Machine learning-based analysis method, system, device, and media for sei images

Abstract

The application provides a machine learning-based SEI image analysis method, system, device and medium. The machine learning-based SEI image analysis method comprises the following steps: obtaining a target SEI image to be analyzed; performing topography analysis according to a trained image classification model; performing decision influence analysis through a trained image interpretation model; performing image analysis on the target SEI image according to at least one pre-set image analysis method, and outputting an image analysis result; wherein the topography analysis, the decision influence analysis and the image analysis are triggered and executed according to first selection information of a user, and the first selection information comprises a combination of one or more of the topography analysis, the decision influence analysis and the image analysis. The application can improve the objectivity, accuracy and efficiency of SEI image analysis, and improve the comfort and convenience of user use, thereby helping researchers to deeply understand the topography information and potential information of the solid-state electrolyte image.

Description

Technical Field

[0001] This application belongs to the field of image processing technology, and in particular relates to a method, system, device and medium for analyzing SEI images based on machine learning. Background Technology

[0002] A lithium-ion battery is a device that stores and releases electricity by moving lithium ions between two electrodes, one positive and one negative. The electrodes are separated by a liquid or gel-like substance called an electrolyte, allowing the lithium ions to move within it. When the battery is first charged, some of the electrolyte reacts with the negative electrode, forming a thin layer on its surface. This thin layer is called the solid electrolyte interface (SEI), which acts as a barrier to prevent further reactions between the negative electrode and the electrolyte, protecting the negative electrode from wear and reducing short circuits. Many factors influence SEI formation, and its stability is relatively poor after formation (it changes with battery cycle time). Therefore, controlling the formation and changes of the SEI is crucial to ensuring and improving the performance and safety of lithium-ion batteries.

[0003] To observe the microstructure of SEIs and enhance our understanding and analysis of them, researchers have begun using cryo-electron microscopy (cryo-EM) to capture SEI images. However, cryo-EM images of SEIs differ from traditional images in terms of pixel distribution and noise, making analysis and processing more challenging. Current techniques typically employ manual methods to assess the morphology of SEIs and measure the gaps between morphologies manually to provide qualitative density results. However, for high-throughput SEI images, these manual methods suffer from high subjectivity, low accuracy, and low efficiency. Summary of the Invention

[0004] In view of the shortcomings of the prior art described above, the purpose of this application is to provide a method, system, device and medium for analyzing SEI images based on machine learning, so as to solve the technical problems of strong subjectivity, low accuracy and low efficiency in the analysis of SEI images in the prior art.

[0005] In a first aspect, this application provides a machine learning-based method for analyzing SEI images, including:

[0006] Obtain the target SEI image to be analyzed;

[0007] The morphology analysis is performed based on the trained image classification model, and the morphology analysis is used to determine the morphology type of the target SEI image;

[0008] The trained image interpretation model is used to perform decision impact analysis, which generates decision impact information that influences the image classification model to determine the shape type of the target SEI image.

[0009] The target SEI image is analyzed according to at least one pre-defined image analysis method, and the image analysis results are output. The image analysis results are used to display hidden information in the target SEI image.

[0010] The morphology analysis, the decision impact analysis, and the image analysis are triggered and executed based on the user's first selection information, which includes one or more combinations of the morphology analysis, the decision impact analysis, and the image analysis.

[0011] In one implementation of the first aspect, before performing morphology analysis based on the trained image classification model, the method further includes:

[0012] Based on the image dataset to be trained, the pre-determined original training model is trained until convergence, resulting in a trained image classification model.

[0013] The step of performing training operations on a pre-determined original training model based on the image dataset to be trained until convergence, to obtain a trained image classification model, includes:

[0014] Obtain the image dataset to be trained;

[0015] Perform data augmentation on the image dataset to be trained to obtain an augmented image dataset;

[0016] All image data in the enhanced image dataset are input into a pre-determined original training model to perform training operations on the original training model until convergence, thereby obtaining a trained image classification model.

[0017] In one implementation of the first aspect, the image dataset to be trained includes a plurality of first SEI images to be trained; the step of performing data augmentation on the image dataset to be trained to obtain an augmented image dataset includes:

[0018] Calculate the gradient histogram information of each first SEI image in the image dataset to be trained, so as to determine the main part of each first SEI image respectively;

[0019] A target transformation operation is performed on the main body of each first SEI image to obtain a second SEI image corresponding to each first SEI image, and the image dataset composed of all the second SEI images is determined as the enhanced image dataset.

[0020] In one implementation of the first aspect, all image data from the enhanced image dataset are input into a pre-determined original training model, including:

[0021] The pixels in all the second SEI images in the enhanced image dataset are converted into the target input data type to obtain the converted image data.

[0022] The converted image data is input into a pre-determined original training model.

[0023] In one implementation of the first aspect, the decision impact analysis using a trained image interpretation model includes:

[0024] Using the trained image interpretation model, the score gradient of the target SEI image relative to the feature map of each convolutional layer in the image classification model is calculated respectively.

