Phase evolution analysis method of battery SEI film, electronic device and storage medium

Abstract

This application discloses a method, electronic device, and storage medium for analyzing the phase evolution of a battery SEI membrane. The method includes: acquiring multiple consecutive TEM images of the battery SEI membrane during its evolution; individually annotating the multiple TEM images to obtain multiple SEI component annotated images; performing instance segmentation on the multiple SEI component annotated images to obtain multiple SEI electron microscope images, wherein the SEI electron microscope images are labeled with the SEI components of the battery SEI membrane; and performing target tracking of the SEI components on the multiple SEI electron microscope images. This application enables automatic, efficient, and accurate analysis of the phase of a battery SEI membrane.

Description

Technical Field

[0001] The present application relates to the field of battery technology, and in particular to a method for analyzing the phase evolution of a battery SEI film, an electronic device, and a storage medium. Background Art

[0002] The Solid Electrolyte Interface (SEI) film is a crucial component of batteries. Currently, manual analysis of its phase evolution is primarily performed, which has significant efficiency and accuracy issues. Instance segmentation, primarily targeting SEM (Scanning Electron Microscope) images, is currently performed using convolutional layer feature extraction and a U-Net architecture. This approach is sensitive to noise and, while fast, suffers from low accuracy. Summary of the Invention

[0003] The present application at least provides a method for analyzing the phase evolution of a battery SEI film, an electronic device, and a storage medium.

[0004] The present application provides a method for phase analysis of a battery SEI film, comprising: obtaining a plurality of continuous TEM images during the evolution of the battery SEI film; separately annotating the plurality of TEM images to obtain a plurality of SEI component annotated images; performing instance segmentation on the plurality of SEI component annotated images to obtain a plurality of SEI electron microscope images, wherein the SEI electron microscope images are marked with SEI components of the battery SEI film; and performing target tracking of the SEI components on the plurality of SEI electron microscope images.

[0005] In the above scheme, by obtaining multiple consecutive TEM images during the evolution of the battery SEI film, the multiple TEM images are separately annotated to obtain multiple SEI component annotated images, and the multiple SEI component annotated images are instance segmented to obtain multiple SEI electron microscope images, wherein the SEI components of the battery SEI film are marked on the SEI electron microscope images, and the SEI components are tracked on the multiple SEI electron microscope images to achieve high-accuracy nanoscale phase segmentation on the TEM images, providing a reliable basis for subsequent target tracking, effectively tracking the dynamic changes of the SEI film during the battery charging and discharging process, and realizing automatic, efficient and accurate analysis of the phase evolution process of the battery SEI film, thereby improving the accuracy of SEI component analysis and helping to understand the formation and evolution of the SEI film and its impact on battery performance, which is of great significance for improving the safety and cycle life of the battery.

[0006] In some embodiments, target tracking of SEI components on multiple SEI electron microscope images includes: detecting multiple SEI component regions from the multiple SEI electron microscope images; extracting appearance features of the multiple SEI component regions based on the multiple SEI component regions; and matching the multiple SEI component regions and the appearance features to perform target tracking.

[0007] In the above scheme, multiple SEI component regions are detected from multiple SEI electron microscope images respectively, and the appearance features of the multiple SEI component regions are extracted based on the multiple SEI component regions. Matching is performed based on the multiple SEI component regions and the appearance features to perform target tracking, thereby preventing continuous jumping of components of the same category during SEI component tracking, achieving continuous stability of target tracking, and realizing accurate monitoring and following.

[0008] In some embodiments, target tracking of SEI components on multiple SEI electron microscope images includes: obtaining a motion model of the corresponding SEI component on the SEI electron microscope image; in response to failure of the motion model of the SEI component, using a component segmentation mask to remove the virtual box corresponding to the motion model of the SEI component on the SEI electron microscope image.

[0009] In the above scheme, the SEI components are tracked by obtaining the motion model of the corresponding SEI components on the SEI electron microscope image. In response to the failure of the motion model of the SEI components, the virtual frame corresponding to the motion model of the SEI components on the SEI electron microscope image can be removed using the component segmentation mask, thereby achieving effective management, significantly improving the accuracy and reliability of target tracking, and achieving continuous stability of target tracking, so as to realize accurate monitoring and following.

[0010] In some embodiments, performing target tracking of SEI components on a plurality of SEI electron microscope images includes adjusting a maximum lifespan of a tracker performing target tracking from a first preset value to a second preset value, wherein the first preset value is less than the second preset value.

[0011] In the above solution, by adjusting the maximum lifespan of the tracker performing target tracking from a first preset value to a second preset value, wherein the first preset value is less than the second preset value, the tracker is prevented from being deleted prematurely, thereby achieving effective management, significantly improving the accuracy and reliability of target tracking, and achieving continuous stability of target tracking, thereby realizing accurate and correct monitoring and tracking.

[0012] In some embodiments, during target tracking, the cost functions of different categories of multiple SEI component regions are infinite.

[0013] In the above scheme, by setting the value of the cost function of different categories of multiple SEI component areas to infinity, the jumping of the same ID between different categories is avoided, effective management is achieved, the accuracy and reliability of target tracking are significantly improved, and the continuous stability of target tracking is achieved, so as to realize accurate monitoring and tracking.

[0014] In some embodiments, annotating a TEM image to obtain an SEI component annotated image includes: pre-annotating the TEM image to obtain a pre-annotated image, wherein the pre-annotated image includes pre-annotation results of SEI components; confirming component boundaries of the pre-annotation results, and in response to receiving a modification operation on the component boundaries of the pre-annotation results, obtaining the modified component boundaries of the pre-annotation results, thereby obtaining the SEI component annotated image.

[0015] In the above scheme, a pre-annotated image is obtained by pre-annotating the TEM image, wherein the pre-annotated image includes the pre-annotated results of the SEI components, the component boundaries of the pre-annotated results are confirmed, and in response to receiving a modification operation on the component boundaries of the pre-annotated results, the component boundaries of the modified pre-annotated results are obtained, thereby obtaining an SEI component annotated image, realizing effective semi-automatic annotation of TEM images, solving the problems of difficulty in determining the boundaries of the SEI component regions and low annotation efficiency during manual annotation, and being able to significantly improve the efficiency and accuracy of SEI component annotation, with high flexibility, which is conducive to achieving highly accurate nanoscale phase segmentation on TEM images.

