Method for constructing GAN-based defect detection model for pole pieces of blade battery

Abstract

Disclosed in the present invention is a method for constructing a GAN-based defect detection model for pole pieces of a blade battery. The method comprises the following steps: collecting several images of defective target pole pieces; pre-processing the images to obtain pre-processed images, and extracting valid defective regions from the images; acquiring contour information for the valid regions, and accurately classifying the contour information on the basis of characteristic parameters; performing data augmentation on classified data, so as to obtain an augmented dataset; and in the dataset, using defect types and position information as labels to train a neural network, and using a neural network model as a defect detection model for pole pieces of a blade battery. In the present invention, a dataset that has undergone data augmentation is inputted into a network as a training set, such that the problem of a severe shortage of training samples caused by numerous types of battery pole piece defects and low occurrence probabilities of individual samples is solved, and the detection accuracy of a defect detection model for pole pieces of a battery is greatly improved, thereby enabling the rapid and accurate detection of various defects and position information of pole pieces of a blade battery.

Description

GAN-based blade battery electrode sheet defect detection model construction method TECHNICAL FIELD

[0001] The application belongs to the field of battery defect detection, and more particularly relates to a GAN-based blade battery electrode sheet defect detection model construction method. BACKGROUND

[0002] The reliability and performance of blade batteries are crucial for the growing market of new energy vehicles, and in this context, ensuring the quality of battery electrode sheets has become a key challenge in the manufacturing process. However, during the manufacturing process of lithium batteries, various factors can cause defects such as scratches and dark spots on the battery electrode sheet, which affect the performance and lifespan of the battery, and in severe cases can even cause short circuits and fires and other safety accidents.

[0003] Now there are methods using deep learning to realize the detection of battery electrode sheet defects, but traditional deep learning technology needs to rely on a large number of manually labeled sample data to train the detection model, and the noise of the electrode sheet is high, the defect sample types are many, and the probability of individual samples is low, which leads to the problem of serious lack of training samples. At the same time, the defect sample is usually small in size and lacks appearance features, increasing the difficulty of manual labeling and easily introducing labeling errors. SUMMARY

[0004] The application provides a GAN-based blade battery electrode sheet defect detection model construction method, which aims to realize automatic, accurate and rapid detection of blade battery defects under small sample conditions.

[0005] The application provides a GAN-based blade battery electrode sheet defect detection model construction method, which includes the following steps:

[0006] S1, a plurality of target electrode sheet images with defects are collected;

[0007] S2, the collected images are preprocessed to obtain preprocessed images to extract the effective area with defects in the images;

[0008] S3, the extracted effective area is used to obtain contour information and accurately classify the defects in the effective area according to the feature parameters;

[0009] S4, the classified image data is subjected to data augmentation to obtain an augmented image data set;

[0010] S5, in the data set, the defect type and location information are taken as labels to form a training set, and the neural network is trained through the training set, and the neural network model after training is taken as a defect detection model of the blade battery pole piece.

[0011] Further, in step S2, the collected image is preprocessed, including:

[0012] The collected image is converted into a gray-scale image and gamma enhanced to improve the extractability of defects; the gray-scale image after gamma enhancement is binarized, and the binarized image is subjected to open operation and close operation to remove the influence of small objects and connect the disconnected defects; the binarized image after open operation and close operation is subjected to pixel traversal to extract the effective area containing defects.

[0013] Further, in step S3, the contour information of the extracted effective area is obtained and accurately classified according to the characteristic parameters, including:

[0014] The extracted effective area containing defects is subjected to region filling, the contour information is obtained according to the filled image, and the defects in the effective area are classified according to the characteristic parameters based on the contour information;

[0015] The characteristic parameters include the aspect ratio of the defect contour, the circularity, the area of the defect, the gray value and the gray mean value of the defect.

[0016] Further, in step S4, the data enhancement includes:

[0017] The classified data is subjected to cropping and random flipping processing;

[0018] The data after cropping and random flipping processing is subjected to a tab defect data enhancement method based on a generative adversarial network GAN to obtain a new training data set.

[0019] Further, the data after cropping and random flipping processing is subjected to a tab defect data enhancement method based on a generative adversarial network GAN to obtain a new training data set, including:

[0020] A target defect image is generated based on an improved GAN generative adversarial network to expand the training data set, the improved GAN generative adversarial network is based on the bottom layer architecture of DCGAN, a multi-scale residual block and a CA module are embedded in the generator G of the network to realize enhancement of feature channels and spatial attention; a spectral normalization module is embedded in the discriminator D of the network to realize stability of the discriminator training and improvement of the quality of the generated image.

[0021] Further, the network generator G is composed of a multi-layer convolutional structure, including an up-sampling layer, a fully connected layer, a multi-layer convolution, a multi-scale residual block and a CA module.

[0022] Further, the network discriminator D includes a series of spectral normalization layers, convolutional layers and random dropout layers, and Mish functions and Dropout layers.

[0023] Further, the network loss function is a HingeLoss loss function.

[0024] Further, in step S5, the neural network is a YOLOV5s network.