[0025] The contribution level of each fractional gradient is determined based on the pixel corresponding to each fractional gradient.

[0026] Decision impact information is generated based on the contribution of at least some of the said score gradients.

[0027] In one implementation of the first aspect, generating decision influence information based on the contribution of at least a portion of the score gradient includes:

[0028] Based on the ReLU activation function, it is determined whether the contribution of each fractional gradient is the contribution of the target gradient, wherein the contribution of the target gradient is the positive contribution among all the contributions of the fractional gradients.

[0029] If the contribution of the fractional gradient is equal to the contribution of the target gradient, then the contribution of the target gradient is retained.

[0030] Decision impact information is generated based on the contribution levels of all the aforementioned target gradients.

[0031] In one implementation of the first aspect, the image analysis of the target SEI image according to at least one pre-defined image analysis method includes:

[0032] Receive second selection information input by the user, the second selection information including the selection of at least one pre-set image analysis method;

[0033] The target SEI image is analyzed based on the second selection information.

[0034] Secondly, this application provides a machine learning-based SEI image analysis system, comprising:

[0035] The acquisition module is used to acquire the target SEI image to be analyzed.

[0036] The morphology analysis module is used to perform morphology analysis based on the trained image classification model, and the morphology analysis is used to determine the morphology type of the target SEI image;

[0037] The decision impact analysis module is used to perform decision impact analysis using a trained image interpretation model. The decision impact analysis is used to generate decision impact information that affects the image classification model in determining the shape type of the target SEI image.

[0038] An image analysis module is used to perform image analysis on the target SEI image according to at least one pre-defined image analysis method and output the image analysis results, which are used to display hidden information in the target SEI image;

[0039] The morphology analysis, the decision impact analysis, and the image analysis are triggered and executed based on the user's first selection information, which includes one or more combinations of the morphology analysis, the decision impact analysis, and the image analysis.

[0040] Thirdly, this application provides an electronic device, including: a processor and a memory;

[0041] The memory is used to store computer programs;

[0042] The processor is configured to execute a computer program stored in the memory to cause the electronic device to perform any of the methods described above.

[0043] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the method described in any of the preceding claims.

[0044] Compared with existing technologies, the machine learning-based SEI image analysis method, system, device, and medium of this application have the following advantages:

[0045] (1) The present invention can automatically predict the shape type of SEI image through image classification model, thereby improving the objectivity and accuracy of SEI image shape type recognition;

[0046] (2) This invention can reveal which pixels / parts in the SEI image have a key impact on the prediction results of the morphology type through the image interpretation model, thereby further ensuring the objectivity and accuracy of morphology type identification, which is conducive to researchers' in-depth understanding of the morphology information of solid electrolyte images;

[0047] (3) The present invention can perform image analysis on the target SEI image according to at least one pre-set image analysis method and output the analysis results, so as to realize various digital image processing techniques for solid electrolyte images, thereby helping researchers to identify potential or hidden information in solid electrolyte images and thus improve the efficiency of solid electrolyte image analysis.

[0048] (4) The present invention provides a user-friendly and interactive selection method, which allows users to arbitrarily select analysis methods such as morphology analysis, decision influence analysis and image analysis according to their needs, thereby improving the convenience and comfort of users. Attached Figure Description

[0049] Figure 1 The diagram shown is a flowchart of an embodiment of the machine learning-based SEI image analysis method described in this application.

[0050] Figure 2 The diagram shown is a flowchart of another embodiment of the machine learning-based SEI image analysis method described in this application.

[0051] Figure 3 The diagram shown is a structural schematic of one embodiment of the machine learning-based SEI image analysis system described in this application.

[0052] Figure 4 The diagram shown is a structural schematic of the electronic device described in this application in one embodiment.

[0053] Component designation explanation:

[0054] 1. Obtaining the module

[0055] 2. Morphology Analysis Module

[0056] 3. Decision Impact Analysis Module

[0057] 4 Image Analysis Module

[0058] 5. Memory

[0059] 6 processors Detailed Implementation

[0060] The following specific examples illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. This application can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this application. It should be noted that, unless otherwise specified, the following embodiments and features in the embodiments can be combined with each other.

[0061] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of this application. Therefore, the drawings only show the components related to this application and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0062] Furthermore, the use of terms such as "first" and "second" in this application is for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the technical solutions of the various embodiments can be combined with each other, but only on the basis of being achievable by those skilled in the art. If the combination of technical solutions is contradictory or impossible to implement, such a combination of technical solutions should be considered non-existent and not within the scope of protection claimed in this application.

[0063] The following embodiments of this application provide a method, system, device, and medium for SEI image analysis based on machine learning. The application scenarios corresponding to this technical solution include, but are not limited to, image management systems, image analysis systems, interactive intelligent application platforms, etc. The technical solutions in the embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0064] like Figure 1 As shown, this embodiment provides a machine learning-based SEI image analysis method, including steps S1-S4.

[0065] Step S1: Obtain the target SEI image to be analyzed.

