A method and device for detecting defects of a pole piece

Abstract

The present invention discloses a method and device for detecting pole piece defects, including: calculating the image edge of the first image data including the pole piece to be detected, segmenting the first image data according to the image edge to obtain a first pole piece image, performing image reconstruction on the first pole piece image to obtain a second pole piece image, and performing image difference between the second pole piece image and the first pole piece image to obtain the difference between the first pole piece image and the second pole piece image and the difference image corresponding to the difference, and performing Gaussian filtering on the difference image to further obtain a defect heat map, segmenting the defect heat map according to the difference to obtain the defect position and extracting the defect image in the first pole piece image according to the defect position, and classifying the defect image to obtain the defect type of the pole piece to be detected, thereby improving the accuracy and efficiency of pole piece defect detection.

Description

Technical Field

[0001] The present invention relates to the technical field of battery manufacturing, and particularly to a method and device for detecting defects of electrode sheets. Background Art

[0002] With the continuous development of the economy, the demand for energy in various industries is increasing. Lithium batteries have been widely used in the fields of energy storage and electric vehicles due to their high working voltage, large energy density, low self-discharge rate, small size, and customizable shape. As the core component of lithium batteries, the performance of electrode sheets directly determines the electrochemical performance of lithium batteries and has a great impact on the safety of lithium batteries. Defects will occur in the slurry preparation, coating, drying, rolling, slitting, etc. of electrode sheets. By detecting defects in lithium battery electrode sheets, it is possible to prevent defective electrode sheets from flowing into the next process, ensuring the safety performance and service life of lithium batteries. At the same time, analyze the causes of electrode sheet defects, reduce the generation of defects, and avoid economic losses.

[0003] Traditional inspectors rely on the human eye for defect detection. However, this method has a high detection intensity, requires a large amount of time and energy, and the detection results are easily affected by the subjectivity of the inspectors, and there is a lack of standard and quantitative index evaluation. At the same time, traditional electrode sheet machine vision detection algorithms need to manually design features to extract defects, which requires high requirements for staff and has poor recognition of new defects. Summary of the Invention

[0004] The present invention provides a method and device for detecting defects of electrode sheets, which improves the accuracy of electrode sheet defect detection, reduces the requirements for staff, and improves the detection efficiency.

[0005] In order to achieve the above object, the present invention discloses a method for detecting defects of electrode sheets, including:

[0006] Obtain first image data including the electrode sheet to be detected, calculate the image edge of the first image data, and segment the first image data according to the image edge to obtain a first electrode sheet image of the electrode sheet to be detected;

[0007] Perform image reconstruction on the first electrode sheet image to obtain a second electrode sheet image after image reconstruction, and perform image difference between the second electrode sheet image and the first electrode sheet image to obtain the difference between the first electrode sheet image and the second electrode sheet image and the defect heat map corresponding to the first electrode sheet image;

[0008] Segment the defect heat map according to the difference to obtain the defect position in the defect heat map, extract the defect image corresponding to the first electrode sheet image according to the defect position, and classify the defect image to obtain the defect type of the electrode sheet to be detected.

[0009] The present invention discloses a method for detecting defects on a pole piece, which includes obtaining first image data containing the pole piece to be measured, and then performing image edge detection on the first image data to facilitate extracting the pole piece image from the image data, so as to perform defect detection based on the pole piece image. After obtaining the pole piece image, perform image reconstruction on the first pole piece image, and perform image difference between the reconstructed image and the first pole piece image. Furthermore, construct a defect heat map of the pole piece according to the difference value of the image difference, so as to judge whether there are defects on the pole piece according to the defect heat map. After obtaining the defect heat map, segment the defect heat map to facilitate obtaining the defect position of the pole piece, and then extract the defect image according to the defect position, and classify the defect image to obtain the type of the defect image, and further judge the type of the pole piece defect. The present invention performs image reconstruction to reconstruct the defect image into an image close to the preset pole piece image, and performs subtraction on the reconstructed image and the preset image to obtain a defect map, and then performs threshold segmentation on the defect map to obtain the final defect location image. This method solves the problem of difficult manual positioning of defect images, reduces the detection difficulty, improves the efficiency, and at the same time facilitates identifying any new defects appearing in the pole piece according to the difference of the images, improving the accuracy of pole piece defect detection.

[0010] As a preferred example, in the step of obtaining the first image data containing the pole piece to be measured, it specifically includes:

[0011] Supplementary light is provided to the pole piece to be measured through a preset light source, so as to obtain image data containing the pole piece to be measured according to a preset image acquisition device; the image data includes the upper surface image and the lower surface image of the pole piece to be measured;

[0012] The first image data is obtained by performing Gaussian smoothing processing on the image data.

