CN114882021A

CN114882021A - Efficient processing method and system for battery lithium film

Info

Publication number: CN114882021A
Application number: CN202210791330.1A
Authority: CN
Inventors: 韦新松; 张勇; 洪布双; 高昂
Original assignee: Jiangsu Zhongqing Advanced Battery Manufacturing Co ltd
Current assignee: Jiangsu Zhongqing Advanced Battery Manufacturing Co ltd
Priority date: 2022-07-07
Filing date: 2022-07-07
Publication date: 2022-08-09
Anticipated expiration: 2042-07-07
Also published as: CN114882021B

Abstract

The application relates to the field of intelligent manufacturing, in particular to a high-efficiency processing method of a battery lithium film and a system thereof, the method comprises the steps of collecting images of a battery lithium film passing through a roller shaft by an area-array camera, processing the lithium film images by using a trained generator model serving as a reflective filter, so as to remove the reflection effect in the original image and ensure that the characteristic distribution of the lithium film generated image in the target domain and the characteristic distribution of the lithium film image in the source domain keep higher consistency, and then, encoding the lithium film generation image by using a convolutional neural network as a feature extractor to extract local high-dimensional implicit features of the lithium film generation image, and further generating a feature representation containing irregular topological information and high-dimensional lithium film image information by using learnable neural network parameters through a graph neural network so as to obtain a more accurate classification result. Thus, the detection accuracy of the battery separator can be improved.

Description

Efficient processing method and system for battery lithium film

Technical Field

The present application relates to the field of smart manufacturing, and more particularly, to a method for efficiently processing a lithium film of a battery and a system thereof.

Background

In the construction of lithium batteries, the separator is one of the key internal components. The performance of the diaphragm determines the interface structure, internal resistance and the like of the battery, directly influences the capacity, circulation, safety performance and other characteristics of the battery, and the diaphragm with excellent performance plays an important role in improving the comprehensive performance of the battery. The separator has a main function of separating the positive electrode and the negative electrode of the battery to prevent short circuit caused by contact between the two electrodes, and also has a function of allowing electrolyte ions to pass.

It will be appreciated that the key to efficient production of lithium battery separators is the ability to improve the accuracy of detection of the separator. At present, lithium battery diaphragm detects the diaphragm to lithium battery diaphragm detection machine commonly used in the market, and lithium battery diaphragm detection machine includes line array camera structure and light source structure, and line array camera is used for shooing the image of lithium battery diaphragm when the roller, and the light source structure is used for providing light when shooing for line array camera, and current lithium battery diaphragm detection machine has following technical problem: (1) the position of a linear array camera is fixed, when a certain small-angle offset deviation exists in the transmission of a lithium battery diaphragm, the conventional linear array camera cannot be rotated and adjusted by a corresponding small angle, so that the normal shooting of images is influenced, and the detection precision is influenced; (2) the light source of the linear array camera is directly opposite to the surface of the lithium battery diaphragm, so that the light reflection phenomenon is serious, the image shooting quality of the linear array camera is influenced, and the detection precision is influenced.

Therefore, an optimized lithium battery separator detector is expected to improve the detection precision of the lithium battery separator so as to improve the processing efficiency of the lithium battery.

Disclosure of Invention

The present application is proposed to solve the above-mentioned technical problems. The embodiment of the application provides a high-efficiency processing method and a system of a battery lithium film, wherein an area array camera is used for collecting an image of the battery lithium film passing through a roll shaft, a generator model which is trained and used as a reflective filter is used for processing the lithium film image so as to remove a reflective effect in an original image and enable the feature distribution of the lithium film generated image in a target domain to keep higher consistency with the feature distribution of the lithium film image in a source domain, then a convolutional neural network is used as a feature extractor for coding the lithium film generated image so as to extract local high-dimensional implicit features of the lithium film generated image, and further a characteristic expression containing irregular topology information and high-dimensional lithium film image information is generated through learnable neural network parameters through a graph neural network so as to obtain a more accurate classification result. Thus, the detection accuracy of the battery separator can be improved.

According to one aspect of the present application, there is provided a method for efficiently processing a lithium film for a battery, comprising: obtaining a plurality of lithium film images of a battery to be detected when a lithium film passes through a roller shaft by an area array camera; respectively passing the plurality of lithium film images through a generator model serving as a light reflecting filter to obtain a plurality of lithium film generation images; respectively enabling each lithium film generation image in the plurality of lithium film generation images to pass through a first convolution neural network to obtain a plurality of lithium film eigenvectors, and two-dimensionally arranging the plurality of lithium film eigenvectors into a lithium film eigenvector matrix according to a sample dimension; enabling the topological matrix of the area-array camera to pass through a second convolutional neural network to obtain a topological characteristic matrix, wherein characteristic values of positions on non-diagonal positions in the topological matrix of the area-array camera are used for representing the distance between the two corresponding cameras, and the characteristic values of the positions on the diagonal positions in the topological matrix of the area-array camera are zero; passing the topological characteristic matrix and the lithium film characteristic matrix through a neural network to obtain a lithium film topological characteristic matrix containing irregular topological information and high-dimensional lithium film image information; and enabling the lithium film topological characteristic matrix to pass through a classifier to obtain a classification result, wherein the classification result is used for indicating whether the processing quality of the lithium film of the battery to be detected is qualified or not.

According to another aspect of the present application, there is provided a system for efficient processing of a lithium film for a battery, comprising: the lithium film image acquisition unit is used for acquiring a plurality of lithium film images of the battery to be detected when the lithium film passes through the roller shaft through the area array camera; a generated image acquisition unit, configured to pass the plurality of lithium film images obtained by the lithium film image acquisition unit through a generator model as a light reflection filter, respectively, to obtain a plurality of lithium film generated images; the first convolution unit is used for enabling each lithium film generation image in the plurality of lithium film generation images obtained by the generation image obtaining unit to pass through a first convolution neural network respectively to obtain a plurality of lithium film characteristic vectors, and arranging the plurality of lithium film characteristic vectors into a lithium film characteristic matrix in a two-dimensional mode according to the sample dimension; the second convolution unit is used for enabling the topological matrix of the area-array camera to pass through a second convolution neural network so as to obtain a topological characteristic matrix, wherein characteristic values of positions on non-diagonal positions in the topological matrix of the area-array camera are used for representing the distance between two corresponding cameras, and the characteristic values of the positions on the diagonal positions in the topological matrix of the area-array camera are zero; the graph neural network unit is used for enabling the topological characteristic matrix obtained by the second convolution unit and the lithium film characteristic matrix obtained by the first convolution unit to pass through a graph neural network so as to obtain a lithium film topological characteristic matrix containing irregular topological information and high-dimensional lithium film image information; and the classification unit is used for enabling the lithium film topological characteristic matrix obtained by the graph neural network unit to pass through a classifier to obtain a classification result, and the classification result is used for indicating whether the processing quality of the lithium film of the battery to be detected is qualified or not.