[0066] Specifically, the target SEI image refers to an image that requires target analysis operations, which may include: shape type analysis; analysis of information that has a key impact on the shape type analysis prediction results; and other digital image processing analyses, such as image rescaling, image segmentation, image texture analysis, and image clustering analysis, etc.

[0067] It is important to emphasize that the SEI image in this application refers to an image of the solid electrolyte interface. The SEI is a thin layer formed when the negative electrode of a lithium-ion battery reacts with the electrolyte during the first charge. It acts as a barrier to prevent further reactions between the negative electrode and the electrolyte, thereby protecting the negative electrode from wear and reducing short circuits. The SEI is composed of various organic and inorganic compounds formed by the decomposition of the electrolyte, and its composition and structure depend on many factors, such as the type of electrolyte solvent, voltage, current density, temperature, and additives.

[0068] Specifically, the SEI images in this application were taken using cryo-electron microscopy. Cryo-electron microscopy can not only resolve the structure of biological macromolecules (such as proteins and DNA) at the atomic level, but also provide high-quality imaging of interfaces in lithium-ion batteries. Through cryo-electron microscopy, the microstructure of solid electrolyte interfaces can be observed, helping researchers to enhance their understanding of solid electrolyte surface phenomena and better study how to improve the morphology of solid electrolytes.

[0069] Step S2: Perform morphological analysis based on the trained image classification model. The morphological analysis is used to determine the morphological type of the target SEI image.

[0070] Specifically, the image classification model is pre-trained and is a binary classification model. The morphology types of the target SEI image mainly include two categories: dendritic and spherical. Among them, spherical shapes are formed by the uniform and rapid diffusion and deposition of lithium; dendritic shapes are formed due to the uneven deposition and stripping of lithium during battery cycling of the lithium metal anode. The formation and growth of dendrites can lead to thermal runaway and internal short circuits in lithium batteries, thereby shortening battery life.

[0071] It should be noted that when training the image classification model, an image dataset containing morphology type classification labels is required. That is, before training the model, an SEI image dataset needs to be obtained using cryo-electron microscopy, and each SEI image is classified and labeled into two groups (dendritic and globular) according to the morphology of the interface. The scale of each SEI image in the SEI image dataset is preferably 1 to 5 micrometers.

[0072] Step S3: Perform decision impact analysis using the trained image interpretation model. The decision impact analysis is used to generate decision impact information that influences the image classification model to determine the shape type of the target SEI image.

[0073] Specifically, the image interpretation model is mainly used to generate and output the basis for the decision-making of the image classification model, namely the decision influence information. It can display and analyze the most important part of the input target SEI image to be analyzed, that is, which pixels / parts in the image have a key impact on the prediction result of the morphology type, and display it in a visual way, thereby helping to understand the morphology information of solid electrolyte images in depth.

[0074] Step S4: Perform image analysis on the target SEI image according to at least one pre-defined image analysis method, and output the image analysis results. The image analysis results are used to display hidden information in the target SEI image.

[0075] The morphology analysis, the decision impact analysis, and the image analysis are triggered and executed based on the user's first selection information, which includes one or more combinations of the morphology analysis, the decision impact analysis, and the image analysis.

[0076] Specifically, hidden information refers to comprehensive and reliable SEI layer characterization information in the target SEI image, such as the density result derived from the gaps between morphologies.

[0077] Specifically, among the three techniques described in steps S2-S4—morphology analysis, decision influence analysis, and image analysis—users can choose according to their needs. For example, they can interact with the system through a pre-set graphical user interface (GUI), allowing them to select one or more techniques from a pre-set menu to analyze the target SEI image. Furthermore, this application designs an application to automate and simplify solid electrolyte processing, including an interactive user interface capable of browsing and loading SEI image files, selecting analysis techniques on demand, and displaying output images available for analysis. The application's code has been compiled and deployed, ensuring compatibility with different operating systems (MacOS, Windows, Linux) and enhancing the user experience.

[0078] Optionally, users can choose only morphology analysis to determine whether the morphology type of their input target SEI image is dendritic or globular. Optionally, based on morphology analysis, users can further choose decision influence analysis and / or image analysis to gain a deeper understanding of why the image classification model identifies the target input SEI image as dendritic or globular, such as which pixels or parts of the image significantly influence the result, and / or, identify potential information in the SEI image. Optionally, users can also choose only image analysis. This embodiment of the invention does not limit the combination and selection of the three methods.

[0079] It should be noted that since the decision impact analysis performed by the image interpretation model is based on the shape analysis performed by the image analysis model, when the user only selects decision impact analysis, the computer device will default to performing shape analysis on the image model first and then performing decision impact analysis, or it will output a prompt to the user to inform the user that shape analysis needs to be selected first.

[0080] Specifically, the pre-defined image analysis method is a digital image processing analysis method, including but not limited to image rescaling, image segmentation, image texture analysis, and image clustering analysis. The hidden information in the target SEI image includes comprehensive and reliable SEI layer representation information, such as density results obtained by analyzing the gaps between morphologies. Furthermore, based on pre-built image classification and image interpretation models, this application has developed a Python program that can implement the various digital image processing analysis methods described above. The relevant code has been compiled and deployed to ensure compatibility with different operating systems (MacOS, Windows, Linux).