[0013] The present invention provides supplementary light to the pole piece to be measured to facilitate the image acquisition device to clearly acquire the image data containing the pole piece to be measured, improving the accuracy of pole piece defect detection. Then, Gaussian smoothing processing is performed on the image data to remove the noise in the image, eliminate the influence on defect detection, and improve the accuracy of pole piece defect detection.

[0014] As a preferred example, in the step of calculating the image edge of the first image data to segment the first image data according to the image edge and obtain the first pole piece image of the pole piece to be measured, it specifically includes:

[0015] Use a preset edge detection algorithm to perform edge detection on the first image data to obtain the image edge between the pole piece to be measured and the background in the first image data;

[0016] Perform edge segmentation on the first image data according to the image edge to obtain the first electrode image of the electrode to be measured.

[0017] In the present invention, the obtained first image data is subjected to edge detection through a preset edge detection algorithm, so as to completely obtain the image data of the electrode to be measured, and the defect detection of the electrode is performed in sequence to improve the accuracy of defect detection.

[0018] As a preferred example, obtaining the difference between the first electrode image and the second electrode image and the defect heat map corresponding to the first electrode image specifically includes:

[0019] Perform image reconstruction on the first electrode image according to a preset image reconstruction algorithm to obtain the second electrode image after reconstruction of the first electrode image;

[0020] Perform image difference between the second electrode image and the first electrode image to respectively obtain a plurality of differences between a plurality of second pixels in the second electrode image and a plurality of corresponding first pixels in the first electrode image, and construct a difference image according to the plurality of differences;

[0021] Obtain the defect heat map corresponding to the first electrode image by performing Gaussian filtering on the difference image.

[0022] The present invention discloses reconstructing the obtained first electrode image by using a preset image reconstruction algorithm, performing image difference between the inverse image reconstruction and the first electrode image to obtain a difference image after image difference, and then performing Gaussian filtering on the difference image to improve the accuracy of the difference image, and further obtaining a defect heat map, so as to judge whether there is a defect in the electrode and determine the defect position according to the defect heat map.

[0023] As a preferred example, segmenting the defect heat map according to the difference to obtain the defect position in the defect heat map, and extracting the defect image corresponding to the first electrode image according to the defect position specifically includes:

[0024] Compare each of the plurality of differences with a preset threshold, aggregate a plurality of second pixels corresponding to the differences greater than the threshold, and construct a first bounding box according to the second pixels;

[0025] Determine the coordinates of the first bounding box, and obtain the defect position of the first electrode image according to the coordinates, so as to extract the defect image in the first electrode image according to the defect position.

[0026] The present invention determines whether there are defects in the pole piece based on the difference after comparing the pole piece images. If all the differences are less than or equal to the threshold value, it means that the current pole piece has no defects. If there are differences greater than the set threshold value, the pixels corresponding to the differences are aggregated to determine the position of the defect according to the positions of the pixels. By utilizing the differences between the images, it determines the existence of defects in the pole piece, solves the difficulty of manually determining new types of defects, and improves the detection efficiency and accuracy.

[0027] As a preferred example, classifying the defect image to obtain the defect type of the pole piece to be measured specifically includes:

[0028] Based on the defect image, the trained classification model is used to identify the defect image to obtain the type of the defect image.

[0029] The present invention automatically classifies the defect image by using the pre-trained classification model, solves the problem of the large amount of energy and time invested in manual classification in the prior art, and improves the defect detection efficiency.

[0030] On the other hand, the present invention also discloses a pole piece defect detection device, including an image segmentation module, an image reconstruction module, and a defect detection module;

[0031] The image segmentation module is used to obtain the first image data including the pole piece to be measured and calculate the image edge of the first image data, so as to segment the first image data according to the image edge to obtain the first pole piece image of the pole piece to be measured;

[0032] The image reconstruction module is used to perform image reconstruction on the first pole piece image to obtain the second pole piece image after image reconstruction, and perform image difference between the second pole piece image and the first pole piece image to obtain the difference between the first pole piece image and the second pole piece image and the defect heat map corresponding to the first pole piece image;

[0033] The defect detection module is used to segment the defect heat map according to the difference to obtain the defect position in the defect heat map, extract the defect image corresponding to the first pole piece image according to the defect position, and classify the defect image to obtain the defect type of the pole piece to be measured.