[0016] In some embodiments, pre-labeling the TEM image includes: inputting the TEM image into a semantic segmentation model to obtain the SEI component region in the TEM image; and scanning the SEI component region using a component classification model to obtain a pre-labeled image.

[0017] In the above scheme, the SEI area in the TEM image is obtained by inputting the TEM image into the semantic segmentation model; the SEI area is scanned using the component classification model to obtain a pre-labeled image, that is, the dual processing of the semantic segmentation model and the component classification model is used to achieve automatic pre-labeling, which is beneficial to significantly improve the efficiency and accuracy of TEM image labeling and reduce the time and error rate of manual operation.

[0018] In some embodiments, pre-labeling the TEM image includes: inputting the TEM image into a semantic segmentation model to obtain SEI regions in the TEM image, so as to obtain a pre-labeled image based on the SEI regions.

[0019] In the above scheme, the SEI region in the TEM image is obtained by inputting the TEM image into the semantic segmentation model, and a pre-labeled image is obtained based on the SEI region. Pre-labeling through the semantic segmentation model can more accurately identify and analyze the SEI region, significantly improving the efficiency and accuracy of TEM image labeling and reducing the time and error rate of manual operation.

[0020] In some embodiments, pre-labeling the TEM image includes: pre-labeling the TEM image using an instance segmentation model to obtain a pre-labeled image.

[0021] In the above scheme, the instance segmentation model is used to pre-annotate TEM images to obtain pre-annotated images and realize automatic pre-annotation, which can significantly improve the annotation efficiency and automatically identify and locate complex structural features, which is conducive to the realization of highly accurate nanoscale phase segmentation on TEM images.

[0022] In some embodiments, a method for analyzing the phase evolution of a battery SEI film further includes: using multiple SEI component labeled images to train an instance segmentation model.

[0023] In the above scheme, by using multiple SEI component annotated images to train the instance segmentation model, a reliable basis is provided for achieving highly accurate nanoscale phase segmentation on TEM images, thereby enabling automatic, efficient and accurate analysis of the phase evolution process of the battery SEI film.

[0024] The present application provides an electronic device, including a memory and a processor, wherein the processor is used to execute program instructions stored in the memory to implement the above-mentioned battery SEI film phase evolution analysis method.

[0025] The present application provides a computer-readable storage medium having program instructions stored thereon. When the program instructions are executed by a processor, the above-mentioned battery SEI film phase evolution analysis method is implemented.

[0026] In the above scheme, a TEM image of the battery SEI film is obtained and the TEM image is annotated to obtain an SEI component annotated image. The SEI component annotated image is instance segmented using an instance segmentation model to obtain an SEI electron microscope image, wherein the SEI components of the battery SEI film are marked on the SEI electron microscope image. The SEI electron microscope image is input into the target tracking model, and the SEI components are managed to analyze the SEI components, thereby achieving highly accurate nanoscale phase segmentation on the TEM image, providing a reliable basis for subsequent target tracking, effectively tracking the dynamic changes of the SEI film during the battery charging and discharging process, and realizing automatic, efficient and accurate analysis of the phase evolution process of the battery SEI film, thereby improving the accuracy of SEI component analysis and helping to understand the formation and evolution of the SEI film and its impact on battery performance, which is of great significance for improving battery safety and cycle life.

[0027] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] The drawings herein are incorporated into and constitute a part of the specification. These drawings illustrate embodiments consistent with the present application and, together with the specification, are used to illustrate the technical solutions of the present application.

[0029] Figure 1 is a flow chart of a method for analyzing the phase evolution of a battery SEI film according to some embodiments of the present application;

[0030] Figure 2 This is a structural diagram of an instance segmentation model based on a general R-CNN framework in some embodiments of the present application;

[0031] Figure 3 is a schematic diagram of a target tracking model in some embodiments of the present application;

[0032] Figure 4 is a flow chart of the SEI component evolution analysis process of some embodiments of the present application;

[0033] Figure 5 is a flowchart of the SEI component labeling process in some embodiments of the present application;

[0034] Figure 6 Schematic diagram of the structure of the battery SEI film phase evolution analysis system provided in some embodiments of the present application;

[0035] Figure 7 is a schematic structural diagram of an electronic device provided in some embodiments of the present application;

[0036] Figure 8It is a schematic diagram of the structure of the computer-readable storage medium provided in some embodiments of the present application. DETAILED DESCRIPTION

[0037] The following describes the embodiments of the present application in detail with reference to the accompanying drawings.

[0038] In the following description, for the purpose of explanation rather than limitation, specific details such as specific system structures, interfaces, and technologies are provided to facilitate a thorough understanding of the present application.

[0039] The term "and / or" in this article is simply a description of the association relationship of associated objects, indicating that three relationships can exist. For example, A and / or B can represent three situations: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated objects are in an "or" relationship. In addition, "many" in this article means two or more than two. In addition, the term "at least one" in this article means any combination of at least two of any one or more of a plurality of. For example, including at least one of A, B, and C can mean including any one or more elements selected from the set consisting of A, B, and C.

[0040] Currently, the analysis of SEI film phase evolution is primarily done manually, but this method suffers from significant efficiency and accuracy issues. Instance segmentation, which primarily targets SEM images, is currently based on convolutional feature extraction and a U-Net architecture. This approach is sensitive to noise and, while fast, suffers from low precision. This makes it unsuitable for nanoscale instance segmentation in TEM (Transmission Electron Microscope) images, as it results in numerous artifacts and inaccurate segmentation. In other words, there is currently no method for nanoscale phase segmentation in TEM images, and in the field of target tracking, there is no specific target tracking technology for SEI TEM images.

[0041] To this end, by obtaining multiple consecutive TEM images during the evolution of the battery SEI film, the multiple TEM images are separately annotated to obtain multiple SEI component annotated images, and the multiple SEI component annotated images are instance segmented to obtain multiple SEI electron microscope images, wherein the SEI components of the battery SEI film are marked on the SEI electron microscope images. The SEI components are tracked on the multiple SEI electron microscope images to analyze the SEI components and achieve highly accurate nanoscale phase segmentation on the TEM images, providing a reliable basis for subsequent target tracking, effectively tracking the dynamic changes of the SEI film during the battery charging and discharging process, and achieving automatic, efficient and accurate analysis of the phase evolution process of the battery SEI film, thereby improving the accuracy of SEI component analysis and helping to understand the formation and evolution of the SEI film and its impact on battery performance, which is of great significance for improving battery safety and cycle life.