[0081] As an optional implementation, the step of performing image analysis on the target SEI image according to at least one pre-defined image analysis method and outputting the image analysis results includes:

[0082] Receive the image clustering analysis method selected by the user;

[0083] Based on the K-Means clustering model, cluster analysis is performed on all pixels in the target SEI image according to the pixel intensity values, and the cluster analysis results are output.

[0084] Specifically, the K-Means clustering model iteratively finds a partitioning scheme of K clusters to minimize the loss function corresponding to the clustering result, and the K-Means clustering model has the characteristic of fast convergence speed.

[0085] In this embodiment, a low-cost and efficient clustering method is proposed. By using the K-Means clustering model to cluster pixels according to the pixel intensity values ​​in the image, the boundaries of the main body in the solid electrolyte interface can be quickly and objectively identified, thereby revealing the hidden information in the solid electrolyte interface.

[0086] As can be seen, the SEI image analysis method based on machine learning described in the embodiments of the present invention can automatically predict the morphology type of SEI images through image classification models, thereby improving the objectivity and accuracy of SEI image morphology type identification; it can reveal which pixels / parts in the SEI image have a key impact on the morphology type prediction results through image interpretation models, thereby further ensuring the objectivity and accuracy of morphology type identification, which is beneficial for researchers to deeply understand the morphology information of solid electrolyte images; it can also perform image analysis on target SEI images according to at least one pre-set image analysis method and output the analysis results, so as to implement various digital image processing techniques for solid electrolyte images, thereby helping researchers to identify potential or hidden information in solid electrolyte images, and thus improving the efficiency of solid electrolyte image analysis; and it can provide a user-friendly and interactive selection method, so that users can arbitrarily select analysis methods such as morphology analysis, decision influence analysis, and image analysis according to their needs, thereby improving the convenience and comfort of users.

[0087] In one embodiment of this application, before step S2, which involves morphological analysis based on a trained image classification model, the SEI image analysis method based on machine learning further includes:

[0088] Step S5: Based on the image dataset to be trained, perform training operations on the pre-determined original training model until convergence, and obtain the trained image classification model.

[0089] Specifically, before model training, it is necessary to obtain an SEI image dataset using cryo-electron microscopy, and classify and label each SEI image into two groups (dendritic and globular) according to the morphology of the interface, thus obtaining the image dataset to be trained.

[0090] Preferably, the original training model is based on the AlexNet convolutional neural network architecture for image classification. Specifically, AlexNet has five convolutional layers and three connected layers. The first convolutional layer uses 96 filters of size 11x11 with a stride of 4 pixels, followed by a 3x3 max-pooling layer with a stride of 2 pixels. The second convolutional layer uses 256 filters of size 5x5 with a stride of 1 pixel, followed by another 3x3 max-pooling layer with a stride of 2 pixels. The third, fourth, and fifth convolutional layers use 384, 384, and 256 filters, respectively, all of size 3x3 with a stride of 1 pixel. The third, fourth, and fifth convolutional layers are interconnected without any pooling or normalization layers in between. Following the fourth and fifth convolutional layers is the third and final max-pooling layer, of size 3x3 with a stride of 2 pixels. The three fully connected layers have 4096, 4096, and 1000 neurons, respectively. By leveraging AlexNet, the model can better learn the features of image datasets, thereby improving the accuracy of image classification, especially for high-throughput, low-contrast, noisy, and complex SEI images. It should be noted that existing image processing software cannot be customized for the specific features and challenges of SEI images. Therefore, before training the model, the input layer was modified based on the AlexNet convolutional neural network architecture to adapt to the image data of solid electrolyte interfaces.

[0091] As an optional implementation, step S5 above, which involves training the pre-determined original training model based on the image dataset to be trained until convergence, to obtain the trained image classification model, includes steps S51-S53:

[0092] Step S51: Obtain the image dataset to be trained;

[0093] Step S52: Perform data augmentation on the image dataset to be trained to obtain the augmented image dataset;

[0094] Step S53: Input all image data in the enhanced image dataset into the predetermined original training model to perform training operations on the original training model until convergence, and obtain the trained image classification model.

[0095] Specifically, existing deep learning frameworks are complex, requiring significant expertise and resources to implement and train SEI image analysis. To avoid increasing the dataset size and ensure the generalization performance of the machine learning model, this application augments the image dataset to be trained, ensuring sufficient samples for training the machine learning model. Preferably, the augmented image dataset is 100 times larger than the image dataset to be trained.

[0096] Optionally, the following hyperparameters are used for model training in this application:

[0097] Model type: Binary classification;

[0098] Optimization algorithm: Adam;

[0099] Learning rate: 0.001;

[0100] Cost function: Cross-entropy loss algorithm;

[0101] Batch size: 100.

[0102] As can be seen, the machine learning-based SEI image analysis method described in the embodiments of the present invention proposes a data augmentation method for cryo-electron microscopy images of solid electrolytes, which is suitable for large-scale machine learning model training and can improve the accuracy and depth of model training.