Compared with the prior art, the application provides a high-efficiency processing method and a system for a battery lithium film, the method comprises the steps of collecting images of a battery lithium film passing through a roller shaft by an area-array camera, processing the lithium film images by using a trained generator model serving as a reflective filter, so as to remove the reflection effect in the original image and ensure that the characteristic distribution of the lithium film generated image in the target domain and the characteristic distribution of the lithium film image in the source domain keep higher consistency, and then, encoding the lithium film generation image by using a convolutional neural network as a feature extractor to extract local high-dimensional implicit features of the lithium film generation image, and further generating a feature representation containing irregular topological information and high-dimensional lithium film image information by using learnable neural network parameters through a graph neural network so as to obtain a more accurate classification result. Thus, the detection accuracy of the battery separator can be improved.

Drawings

The above and other objects, features and advantages of the present application will become more apparent by describing in more detail embodiments of the present application with reference to the attached drawings. The accompanying drawings are included to provide a further understanding of the embodiments of the application and are incorporated in and constitute a part of this specification, illustrate embodiments of the application and together with the description serve to explain the principles of the application. In the drawings, like reference numbers generally represent like parts or steps.

Fig. 1 is a view of an application scenario of a method for efficiently processing a lithium film of a battery according to an embodiment of the present application.

Fig. 2 is a flow chart of a method for efficient processing of a lithium film of a battery according to an embodiment of the present application.

Fig. 3 is a schematic diagram of an architecture of a method for efficiently processing a lithium film of a battery according to an embodiment of the present application.

Fig. 4 is a flowchart of a training process of a generator model as a reflective filter in a method for efficiently processing a lithium battery membrane according to an embodiment of the present application.

Fig. 5 is a schematic diagram of a training process of a generator model as a reflective filter in a method for efficiently processing a lithium battery membrane according to an embodiment of the present application.

Fig. 6 is a block diagram of a system for efficient processing of a lithium film of a battery according to an embodiment of the present application.

Fig. 7 is a block diagram of a training module as a generator model of a reflectorized filter in a system for efficient processing of battery lithium membranes according to an embodiment of the present application.

Detailed Description

Hereinafter, example embodiments according to the present application will be described in detail with reference to the accompanying drawings. It should be understood that the described embodiments are only some embodiments of the present application and not all embodiments of the present application, and that the present application is not limited by the example embodiments described herein.

Overview of the scenario as described above, the separator is one of the key internal layer components in the construction of lithium batteries. The performance of the diaphragm determines the interface structure, internal resistance and the like of the battery, directly influences the capacity, circulation, safety performance and other characteristics of the battery, and the diaphragm with excellent performance plays an important role in improving the comprehensive performance of the battery. The separator has a main function of separating the positive electrode and the negative electrode of the battery to prevent short circuit due to contact between the two electrodes, and also has a function of allowing electrolyte ions to pass therethrough.

Accordingly, in the technical scheme of the application, the inventor replaces a linear camera with an area-array camera to acquire images of the detected diaphragm from more shooting angles. In order to solve the problem of detection accuracy reduction caused by transmission offset during lithium battery diaphragm transmission, in the technical scheme of the application, a convolutional neural network model is used as a feature extractor to encode a diaphragm image, wherein the convolutional neural network is insensitive to offset and rotation of an object in the image. Aiming at the problem of light reflection during image acquisition, the inventor trains an image generator capable of removing the influence of light reflection through the idea of countertraining. In turn, the detection accuracy of the battery diaphragm is improved, thereby improving the processing efficiency of the lithium battery.

Specifically, in the technical scheme of the application, a plurality of lithium film images of a to-be-detected battery lithium film passing through a roller shaft are obtained through an area-array camera. Then, the plurality of lithium film images are respectively passed through a generator model as a light reflecting filter to obtain a plurality of lithium film generation images. In an embodiment of the present application, the training process of the generator model includes: and respectively enabling the training image and the reference image to pass through the generator model serving as the reflecting filter to obtain a first characteristic matrix and a second characteristic matrix, wherein the training image is a camera image acquired by a camera facing a light source, and the reference image is a camera image to be replaced. In a specific example of the present application, the generator model is a deep convolutional neural network model. The generator model may then be trained by computing discriminator loss function values between the first feature matrix and the second feature matrix to "fool" features learned by the generator model from the training images more toward features of the reference image.

However, when training the first convolutional neural network by a discriminator loss function between the first feature matrix and the second feature matrix, since the discriminator loss function is mainly consistent from the perceptual aspect of the image appearance, i.e., as a perceptual loss function, it is also desirable that the first feature matrix and the second feature matrix are as consistent as possible in feature distribution.

Therefore, a smooth motion matrix of the first feature matrix relative to the second feature matrix is calculated as the modified first feature matrix, that is:

wherein

Representing the first feature matrix in a first order,

representing the second feature matrix in a second order,

representing the modified first feature matrix,

expressing exponential operation with the matrix as power, wherein the exponential operation with the matrix as power expresses that the value of each position of the matrix is used as power exponent, and then filling the result into each position of the matrix to obtain the matrix operation result,

and

respectively represent subtraction and addition by position of the matrix, an

Representing a dot multiplication of a number with a matrix,

is a hyper-parameter controlling the posterior weight.