[0034] The present invention discloses a pole piece defect detection device, which includes obtaining first image data containing a pole piece to be detected, and then performing image edge detection on the first image data to facilitate extracting the pole piece image in the image data, so as to perform defect detection based on the pole piece image. After obtaining the pole piece image, image reconstruction is performed on the first pole piece image, and image difference is performed between the reconstructed image and the first pole piece image. Furthermore, a defect heat map of the pole piece is constructed based on the difference value of the image difference, so as to judge whether the pole piece has defects according to the defect heat map. After obtaining the defect heat map, the defect heat map is segmented to facilitate obtaining the defect position of the pole piece. Furthermore, a defect image is extracted according to the defect position, and then the defect image is classified to obtain the type of the defect image, and further judge the type of the pole piece defect. The present invention performs image reconstruction, reconstructs the defect image into an image close to a preset pole piece image, performs subtraction on the reconstructed image and the preset image to obtain a defect map, and then performs threshold segmentation on the defect map to obtain a final defect localization image. This method solves the problem of difficult manual localization of defect images, reduces the detection difficulty, improves the efficiency, and at the same time facilitates identifying any new defects appearing in the pole piece according to the image difference, improving the accuracy of pole piece defect detection.

[0035] As a preferred example, the image segmentation module includes a collection unit, an edge calculation unit, and a segmentation unit;

[0036] The collection unit is used to perform supplementary lighting on the pole piece to be detected through a preset light source, so as to obtain image data containing the pole piece to be detected according to a preset image acquisition device; the image data includes the upper surface image and the lower surface image of the pole piece to be detected; the first image data is obtained by performing Gaussian smoothing processing on the image data;

[0037] The edge calculation unit is used to perform edge detection on the first image data by using a preset edge detection algorithm to obtain the image edge between the pole piece to be detected and the background in the first image data;

[0038] The segmentation unit is used to perform edge segmentation on the first image data according to the image edge to obtain the first pole piece image of the pole piece to be detected.

[0039] The present invention performs supplementary lighting through the polar plate to be measured, so that the image acquisition device can clearly acquire the image data including the polar plate to be measured, improving the accuracy of polar plate defect detection. Then, Gaussian smoothing processing is performed on the image data to remove the noise in the image, eliminate the influence on defect detection, and improve the accuracy of polar plate defect detection. Then, a preset edge detection algorithm is used to perform edge detection on the obtained first image data, so as to completely obtain the image data of the polar plate to be measured, and defect detection of the polar plate is performed in sequence, improving the accuracy of defect detection.

[0040] As a preferred example, in the image reconstruction module, it includes an image unit, a difference unit, and a processing unit;

[0041] The image unit is used to perform image reconstruction on the first polar plate image according to a preset image reconstruction algorithm to obtain a second polar plate image after the first polar plate image is reconstructed;

[0042] The difference unit is used to perform image difference between the second polar plate image and the first polar plate image to respectively obtain a number of differences between a number of second pixels in the second polar plate image and a corresponding number of first pixels in the first polar plate image, and construct a difference image according to the number of differences;

[0043] The processing unit is used to perform Gaussian filtering on the difference image to obtain a defect heat map corresponding to the first polar plate image.

[0044] The present invention discloses reconstructing the obtained first polar plate image by using a preset image reconstruction algorithm, performing image difference between the reverse image reconstruction and the first polar plate image to obtain a difference image after image difference, and then performing Gaussian filtering on the difference image to improve the accuracy of the difference image, and further obtaining a defect heat map, so as to determine whether there is a defect in the polar plate and determine the defect position according to the defect heat map.

[0045] As a preferred example, in the defect detection module, it includes a positioning unit, an extraction unit, and a classification unit;

[0046] The positioning unit is used to compare each of the number of differences with a preset threshold, aggregate a number of second pixels corresponding to the differences greater than the threshold, and construct a first bounding box according to the second pixels;

[0047] The extraction unit is used to determine the coordinates of the first bounding box and obtain the defect position of the first polar plate image according to the coordinates, so as to extract the defect image in the first polar plate image according to the defect position;

[0048] The classification unit is used to identify the defective image through a trained classification model according to the defective image, and obtain the type of the defective image.

[0049] In the present invention, whether there is a defect in the pole piece is determined by the difference after comparing the pole piece images. If all the differences are less than or equal to the threshold, it means that there is no defect in the current pole piece. If there is a difference greater than the set threshold, the pixels corresponding to the differences are aggregated, so that the position of the defect can be determined according to the position of the pixels. By utilizing the difference between images, it is determined that there is a defect in the pole piece, which solves the difficulty of manually determining new types of defects, improves the detection efficiency and accuracy, and automatically classifies the defective image by using a pre-trained classification model, which solves the problem of a large amount of energy and time invested in manual classification in the prior art and improves the efficiency of defect detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 : is a schematic flow chart of a method for detecting pole piece defects provided by an embodiment of the present invention;

[0051] Figure 2 : is a schematic structural diagram of a device for detecting pole piece defects provided by an embodiment of the present invention;

[0052] Figure 3 : is a schematic flow chart of another method for detecting pole piece defects provided by another embodiment of the present invention;

[0053] Figure 4 : is a schematic structural diagram of a device for detecting pole piece defects provided by another embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0054] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0055] Embodiment 1

[0056] An embodiment of the present invention provides a method for detecting pole piece defects. The specific implementation process of this method is as follows for reference Figure 1 , mainly including steps 101 to 103. Each step specifically includes:

[0057] Step 101: Obtain first image data including a pole piece to be detected, and calculate the image edge of the first image data, so as to segment the first image data according to the image edge and obtain a first pole piece image of the pole piece to be detected.