[0103] In one embodiment of this application, the image dataset to be trained includes a plurality of first SEI images to be trained.

[0104] Step S52 above, the data augmentation operation performed on the image dataset to be trained to obtain the augmented image dataset, includes steps S521-S522:

[0105] Step S521: Calculate the gradient histogram information of each first SEI image in the image dataset to be trained, so as to determine the main body of each first SEI image respectively;

[0106] Step S522: Perform a target transformation operation on the main body of each first SEI image to obtain a second SEI image corresponding to each first SEI image, and determine the image dataset composed of all second SEI images as the enhanced image dataset.

[0107] Specifically, based on the gradient histogram information of the first SEI image, the main pixel distribution of the first SEI image can be obtained, and thus the main body of the first SEI image can be determined. The main body refers to lithium dendrites of different sizes and shapes in the SEI image.

[0108] Optionally, the target transformation operation refers to image transformation operations, including but not limited to image translation and rotation, image flipping, image magnification, and image reduction, to enrich the dataset while expanding its size. It should be noted that when performing the target transformation operation on the main body of the first SEI image in this application, the operation must be performed within the dimensions of the original image. Specifically, the original image refers to a directly obtained image dataset without data augmentation; typically, the original image size is 256*256.

[0109] As can be seen, the machine learning-based SEI image analysis method described in this embodiment of the invention proposes a data augmentation method for cryo-electron microscopy images of solid electrolytes. It uses gradient histograms to determine the main part of the SEI image and performs image transformation on the main part of the SEI image to perform data augmentation, thereby ensuring that there are enough samples to train the machine learning model.

[0110] In one embodiment of this application, step S53, inputting all image data from the enhanced image dataset into a pre-determined original training model, includes steps S531-S532:

[0111] Step S531: Convert all pixels in the second SEI image in the enhanced image dataset into the target input data type to obtain the converted image data;

[0112] Step S532: Input the converted image data into the pre-determined original training model.

[0113] Specifically, the target input data type can be matrix-like data suitable for model input.

[0114] In one embodiment of this application, as Figure 2 As shown, step S3 above, which involves performing decision impact analysis using a trained image interpretation model, includes steps S31-S33:

[0115] Step S31: Using the trained image interpretation model, calculate the score gradient of the target SEI image relative to the feature map of each convolutional layer in the image classification model.

[0116] Specifically, as mentioned earlier, the image interpretation model is used to analyze which pixels / parts in the target SEI image have a crucial impact on the morphology type prediction results of the image classification model, and displays this information visually, thereby contributing to a deeper understanding of the morphology information of solid electrolyte images. It should be noted that the image interpretation model in this application is based on an improvement of Grad-CAM (Gradient Weighted Class Activation Mapping), a method for creating heatmaps that show the most important parts of the input image for the predicted category. This helps researchers understand how image classification models work and what they learn from SEI images. It also helps identify potential biases or errors in image classification models, thereby improving the performance and reliability of image classification models.

[0117] Specifically, the fractional gradient represents the contribution of the feature map of each convolutional layer to the target class score. Traditional Grad-CAM calculates the fractional gradient of the target class relative to the feature map of the last convolutional layer. However, in this application, for each input cryo-electron microscopy SEI image, the fractional gradient of that image relative to the feature map of each convolutional layer in the image classification model is calculated separately. The calculated information is more representative of complex cryo-electron microscopy SEI images, making the analysis of SEI more in-depth and detailed.

[0118] Step S32: Determine the contribution level of each fraction gradient based on the pixel corresponding to each fraction gradient.

[0119] Specifically, since the image passes through different convolutional layers, by calculating the average, maximum, and minimum values ​​of the pixels corresponding to the fractional gradients of different convolutional layers, we can determine the effect of each pixel on gradient updates during model training, thereby judging the degree of contribution of each fractional gradient.

[0120] Step S33: Generate decision impact information based on the contribution of at least a portion of the score gradients.

[0121] In an optional implementation, step S33, generating decision influence information based on the contribution of at least a portion of the score gradient, includes steps S331-S333:

[0122] Step S331: Based on the ReLU activation function, determine whether the contribution of each fractional gradient is the same as the contribution of the target gradient.

[0123] Step S332: If the contribution of the fractional gradient is the same as the contribution of the target gradient, then retain the contribution of the target gradient.

[0124] Step S333: Generate decision impact information based on the contribution levels of all the target gradients.

[0125] Specifically, the target gradient contribution is the positive contribution among all the fractional gradient contributions. This application uses the ReLU activation function to filter the gradient contributions, selecting the positive target gradient contribution from all fractional gradient contributions.

[0126] Furthermore, after the ReLU activation function filters out the target gradient contribution from all fractional gradient contributions, the filtered target gradient contributions are retained. It should be noted that the number of target gradient contributions is at least a fraction of the contributions from all fractional gradients.

[0127] Specifically, the decision impact information can be a coarse localization map showing which parts / pixels of the SEI image are important for the prediction of the topography type.