The first feature matrix

As a posterior feature matrix, by basing it on a posterior feature, i.e. a first feature matrix

With respect to a prior feature, i.e. a second feature matrix

The smooth motion model of (2) represents transition information of image semantics brought by different light source angles, thereby realizing posterior feature distribution expressed as prior hidden feature distribution

For prior feature distribution

So that the first feature matrix is more smoothly responsive

And a second feature matrix

As consistent as possible in feature distribution. In this way, when the trained generator model is used for processing the lithium film image, not only can the light reflection effect in the original image be removed, but also the characteristic distribution of the lithium film generated image in the target domain and the characteristic distribution of the lithium film image in the source domain are kept to be higher in consistency.

Then, enabling each lithium film generation image in the plurality of lithium film generation images to pass through a first convolution neural network respectively to obtain a plurality of lithium film characteristic vectors, and arranging the plurality of lithium film characteristic vectors into a lithium film characteristic matrix in a two-dimensional mode according to the sample dimension. That is, the lithium film generation image is encoded using a convolutional neural network as a feature extractor to extract local high-dimensional implicit features of the lithium film generation image. It is worth mentioning that the convolutional neural network has insensitivity to position in terms of feature extraction, that is, insensitivity to rotation and offset of the lithium film object in the lithium film generated image, and therefore, even if there is a positional offset at the time of lithium film transmission, it does not adversely affect the detection accuracy.

It should be noted that, in the technical solution of the present application, the cameras for acquiring the lithium film images are arranged in an array (i.e., the area-array camera), so that the lithium film images acquired by the cameras in the area-array camera have a predetermined spatial topological relationship, and fusing this information is beneficial to improving the detection accuracy. Specifically, the topological characteristic matrix is obtained by passing the topological matrix of the area-array camera through a second convolutional neural network, wherein the characteristic value of each position at a non-diagonal position in the topological matrix of the area-array camera is used for representing the distance between the two corresponding cameras, and the characteristic value of each position at a diagonal position in the topological matrix of the area-array camera is zero. And then, passing the topological characteristic matrix and the lithium film characteristic matrix through a neural network to obtain a lithium film topological characteristic matrix containing irregular topological information and high-dimensional lithium film image information. Wherein the graph convolution neural network generates a feature representation containing irregular topology information and high-dimensional lithium film image information through learnable neural network parameters. And finally, passing the lithium film topological characteristic matrix through a classifier to obtain a classification result, wherein the classification result is used for indicating whether the processing quality of the lithium film of the battery to be detected is qualified or not.

Based on this, the present application proposes a method for efficiently processing a lithium film of a battery, which includes: obtaining a plurality of lithium film images of a battery to be detected when a lithium film passes through a roller shaft by an area array camera; respectively passing the plurality of lithium film images through a generator model serving as a light reflecting filter to obtain a plurality of lithium film generation images; respectively enabling each lithium film generation image in the plurality of lithium film generation images to pass through a first convolution neural network to obtain a plurality of lithium film eigenvectors, and two-dimensionally arranging the plurality of lithium film eigenvectors into a lithium film eigenvector matrix according to a sample dimension; enabling the topological matrix of the area-array camera to pass through a second convolutional neural network to obtain a topological characteristic matrix, wherein characteristic values of positions on non-diagonal positions in the topological matrix of the area-array camera are used for representing the distance between the two corresponding cameras, and the characteristic values of the positions on the diagonal positions in the topological matrix of the area-array camera are zero; passing the topological characteristic matrix and the lithium film characteristic matrix through a neural network to obtain a lithium film topological characteristic matrix containing irregular topological information and high-dimensional lithium film image information; and enabling the lithium film topological characteristic matrix to pass through a classifier to obtain a classification result, wherein the classification result is used for indicating whether the processing quality of the lithium film of the battery to be detected is qualified or not.

Fig. 1 illustrates an application scenario of a method for efficiently processing a lithium film of a battery according to an embodiment of the present application. As shown in fig. 1, in the training phase of the application scenario, first, a camera image of a lithium battery diaphragm (e.g., D as illustrated in fig. 1) passing through a roller (e.g., R as illustrated in fig. 1) is acquired by a camera (e.g., C as illustrated in fig. 1) facing a light source (e.g., L as illustrated in fig. 1), and a camera image that needs to be replaced is acquired as a reference image. Then, the obtained camera image and the reference image are input into a server (e.g., a cloud server S as illustrated in fig. 1) in which a high efficiency processing algorithm for a battery lithium film is deployed, wherein the server is capable of training the generator model as a reflective filter of a high efficiency processing method for a battery lithium film with the camera image and the reference image.

After the training is completed, in the inspection stage, first, a plurality of lithium film images of the battery to be inspected (e.g., D as illustrated in fig. 1) as it passes through the roll shaft (e.g., R as illustrated in fig. 1) are obtained by the area-array camera (e.g., C as illustrated in fig. 1). Then, the obtained multiple lithium film images are input into a server (for example, S as illustrated in fig. 1) deployed with a high-efficiency processing algorithm of the battery lithium film, wherein the server can process the multiple lithium film images with the high-efficiency processing algorithm of the battery lithium film to generate a classification result for indicating whether the processing quality of the battery lithium film to be detected is qualified.

Having described the general principles of the present application, various non-limiting embodiments of the present application will now be described with reference to the accompanying drawings.

Exemplary method

Fig. 2 illustrates a flow chart of a method for efficient processing of a lithium film for a battery according to an embodiment of the present application. As shown in fig. 2, the method for efficiently processing a lithium film of a battery according to an embodiment of the present application includes the steps of: s110, obtaining a plurality of lithium film images of a to-be-detected battery lithium film passing through a roll shaft through an area array camera; s120, respectively enabling the plurality of lithium film images to pass through a generator model serving as a light reflecting filter to obtain a plurality of lithium film generation images; s130, enabling each lithium film generation image in the plurality of lithium film generation images to pass through a first convolution neural network respectively to obtain a plurality of lithium film characteristic vectors, and arranging the plurality of lithium film characteristic vectors into a lithium film characteristic matrix in a two-dimensional mode according to a sample dimension; s140, passing the topological matrix of the area array camera through a second convolutional neural network to obtain a topological characteristic matrix, wherein characteristic values of positions on non-diagonal positions in the topological matrix of the area array camera are used for representing the distance between the two corresponding cameras, and the characteristic values of the positions on the diagonal positions in the topological matrix of the area array camera are zero; s150, passing the topological characteristic matrix and the lithium film characteristic matrix through a neural network to obtain a lithium film topological characteristic matrix containing irregular topological information and high-dimensional lithium film image information; and S160, passing the topological characteristic matrix of the lithium film through a classifier to obtain a classification result, wherein the classification result is used for indicating whether the processing quality of the lithium film of the battery to be detected is qualified or not.