[0058] In this embodiment, this step specifically includes: supplementing light to the to-be-tested electrode sheet through a preset light source so as to obtain image data including the to-be-tested electrode sheet according to a preset image acquisition device; the image data includes the upper surface image and the lower surface image of the to-be-tested electrode sheet; performing Gaussian smoothing processing on the image data to obtain the first image data; using a preset edge detection algorithm to perform edge detection on the first image data to obtain the image edge between the to-be-tested electrode sheet and the background in the first image data; performing edge segmentation on the first image data according to the image edge to obtain the first electrode sheet image of the to-be-tested electrode sheet.

[0059] In this embodiment, by supplementing light to the to-be-tested electrode sheet, it is convenient for the image acquisition device to clearly collect the image data including the to-be-tested electrode sheet, improving the accuracy of electrode sheet defect detection. Then, Gaussian smoothing processing is performed on the image data to remove the noise in the image, eliminating the influence on defect detection and improving the accuracy of electrode sheet defect detection. Then, edge detection is performed on the obtained first image data through a preset edge detection algorithm to completely obtain the image data of the to-be-tested electrode sheet, and defect detection of the electrode sheet is sequentially performed to improve the accuracy of defect detection.

[0060] Step 102: Perform image reconstruction on the first electrode sheet image to obtain a second electrode sheet image after image reconstruction, and perform image difference between the second electrode sheet image and the first electrode sheet image to obtain the difference between the first electrode sheet image and the second electrode sheet image and the defect heat map corresponding to the first electrode sheet image.

[0061] In this embodiment, this step specifically includes: performing image reconstruction on the first electrode sheet image according to a preset image reconstruction algorithm to obtain a second electrode sheet image after reconstruction of the first electrode sheet image; performing image difference between the second electrode sheet image and the first electrode sheet image to respectively obtain a plurality of differences between a plurality of second pixels in the second electrode sheet image and corresponding a plurality of first pixels in the first electrode sheet image, and constructing a difference image according to the plurality of differences; performing Gaussian filtering on the difference image to obtain the defect heat map corresponding to the first electrode sheet image.

[0062] This embodiment discloses reconstructing the obtained first electrode sheet image by using a preset image reconstruction algorithm, performing image difference between the reverse image reconstruction and the first electrode sheet image to obtain a difference image after image difference, and then performing Gaussian filtering on the difference image to improve the accuracy of the difference image, and further obtaining a defect heat map, so as to judge whether the electrode sheet has defects and determine the defect position according to the defect heat map.

[0063] Step 103: Segment the defective heat map according to the difference value, obtain the defective positions in the defective heat map, extract the defective images corresponding to the first pole piece image according to the defective positions, and classify the defective images to obtain the defective types of the pole piece to be measured.

[0064] In this embodiment, this step specifically includes: comparing each of the several difference values with a preset threshold, aggregating several second pixels corresponding to the difference values greater than the threshold, and constructing a first bounding box according to the second pixels; determining the coordinates of the first bounding box, and obtaining the defective positions of the first pole piece image according to the coordinates, so as to extract the defective images in the first pole piece image according to the defective positions; according to the defective images, identifying the defective images through a trained classification model to obtain the types of the defective images.

[0065] In this embodiment, it is determined whether the pole piece has defects through the difference value after comparing the pole piece images. If all the difference values are less than or equal to the threshold, it means that the current pole piece has no defects. If there are difference values greater than the set threshold, the pixels corresponding to the difference values are aggregated, so as to determine the positions of the defects according to the positions of the pixels. By using the differences between the images, it is determined that the pole piece has defects, which solves the difficulty of manually determining new types of defects and improves the detection efficiency and accuracy. Then, the pre-trained classification model is used to automatically classify the defective images, which solves the problem of the large amount of energy and time invested in manual classification in the prior art and improves the efficiency of defect detection.

[0066] On the other hand, an embodiment of the present invention further provides a pole piece defect detection device. For the specific structural schematic diagram of the device, please refer to Figure 2 and mainly includes an image segmentation module 201, an image reconstruction module 202 and a defect detection module 203.