[0128] As can be seen, the machine learning-based SEI image analysis method described in the embodiments of the present invention can calculate the gradients of the feature maps of all convolutional layers for each input cryo-electron microscopy SEI image, and obtain the gradient contribution based on the pixels corresponding to these gradients. By using the information generated by retaining the target gradient contribution, the image classification model can be visualized to see how it makes decisions and which regions in the SEI image are important for determining the morphology type. This enhances the understanding and control of SEI formation and evolution, and helps researchers establish a best practice protocol to effectively obtain future SEI images.

[0129] In one embodiment of this application, step S4, performing image analysis on the target SEI image according to at least one pre-defined image analysis method, includes steps S41-S42:

[0130] Step S41: Receive second selection information input by the user, the second selection information including the selection of at least one pre-set image analysis method;

[0131] Step S42: Analyze the target SEI image based on the second selection information.

[0132] Specifically, the image analysis in this application is also interactive. Users can select one or more analysis methods from the pre-set digital image processing analysis method menu to perform image analysis on the target SEI image. After receiving the second selection information input by the user, the electronic device performs image analysis on the target SEI image and displays the analysis results, such as outputting image texture analysis results and / or image clustering analysis results.

[0133] As can be seen, the embodiments of the present invention can improve user comfort and convenience by providing users with a user-friendly graphical interface (GUI) for selecting analysis methods.

[0134] The scope of protection for the SEI image analysis method based on machine learning described in this application is not limited to the execution order of the steps listed in this embodiment. Any solution implemented by adding, subtracting, or replacing steps in the prior art based on the principles of this application is included within the scope of protection of this application.

[0135] like Figure 3 As shown, this embodiment provides a machine learning-based SEI image analysis system, including:

[0136] Module 1 is used to acquire the target SEI image to be analyzed;

[0137] The morphology analysis module 2 is used to perform morphology analysis based on the trained image classification model, and the morphology analysis is used to determine the morphology type of the target SEI image;

[0138] The decision impact analysis module 3 is used to perform decision impact analysis through a trained image interpretation model. The decision impact analysis is used to generate decision impact information that affects the image classification model in determining the shape type of the target SEI image.

[0139] Image analysis module 4 is used to perform image analysis on the target SEI image according to at least one pre-set image analysis method and output the image analysis results, which are used to display hidden information in the target SEI image;

[0140] The morphology analysis, the decision impact analysis, and the image analysis are triggered and executed based on the user's first selection information, which includes one or more combinations of the morphology analysis, the decision impact analysis, and the image analysis.

[0141] visible, Figure 3The described machine learning-based SEI image analysis system can automatically predict the morphology type of SEI images through an image classification model, thereby improving the objectivity and accuracy of SEI image morphology type identification. It can also reveal which pixels / parts in the SEI image have a key impact on the morphology type prediction results through an image interpretation model, further ensuring the objectivity and accuracy of morphology type identification and facilitating researchers' in-depth understanding of the morphology information of solid electrolyte images. Furthermore, it can perform image analysis on target SEI images according to at least one pre-defined image analysis method and output the analysis results, enabling various digital image processing techniques for solid electrolyte images. This helps researchers identify potential or hidden information in solid electrolyte images, thereby improving the efficiency of solid electrolyte image analysis. Finally, it provides a user-friendly and interactive selection method, allowing users to arbitrarily choose analysis methods such as morphology analysis, decision influence analysis, and image analysis according to their needs, thus improving user convenience and comfort.

[0142] In one embodiment of this application, before the morphology analysis module 2 performs morphology analysis based on the trained image classification model, the system further includes:

[0143] The execution module is used to perform training operations on a pre-determined original training model based on the image dataset to be trained until convergence, thereby obtaining a trained image classification model.

[0144] The execution module performs training operations on a pre-determined original training model based on the image dataset to be trained until convergence, and obtains the trained image classification model in the following specific way:

[0145] Obtain the image dataset to be trained;

[0146] Perform data augmentation on the image dataset to be trained to obtain an augmented image dataset;

[0147] All image data in the enhanced image dataset are input into a pre-determined original training model to perform training operations on the original training model until convergence, thereby obtaining a trained image classification model.

[0148] It is evident that the system implemented in this embodiment of the invention is applicable to large-scale machine learning model training, thereby improving the accuracy and depth of model training.

[0149] In one embodiment of this application, the image dataset to be trained includes multiple first SEI images to be trained; the execution module performs data augmentation operations on the image dataset to be trained to obtain the augmented image dataset in the following specific manner:

[0150] Calculate the gradient histogram information of each first SEI image in the image dataset to be trained, so as to determine the main part of each first SEI image respectively;

[0151] A target transformation operation is performed on the main body of each first SEI image to obtain a second SEI image corresponding to each first SEI image, and the image dataset composed of all the second SEI images is determined as the enhanced image dataset.

[0152] As can be seen, the system implemented in this embodiment of the invention can use gradient histograms to determine the main part of the SEI image, thereby performing image transformation on the main part of the SEI image for data augmentation, thus ensuring that there are enough samples to train the machine learning model.