Fig. 3 illustrates an architectural schematic of a method for efficient processing of a battery lithium film according to an embodiment of the present application. As shown in fig. 3, in the network architecture, first, the plurality of lithium film images (e.g., P as illustrated in fig. 3) are respectively passed through a generator model (e.g., CN as illustrated in fig. 3) as a reflex filter to obtain a plurality of lithium film generation images (e.g., Q as illustrated in fig. 3); then, respectively passing each of the obtained plurality of lithium film generation images through a first convolutional neural network (e.g., CNN1 as illustrated in fig. 3) to obtain a plurality of lithium film eigenvectors (e.g., VF1 as illustrated in fig. 3), and two-dimensionally arranging the plurality of lithium film eigenvectors into a lithium film eigenvector matrix (e.g., MF1 as illustrated in fig. 3) according to a sample dimension; then, passing the area-array camera's topology matrix (e.g., M as illustrated in fig. 3) through a second convolutional neural network (e.g., CNN2 as illustrated in fig. 3) to obtain a topology feature matrix (e.g., MF2 as illustrated in fig. 3); then, passing the topological feature matrix and the lithium film feature matrix through a graphical neural network (e.g., GNN as illustrated in fig. 3) to obtain a lithium film topological feature matrix (e.g., MF as illustrated in fig. 3) containing irregular topological information and high-dimensional lithium film image information; and finally, passing the lithium film topological characteristic matrix through a classifier (for example, a classifier as illustrated in fig. 3) to obtain a classification result, wherein the classification result is used for indicating whether the processing quality of the battery lithium film to be detected is qualified or not.

In steps S110 and S120, a plurality of lithium film images of the lithium film of the battery to be detected passing through the roller are obtained by the area-array camera, and the plurality of lithium film images are respectively passed through a generator model as a reflective filter to obtain a plurality of lithium film generation images. As described above, it can be understood that the key to efficient production of the lithium battery diaphragm is to improve the detection accuracy of the diaphragm, while the lithium battery diaphragm detector commonly used in the market detects the diaphragm, which results in lower detection accuracy due to the fixed position of the line camera and the light source facing the surface of the lithium battery diaphragm. Therefore, in the technical scheme of this application, thereby it is expected to optimize lithium battery diaphragm detector in order to improve the detection precision of lithium battery diaphragm and improve the machining efficiency of lithium cell.

That is, specifically, in the technical solution of the present application, the linear camera is replaced with an area-array camera to acquire images of the detected diaphragm from more shooting angles. In order to solve the problem of detection accuracy reduction caused by transmission offset during lithium battery diaphragm transmission, in the technical scheme of the application, a convolutional neural network model is used as a feature extractor to encode a diaphragm image, wherein the convolutional neural network is insensitive to offset and rotation of an object in the image. Aiming at the problem of light reflection during image acquisition, the image generator capable of removing the influence of light reflection is trained through the idea of countertraining. And in turn, the detection accuracy of the battery separator is improved, thereby improving the processing efficiency of the lithium battery.

Specifically, in the technical scheme of the application, firstly, a plurality of lithium film images of a battery to be detected when the lithium film passes through a roller shaft are obtained through an area-array camera. Then, the plurality of lithium film images are respectively passed through a generator model as a light reflecting filter to obtain a plurality of lithium film generation images. In an embodiment of the present application, the training process of the generator model includes: and respectively passing the training image and the reference image through the generator model serving as a reflection filter to obtain a first characteristic matrix and a second characteristic matrix, wherein the training image is a camera image acquired by a camera facing a light source, and the reference image is a camera image to be replaced. In a specific example of the present application, the generator model is a deep convolutional neural network model. The generator model may then be trained by computing discriminator loss function values between the first feature matrix and the second feature matrix to "fool" features learned by the generator model from the training images toward features of the reference image.

Specifically, in the embodiment of the present application, the process of training the generator model as the reflex filter includes: firstly, a camera image of the lithium battery diaphragm passing through the roller shaft is collected through a camera facing a light source, and the camera image to be replaced is acquired as a reference image. And then, respectively passing the training image and the reference image through the generator model serving as a reflection filter to obtain a first characteristic matrix and a second characteristic matrix, wherein the generator model serving as the reflection filter is a third convolutional neural network.

It will then be appreciated that in training the first convolutional neural network by a discriminator loss function between said first feature matrix and said second feature matrix, it is also desirable that said first feature matrix and said second feature matrix are as consistent in feature distribution as possible, since the discriminator loss function is mainly consistent from a perceptual level of image appearance, i.e. as a perceptual loss function. Therefore, a smooth motion matrix of the first feature matrix with respect to the second feature matrix is further calculated as a modified first feature matrix, whichThe smooth motion matrix is constructed based on a difference matrix between the first feature matrix and the second feature matrix. Accordingly, in a specific example, a smooth motion matrix of the first feature matrix relative to the second feature matrix is calculated as the modified first feature matrix in the following formula; wherein the formula is:

wherein

Representing the first feature matrix in a first order,

representing the second feature matrix in a second order,

representing the modified first feature matrix,

and

respectively represent subtraction and addition by position of the matrix, an

Representing a dot multiplication of a number with a matrix,

is a hyper-parameter controlling the posterior weight.

Then, a discriminator loss function value between the corrected first feature matrix and the second feature matrix is calculated. In one specific example, the modified first feature matrix is input to the discriminator neural network to obtain a third feature matrix; inputting the second feature matrix into the discriminator neural network to obtain a fourth feature matrix; determining whether values of predetermined positions in the third feature matrix and the fourth feature matrix are the same; in response to the values of the predetermined positions in the third feature matrix and the fourth feature matrix being the same, calculating a negative value of a base two logarithm of the values of the predetermined positions as a first value; in response to the values of the predetermined positions in the third feature matrix and the fourth feature matrix being different, calculating a base two logarithm value of the values of the predetermined positions as a second value; and calculating the sum of the average value of the positions where the first values are the same in value and the average value of the positions where the second values are different in value as the discriminator loss function value.

