A method for predicting material microscopic images based on improved sample-free class incremental learning

Abstract

This application relates to an improved sample-free, incremental learning method for predicting materials microscopy images. The method includes: constructing a materials microscopy image prediction model; training the model using a set of materials microscopy images; extracting initial features from a first dataset using a feature extractor; expanding the initial features in a feature and label alignment module and determining the parameters of the linear layer while preserving the feature backbone of the feature extractor; optimizing the feature extractor and determining the parameters of the second linear layer in a feature optimization and alignment module; performing incremental alignment and saving the prototype of the new class; performing label alignment on the expanded pseudo-features based on the parameters of the second linear layer in a forgetting compensation and testing module; and using the trained materials microscopy image prediction model to predict the materials microscopy image to be predicted. This method can improve prediction accuracy.

Description

Technical Field

[0001] This application relates to the field of image processing technology, and in particular to a method for predicting material microscopic images based on improved sample-free class incremental learning. Background Technology

[0002] In materials science, microscopic image analysis techniques (such as scanning electron microscopy, transmission electron microscopy, and atomic force microscopy) have become important research tools, providing a wealth of microscopic information for understanding material properties and behavior. However, with the explosive growth of materials microscopic image data, researchers face increasing challenges in image analysis. This not only increases the complexity of the analysis but may also lead to errors and omissions in understanding material properties. To improve the accuracy and efficiency of microscopic image analysis, researchers have begun to explore materials microscopic image prediction methods based on artificial intelligence and machine learning. These methods can automatically analyze large amounts of microscopic image data and, by learning features extracted from historical samples, help researchers predict material categories more quickly and accurately.

[0003] However, current methods still face problems in practical applications, such as insufficient interpretability, strong dependence on datasets, and low prediction accuracy. Summary of the Invention

[0004] Therefore, it is necessary to provide a material microscopy image prediction method based on improved sample-free class incremental learning that can improve prediction accuracy in response to the above-mentioned technical problems.

[0005] A method for predicting material microscopic images based on improved sample-free class incremental learning, the method comprising:

[0006] Obtain a set of labeled microscopic images of common materials; the set of material microscopic images includes a first dataset and a second dataset; construct a material microscopic image prediction model; the material microscopic image prediction model includes a feature extractor, a feature and label alignment module, a feature optimization and alignment module, and a forgetting compensation and testing module;

[0007] The material microscopy image prediction model is trained based on a set of material microscopy images. The feature extractor extracts features from the first dataset to obtain initial features.

[0008] In the feature and label alignment module, the initial features are expanded using the extended layer of the analysis classifier, and the parameters of the linear layer are determined. The extended features and labels are then aligned in the initial stage using the linear layer after the parameters are determined to preserve the feature backbone and important information of the feature extractor. The important information includes the autocorrelation matrix and the prototype of the initial class.

[0009] The feature extractor is optimized in the feature optimization and alignment module. Based on the optimized feature extractor and feature backbone, features are extracted from the second dataset and passed through the extension layer to obtain optimized extended features. The parameters of the second linear layer are determined using the optimized extended features. Based on the parameters of the second linear layer, the optimized extended features and labels are aligned in the incremental stage and the prototype of the new class is saved.

[0010] In the forgetting compensation and testing module, the linear layer and autocorrelation matrix in the analysis classifier are copied. The prototype features of each new class are calculated using the feature backbone. Pseudo-features of the old class are generated based on the similarity between the new class and the old class. The pseudo-features are then passed through an optimized feature extractor and an expansion layer to obtain expanded pseudo-features. The expanded pseudo-features are then labeled according to the parameters of the second linear layer to obtain the trained material microscopy image prediction model.

[0011] The trained material microscopy image prediction model is used to predict the material microscopy image to be predicted, and the material category prediction result is obtained.