[0153] In one embodiment of this application, the execution module inputs all image data from the enhanced image dataset into a pre-determined original training model in the following manner:

[0154] The pixels in all the second SEI images in the enhanced image dataset are converted into the target input data type to obtain the converted image data.

[0155] The converted image data is input into a pre-determined original training model.

[0156] In one embodiment of this application, the decision impact analysis module 3 performs decision impact analysis using a trained image interpretation model in the following specific manner:

[0157] Using the trained image interpretation model, the score gradient of the target SEI image relative to the feature map of each convolutional layer in the image classification model is calculated respectively.

[0158] The contribution level of each fractional gradient is determined based on the pixel corresponding to each fractional gradient.

[0159] Decision impact information is generated based on the contribution of at least some of the said score gradients.

[0160] In one embodiment of this application, the decision impact analysis module 3 generates decision impact information based on the contribution of at least a portion of the score gradients in the following specific manner:

[0161] Based on the ReLU activation function, it is determined whether the contribution of each fractional gradient is the contribution of the target gradient, wherein the contribution of the target gradient is the positive contribution among all the contributions of the fractional gradients.

[0162] If the contribution of the fractional gradient is equal to the contribution of the target gradient, then the contribution of the target gradient is retained.

[0163] Decision impact information is generated based on the contribution levels of all the aforementioned target gradients.

[0164] As can be seen, the system implementing the embodiments of the present invention can calculate the gradient of the feature maps of all convolutional layers for each input cryo-electron microscopy SEI image, and obtain the gradient contribution based on the pixels corresponding to these gradients. By using the information generated by retaining the target gradient contribution, the system can visualize how the image classification model makes decisions and which regions in the SEI image are important for morphology type determination. This enhances the understanding and control of SEI formation and evolution, and helps researchers establish a best practice protocol to effectively obtain future SEI images.

[0165] In one embodiment of this application, the image analysis module 4 analyzes and processes the target SEI image according to at least one pre-set analysis method in the following specific manner:

[0166] The system receives analysis method selection information input by the user, wherein the analysis method selection information includes the selection of at least one pre-set analysis method;

[0167] The target SEI image is analyzed and processed based on the information selected according to the analysis method.

[0168] As can be seen, the system implementing the embodiments of the present invention can improve user comfort and convenience by providing users with a user-friendly graphical interface (GUI) for selecting analysis methods.

[0169] The SEI image analysis system based on machine learning described in this application can implement the SEI image analysis method based on machine learning described in this application. However, the implementation device of the SEI image analysis method based on machine learning described in this application includes, but is not limited to, the structure of the SEI image analysis system based on machine learning listed in this embodiment. All structural modifications and substitutions of the prior art made in accordance with the principles of this application are included within the protection scope of this application.

[0170] like Figure 4 As shown, this embodiment provides an electronic device, including a processor and a memory.

[0171] The memory is used to store computer programs.

[0172] The processor is used to execute the computer program stored in the memory to cause the electronic device to perform the machine learning-based SEI image analysis method described above.

[0173] In the embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, or methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of modules / units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or units may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection of apparatuses or modules or units may be electrical, mechanical, or other forms.

[0174] The modules / units described as separate components may or may not be physically separate. The components shown as modules / units may or may not be physical modules; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules / units can be selected to achieve the objectives of the embodiments of this application, depending on actual needs. For example, the functional modules / units in the various embodiments of this application may be integrated into one processing module, or each module / unit may exist physically separately, or two or more modules / units may be integrated into one module / unit.

[0175] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0176] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned machine learning-based SEI image analysis method. Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing a processor. The program can be stored in a computer-readable storage medium, which is a non-transitory medium, such as random access memory, read-only memory, flash memory, hard disk, solid-state drive, magnetic tape, floppy disk, optical disk, and any combination thereof. The aforementioned storage medium can be any available medium accessible to a computer or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., digital video disc (DVD)), or a semiconductor medium (e.g., solid-state disk (SSD)).

[0177] This application embodiment may also provide a computer program product comprising one or more computer instructions. When the computer instructions are loaded and executed on a computing device, all or part of the processes or functions described in this application embodiment are generated. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, or data center to another via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means.

[0178] When the computer program product is executed by a computer, the computer performs the method described in the foregoing method embodiments. The computer program product can be a software installation package; when the foregoing method is required, the computer program product can be downloaded and executed on the computer.

[0179] The above embodiments are merely illustrative of the principles and effects of this application and are not intended to limit this application. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of this application. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in this application should still be covered by the claims of this application.