Finally, the generator model as a reflectron filter is trained with the discriminator loss function values and counter-propagating through gradient descent. It should be understood that the first feature is obtained by matrixing the first feature

As a posterior feature matrix, by basing it on the posterior features, i.e. the first feature matrix

With respect to a prior feature, i.e. the second feature matrix

For prior feature distribution

So that the first feature matrix is more smoothly responsive

And the second feature matrix

Fig. 4 illustrates a flowchart of a training process of a generator model as a reflective filter in a method for efficient processing of a battery lithium membrane according to an embodiment of the present application. As shown in fig. 4, in the embodiment of the present application, the training process of the generator model as the reflex filter includes: s210, a training image and a reference image are obtained, wherein the training image is a camera image acquired by a camera which is over against a light source, and the reference image is a camera image to be replaced. S220, respectively passing the training image and the reference image through the generator model serving as the reflection filter to obtain a first characteristic matrix and a second characteristic matrix, wherein the generator model serving as the reflection filter is a third convolutional neural network. And S230, calculating a smooth motion matrix of the first feature matrix relative to the second feature matrix as a modified first feature matrix, wherein the smooth motion matrix is constructed based on a difference matrix between the first feature matrix and the second feature matrix. And S240, calculating a discriminator loss function value between the corrected first feature matrix and the second feature matrix. S250, training the generator model as a reflex filter with the discriminator loss function values and by back propagation of gradient descent.

Fig. 5 illustrates an architecture diagram of a training process of a generator model as a reflective filter in a method for efficient processing of a battery lithium membrane according to an embodiment of the present application. As shown in fig. 5, in the training phase, first, in the network architecture, the training image (e.g., P1 as illustrated in fig. 5) and the reference image (e.g., P2 as illustrated in fig. 5) are respectively passed through the generator model (e.g., CNN as illustrated in fig. 5) as a reflex filter to obtain a first feature matrix (e.g., MF1 as illustrated in fig. 5) and a second feature matrix (e.g., MF2 as illustrated in fig. 5); then, a smooth motion matrix of the first feature matrix relative to the second feature matrix is calculated as a modified first feature matrix (e.g., MF as illustrated in fig. 5); then, a discriminator loss function value (for example, LV as illustrated in fig. 5) between the corrected first feature matrix and the second feature matrix is calculated; finally, the generator model as a reflectron filter is trained with the discriminator loss function values and counter-propagating through gradient descent.

In step S130, each of the plurality of lithium film generation images is respectively passed through a first convolutional neural network to obtain a plurality of lithium film eigenvectors, and the plurality of lithium film eigenvectors are two-dimensionally arranged into a lithium film eigenvector matrix according to a sample dimension. That is, in the technical solution of the present application, further, each lithium film generation image in the plurality of lithium film generation images is respectively passed through a first convolutional neural network to obtain a plurality of lithium film eigenvectors, and the plurality of lithium film eigenvectors are two-dimensionally arranged as a lithium film eigenvector matrix according to a sample dimension. That is, the lithium film generated image is encoded using the first convolutional neural network as a feature extractor to extract local high-dimensional implicit features of the lithium film generated image. It is worth mentioning that the first convolutional neural network has insensitivity to position in terms of feature extraction, that is, insensitivity to rotation and offset of the lithium film object in the lithium film generated image, and therefore, even if there is a position offset at the time of lithium film transfer, it does not adversely affect the detection accuracy.

Specifically, in the embodiment of the present application, each layer of the first convolutional neural network performs in the forward pass of the layer: performing convolution processing on input data to obtain a convolution characteristic diagram; performing mean pooling based on a local feature matrix on the convolution feature map to obtain a pooled feature map; and performing nonlinear activation on the pooled feature map to obtain an activated feature map; wherein the output of the last layer of the first convolutional neural network is the plurality of lithium film eigenvectors, and the input of the first layer of the first convolutional neural network generates an image for the plurality of lithium films.

In step S140, the topological matrix of the area-array camera is passed through a second convolutional neural network to obtain a topological feature matrix, where feature values of positions at non-diagonal positions in the topological matrix of the area-array camera are used to represent distances between the two corresponding cameras, and the feature values of positions at diagonal positions in the topological matrix of the area-array camera are zero. It should be understood that, in the technical solution of the present application, the cameras for acquiring the lithium film images are arranged in an array (i.e., the area-array camera), so that the lithium film images acquired by the cameras in the area-array camera have a predetermined spatial topological relationship, and fusing this information is beneficial to improving the detection accuracy. Specifically, the topological characteristic matrix is obtained by passing the topological matrix of the area-array camera through a second convolutional neural network, wherein the characteristic value of each position at a non-diagonal position in the topological matrix of the area-array camera is used for representing the distance between the two corresponding cameras, and the characteristic value of each position at a diagonal position in the topological matrix of the area-array camera is zero.

Specifically, in the embodiment of the present application, each layer of the second convolutional neural network performs in the forward pass of the layer: performing convolution processing on input data to obtain a convolution characteristic diagram; performing mean pooling based on local channel dimensions on the convolution feature map to obtain a pooled feature map; and performing nonlinear activation on the pooled feature map to obtain an activated feature map; and the output of the last layer of the second convolutional neural network is the topological characteristic matrix, and the input of the first layer of the second convolutional neural network is the topological matrix of the area-array camera.

In step S150 and step S160, the topological feature matrix and the lithium film feature matrix are passed through a neural network to obtain a lithium film topological feature matrix containing irregular topological information and high-dimensional lithium film image information, and the lithium film topological feature matrix is passed through a classifier to obtain a classification result, wherein the classification result is used for indicating whether the processing quality of the lithium film of the battery to be detected is qualified. That is, in the technical solution of the present application, the topological feature matrix and the lithium film feature matrix are then passed through a graph neural network to obtain a lithium film topological feature matrix containing irregular topological information and high-dimensional lithium film image information, where the graph convolutional neural network generates a feature representation containing irregular topological information and high-dimensional lithium film image information by learnable neural network parameters. And finally, passing the lithium film topological characteristic matrix through a classifier to obtain a classification result, wherein the classification result is used for indicating whether the processing quality of the lithium film of the battery to be detected is qualified or not.