[0025] The application also claims to protect a GAN-based blade battery pole piece defect detection model, which is constructed by the above-mentioned GAN-based blade battery pole piece defect detection model construction method.

[0026] Advantage: The GAN-based blade battery pole piece defect detection model construction method provided by the application includes collecting a plurality of target pole piece images with defects; the collected images are preprocessed to extract effective regions with defects in the images; contour information is obtained for the extracted effective regions and accurately classified according to characteristic parameters; the classified data is data enhanced to obtain an enhanced data set; in the data set, the defect type and position information are taken as labels to form a training set, and a neural network is trained by the training set to obtain a defect detection model of the blade battery pole piece, which can quickly and accurately detect a plurality of defects and position information of the blade battery pole piece.

[0027] The GAN-based pole piece defect data enhancement method in the application realizes enhancement of feature channels and spatial attention by embedding a multi-scale residual block and a CA module in the generator G of the DCGAN bottom-layer architecture; and realizes stability of discriminator training and improvement of generated image quality by embedding a spectral normalization module in the discriminator D. The method solves the problem of serious shortage of training samples caused by multiple noises of the battery pole piece, multiple defect sample types and low probability of occurrence of individual samples, and realizes small-sample training of the blade battery pole piece detection model. BRIEF DESCRIPTION OF DRAWINGS

[0028] Fig. 1 is a flowchart of a GAN-based blade battery pole piece defect detection model construction method according to an embodiment of the application;

[0029] Fig. 2 is a schematic diagram of a blade battery pole piece image acquisition system according to an embodiment of the application;

[0030] Fig. 3 is a structure diagram of an improved GAN generator G according to an embodiment of the application;

[0031] FIG. 4 is a structure diagram of a discriminator D based on an improved GAN generated adversarial network according to an embodiment of the present application. DETAILED DESCRIPTION

[0032] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and should not be used to limit the present application. In addition, the technical features involved in the various embodiments of the present application described below can be combined with each other as long as they do not conflict with each other.

[0033] Referring to FIG. 1, the method for constructing a GAN-based blade battery pole piece defect detection model according to an embodiment of the present application includes the following steps:

[0034] S1, collecting a plurality of target pole piece images with defects;

[0035] Further, the image acquisition system for blade battery pole piece images is used for image acquisition according to the embodiment of the present application, as shown in FIG. 2, the system includes a camera control unit, a CCD camera, a light source control unit, a light source, a pole piece conveying mechanism, and a pole piece overturning mechanism. The pole piece conveying mechanism is used to transport the pole piece to be detected to the required position, and the camera control unit controls the camera to cooperate with the light source controlled by the light source control unit to acquire the single-side image of the pole piece. After completion, the pole piece overturning mechanism is used to overturn the pole piece, so as to realize the acquisition of the image of the other side of the pole piece.

[0036] S2, preprocessing the collected images to obtain preprocessed images, so as to extract the effective area with defects in the images;

[0037] S21, converting the collected images into grayscale images;

[0038] S22, performing gamma enhancement on the grayscale images to improve the extractability of defects;

[0039] S23, performing binaryzation processing on the gamma-enhanced grayscale images;

[0040] S24, performing open-close operation on the binaryzation images to remove the influence of small objects and connect the disconnected defects between adjacent defects;

[0041] S25, performing pixel traversal on the binaryzation images processed by open operation and close operation to extract the effective area containing defects.

[0042] In some embodiments, step S24, an opening and closing operation is performed on the binary image, the opening operation is a combination operation of erosion followed by dilation, and the closing operation is a combination operation of dilation followed by erosion. According to the actual size of the image, different structure elements B and different opening and closing operation times are selected. The principle is to first use a large structure element to perform several rounds of closing operation to connect multiple regions that may be slightly far apart but still belong to the same defect influence range; then, a small structure element is used to perform several rounds of opening operation to remove small islands that do not affect the judgment of the region; finally, a small structure element is used to perform several rounds of closing operation to reconnect the fine connections of the regions that may have been disconnected in the last opening operation.

[0043] S3, obtaining contour information for the extracted effective area and accurately classifying defects in the effective area according to feature parameters;

[0044] performing region filling on the extracted effective area containing defects, obtaining contour information according to the filled image, and classifying defects in the effective area according to feature parameters based on the contour information;

[0045] The feature parameters include the aspect ratio of the defect contour, the circularity, the area of the defect, the gray value and the gray mean value of the defect.

[0046] S4, data augmentation is performed on the classified data to obtain an augmented data set;

[0047] S41, the classified data is cropped and randomly flipped;

[0048] S42, the data after cropping and random flipping is processed using a GAN-based polar slice defect data augmentation method to obtain a new training data set.

[0049] Further, the GAN-based polar slice defect data augmentation method in step S42 includes:

[0050] A target defect image is generated based on an improved GAN for expanding the training data set. The improved GAN-based generative adversarial network is based on the underlying architecture of DCGAN, and multi-scale residual blocks and CA modules are embedded in the generator G of the network to enhance feature channels and spatial attention. The spectral normalization module is embedded in the discriminator D of the network to realize the stability of the discriminator training and the improvement of the quality of the generated image.