[0067] The image segmentation module 201 is used to obtain the first image data including the pole piece to be measured, calculate the image edge of the first image data, so as to segment the first image data according to the image edge to obtain the first pole piece image of the pole piece to be measured.

[0068] The image reconstruction module 202 is used to perform image reconstruction on the first pole piece image to obtain a second pole piece image after image reconstruction, and perform image difference between the second pole piece image and the first pole piece image to obtain the difference value between the first pole piece image and the second pole piece image and the defective heat map corresponding to the first pole piece image.

[0069] The defect detection module 203 is used to segment the defect heat map according to the difference value, obtain the defect positions in the defect heat map, extract the defect images corresponding to the first pole piece image according to the defect positions, and classify the defect images to obtain the defect types of the to-be-tested pole piece.

[0070] In this embodiment, the image segmentation module 201 further includes an acquisition unit, an edge computing unit, and a segmentation unit.

[0071] The acquisition unit is used to supplement light to the to-be-tested pole piece through a preset light source, so as to obtain image data including the to-be-tested pole piece according to a preset image acquisition device; the image data includes the upper surface image and the lower surface image of the to-be-tested pole piece; the first image data is obtained by performing Gaussian smoothing processing on the image data.

[0072] The edge computing unit is used to perform edge detection on the first image data by using a preset edge detection algorithm to obtain the image edge between the to-be-tested pole piece and the background in the first image data.

[0073] The segmentation unit is used to perform edge segmentation on the first image data according to the image edge to obtain the first pole piece image of the to-be-tested pole piece.

[0074] In this embodiment, the image reconstruction module 202 further includes an image unit, a difference unit, and a processing unit;

[0075] The image unit is used to perform image reconstruction on the first pole piece image according to a preset image reconstruction algorithm to obtain the second pole piece image after the first pole piece image is reconstructed;

[0076] The difference unit is used to perform image difference between the second pole piece image and the first pole piece image to respectively obtain a plurality of differences between a plurality of second pixels in the second pole piece image and corresponding a plurality of first pixels in the first pole piece image, and construct a difference image according to the plurality of differences;

[0077] The processing unit is used to perform Gaussian filtering on the difference image to obtain the defect heat map corresponding to the first pole piece image.

[0078] In this embodiment, the defect detection module 203 further includes a positioning unit, an extraction unit, and a classification unit.

[0079] The positioning unit is used to compare each of the plurality of differences with a preset threshold, aggregate the plurality of second pixels corresponding to the differences greater than the threshold, and construct a first bounding box according to the second pixels.

[0080] The extraction unit is used to determine the coordinates of the first bounding box, and obtain the defect positions of the first pole piece image according to the coordinates, so as to extract the defect images in the first pole piece image according to the defect positions.

[0081] The classification unit is used to identify the defect images according to the defect images through a trained classification model, and obtain the types of the defect images.

[0082] A method and device for detecting pole piece defects disclosed in this embodiment include obtaining first image data containing a to-be-detected pole piece, then performing image edge detection on the first image data to facilitate extracting the pole piece image from the image data, so as to perform defect detection according to the pole piece image. After obtaining the pole piece image, perform image reconstruction on the first pole piece image, and perform image difference between the reconstructed image and the first pole piece image. Furthermore, construct a defect heat map of the pole piece according to the difference value of the image difference, so as to judge whether the pole piece has defects according to the defect heat map. After obtaining the defect heat map, perform segmentation on the defect heat map to facilitate obtaining the defect positions of the pole piece, and then extract defect images according to the defect positions, and then classify the defect images to obtain the types of the defect images, and further judge the types of the pole piece defects. The present invention performs image reconstruction, reconstructs the defect image into an image close to a preset pole piece image, subtracts the reconstructed image and the preset image to obtain a defect map, and then performs threshold segmentation on the defect map to obtain a final defect positioning image. This method solves the problem of difficult manual positioning of defect images, reduces the detection difficulty, improves the efficiency, and at the same time facilitates identifying any new defects in the pole piece according to the difference of the images, improving the accuracy of pole piece defect detection.

[0083] Embodiment 2

[0084] This embodiment provides another method for detecting pole piece defects. The specific implementation process of this method is referred to Figure 3 , and mainly includes steps 301 to 304. Each step specifically includes:

[0085] Step 301: Obtain image information of a lithium battery pole piece, and perform Gaussian smoothing processing on the image information to obtain image data containing the lithium battery pole piece.

[0086] In this embodiment, this step specifically includes: supplementing light to the lithium battery pole piece through a preset light source, so as to obtain image information containing the lithium battery pole piece according to a preset image acquisition device. The image information includes the upper surface image and the lower surface image of the to-be-detected pole piece, and then perform Gaussian smoothing processing on the image information to obtain the image data.