[0042] See also Figure 1 , Figure 1 Flowchart of the method for analyzing the phase evolution of the battery SEI film according to some embodiments of the present application. Figure 1 As shown, the method for analyzing the phase evolution of the battery SEI film includes: step S11: acquiring multiple consecutive TEM images of the battery SEI film during its evolution process; step S12: separately annotating the multiple TEM images to obtain multiple SEI component annotated images; step S13: performing instance segmentation on the multiple SEI component annotated images to obtain multiple SEI electron microscopy images, wherein the SEI electron microscopy images are labeled with the SEI components of the battery SEI film; and step S14: performing target tracking of the SEI components on the multiple SEI electron microscopy images.

[0043] Acquire multiple consecutive TEM images of the battery SEI film during its evolution. TEM images are high-resolution electron microscope images that clearly show the material's microstructure. Label each of these images separately, using different colors. For example, labeling copper oxide (CuO) in the battery SEI film with blue and lithium (Li) in the battery SEI film with red, yields multiple labeled images of the SEI components.

[0044] Instance segmentation is performed on multiple SEI component annotated images to obtain multiple SEI electron microscope images, on which SEI components of the battery SEI film are marked. For example, an instance segmentation model can be used to perform instance segmentation on each SEI component annotated image, wherein the instance segmentation model can adopt a generalized R-CNN (Generalized R-CNN) framework, such as Figure 2As shown, the Generalized R-CNN (GRC) framework can include a feature backbone network 21, a region proposal network (RPN) 22, and a cascaded classification and regression network 23. The feature backbone network 21 converts the input image into a high-dimensional feature map. Using a feature pyramid network (SFP), for example, a lightweight feature pyramid network, it achieves multi-scale feature fusion in a simpler manner. This can be built using a vision transformer (ViT). This gives the large instance segmentation model, Generalized R-CNN, good generalization and enables high-precision SEI component identification with minimal data fine-tuning. The RPN 22 is used to generate candidate SEI component regions that may contain SEI components on the high-dimensional feature map. In the cascaded classification and regression network 23, the features of the candidate SEI component regions are first cropped and pooled. For example, the features of candidate SEI component regions of different sizes are unified to a fixed size. Category prediction and bounding box optimization are then performed on the candidate SEI component regions. This results in an SEI electron microscope image labeled with the SEI components of the battery SEI membrane, such as copper oxide (CuO) and lithium (Li). After constructing the instance segmentation model architecture, the model is trained using a training set. The TEM images of the battery SEI membrane in the training set are also referred to as training TEM images. During the training process, the training TEM images in the training set are annotated to produce training SEI component annotated images, i.e., SEI component annotated images used for training. The annotations for the training TEM images can be the same as those for the TEM images, for example, both can be annotated using different colors.

[0045] A target tracking model can be used to track the target of SEI components in multiple SEI electron microscope images. For example, the target tracking model can be DeepSORT, which is a detection-based multi-object tracking (MOT) algorithm that achieves efficient and accurate target tracking by combining deep learning features and Kalman filtering. Figure 3 As shown, the target tracking model 30 may include a target detection module 31 and an appearance feature extractor 32. The target detection module 31 may detect multiple SEI component regions from multiple SEI electron microscope images. The SEI component regions may be represented by bounding boxes of the SEI component regions or features of the SEI components. The appearance feature extractor 32 may extract appearance features of the multiple SEI component regions based on the detected multiple SEI component regions, thereby managing the SEI components. The appearance feature extractor 32 may be a MobileNetv2 appearance feature extractor.

[0046] In the above scheme, by obtaining multiple consecutive TEM images during the evolution of the battery SEI film, the multiple TEM images are separately annotated to obtain multiple SEI component annotated images, and the multiple SEI component annotated images are instance segmented to obtain multiple SEI electron microscope images, wherein the SEI components of the battery SEI film are marked on the SEI electron microscope images, and the SEI components are tracked on the multiple SEI electron microscope images to achieve high-accuracy nanoscale phase segmentation on the TEM images, providing a reliable basis for subsequent target tracking, effectively tracking the dynamic changes of the SEI film during the battery charging and discharging process, and realizing automatic, efficient and accurate analysis of the phase evolution process of the battery SEI film, thereby improving the accuracy of SEI component analysis and helping to understand the formation and evolution of the SEI film and its impact on battery performance, which is of great significance for improving the safety and cycle life of the battery.

[0047] In some embodiments, target tracking of SEI components on multiple SEI electron microscope images includes: detecting multiple SEI component regions from the multiple SEI electron microscope images; extracting appearance features of the multiple SEI component regions based on the multiple SEI component regions; and matching the multiple SEI component regions and the appearance features to perform target tracking.

[0048] like Figure 3As shown, target tracking of SEI component regions can be achieved using a target tracking model 30. Target tracking model 30 can be DeepSORT, a detection-based multi-object tracking (MOT) algorithm that achieves efficient and accurate target tracking by combining deep learning features with Kalman filtering. Target tracking model 30 includes a target detection module 31, an appearance feature extractor 32, and a feature matching module 33. Target detection module 31 can detect multiple SEI component regions from an SEI electron microscope image, such as copper oxide regions and lithium regions in a battery SEI membrane. SEI component regions are detected based on their features and represented by bounding boxes on the SEI electron microscope image. For example, copper oxide regions are detected based on the features of copper oxide in a battery SEI membrane and represented by corresponding bounding boxes. Appearance feature extractor 32 can extract appearance features of a detected SEI component region, such as the appearance features of copper oxide in a battery SEI membrane. The feature matching module 33 matches the appearance features of the extracted SEI component regions with the features of the SEI component regions to track the SEI components. For example, the feature matching module 33 matches the appearance features of the copper oxide in the battery SEI membrane with the features of the copper oxide in the battery SEI membrane to confirm that they are both copper oxide in the battery SEI membrane and are of the same type. This allows for continuous tracking of these components to prevent continuous jumps in the same type of components during SEI component tracking.

[0049] In the above scheme, multiple SEI component regions are detected from multiple SEI electron microscope images respectively, and the appearance features of the multiple SEI component regions are extracted based on the multiple SEI component regions. Matching is performed based on the multiple SEI component regions and the appearance features to perform target tracking, thereby preventing continuous jumping of components of the same category during SEI component tracking, achieving continuous stability of target tracking, and realizing accurate monitoring and following.