[0051] In some embodiments, the generator G structure in the improved GAN-based generative adversarial network is shown in Figure 3, which is composed of a multi-layer convolutional structure and combines up-sampling layers, fully connected layers, multi-layer convolutions, multi-scale residual blocks and CA modules. The generation process starts with a fully connected layer, then gradually increases the spatial dimension and reduces the depth of the feature map through batch normalization, up-sampling and convolution layers, and finally generates the target image by applying multi-scale residual blocks and CA modules for feature optimization.

[0052] The structure of the discriminator D is shown in Figure 4, which is composed of a series of spectral normalization layers, convolution layers and random dropout layers. An image with a fixed size is input, and the depth is processed through the fully connected layers of multiple convolution layers. In addition, batch normalization, Mish nonlinear activation function and Dropout layer are added to prevent overfitting. Finally, the Flatten layer and a fully connected layer output a vector, which ensures the model's ability to identify the authenticity of the generated image.

[0053] In some embodiments, the loss function of the generative adversarial network is a HingeLoss loss function.

[0054] S5, in the data set, the defect type and position information are taken as labels to form a training set, and the neural network is trained through the training set, and the trained neural network model is taken as a defect detection model of the blade battery pole piece.

[0055] Further, in step S5, the neural network is a YOLOV5s network.

[0056] The application also claims to protect a GAN-based blade battery pole piece defect detection model, which is constructed by the above-mentioned GAN-based blade battery pole piece defect detection model construction method.

[0057] Those skilled in the art will readily understand that the above description is only of the preferred embodiments of the present application and is not intended to limit the present application, and any modifications, equivalent replacements and improvements made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for constructing a GAN-based blade battery electrode tab defect detection model, characterized in that, The method comprises the following steps: S1, collecting a plurality of target electrode sheet images with defects; S2, preprocessing the collected images to obtain preprocessed images to extract effective regions with defects in the images; S3, obtaining contour information for the extracted effective regions and accurately classifying defects in the effective regions according to characteristic parameters; S4, performing data enhancement on the classified image data to obtain an enhanced image data set; S5, forming a training set by taking defect types and position information as labels in the data set, training a neural network through the training set, and taking the trained neural network model as a defect detection model for the electrode sheet of the blade battery.

2. The GAN-based blade battery electrode tab defect detection model construction method of claim 1, wherein, Step S2, preprocessing the collected images comprises: Converting the collected images into grayscale images and performing gamma enhancement to improve the extractability of defects; performing binaryzation processing on the gamma-enhanced grayscale images, performing open operation and close operation on the binaryzation images to remove the influence of small objects and connect the disconnected defects; and performing pixel traversal on the binaryzation images processed by open operation and close operation to extract the effective regions containing defects.

3. The GAN-based blade battery electrode tab defect detection model construction method of claim 1, wherein, Step S3, obtaining contour information for the extracted effective regions and accurately classifying defects in the effective regions according to characteristic parameters comprises: Performing region filling on the extracted effective regions containing defects, obtaining contour information according to the filled images, and classifying defects in the effective regions according to characteristic parameters based on the contour information; The characteristic parameters include the aspect ratio of the defect contour, the circularity, the area of the defect, the gray value and the gray mean value of the defect.

4. The GAN-based blade battery electrode tab defect detection model construction method of claim 1, wherein, Step S4, the data enhancement comprises: Performing cropping and random flipping processing on the classified data; Using a GAN-based electrode sheet defect data enhancement method to obtain a new training data set from the data processed by cropping and random flipping. 5.The GAN-based blade battery electrode tab defect detection model construction method of claim 4, wherein, The GAN-based electrode sheet defect data enhancement method for the data processed by cropping and random flipping comprises: Generating target defect images based on an improved GAN to expand the training data set, wherein the improved GAN is based on the underlying architecture of DCGAN, a multi-scale residual block and a CA module are embedded in the generator G of the network to enhance the feature channel and spatial attention; a spectral normalization module is embedded in the discriminator D of the network to realize the stability of the discriminator training and the improvement of the generated image quality. 6.The GAN-based blade battery electrode tab defect detection model construction method of claim 5, wherein, The network generator G is composed of a multi-layer convolutional structure, including an up-sampling layer, a full connection layer, a multi-layer convolution, a multi-scale residual block and a CA module. 7.The GAN-based blade battery electrode tab defect detection model construction method of claim 5, wherein, The network discriminator D includes a series of spectral normalization layers, convolutional layers and random dropout layers, as well as Mish functions and Dropout layers.

8. The improved GAN-based generative adversarial network of claim 5, wherein, The network loss function is a HingeLoss loss function. 9.The GAN-based blade battery electrode tab defect detection model construction method of claim 5, wherein, Step S5, the neural network is a YOLOV5s network. 10.A GAN-based model for detecting defects of a blade battery electrode tab, characterized in that, The detection model is constructed by the GAN-based electrode sheet defect detection model construction method according to any one of claims 1-9.

Images