[0087] In this embodiment, this step specifically includes: using the parallel light source provided by a device for detecting defects on the electrode sheet in this embodiment, that is, the preset light source, to supplement light to the lithium battery electrode sheet. For the specific structure of the device for monitoring electrode sheet defects, please refer to Figure 4 , the Figure 4 includes rollers 1 and 2, coating, linear array cameras 1 and 2, two LED linear light sources, a PLC, an encoder, and an industrial computer. The PLC includes an actuator, and the industrial computer includes an image acquisition card and an I / O card. According to the PLC and the actuator, the defect detection operation of the lithium battery electrode sheet is controlled. The rollers and coating included in Figure 4 are used to realize the transmission of the lithium battery electrode sheet. The two LED linear light sources provided in Figure 4 are used to supplement light to the upper surface and the lower surface of the lithium battery electrode sheet respectively, so that the linear array camera 2 parallel to the light propagation direction of the LED linear light source and perpendicular to the lithium battery electrode sheet can collect the upper surface image of the lithium battery electrode sheet, and the linear array camera 1 parallel to the light propagation direction of the LED linear light source and perpendicular to the lithium battery electrode sheet can collect the lower surface image of the lithium battery electrode sheet. Then, the upper surface image and the lower surface image are collected by means of image scanning. Then, the multi-line scanned images are integrated into a surface image by the image acquisition card included in the industrial computer and stored in the computer, so as to obtain the image data corresponding to the upper surface image and the lower surface image. Then, the image data of the upper surface and the lower surface of the collected lithium battery electrode sheet is subjected to Gaussian smoothing processing to remove the noise in the image data, and the image data including the lithium battery electrode sheet is obtained.

[0088] Step 302: Use the preset edge detection technology to obtain the electrode sheet edge in the image data, and then segment the lithium battery electrode sheet from the image data according to the electrode sheet edge to obtain the first electrode sheet image corresponding to the lithium battery electrode sheet.

[0089] In this embodiment, this step specifically includes: calculating the image edge of the image data, that is, using the preset edge detection algorithm to perform edge detection on the image data to obtain the image edge between the electrode sheet to be measured and the background in the image data, and then segmenting the image data according to the image edge to obtain the first electrode sheet image of the lithium battery electrode sheet.

[0090] In this embodiment, this step is specifically: using the Figure 4The edge detection technology preset in the industrial control computer provided in [reference] is used to detect the image edges in the image data. The image edges are the boundaries between the lithium battery pole piece and the background. Then, the image data is segmented according to the obtained image edges to separate the lithium battery pole piece from the background, and the first pole piece image corresponding to the lithium battery pole piece is obtained for subsequent defect detection.

[0091] Step 303: Perform image reconstruction on the first pole piece image through a preset image reconstruction algorithm to obtain a second pole piece image after image reconstruction, and perform image difference on the second pole piece image and the first pole piece image to obtain the difference between the first pole piece image and the second pole piece image and the defect heat map corresponding to the first pole piece image.

[0092] In this embodiment, this step specifically includes: performing image reconstruction on the first pole piece image and performing image reconstruction on the first pole piece image through a preset image reconstruction algorithm to obtain a second pole piece image after the first pole piece image is reconstructed, and performing image difference on the second pole piece image and the first pole piece image to respectively obtain a number of differences between a number of first pixels in the second pole piece image and a number of second pixels in the first pole piece image, and constructing the defect heat map according to the number of differences.

[0093] In this embodiment, this step is specifically: according to the first pole piece image, the first pole piece image is reconstructed through an image reconstruction algorithm pre-trained in the industrial control computer to obtain a reconstructed second pole piece image. If the first pole piece image is a normal image, it can be normally reconstructed. If the first pole piece image is a defective image, there will be a large difference from the normal pole piece image after reconstruction. The reconstructed second pole piece image and the first pole piece image are used for difference, a score map is constructed according to the difference between the reconstructed image and the first pole piece image, and the score map is Gaussian filtered to obtain the defect heat map. The difference is the number of differences between a number of pixels in the second pole piece image and the corresponding number of pixels in the first pole piece image, and the defect heat map is constructed according to the number of differences.