[0050] In some embodiments, target tracking of SEI components on multiple SEI electron microscope images includes: obtaining a motion model of the corresponding SEI component on the SEI electron microscope image; in response to failure of the motion model of the SEI component, using a component segmentation mask to remove the virtual box corresponding to the motion model of the SEI component on the SEI electron microscope image.

[0051] During the target tracking process of the SEI component, a component segmentation mask is provided, that is, a component segmentation mask is provided in the target tracking model 30, which can remove the virtual box corresponding to the motion model of the SEI component. First, during the target tracking process, the target tracking model 30 sequentially obtains the virtual box corresponding to the motion model of the corresponding SEI component on each SEI electron microscope image. The virtual box corresponding to the motion model is used to track the SEI component. In the target tracking model 30, the component segmentation mask is used to detect and remove the virtual box generated by the motion model of the failed SEI component. During the target tracking process, the target tracking model 30 analyzes and extracts the motion model of the SEI component. If the motion model of a certain SEI component fails during the monitoring process, for example, the SEI component does not have a corresponding SEI component area, indicating that the SEI component fails, the target tracking model 30 will automatically activate the component segmentation mask, and remove the virtual box corresponding to the failed motion model from the corresponding image through the component segmentation mask.

[0052] In the above scheme, the SEI components are tracked by obtaining the motion model of the corresponding SEI components on the SEI electron microscope image. In response to the failure of the motion model of the SEI components, the virtual frame corresponding to the motion model of the SEI components on the SEI electron microscope image can be removed using the component segmentation mask, thereby achieving effective management, significantly improving the accuracy and reliability of target tracking, and achieving continuous stability of target tracking, so as to realize accurate monitoring and following.

[0053] In some embodiments, performing target tracking of SEI components on a plurality of SEI electron microscope images includes adjusting a maximum lifespan of a tracker performing target tracking from a first preset value to a second preset value, wherein the first preset value is less than the second preset value.

[0054] During the target tracking process, the tracker in the target tracking model 30 can be a state estimate, and a Kalman filter is used for prediction. The maximum life refers to the length of time the tracker can continue to work without failure. By adjusting the maximum life from a first preset value to a second preset value, wherein the first preset value is less than the second preset value, the working time or performance duration of the tracker is extended. For example, the first preset value can be 30 frames, and the second preset value can be 100 frames. For a certain SEI component. For example, the lithium of the battery SEI film, if the maximum life of the tracker is set to 30 frames, the ID corresponding to the motion model of the lithium has appeared in the previous 30 frames, which means that the lithium in the previous 30 frames is the same lithium. If lithium appears in the 31st frame, the lithium in the previous 30 frames is deleted, and it is impossible to determine whether it is the same lithium as the lithium in the previous 30 frames. If the maximum life of the tracker is set to 100 frames, and the ID corresponding to the motion model of the lithium has appeared in the first 100 frames, it means that the lithium in the first 100 frames is the same lithium. If the lithium appears in the 101st frame, the lithium in the previous 100 frames will be deleted to avoid the tracker being deleted prematurely.

[0055] In the above solution, by adjusting the maximum lifespan of the tracker performing target tracking from a first preset value to a second preset value, wherein the first preset value is less than the second preset value, the tracker is prevented from being deleted prematurely, thereby achieving effective management, significantly improving the accuracy and reliability of target tracking, and achieving continuous stability of target tracking, thereby realizing accurate and correct monitoring and tracking.

[0056] In some embodiments, during target tracking, the cost functions of different categories of multiple SEI component regions are infinite.

[0057] In target tracking, different categories of SEI components have different categories of cost functions, and the values ​​of the cost functions of different categories are used to track each category. During the tracking process, assuming that the value of the cost function of different categories is a certain value, at this time, for example, based on the appearance extracted by the appearance feature extractor 32, such as the shape, if the ID of the motion model of a certain SEI component, such as lithium, is tracked as the ID of another SEI component, such as copper oxide, that is, at this time both SEI components use the same ID, resulting in the same ID jumping between different categories. Assuming that the value of the cost function of different categories is infinite, then based on the value of the cost function of different categories, the ID of the motion model of lithium will not be tracked as the ID of copper oxide, and such matches will be automatically excluded to prevent the same ID from jumping between different categories.

[0058] In the above scheme, by setting the value of the cost function of different categories of multiple SEI component areas to infinity, the jumping of the same ID between different categories is avoided, effective management is achieved, the accuracy and reliability of target tracking are significantly improved, and the continuous stability of target tracking is achieved, so as to realize accurate monitoring and tracking.

[0059] In some embodiments, annotating a TEM image to obtain an SEI component annotated image includes: pre-annotating the TEM image to obtain a pre-annotated image, wherein the pre-annotated image includes pre-annotation results of SEI components; confirming component boundaries of the pre-annotation results, and in response to receiving a modification operation on the component boundaries of the pre-annotation results, obtaining the modified component boundaries of the pre-annotation results, thereby obtaining the SEI component annotated image.

[0060] The pre-annotated image can be in the form of the original TEM image plus the pre-annotated results of the SEI components. The component boundaries of the pre-annotated results can be component region boundaries, for example, the region boundaries of copper oxide. The component boundaries of the pre-annotated results can be confirmed in reciprocal space using a fast Fourier transform (FFT). Reciprocal space corresponds to real space and is connected through Fourier transform. Reciprocal space is the Fourier conjugate space of real space. Modifying the component boundaries of the pre-annotated results can mean manually modifying the component boundaries of the pre-annotated results, that is, manually fine-tuning the component boundaries of the pre-annotated results. After manual fine-tuning, the modified component boundaries of the pre-annotated results are obtained, thereby obtaining the SEI component annotated image.

[0061] In the above scheme, a pre-annotated image is obtained by pre-annotating the TEM image, wherein the pre-annotated image includes the pre-annotated results of the SEI components, the component boundaries of the pre-annotated results are confirmed, and in response to receiving a modification operation on the component boundaries of the pre-annotated results, the component boundaries of the modified pre-annotated results are obtained, thereby obtaining an SEI component annotated image, realizing effective semi-automatic annotation of TEM images, solving the problems of difficulty in determining the boundaries of the SEI component regions and low annotation efficiency during manual annotation, and being able to significantly improve the efficiency and accuracy of SEI component annotation, with high flexibility, which is conducive to achieving highly accurate nanoscale phase segmentation on TEM images.

[0062] In some embodiments, pre-labeling the TEM image includes: inputting the TEM image into a semantic segmentation model to obtain the SEI region in the TEM image; and scanning the SEI region using a component classification model to obtain a pre-labeled image.

