-12-29 2021-04-09 北京市商汤科技开发有限公司 绝缘子缺陷检测方法及相关装置
US20210374940A1 (en) * 2019-12-30 2021-12-02 Goertek Inc. Product defect detection method, device and system
US20210390682A1 (en) * 2020-06-12 2021-12-16 Beijing Baidu Netcom Science And Technology Co., Ltd. Method for detecting surface defect, method for training model, apparatus, device, and media
CN114782310A (zh) * 2022-03-08 2022-07-22 江苏立导科技有限公司 表面缺陷检测方法、装置、设备和存储介质
CN115829912A (zh) * 2022-07-29 2023-03-21 宁德时代新能源科技股份有限公司 检测电芯表面缺陷的方法和装置

Patent Citations (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109472284A (zh) * 2018-09-18 2019-03-15 浙江大学 一种基于无偏嵌入零样本学习的电芯缺陷分类方法
CN109598721A (zh) * 2018-12-10 2019-04-09 广州市易鸿智能装备有限公司 电池极片的缺陷检测方法、装置、检测设备和存储介质
CN111060514A (zh) * 2019-12-02 2020-04-24 精锐视觉智能科技(上海)有限公司 缺陷检测方法、装置及终端设备
US20210374940A1 (en) * 2019-12-30 2021-12-02 Goertek Inc. Product defect detection method, device and system
CN111640091A (zh) * 2020-05-14 2020-09-08 阿丘机器人科技(苏州)有限公司 产品缺陷的检测方法及计算机存储介质
CN111598877A (zh) * 2020-05-18 2020-08-28 河北工业大学 一种基于生成对抗网络的锂电池表面缺陷检测方法
US20210390682A1 (en) * 2020-06-12 2021-12-16 Beijing Baidu Netcom Science And Technology Co., Ltd. Method for detecting surface defect, method for training model, apparatus, device, and media
CN112241699A (zh) * 2020-10-13 2021-01-19 无锡先导智能装备股份有限公司 物体缺陷类别识别方法、装置、计算机设备和存储介质
CN112634254A (zh) * 2020-12-29 2021-04-09 北京市商汤科技开发有限公司 绝缘子缺陷检测方法及相关装置
CN114782310A (zh) * 2022-03-08 2022-07-22 江苏立导科技有限公司 表面缺陷检测方法、装置、设备和存储介质
CN115829912A (zh) * 2022-07-29 2023-03-21 宁德时代新能源科技股份有限公司 检测电芯表面缺陷的方法和装置

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117723491A (zh) * 2024-02-07 2024-03-19 宁德时代新能源科技股份有限公司 电芯防爆阀的检测系统及检测方法
CN117929280A (zh) * 2024-03-21 2024-04-26 宁德时代新能源科技股份有限公司 电池蓝膜包覆系统及电池蓝膜包覆的检测方法
CN117929280B (zh) * 2024-03-21 2024-09-06 宁德时代新能源科技股份有限公司 电池蓝膜包覆系统及电池蓝膜包覆的检测方法
WO2025194675A1 (zh) * 2024-03-21 2025-09-25 宁德时代新能源科技股份有限公司 电池蓝膜包覆系统及电池蓝膜包覆的检测方法
WO2025255946A1 (zh) * 2024-06-13 2025-12-18 宁德时代新能源科技股份有限公司 电芯绝缘膜检测系统及其方法
CN118570554A (zh) * 2024-07-17 2024-08-30 敬业钢铁有限公司 基于图像处理的钢铁缺陷预测方法及系统
CN119780093A (zh) * 2024-11-05 2025-04-08 比亚迪股份有限公司 金属异物检测系统、方法、装置、设备及存储介质
CN120122578A (zh) * 2025-02-24 2025-06-10 青岛贝斯兰半导体科技有限公司 一种隔膜泵的生产过程控制方法、设备及介质

Also Published As

Publication number Publication date
HUE071907T2 (hu) 2025-10-28
CN115829912A (zh) 2023-03-21

Similar Documents

Publication Publication Date Title
WO2024021662A1 (zh) 检测电芯表面缺陷的方法和装置
US20240077432A1 (en) Cell detection method, apparatus, and system, computer device, and storage medium
KR20240050461A (ko) 배터리 전극 시트 절연 코팅층 결함의 검출 방법, 장치 및 컴퓨터 장치
CN112330623B (zh) 电芯极组极片对齐度检测方法和检测装置
CN108398649B (zh) 析锂检测方法及装置
US11763549B1 (en) Method and apparatus for training cell defect detection model
CN111141755B (zh) 一种电池电芯内部缺陷的检测方法
Zhao et al. Analysis of polarization and thermal characteristics in lithium-ion battery with various electrode thicknesses
CN114897889B (zh) 电池组点焊自动化全检方法及其系统
EP4332555A1 (en) Electrode sheet wrinkling detection apparatus and battery cell production equipment
CN110992321B (zh) 一种太阳电池片栅线提取方法
EP4336443B1 (en) Method and apparatus for detecting defect of battery cell surface
CN117173100A (zh) 聚合物锂离子电池生产控制系统及其方法
US20240221150A1 (en) Electrode plate wrinkling detection method and system, terminal, and storage medium
Wang et al. Hotspot detection of photovoltaic modules in infrared thermal image based on saliency analysis
WO2024109396A1 (zh) 转接片的检测方法、装置、电池单体的生产方法和设备
CN114387466A (zh) 一种太阳能电池片色差检测方法及系统
CN109127478A (zh) 用于筛选电池组的单体电池的方法
CN107037365A (zh) 一种动力锂电池电芯的测量方法
CN118314141B (zh) 检测方法、装置、存储介质及程序产品
KR20250052876A (ko) 배터리 이미지 분석 장치 및 그의 동작 방법
CN206876838U (zh) 一种动力锂电池电芯的测量系统
CN119559120B (zh) 一种基于plc控制系统的锂电池缺陷检测系统及方法
CN212433756U (zh) 一种线路巡检照片整理装置
EP4417960A1 (en) Yield rate measurement method and apparatus for battery electrode sheets, device, and medium

Legal Events

Date Code Title Description
ENP Entry into the national phase

Ref document number: 2023744030

Country of ref document: EP

Effective date: 20230731

121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 23744030

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

WWG Wipo information: grant in national office

Ref document number: 2023744030

Country of ref document: EP