Specifically, in an embodiment of the present application, the process of passing the lithium film topological feature matrix through a classifier to obtain a classification result includes: the classifier processes the lithium film topological characteristic matrix by the following formula to generate a classification result, wherein the formula is as follows:

wherein

Represents projecting the topological feature matrix of the lithium film into vectors,

to is that

Is a weight matrix of the fully connected layers of each layer,

to

A bias matrix representing the layers of the fully connected layer.

In summary, the efficient processing method for the battery lithium film based on the embodiment of the present application is elucidated, the method includes acquiring an image of the battery lithium film passing through a roll shaft by using an area-array camera, processing the lithium film image by using a trained generator model serving as a reflective filter to remove a reflective effect in an original image and enable a feature distribution of the lithium film generated image in a target domain to keep higher consistency with a feature distribution of the lithium film image in a source domain, then encoding the lithium film generated image by using a convolutional neural network as a feature extractor to extract a local high-dimensional implicit feature of the lithium film generated image, and further generating a feature representation including irregular topology information and high-dimensional lithium film image information by using a neural network through learnable neural network parameters to obtain a more accurate classification result. Thus, the detection accuracy of the battery separator can be improved.

An exemplary system: fig. 6 illustrates a block diagram of a system for efficient processing of battery lithium films in accordance with an embodiment of the present application. As shown in fig. 6, the system 500 for efficiently processing a lithium film for a battery according to an embodiment of the present application includes: a lithium film image obtaining unit 510, configured to obtain, by an area-array camera, a plurality of lithium film images of a battery to be detected when a lithium film passes through a roller; a generated image acquiring unit 520, configured to pass the plurality of lithium film images acquired by the lithium film image acquiring unit 510 through a generator model as a light reflecting filter, respectively, to obtain a plurality of lithium film generated images; a first convolution unit 530, configured to pass each lithium film generation image in the multiple lithium film generation images obtained by the generation image obtaining unit 520 through a first convolution neural network to obtain multiple lithium film feature vectors, and two-dimensionally arrange the multiple lithium film feature vectors into a lithium film feature matrix according to a sample dimension; a second convolution unit 540, configured to pass the topology matrix of the area-array camera through a second convolution neural network to obtain a topology feature matrix, where feature values of positions at non-diagonal positions in the topology matrix of the area-array camera are used to represent a distance between two corresponding cameras, and the feature values of positions at diagonal positions in the topology matrix of the area-array camera are zero; a graph neural network unit 550, configured to pass the topological feature matrix obtained by the second convolution unit 540 and the lithium film feature matrix obtained by the first convolution unit 530 through a graph neural network to obtain a lithium film topological feature matrix containing irregular topological information and high-dimensional lithium film image information; and a classification unit 560, configured to pass the lithium film topological characteristic matrix obtained by the graph neural network unit 550 through a classifier to obtain a classification result, where the classification result is used to indicate whether the processing quality of the lithium film of the battery to be detected is qualified.

In one example, in the above-mentioned high efficiency processing system 500 for battery lithium membrane, as shown in fig. 7, the training module 52 as a generator model of the reflective filter includes: a training image acquisition unit 521 for acquiring a training image which is a camera image captured by a camera facing the light source and a reference image which is a camera image to be replaced; a reflection filtering unit 522, configured to pass the training image obtained by the training image obtaining unit 521 and the reference image obtained by the training image obtaining unit 521 through the generator model serving as a reflection filter to obtain a first feature matrix and a second feature matrix, where the generator model serving as a reflection filter is a third convolutional neural network; a correcting unit 523, configured to calculate a smooth motion matrix of the first feature matrix obtained by the reflection filtering unit 522 relative to the second feature matrix obtained by the reflection filtering unit 522 as a corrected first feature matrix, where the smooth motion matrix is constructed based on a difference matrix between the first feature matrix and the second feature matrix; a discriminator loss function value calculation unit 524, configured to calculate a discriminator loss function value between the corrected first feature matrix obtained by the correction unit 523 and the second feature matrix obtained by the reflection filter unit 522; and a training unit 525 for training the generator model as a reflex filter with the discriminator loss function value obtained by the discriminator loss function value calculation unit 524 and by back propagation of gradient descent.

In one example, in the above system 500 for efficiently processing a lithium battery film, the correcting unit 523 is further configured to: calculating a smooth motion matrix of the first feature matrix relative to the second feature matrix as the modified first feature matrix according to the following formula; wherein the formula is:

wherein

Representing the first feature matrix in a first order,

representing the second feature matrix in a second order,

representing the modified first feature matrix,

and

respectively represent subtraction and addition by position of the matrix, an

Representing a dot multiplication of a number with a matrix,

is a hyper-parameter.

In one example, in the above-described system 500 for high-efficiency processing of a battery lithium film, the discriminator loss function value calculation unit 524 includes: a third feature matrix generation subunit, configured to input the first feature matrix into the discriminator neural network to obtain a third feature matrix; a fourth feature matrix generation subunit, configured to input the second feature matrix into the discriminator neural network to obtain a fourth feature matrix; a determining subunit, configured to determine whether values of predetermined positions in the third feature matrix and the fourth feature matrix are the same; a first value subunit, configured to calculate, as a first value, a negative value of a base two logarithm value of the predetermined position in response to the predetermined position in the third feature matrix and the fourth feature matrix having the same value; a second value subunit, configured to calculate, as a second value, a base-two logarithmic value of the predetermined position in response to a difference in the value of the predetermined position in the third feature matrix and the fourth feature matrix; and a sum of average value calculating subunit configured to calculate a sum of an average value of positions where the first values are the same in value and an average value of positions where the second values are different in value as the discriminator loss function value.

In one example, in the above-described system 500 for efficient processing of battery lithium film, the layers of the first convolutional neural network are separately performed in a forward pass of the layers: performing convolution processing on input data to obtain a convolution characteristic diagram; performing mean pooling based on a local feature matrix on the convolution feature map to obtain a pooled feature map; and performing nonlinear activation on the pooled feature map to obtain an activated feature map; wherein the output of the last layer of the first convolutional neural network is the plurality of lithium film eigenvectors, and the input of the first layer of the first convolutional neural network generates an image for the plurality of lithium films.