[0094] In this embodiment, the training process of the image reconstruction algorithm is as follows: A plurality of pole piece images are collected by the linear array camera and the LED linear light source arranged on the device provided in this embodiment. After edge segmentation of the plurality of pole piece images, the image data of the pole piece is obtained. Then, professionals annotate the pole pieces, classifying them into defective images and non-defective images. At the same time, the professionals classify the pole piece defects corresponding to the defective images and label the classified defective images. Among them, the non-defective images are positive sample images, and the positive samples are used to train the image reconstruction algorithm. Thus, normal images can be reconstructed normally, and there will be a large difference between the reconstructed defective images and the original images. An image reconstruction network of the image reconstruction algorithm is constructed according to the positive sample images used for training, the ability of the training image intermediate algorithm to reconstruct the samples is trained, and the parameters of the image reconstruction algorithm are continuously optimized using a preset loss function to obtain the trained image reconstruction algorithm.

[0095] In this embodiment, the preset loss function of the image reconstruction algorithm is:

[0096]

[0097] Step 304: Segment the defective heat map according to the difference value to obtain the defective position, then extract an image from the first pole piece image according to the defective position to obtain a defective image, and further classify the defective image through a preset defective classification model to obtain the defective type of the defective image.

[0098] In this embodiment, this step specifically includes: comparing each of the plurality of difference values with a preset threshold, aggregating a plurality of second pixels corresponding to the difference values greater than the threshold, constructing a first bounding box according to the second pixels, determining the coordinates of the first bounding box, and obtaining the defective position of the first pole piece image according to the coordinates, so as to extract the defective image in the first pole piece image according to the defective position, and identify the defective image through a trained classification model to obtain the type of the defective image.

[0099] In this embodiment, this step is specifically: In the Figure 4In the provided industrial control computer, several differences obtained according to the image difference are respectively compared with a preset threshold. Pixels corresponding to differences greater than the threshold are set to 1, and pixels corresponding to differences less than the preset threshold are set to 0. Among them, a minimum bounding box is drawn for the set of pixels set to 1, the coordinates of the bounding box are obtained, and the defect heat map is segmented according to the coordinates to obtain the defect position. The defect image is extracted according to the position to enable image classification work. According to the defect image, the type of the defect is identified through a pre-trained classification model to obtain the defect type corresponding to the defect image. The defect type includes various types such as scratches, black spots, and metal leakage.

[0100] In this embodiment, the training process of the classification model is as follows: The initial classification model pre-constructed is trained according to the classified and labeled defect images obtained above. The initial classification model is a deep learning method. The initial classification model is continuously trained according to the classified defect images so that the classification model learns the features of each defect type, and the parameters of the initial classification model are continuously optimized using a preset loss function to obtain the trained classification model.

[0101] The preset loss function is:

[0102]

[0103] When the industrial control computer detects a defective pole piece, the encoder predicts and delays the position of the pole piece. According to the position of the defective pole piece judged by the encoder, a label is attached at the labeling position, and the pole piece with the defective label is removed in the subsequent process.

[0104] The present invention proposes a method for detecting pole piece defects aiming at the surface defects of pole pieces. This method uses deep learning technology to construct an image reconstruction algorithm and a classification model, reconstructs the defective image into an image close to the positive sample image, subtracts the reconstructed image from the defective image to obtain a defect score map, and performs threshold segmentation on the defective image to obtain the final defect localization image. This method solves the problem of difficult acquisition of defective images in deep learning, has a good recognition effect on new defects, and is different from traditional deep learning algorithms that have poor recognition of unknown defects or some defects with less data. Traditional algorithms require manual design of classifiers and need algorithm designers to have good professional knowledge and computer literacy. This method only needs to train the image reconstruction algorithm with positive sample images to complete the defect detection problem, has no requirements for defect data, and can reduce the personnel requirements.

[0105] The specific embodiments described above further elaborate on the object, technical solution and beneficial effects of the present invention. It should be understood that the above description is only for the specific embodiments of the present invention and is not used to limit the protection scope of the present invention. In particular, it is pointed out that for those skilled in the art, any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for detecting defects on a pole piece, characterized in that, Including: Obtain first image data including the electrode to be measured, and calculate the image edge of the first image data, so as to segment the first image data according to the image edge to obtain a first electrode image of the electrode to be measured; Perform image reconstruction on the first electrode image to obtain a second electrode image after image reconstruction, and perform image difference between the second electrode image and the first electrode image to obtain the difference between the first electrode image and the second electrode image and the defect heat map corresponding to the first electrode image; wherein, by performing image difference between the second electrode image and the first electrode image, respectively obtain a plurality of differences between a plurality of second pixels in the second electrode image and a plurality of corresponding first pixels in the first electrode image, and construct a difference image according to the plurality of differences; obtain the defect heat map corresponding to the first electrode image by performing Gaussian filtering on the difference image; Segment the defect heat map according to the difference to obtain the defect position in the defect heat map, extract the defect image corresponding to the first electrode image according to the defect position, and classify the defect image to obtain the defect type of the electrode to be measured; wherein, compare each of the plurality of differences with a preset threshold, aggregate a plurality of second pixels corresponding to the differences greater than the threshold, and construct a first bounding box according to the second pixels; determine the coordinates of the first bounding box, and obtain the defect position of the first electrode image according to the coordinates, so as to extract the defect image in the first electrode image according to the defect position.