[0063] The semantic segmentation model can be a U-Net architecture, a convolutional neural network architecture consisting of an encoder, decoder, connection components, and output components. The component classification model can be a CNN-based model or a Transformer model. The semantic segmentation and component classification models can be trained using corresponding training sets. The TEM image is input into the U-Net architecture, which identifies and extracts SEI component regions in the TEM image. The SEI component regions are then scanned using a component classification model, for example, a CNN-based model, for classification, resulting in pre-labeled results and a pre-labeled image.

[0064] In the above scheme, the SEI area in the TEM image is obtained by inputting the TEM image into the semantic segmentation model; the SEI area is scanned using the component classification model to obtain a pre-labeled image, that is, the dual processing of the semantic segmentation model and the component classification model is used to achieve automatic pre-labeling, which is beneficial to significantly improve the efficiency and accuracy of TEM image labeling and reduce the time and error rate of manual operation.

[0065] In some embodiments, pre-labeling the TEM image includes: inputting the TEM image into a semantic segmentation model to obtain SEI regions in the TEM image, so as to obtain a pre-labeled image based on the SEI regions.

[0066] The semantic segmentation model can be a U-Net architecture, which is a convolutional neural network architecture. The TEM image is input into the U-Net architecture, and the U-Net architecture identifies and extracts the SEI area in the TEM image, thereby obtaining a pre-labeled image based on the SEI area.

[0067] In the above scheme, the SEI region in the TEM image is obtained by inputting the TEM image into the semantic segmentation model, and a pre-labeled image is obtained based on the SEI region. Pre-labeling through the semantic segmentation model can more accurately identify and analyze the SEI region, significantly improving the efficiency and accuracy of TEM image labeling and reducing the time and error rate of manual operation.

[0068] In some embodiments, pre-labeling the TEM image includes: pre-labeling the TEM image using an instance segmentation model to obtain a pre-labeled image.

[0069] The instance segmentation model can adopt the generalized R-CNN framework, such as Figure 2As shown, the generalized R-CNN (Generalized R-CNN) framework may include a feature backbone network 21, a region proposal network (RPN) 22, and a cascade classification regression network 23, wherein the generalized R-CNN (Generalized R-CNN) framework is trained using a training set. During the training process, the training TEM images in the training set are annotated to obtain training SEI component annotated images, i.e., SEI component annotated images used for training. The annotations of the training TEM images may be the same as the annotations of the TEM images, for example, both may be annotated using different colors. The generalized R-CNN architecture is used to pre-annotate the TEM images, i.e., the component regions segmented by the generalized R-CNN are pre-annotated to obtain pre-annotated results, thereby obtaining pre-annotated images.

[0070] In the above scheme, the instance segmentation model is used to pre-annotate TEM images to obtain pre-annotated images and realize automatic pre-annotation, which can significantly improve the annotation efficiency and automatically identify and locate complex structural features, which is conducive to the realization of highly accurate nanoscale phase segmentation on TEM images.

[0071] In some embodiments, the method further comprises: using the plurality of SEI component annotated images for training an instance segmentation model.

[0072] Multiple SEI component annotated images obtained by separately annotating multiple TEM images of the battery SEI membrane can be used to train the instance segmentation model, where the multiple TEM images of the battery SEI membrane can be newly collected data. In other words, the multiple SEI component annotated images obtained by separately annotating multiple TEM images of the battery SEI membrane can also be used as training SEI component annotated images in the training set for training the instance segmentation model.

[0073] In the above scheme, by using multiple SEI component annotated images to train the instance segmentation model, a reliable basis is provided for achieving highly accurate nanoscale phase segmentation on TEM images, thereby enabling automatic, efficient and accurate analysis of the phase evolution process of the battery SEI film.

[0074] In some embodiments, as Figure 4 As shown, Figure 4 Flowchart of the SEI component evolution analysis process of some embodiments of the present application, including the following steps:

[0075] Step S41: SEI component labeling.

[0076] Specifically, if Figure 5 As shown, Figure 5This is a flowchart of the SEI component labeling process in some embodiments of the present application. The SEI component labeling includes the following steps:

[0077] Step S51: using a semantic segmentation model to identify SEI component regions, wherein the semantic segmentation model may be a U-net architecture.

[0078] Step S52: Scan the SEI component region using the component classification model.

[0079] Step S53: Obtain a pre-labeled image.

[0080] The pre-labeled image may be obtained based on the SEI component region identified in step S51 , or may be obtained by scanning the SEI component region.

[0081] Step S54: Use FFT to confirm the component boundaries in the reciprocal space and manually fine-tune the pre-labeling.

[0082] Step S55: Obtain the SEI component annotated image.

[0083] SEI component annotated images can be used to train instance segmentation models, and can also be called training SEI component annotated images, which are used in the training set of instance segmentation models.

[0084] The instance segmentation model can adopt the general R-CNN framework, such as Figure 2As shown, the Generalized R-CNN (Generalized R-CNN) framework can include a feature backbone network 21, a region proposal network (RPN) 22, and a cascaded classification and regression network 23. The feature backbone network 21 converts the input image into a high-dimensional feature map. Using a feature pyramid network (SFP), for example, a lightweight feature pyramid network, it achieves multi-scale feature fusion in a simpler manner. This can be built using a vision transformer (ViT). This gives the Generalized R-CNN instance segmentation model good generalization and enables high-precision SEI component identification with minimal data fine-tuning. The RPN 22 is used to generate candidate SEI component regions that may contain SEI components on the high-dimensional feature map. In the cascaded classification and regression network 23, the features of the candidate SEI component regions are first cropped and pooled. For example, the features of candidate SEI component regions of different sizes are unified to a fixed size. Category prediction and bounding box optimization are then performed on the candidate SEI component regions. This results in an SEI electron microscope image labeled with the SEI components of the battery SEI membrane, such as copper oxide (CuO) and lithium (Li). After constructing the instance segmentation model architecture, the model is trained using a training set. The TEM images of the battery SEI membrane in the training set are also referred to as training TEM images. During the training process, the training TEM images in the training set are annotated to produce training SEI component annotated images, i.e., SEI component annotated images used for training. The annotations for the training TEM images can be the same as those for the TEM images, for example, both can be annotated using different colors.

[0085] Since the feature pyramid network of the instance segmentation model uses the Vision Transformer (ViT) module, a small number of training SEI component annotated images can be used for training and fine-tuning to achieve high-precision SEI component recognition.