In one example, in the above system 500 for efficient processing of battery lithium membranes, the layers of the second convolutional neural network are each performed in a forward pass of the layer: performing convolution processing on input data to obtain a convolution characteristic diagram; performing mean pooling based on a local feature matrix on the convolution feature map to obtain a pooled feature map; and performing nonlinear activation on the pooled feature map to obtain an activated feature map; and the output of the last layer of the second convolutional neural network is the topological characteristic matrix, and the input of the first layer of the second convolutional neural network is the topological matrix of the area-array camera.

In one example, in the above system 500 for processing lithium battery film efficiently, the sorting unit 560 is further configured to: the classifier processes the lithium film topological characteristic matrix by the following formula to generate a classification result, wherein the formula is as follows:

wherein

to

Is a weight matrix of the fully connected layers of each layer,

to

A bias matrix representing the layers of the fully connected layer.

Here, it can be understood by those skilled in the art that the specific functions and operations of the respective units and modules in the above-described system 500 for efficiently processing a battery lithium film have been described in detail in the above description of the method for efficiently processing a battery lithium film with reference to fig. 1 to 5, and thus, the repetitive description thereof will be omitted.

As described above, the system 500 for efficiently processing a battery lithium film according to the embodiment of the present application may be implemented in various terminal devices, such as a server for an efficient processing algorithm of a battery lithium film, and the like. In one example, the system 500 for efficient processing of battery lithium films according to embodiments of the present application may be integrated into a terminal device as one software module and/or hardware module. For example, the system 500 for efficient processing of battery lithium film may be a software module in the operating system of the terminal device, or may be an application developed for the terminal device; of course, the system 500 for efficiently processing the lithium film of the battery can also be one of many hardware modules of the terminal device.

Alternatively, in another example, the system 500 for efficient processing of battery lithium film and the terminal device may be separate devices, and the system 500 for efficient processing of battery lithium film may be connected to the terminal device through a wired and/or wireless network and transmit the interactive information according to an agreed data format.

Exemplary computer program product and computer-readable storage Medium

In addition to the methods and apparatus described above, embodiments of the present application may also be a computer program product comprising computer program instructions that, when executed by a processor, cause the processor to perform the steps in the method for efficient processing of battery lithium films according to various embodiments of the present application described in the "exemplary methods" section above of this specification.

The computer program product may be written with program code for performing the operations of embodiments of the present application in any combination of one or more programming languages, including an object oriented programming language such as Java, C + + or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code may execute entirely on the user's computing device, partly on the user's device, as a stand-alone software package, partly on the user's computing device and partly on a remote computing device, or entirely on the remote computing device or server.

Furthermore, embodiments of the present application may also be a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, cause the processor to perform the steps in the method for efficient processing of battery lithium films according to various embodiments of the present application described in the "exemplary methods" section above in this specification.

The computer-readable storage medium may take any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may include, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or a combination of any of the foregoing. More specific examples (a non-exhaustive list) of the readable storage medium include: an electrical connection having at least one wire, a portable disk, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

The foregoing describes the general principles of the present application in conjunction with specific embodiments, however, it is noted that the advantages, effects, etc. mentioned in the present application are merely examples and are not limiting, and they should not be considered essential to the various embodiments of the present application. Furthermore, the foregoing disclosure of specific details is for the purpose of illustration and description and is not intended to be limiting, since the foregoing disclosure is not intended to be exhaustive or to limit the disclosure to the precise details disclosed.

The block diagrams of devices, apparatuses, devices, systems referred to in this application are only used as illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. These devices, apparatuses, devices, systems may be connected, arranged, configured in any manner, as will be appreciated by those skilled in the art. Words such as "including," "comprising," "having," and the like are open-ended words that mean "including, but not limited to," and are used interchangeably therewith. The words "or" and "as used herein mean, and are used interchangeably with, the word" and/or, "unless the context clearly dictates otherwise. The word "such as" is used herein to mean, and is used interchangeably with, the phrase "such as but not limited to".

It should also be noted that in the devices, apparatuses, and methods of the present application, each component or step can be decomposed and/or re-combined. These decompositions and/or recombinations are to be considered as equivalents of the present application.

The previous description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other aspects without departing from the scope of the application. Thus, the present application is not intended to be limited to the aspects shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for efficiently processing a lithium film of a battery, comprising: obtaining a plurality of lithium film images of a battery to be detected when a lithium film passes through a roller shaft by an area array camera; respectively passing the plurality of lithium film images through a generator model serving as a light reflecting filter to obtain a plurality of lithium film generation images; respectively enabling each lithium film generation image in the plurality of lithium film generation images to pass through a first convolution neural network to obtain a plurality of lithium film eigenvectors, and two-dimensionally arranging the plurality of lithium film eigenvectors into a lithium film eigenvector matrix according to a sample dimension; enabling the topological matrix of the area-array camera to pass through a second convolutional neural network to obtain a topological characteristic matrix, wherein characteristic values of positions on non-diagonal positions in the topological matrix of the area-array camera are used for representing the distance between the two corresponding cameras, and the characteristic values of the positions on the diagonal positions in the topological matrix of the area-array camera are zero; passing the topological characteristic matrix and the lithium film characteristic matrix through a neural network to obtain a lithium film topological characteristic matrix containing irregular topological information and high-dimensional lithium film image information; and enabling the lithium film topological characteristic matrix to pass through a classifier to obtain a classification result, wherein the classification result is used for indicating whether the processing quality of the lithium film of the battery to be detected is qualified or not.

2. The method for efficient processing of battery lithium membranes as claimed in claim 1, wherein the training process as a generator model of a reflective filter comprises: acquiring a training image and a reference image, wherein the training image is a camera image acquired by a camera which is over against a light source, and the reference image is a camera image to be replaced; respectively enabling the training image and the reference image to pass through the generator model serving as a reflection filter to obtain a first characteristic matrix and a second characteristic matrix, wherein the generator model serving as the reflection filter is a third convolutional neural network; calculating a smooth motion matrix of the first feature matrix relative to the second feature matrix as a modified first feature matrix, wherein the smooth motion matrix is constructed based on a difference matrix between the first feature matrix and the second feature matrix; calculating a discriminator loss function value between the corrected first feature matrix and the second feature matrix; and training the generator model as a reflectorized filter with the discriminator loss function values and by back propagation of gradient descent.