2. The method for detecting defects of a pole piece according to claim 1, characterized in that, The obtaining of the first image data including the electrode to be measured specifically includes: Perform supplementary lighting on the electrode to be measured through a preset light source, so as to obtain image data including the electrode to be measured according to a preset image acquisition device; the image data includes the upper surface image and the lower surface image of the electrode to be measured; Obtain the first image data by performing Gaussian smoothing processing on the image data.

3. The method for detecting defects of a pole piece according to claim 1, characterized in that, The calculating of the image edge of the first image data, so as to segment the first image data according to the image edge to obtain a first electrode image of the electrode to be measured, specifically includes: Perform edge detection on the first image data by using a preset edge detection algorithm to obtain the image edge between the electrode to be measured and the background in the first image data; Perform edge segmentation on the first image data according to the image edge to obtain a first electrode image of the electrode to be measured.

4. The method for detecting defects of a pole piece according to claim 1, characterized in that, The obtaining of the difference between the first electrode image and the second electrode image and the defect heat map corresponding to the first electrode image further includes: Perform image reconstruction on the first electrode image according to a preset image reconstruction algorithm to obtain a second electrode image after reconstruction of the first electrode image.

5. The method for detecting defects of a pole piece according to claim 1, wherein, The classifying of the defect image to obtain the defect type of the electrode to be measured specifically includes: Recognize the defect image according to the defect image through a trained classification model to obtain the type of the defect image.

6. A device for detecting defects of a pole piece, characterized in that, Including an image segmentation module, an image reconstruction module and a defect detection module; The image segmentation module is used to obtain first image data containing the electrode tab to be measured, and calculate the image edge of the first image data, so as to segment the first image data according to the image edge to obtain a first electrode tab image of the electrode tab to be measured; The image reconstruction module is used to perform image reconstruction on the first electrode tab image to obtain a second electrode tab image after image reconstruction, and perform image difference between the second electrode tab image and the first electrode tab image to obtain the difference between the first electrode tab image and the second electrode tab image and the defect heat map corresponding to the first electrode tab image; wherein, the image reconstruction module includes a difference unit and a processing unit; the difference unit is used to perform image difference between the second electrode tab image and the first electrode tab image to respectively obtain a plurality of differences between a plurality of second pixels in the second electrode tab image and a plurality of corresponding first pixels in the first electrode tab image, and construct a difference image according to the plurality of differences; the processing unit is used to obtain the defect heat map corresponding to the first electrode tab image by performing Gaussian filtering on the difference image; The defect detection module is used to segment the defect heat map according to the difference to obtain the defect position in the defect heat map, extract the defect image corresponding to the first electrode tab image according to the defect position, and classify the defect image to obtain the defect type of the electrode tab to be measured; wherein, the defect detection module includes a positioning unit and an extraction unit; the positioning unit is used to compare each of the plurality of differences with a preset threshold, aggregate a plurality of second pixels corresponding to the differences greater than the threshold, and construct a first bounding box according to the second pixels; the extraction unit is used to determine the coordinates of the first bounding box, and obtain the defect position of the first electrode tab image according to the coordinates, so as to extract the defect image in the first electrode tab image according to the defect position.

7. The anode sheet defect detection device according to claim 6, characterized in that, The image segmentation module includes a collection unit, an edge calculation unit and a segmentation unit The collection unit is used to supplement light to the electrode tab to be measured through a preset light source, so as to obtain image data containing the electrode tab to be measured according to a preset image acquisition device; the image data includes the upper surface image and the lower surface image of the electrode tab to be measured; the first image data is obtained by performing Gaussian smoothing processing on the first image data; The edge calculation unit is used to perform edge detection on the first image data by using a preset edge detection algorithm to obtain the image edge between the electrode tab to be measured and the background in the first image data; The segmentation unit is used to perform edge segmentation on the first image data according to the image edge to obtain a first electrode tab image of the electrode tab to be measured.

8. The polar plate defect detection device according to claim 6, characterized in that, The image reconstruction module further includes an image unit; The image unit is used to perform image reconstruction on the first electrode tab image according to a preset image reconstruction algorithm to obtain a second electrode tab image after reconstruction of the first electrode tab image.

9. The a kind of pole piece defect detection device according to claim 6, characterized in that, The defect detection module further includes a classification unit; The taxonomic unit is used to identify the defect image according to the defect image through a trained classification model to obtain the type of the defect image.

Images