[0086] Step S56: If the TEM image of the battery SEI film is newly acquired, pre-label it using the instance segmentation model. After executing step S56, execute steps S54 and S55.

[0087] It should be noted that, for a newly collected TEM image of the battery SEI film, step S56 can be executed simultaneously with step S52 or step S51 , that is, the execution order of the steps is not based on the step numbers.

[0088] Step S42: instance segmentation.

[0089] The trained instance segmentation model is used to perform instance segmentation on the SEI component annotated image to obtain the SEI electron microscopy image. The instance segmentation model is described above for details.

[0090] Step S43: Target tracking.

[0091] The target tracking model 30 is used to track the SEI components in multiple SEI electron microscope images.

[0092] The target tracking model 30 may be DeepSORT (Deep Simple Online and Realtime Tracking), which is a detection-based multi-object tracking (MOT) algorithm that achieves efficient and accurate target tracking by combining deep learning features and Kalman filtering. For example, Figure 3 As shown, the target tracking model 30 may include a target detection module 31, an appearance feature extractor 32 and a feature matching module 33. The target detection module 31 may detect multiple SEI component regions from the SEI electron microscope image, for example, the copper oxide region and the lithium region of the battery SEI film. The SEI component region is detected by the characteristics of the SEI component and is represented by the bounding box of the SEI component region on the SEI electron microscope image. For example, the copper oxide region is detected by the characteristics of the copper oxide of the battery SEI film and is represented by the corresponding bounding box. The appearance feature extractor 32 may extract the appearance features of the SEI component region based on the detected SEI component region, for example, the appearance features of the copper oxide of the battery SEI film. The appearance feature extractor 32 may be a MobileNetv2 appearance feature extractor.

[0093] Furthermore, the target detection module 31 can detect multiple SEI component regions from a specific SEI electron microscope image, such as copper oxide regions and lithium regions in a battery SEI membrane. The SEI component regions are detected based on SEI component features and represented by bounding boxes on the SEI electron microscope image. For example, a copper oxide region is detected based on the copper oxide features in a battery SEI membrane and represented by a corresponding bounding box. The appearance feature extractor 32 can extract the appearance features of a detected SEI component region, such as the appearance features of copper oxide in a battery SEI membrane. The feature matching module 33 matches the extracted appearance features of the SEI component region with the features of the SEI component region to track the SEI component. For example, the feature matching module 33 matches the appearance features of copper oxide in a battery SEI membrane with the features of copper oxide in a battery SEI membrane to confirm that both are copper oxide in the battery SEI membrane and belong to the same category, allowing for continuous tracking of the components and preventing continuous jumps in the same category during SEI component tracking.

[0094] Furthermore, in the target tracking process of the SEI component, a component segmentation mask is provided, that is, a component segmentation mask is provided in the target tracking model 30, which can remove the virtual box corresponding to the motion model of the SEI component. First, in the target tracking process, the target tracking model 30 sequentially obtains the virtual box corresponding to the motion model of the corresponding SEI component on each SEI electron microscope image. The virtual box corresponding to the motion model is used to track the SEI component. In the target tracking model 30, the component segmentation mask is used to detect and remove the virtual box generated by the motion model of the failed SEI component. In the target tracking process, the target tracking model 30 analyzes and extracts the motion model of the SEI component. If the motion model of a certain SEI component fails during the monitoring process, for example, the SEI component does not have a corresponding SEI component area, indicating that the SEI component fails, the target tracking model 30 will automatically activate the component segmentation mask, and remove the virtual box corresponding to the failed motion model from the corresponding image through the component segmentation mask.

[0095] Furthermore, in the target tracking process of the SEI component, the tracker in the target tracking model 30 can be a state estimate, and a Kalman filter is used for prediction. The maximum life refers to the length of time the tracker can continue to work without failure. By adjusting the maximum life from a first preset value to a second preset value, wherein the first preset value is less than the second preset value, the working time of the tracker or the time the performance is maintained is extended. For example, the first preset value can be 30 frames, and the second preset value can be 100 frames. For a certain SEI component, for example, lithium in the SEI film of a battery, if the maximum life of the tracker is set to 30 frames, and the ID corresponding to the motion model of the lithium has appeared in the previous 30 frames, it means that the lithium in the previous 30 frames is the same lithium. If lithium appears in the 31st frame, the lithium in the previous 30 frames is deleted, and it is impossible to determine whether it is the same lithium as the lithium in the previous 30 frames. If the maximum lifespan of the tracker is set to 100 frames, and the ID corresponding to the motion model of the lithium has appeared in the first 100 frames, it means that the lithium in the first 100 frames is the same lithium. If the lithium appears in the 101st frame, the lithium in the previous 100 frames will be deleted to avoid the tracker being deleted prematurely.

[0096] Furthermore, in the target tracking process of the SEI component, that is, in the target tracking model, different categories of SEI components have different categories of cost functions, and the values ​​of the cost functions of different categories are used to track each category. During the tracking process, assuming that the value of the cost function of different categories is a certain value, at this time, for example, based on the appearance extracted by the appearance feature extractor 32, such as the shape, if the ID of the motion model of a certain SEI component, such as lithium, is tracked as the ID of another SEI component, such as copper oxide, that is, at this time both SEI components use the same ID, resulting in the same ID jumping between different categories. Assuming that the value of the cost function of different categories is infinite, then according to the value of the cost function of different categories, the ID of the motion model of lithium will not be tracked as the ID of copper oxide, and such matching will be automatically excluded to prevent the same ID from jumping between different categories.

[0097] See also Figure 6 , Figure 6: This is a structural schematic diagram of a phase evolution analysis system for a battery SEI film provided in some embodiments of the present application. The phase evolution analysis system includes an image acquisition module 61, an annotation module 62, an instance segmentation module 63, and a target tracking module 64. The image acquisition module 61 is used to acquire a plurality of continuous TEM images during the evolution of the battery SEI film. The annotation module 62 is used to separately annotate the plurality of TEM images to obtain a plurality of SEI component annotated images. The instance segmentation module 63 is used to perform instance segmentation on the plurality of SEI component annotated images to obtain a plurality of SEI electron microscope images, wherein the SEI electron microscope images are marked with SEI components of the battery SEI film. The target tracking module 64 is used to perform target tracking of SEI components on the plurality of SEI electron microscope images.