Claims

1. A method for analyzing SEI images, characterized in that, The method includes: Obtain the target SEI image to be analyzed; A morphological analysis is performed based on a trained image classification model to determine the morphological type of the target SEI image. The image classification model is trained on a pre-determined original training model based on the image dataset to be trained. The original training model is based on the AlexNet convolutional neural network architecture and is used for image classification. Decision impact analysis is performed using a trained image interpretation model. This analysis generates decision impact information that influences the image classification model's determination of the shape type of the target SEI image. The image interpretation model is built based on Grad-CAM. The decision impact analysis using the trained image interpretation model includes: calculating the fractional gradient of the target SEI image relative to the feature map of each convolutional layer in the image classification model using the trained image interpretation model; determining the contribution of each fractional gradient based on the corresponding pixel; and generating decision impact information based on the contribution of at least some of the fractional gradients. The target SEI image is analyzed according to at least one pre-defined image analysis method, and the image analysis results are output. The image analysis results are used to display hidden information in the target SEI image. The morphology analysis, decision influence analysis, and image analysis are triggered and executed based on the user's first selection information, which includes executing the morphology analysis, decision influence analysis, and image analysis. The step of performing image analysis on the target SEI image according to at least one pre-defined image analysis method and outputting the image analysis results includes: receiving the image clustering analysis method selected by the user; performing clustering analysis on all pixels in the target SEI image based on the pixel intensity values ​​in the target SEI image using the K-Means clustering model, and outputting the clustering analysis results.

2. The method according to claim 1, characterized in that, Before performing shape analysis based on the trained image classification model, the method further includes: Based on the image dataset to be trained, the pre-determined original training model is trained until convergence, resulting in a trained image classification model. The step of performing training operations on a pre-determined original training model based on the image dataset to be trained until convergence, to obtain a trained image classification model, includes: Obtain the image dataset to be trained; Perform data augmentation on the image dataset to be trained to obtain an augmented image dataset; All image data in the enhanced image dataset are input into a pre-determined original training model to perform training operations on the original training model until convergence, thereby obtaining a trained image classification model.

3. The method according to claim 2, characterized in that, The image dataset to be trained includes multiple first SEI images to be trained; The step of performing data augmentation on the image dataset to be trained to obtain an augmented image dataset includes: Calculate the gradient histogram information of each first SEI image in the image dataset to be trained, so as to determine the main part of each first SEI image respectively; A target transformation operation is performed on the main body of each first SEI image to obtain a second SEI image corresponding to each first SEI image, and the image dataset composed of all the second SEI images is determined as the enhanced image dataset.

4. The method according to claim 3, characterized in that, Inputting all image data from the enhanced image dataset into a pre-determined original training model includes: The pixels in all the second SEI images in the enhanced image dataset are converted into the target input data type to obtain the converted image data. The converted image data is input into a pre-determined original training model.

5. The method according to claim 1, characterized in that, The step of generating decision impact information based on the contribution of at least a portion of the score gradient includes: Based on the ReLU activation function, it is determined whether the contribution of each fractional gradient is the contribution of the target gradient, wherein the contribution of the target gradient is the positive contribution among all the contributions of the fractional gradients. If the contribution of the fractional gradient is equal to the contribution of the target gradient, then the contribution of the target gradient is retained. Decision impact information is generated based on the contribution levels of all the aforementioned target gradients.

6. The method according to claim 1, characterized in that, The step of performing image analysis on the target SEI image according to at least one pre-defined image analysis method includes: Receive second selection information input by the user, the second selection information including the selection of at least one pre-set image analysis method; The target SEI image is analyzed based on the second selection information.

7. A system for analyzing SEI images, characterized in that, The system includes: The acquisition module is used to acquire the target SEI image to be analyzed. The morphology analysis module is used to perform morphology analysis based on a trained image classification model. The morphology analysis is used to determine the morphology type of the target SEI image. The image classification model is trained on a pre-determined original training model based on the image dataset to be trained. The original training model is based on the AlexNet convolutional neural network architecture and is used for image classification. A decision impact analysis module is used to perform decision impact analysis using a trained image interpretation model. This analysis generates decision impact information that influences the image classification model's determination of the shape type of the target SEI image. The image interpretation model is built based on Grad-CAM. The decision impact analysis using the trained image interpretation model includes: calculating the fractional gradient of the target SEI image relative to the feature map of each convolutional layer in the image classification model using the trained image interpretation model; determining the contribution level of each fractional gradient based on the pixels corresponding to each fractional gradient; and generating decision impact information based on the contribution levels of at least some of the fractional gradients. An image analysis module is used to perform image analysis on the target SEI image according to at least one pre-defined image analysis method and output the image analysis results, which are used to display hidden information in the target SEI image; The morphology analysis, decision influence analysis, and image analysis are triggered and executed based on the user's first selection information, which includes executing the morphology analysis, decision influence analysis, and image analysis. The step of performing image analysis on the target SEI image according to at least one pre-defined image analysis method and outputting the image analysis results includes: receiving the image clustering analysis method selected by the user; performing clustering analysis on all pixels in the target SEI image based on the pixel intensity values ​​in the target SEI image using the K-Means clustering model, and outputting the clustering analysis results.

8. An electronic device, characterized in that, include: Processor and memory; The memory is used to store computer programs; The processor is configured to execute a computer program stored in the memory to cause the electronic device to perform the method of any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method of any one of claims 1 to 6.

Images