3. The method for efficient processing of a battery lithium film as recited in claim 2, wherein calculating a smooth motion matrix of the first feature matrix relative to the second feature matrix as the modified first feature matrix comprises: calculating a smooth motion matrix of the first feature matrix relative to the second feature matrix as the modified first feature matrix according to the following formula; wherein the formula is:

wherein

Representing the first feature matrix in a first order,

representing the second feature matrix in a second order,

representing the modified first feature matrix,

and

respectively represent subtraction and addition by position of the matrix, an

Representing a dot multiplication of a number with a matrix,

is a hyper-parameter.

4. The method for efficient processing of battery lithium films as claimed in claim 3, wherein calculating a discriminator loss function value between the modified first signature matrix and the second signature matrix comprises: inputting the modified first feature matrix into the discriminator neural network to obtain a third feature matrix; inputting the second feature matrix into the discriminator neural network to obtain a fourth feature matrix; determining whether values of predetermined positions in the third feature matrix and the fourth feature matrix are the same; in response to the values of the predetermined positions in the third feature matrix and the fourth feature matrix being the same, calculating a negative value of a base two logarithm of the values of the predetermined positions as a first value; in response to the values of the predetermined positions in the third feature matrix and the fourth feature matrix being different, calculating a base two logarithm value of the values of the predetermined positions as a second value; and calculating the sum of the average value of the positions where the first values are the same in value and the average value of the positions where the second values are different in value as the discriminator loss function value.

5. The method for efficient processing of battery lithium membranes as claimed in claim 4, wherein each layer of the first convolutional neural network is separately performed in the forward pass of the layer: performing convolution processing on input data to obtain a convolution characteristic diagram; performing mean pooling based on a local feature matrix on the convolution feature map to obtain a pooled feature map; and carrying out nonlinear activation on the pooling feature map to obtain an activation feature map; wherein the output of the last layer of the first convolutional neural network is the plurality of lithium film eigenvectors, and the input of the first layer of the first convolutional neural network generates an image for the plurality of lithium films.

6. The method for efficient processing of battery lithium membranes as claimed in claim 5, wherein the layers of the second convolutional neural network are separately performed in the forward pass of the layer: performing convolution processing on input data to obtain a convolution characteristic diagram; performing mean pooling based on local channel dimensions on the convolution feature map to obtain a pooled feature map; and carrying out nonlinear activation on the pooling feature map to obtain an activation feature map; and the output of the last layer of the second convolutional neural network is the topological characteristic matrix, and the input of the first layer of the second convolutional neural network is the topological matrix of the area-array camera.

7. The method for processing the lithium battery film at high efficiency according to claim 6, wherein the step of passing the topological characteristic matrix of the lithium battery film through a classifier to obtain a classification result comprises the following steps: what is needed isThe classifier processes the lithium film topological characteristic matrix by the following formula to generate a classification result, wherein the formula is as follows:

wherein

to

Is a weight matrix of the fully connected layers of each layer,

to

A bias matrix representing the layers of the fully connected layer.

8. A system for efficient processing of a lithium film for a battery, comprising: the lithium film image acquisition unit is used for acquiring a plurality of lithium film images of the battery to be detected when the lithium film passes through the roller shaft through the area array camera; a generated image acquisition unit, configured to pass the plurality of lithium film images obtained by the lithium film image acquisition unit through a generator model as a light reflection filter, respectively, to obtain a plurality of lithium film generated images; the first convolution unit is used for enabling each lithium film generation image in the plurality of lithium film generation images obtained by the generation image obtaining unit to pass through a first convolution neural network respectively to obtain a plurality of lithium film characteristic vectors, and arranging the plurality of lithium film characteristic vectors into a lithium film characteristic matrix in a two-dimensional mode according to the sample dimension; the second convolution unit is used for enabling the topological matrix of the area-array camera to pass through a second convolution neural network so as to obtain a topological characteristic matrix, wherein characteristic values of positions on non-diagonal positions in the topological matrix of the area-array camera are used for representing the distance between two corresponding cameras, and the characteristic values of the positions on the diagonal positions in the topological matrix of the area-array camera are zero; the graph neural network unit is used for enabling the topological characteristic matrix obtained by the second convolution unit and the lithium film characteristic matrix obtained by the first convolution unit to pass through a graph neural network so as to obtain a lithium film topological characteristic matrix containing irregular topological information and high-dimensional lithium film image information; and the classification unit is used for enabling the lithium film topological characteristic matrix obtained by the graph neural network unit to pass through a classifier to obtain a classification result, and the classification result is used for indicating whether the processing quality of the lithium film of the battery to be detected is qualified or not.

9. The system for efficient processing of battery lithium membranes of claim 8, wherein the training module as a generator model of a reflective filter comprises: a training image acquisition unit for acquiring a training image and a reference image, wherein the training image is a camera image acquired by a camera facing the light source, and the reference image is a camera image to be replaced; the reflection filtering unit is used for enabling the training image obtained by the training image obtaining unit and the reference image obtained by the training image obtaining unit to pass through the generator model serving as the reflection filter respectively to obtain a first characteristic matrix and a second characteristic matrix, wherein the generator model serving as the reflection filter is a third convolutional neural network; a correction unit, configured to calculate a smooth motion matrix of the first feature matrix obtained by the reflection filtering unit relative to the second feature matrix obtained by the reflection filtering unit as a corrected first feature matrix, where the smooth motion matrix is constructed based on a difference matrix between the first feature matrix and the second feature matrix; a discriminator loss function value calculation unit configured to calculate a discriminator loss function value between the corrected first feature matrix obtained by the correction unit and the second feature matrix obtained by the reflection filter unit; and a training unit for training the generator model as a reflectorized filter with the discriminator loss function value obtained by the discriminator loss function value calculation unit and by back propagation of gradient descent.

10. The system for efficient processing of battery lithium film of claim 9, wherein the correction unit is further configured to: calculating a smooth motion matrix of the first feature matrix relative to the second feature matrix as the modified first feature matrix according to the following formula; wherein the formula is:

wherein

Representing the first feature matrix in a first order,

representing the second feature matrix in a second order,

representing the modified first feature matrix,

and

respectively represent subtraction and addition by position of the matrix, an

Representing a dot multiplication of a number with a matrix,

is a hyper-parameter.