[0098] In the above scheme, by obtaining multiple consecutive TEM images during the evolution of the battery SEI film, the multiple TEM images are separately annotated to obtain multiple SEI component annotated images, and using the instance segmentation model, the multiple SEI component annotated images are instance segmented to obtain multiple SEI electron microscope images, wherein the SEI components of the battery SEI film are marked on the SEI electron microscope images, and the SEI components are tracked on the multiple SEI electron microscope images to achieve high-accuracy nanoscale phase segmentation on the TEM images, providing a reliable basis for subsequent target tracking, effectively tracking the dynamic changes of the SEI film during the battery charging and discharging process, and realizing automatic, efficient and accurate analysis of the phase evolution process of the battery SEI film, thereby improving the accuracy of SEI component analysis and helping to understand the formation and evolution of the SEI film and its impact on battery performance, which is of great significance for improving battery safety and cycle life.

[0099] In some embodiments, target tracking of SEI components on multiple SEI electron microscope images includes: detecting multiple SEI component regions from the multiple SEI electron microscope images; extracting appearance features of the multiple SEI component regions based on the multiple SEI component regions; and matching the multiple SEI component regions and the appearance features to perform target tracking.

[0100] In the above scheme, multiple SEI component regions are detected from multiple SEI electron microscope images respectively, and the appearance features of the multiple SEI component regions are extracted based on the multiple SEI component regions. Matching is performed based on the multiple SEI component regions and the appearance features to perform target tracking, thereby preventing continuous jumping of components of the same category during SEI component tracking, achieving continuous stability of target tracking, and realizing accurate monitoring and following.

[0101] In some embodiments, target tracking of SEI components on multiple SEI electron microscope images includes: obtaining a motion model of the corresponding SEI component on the SEI electron microscope image; in response to failure of the motion model of the SEI component, using a component segmentation mask to remove the virtual box corresponding to the motion model of the SEI component on the SEI electron microscope image.

[0102] In the above scheme, the SEI components are tracked by obtaining the motion model of the corresponding SEI components on the SEI electron microscope image. In response to the failure of the motion model of the SEI components, the virtual frame corresponding to the motion model of the SEI components on the SEI electron microscope image can be removed using the component segmentation mask, thereby achieving effective management, significantly improving the accuracy and reliability of target tracking, and achieving continuous stability of target tracking, so as to realize accurate monitoring and following.

[0103] In some embodiments, performing target tracking of SEI components on a plurality of SEI electron microscope images includes adjusting a maximum lifespan of a tracker performing target tracking from a first preset value to a second preset value, wherein the first preset value is less than the second preset value.

[0104] In the above solution, by adjusting the maximum lifespan of the tracker performing target tracking from a first preset value to a second preset value, wherein the first preset value is less than the second preset value, the tracker is prevented from being deleted prematurely, thereby achieving effective management, significantly improving the accuracy and reliability of target tracking, and achieving continuous stability of target tracking, thereby realizing accurate and correct monitoring and tracking.

[0105] In some embodiments, during target tracking, the cost functions of different categories of multiple SEI component regions are infinite.

[0106] In the above scheme, by setting the value of the cost function of different categories of multiple SEI component areas to infinity, the jumping of the same ID between different categories is avoided, effective management is achieved, the accuracy and reliability of target tracking are significantly improved, and the continuous stability of target tracking is achieved, so as to realize accurate monitoring and tracking.

[0107] In some embodiments, the annotation module 62 is specifically used to pre-annotate the TEM image to obtain a pre-annotated image, wherein the pre-annotated image includes the pre-annotation results of the SEI components; confirm the component boundaries of the pre-annotation results, and in response to receiving a modification operation on the component boundaries of the pre-annotation results, obtain the modified component boundaries of the pre-annotation results, thereby obtaining the SEI component annotated image.

[0108] In the above scheme, a pre-annotated image is obtained by pre-annotating the TEM image, wherein the pre-annotated image includes the pre-annotated results of the SEI components, the component boundaries of the pre-annotated results are confirmed, and in response to receiving a modification operation on the component boundaries of the pre-annotated results, the component boundaries of the modified pre-annotated results are obtained, thereby obtaining an SEI component annotated image, realizing effective semi-automatic annotation of TEM images, solving the problems of difficulty in determining the boundaries of the SEI component regions and low annotation efficiency during manual annotation, and being able to significantly improve the efficiency and accuracy of SEI component annotation, with high flexibility, which is conducive to achieving highly accurate nanoscale phase segmentation on TEM images.

[0109] In some embodiments, the labeling module 62 pre-labels the TEM image, including: inputting the TEM image into a semantic segmentation model to obtain the SEI region in the TEM image; and scanning the SEI region using a component classification model to obtain a pre-labeled image.

[0110] In the above scheme, the SEI area in the TEM image is obtained by inputting the TEM image into the semantic segmentation model; the SEI area is scanned using the component classification model to obtain a pre-labeled image, that is, the dual processing of the semantic segmentation model and the component classification model is used to achieve automatic pre-labeling, which is beneficial to significantly improve the efficiency and accuracy of TEM image labeling and reduce the time and error rate of manual operation.

[0111] In some embodiments, the labeling module 62 pre-labels the TEM image, including: inputting the TEM image into a semantic segmentation model to obtain SEI regions in the TEM image, so as to obtain a pre-labeled image based on the SEI regions.

[0112] In the above scheme, the SEI region in the TEM image is obtained by inputting the TEM image into the semantic segmentation model, and a pre-labeled image is obtained based on the SEI region. Pre-labeling through the semantic segmentation model can more accurately identify and analyze the SEI region, significantly improving the efficiency and accuracy of TEM image labeling and reducing the time and error rate of manual operation.

[0113] In some embodiments, the labeling module 62 pre-labels the TEM image, including: pre-labeling the TEM image using an instance segmentation model to obtain a pre-labeled image.

[0114] In the above scheme, the instance segmentation model is used to pre-annotate TEM images to obtain pre-annotated images and realize automatic pre-annotation, which can significantly improve the annotation efficiency and automatically identify and locate complex structural features, which is conducive to the realization of highly accurate nanoscale phase segmentation on TEM images.

[0115] In some embodiments, the phase evolution analysis system further includes a training module (not shown in the figure) for using a plurality of SEI component annotated images to train an instance segmentation model.

[0116] In the above scheme, by using multiple SEI component annotated images to train the instance segmentation model, a reliable basis is provided for achieving highly accurate nanoscale phase segmentation on TEM images, thereby enabling automatic, efficient and accurate analysis of the phase evolution process of the battery SEI film.

[0117] See also Figure 7 , Figure 7 is a schematic diagram of the structure of an electronic device provided in some embodiments of the present application. Electronic device 70 includes memory 71 and processor 72. Processor 72 is configured to execute program instructions stored in memory 71 to implement the steps of any of the aforementioned battery SEI film phase evolution analysis methods. In a specific implementation scenario, electronic device 70 may include, but is not limited to, a microcomputer and a server. Furthermore, electronic device 70 may also include a supporting device such as a laptop computer or tablet computer, which is not limited here.

[0118] Specifically, the processor 72 is used to control itself and the memory 71 to implement the steps of the aforementioned method for accessing any of the battery algorithm items. The processor 72 may also be referred to as a CPU (Central Processing Unit). The processor 72 may be an integrated circuit chip with signal processing capabilities. The processor 72 may also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic device, discrete gate or transistor logic device, or discrete hardware component. The general-purpose processor may be a microprocessor or any conventional processor. In addition, the processor 72 may be implemented by an integrated circuit chip.

[0119] See also Figure 8 , Figure 8 Schematic diagram of the structure of a computer-readable storage medium provided in some embodiments of the present application. The computer-readable storage medium 80 stores program instructions 801 thereon. When executed by a processor, the program instructions 801 implement the steps of any of the above-mentioned battery SEI film phase evolution analysis methods.

[0120] In the above scheme, by obtaining multiple consecutive TEM images during the evolution of the battery SEI film, the multiple TEM images are separately annotated to obtain multiple SEI component annotated images, and the multiple SEI component annotated images are instance segmented to obtain multiple SEI electron microscope images, wherein the SEI components of the battery SEI film are marked on the SEI electron microscope images, and the SEI components are tracked on the multiple SEI electron microscope images to achieve high-accuracy nanoscale phase segmentation on the TEM images, providing a reliable basis for subsequent target tracking, effectively tracking the dynamic changes of the SEI film during the battery charging and discharging process, and realizing automatic, efficient and accurate analysis of the phase evolution process of the battery SEI film, thereby improving the accuracy of SEI component analysis and helping to understand the formation and evolution of the SEI film and its impact on battery performance, which is of great significance for improving the safety and cycle life of the battery.

[0121] In some embodiments, the functions or modules included in the device provided by the embodiments of the present disclosure can be used to execute the method described in the above method embodiments. The specific implementation can refer to the description of the above method embodiments. For the sake of brevity, it will not be repeated here.

[0122] The above description of the various embodiments tends to emphasize the differences between the various embodiments. The same or similar aspects can be referenced with each other and will not be repeated herein for the sake of brevity.

[0123] In the several embodiments provided in this application, it should be understood that the disclosed methods and devices can be implemented in other ways. For example, the device implementation methods described above are only schematic. For example, the division of modules or units is only a logical function division. There may be other division methods in actual implementation. For example, units or components can be combined or integrated into another system, or some features can be ignored or not executed. In another image position, the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, and the indirect coupling or communication connection of devices or units can be electrical, mechanical or other forms.

[0124] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in either hardware or software functional units. If the integrated units are implemented as software functional units and sold or used as standalone products, they may be stored on a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the portion that contributes to the prior art, or all or part of the technical solution, may be embodied in the form of a software product. This computer software product, stored on a storage medium, includes instructions for causing a computer device (such as a personal computer, server, or network device) or processor to execute all or part of the steps of the various embodiments of the present application. The aforementioned storage media include various media capable of storing program code, such as USB flash drives, removable hard drives, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical disks.

Claims

1. A method for analyzing the phase evolution of a battery SEI film, characterized in that: include: Acquiring a plurality of continuous TEM images during the evolution of the SEI film of the battery; Annotating the multiple TEM images to obtain multiple SEI component annotated images; Performing instance segmentation on the multiple SEI component annotated images to obtain multiple SEI electron microscope images, wherein the SEI electron microscope images are marked with SEI components of the battery SEI film; performing target tracking of the SEI components on the multiple SEI electron microscope images; The target tracking of the SEI components on the multiple SEI electron microscope images includes: detecting a plurality of SEI component regions from the plurality of SEI electron microscope images; extracting appearance features of the plurality of SEI component regions based on the plurality of SEI component regions; Matching the multiple SEI component regions with the appearance features to perform target tracking, wherein the cost functions of different categories of the multiple SEI component regions have values ​​of infinity; Obtaining a motion model of the SEI component corresponding to the SEI electron microscope image; In response to the failure of the motion model of the SEI component, using a component segmentation mask to remove a virtual box corresponding to the motion model of the SEI component on the SEI electron microscope image; The maximum lifespan of a tracker for target tracking is adjusted from a first preset value to a second preset value, wherein the first preset value is smaller than the second preset value.

2. The method according to claim 1, characterized in that The multiple TEM images are annotated separately to obtain multiple SEI component annotated images, including: Pre-annotating the TEM image to obtain a pre-annotated image, wherein the pre-annotated image includes pre-annotated results of the SEI components; The component boundary of the pre-labeling result is confirmed, and in response to receiving a modification operation on the component boundary of the pre-labeling result, a modified component boundary of the pre-labeling result is obtained, thereby obtaining the SEI component labeled image.

3. The method according to claim 2, characterized in that The pre-labeling of the TEM image comprises: Inputting the TEM image into a semantic segmentation model to obtain the SEI component region in the TEM image; The SEI component region is scanned using a component classification model to obtain the pre-labeled image.

4. The method according to claim 2, characterized in that The pre-labeling of the TEM image comprises: The TEM image is input into a semantic segmentation model to obtain the SEI region in the TEM image, so as to obtain the pre-labeled image based on the SEI region.

5. The method according to claim 2, characterized in that The pre-labeling of the TEM image comprises: The TEM image is pre-labeled using an instance segmentation model to obtain the pre-labeled image.

6. The method according to any one of claims 1 to 5, characterized in that The method further includes using the plurality of SEI component annotated images for training an instance segmentation model.

7. An electronic device, characterized in that: The method comprises a memory and a processor, wherein the processor is used to execute program instructions stored in the memory to implement the phase evolution analysis method of the battery SEI film according to any one of claims 1 to 6.

8. A computer-readable storage medium having program instructions stored thereon, characterized in that: When the program instructions are executed by a processor, the method for analyzing the phase evolution of the battery SEI film according to any one of claims 1 to 6 is implemented.

Images