[0012] The aforementioned improved sample-free incremental learning-based material microscopy image prediction method first constructs a material microscopy image prediction model. This model includes a feature extractor, a feature and label alignment module, a feature optimization and alignment module, and a forgetting compensation and testing module. Addressing the limitations of existing technologies in sample feature generalization and insufficient mitigation of the forgetting problem, this application introduces self-supervised label enhancement and knowledge distillation techniques into the feature extractor for feature optimization. This enables the model to learn more generalizable features, overcoming the limited feature generalization caused by feature extractor freezing in existing technologies. Simultaneously, knowledge distillation constrains the feature optimization process, ensuring that the model does not deviate excessively from learned knowledge when learning new data features, effectively mitigating the forgetting problem. Secondly, to address the issues of insufficient backbone network utilization and limited mitigation of sample imbalance, the feature extractor architecture of this application consists of an invariant feature backbone and an adaptive feature projector. In the feature and label alignment module, the initial features are expanded using the extended layer of the analysis classifier, and the parameters of the linear layer are determined. The linear layer with determined parameters then performs initial alignment between the expanded features and labels to preserve the feature backbone and important information of the feature extractor. The unchanged feature backbone preserves the original and generates pseudo-features, providing a stable foundation for the model and fully utilizing the learning ability of the backbone network. The forgetting compensation and testing module further improves the model's ability to handle different categories of data by generating pseudo-features and performing alignment tests, more effectively alleviating the sample imbalance problem. The feature optimization and alignment module optimizes the feature extractor, enhancing the generalization and plasticity of the features, enabling better fitting of data with complex nonlinear relationships and effectively solving the underfitting problem. Simultaneously, the forgetting compensation and testing module dynamically adjusts the balance between learning new knowledge and retaining old knowledge by replicating the analysis classifier and performing pseudo-feature alignment, more accurately balancing stability and plasticity, improving the overall performance and generalization ability of the model, and effectively improving the prediction accuracy of material microscopy images. This also ensures that the model can not only process new data, but also maintain stable performance and efficient learning ability in the continuous emergence of material microscopic images, thus having important value in material design and application practice. Attached Figure Description

[0013] Figure 1 This is a flowchart illustrating a method for predicting material microscopic images based on improved sample-free class incremental learning in one embodiment.

[0014] Figure 2 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0015] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0016] In one embodiment, such as Figure 2 As shown, an improved sample-free class incremental learning-based method for predicting material microscopic images is provided, comprising the following steps:

[0017] Step 102: Obtain a set of labeled material microscopy images of common materials; the set of material microscopy images includes a first dataset and a second dataset; construct a material microscopy image prediction model; the material microscopy image prediction model includes a feature extractor, a feature and label alignment module, a feature optimization and alignment module, and a forgetting compensation and testing module.

[0018] Obtaining a large collection of labeled microscopic images of common materials, including both the first and second datasets, provides a solid foundation for model training. This large-scale, diverse dataset effectively reduces the risk of overfitting the model to specific samples and improves its ability to recognize microscopic images of different materials. The diversity of the data ensures that the model maintains high accuracy when faced with new samples. The first and second datasets are distinct datasets.

[0019] Step 104: Train the material microscopy image prediction model based on the material microscopy image set, and extract features from the first dataset using the feature extractor to obtain initial features.

[0020] By performing feature extraction on the first dataset, initial features can be learned from the rich array of material microscopy images. These features form the basis for subsequent analysis, and good feature extraction preserves important information from the images.

[0021] Step 106: In the feature and label alignment module, the initial features are expanded using the extended layer of the analysis classifier and the parameters of the linear layer are determined. The extended features and labels are then aligned in the initial stage using the linear layer after the parameters are determined to preserve the feature backbone and important information of the feature extractor. The important information includes the autocorrelation matrix and the prototype of the initial class.

[0022] The input data is processed by a feature extractor to obtain features, which are then passed through an expansion layer to obtain expanded features. The parameters of the linear layer are then determined by solving an optimization problem (such as minimizing a distance metric between the features and the labels), allowing the expanded features to better align with the labels. After alignment, the feature backbone f in the feature extractor is frozen. FBIt also saves some important information, such as the autocorrelation matrix (Save Relation) and the prototype set of the initial class (Save Prototype Set).

[0023] Expanding the initial features through an extension layer and aligning them with the labels not only preserves important information from the feature extractor but also enhances the correlation between features and their labels, thereby improving the model's discriminative ability. The initial class refers to the labeled material features corresponding to the initial features.

[0024] Step 108: Optimize the feature extractor in the feature optimization and alignment module. Extract features from the second dataset based on the optimized feature extractor and feature backbone, and obtain optimized extended features through the extension layer. Use the optimized extended features to determine the parameters of the second linear layer. Align the optimized extended features and labels in the incremental stage based on the parameters of the second linear layer and save the prototype of the new class.

[0025] Using invariant feature trunk (f) FB ) and the trained adaptive feature projector f FP (SSLA Projector) and the current dataset Obtain the characteristic matrix The extended feature matrix is ​​then obtained through the extension layer. The proposal of innovation is based on the invariant feature f FB and adaptive feature projector f FP The optimized feature extractor architecture consists of an invariant feature backbone for preserving the prototype and generating pseudo-features, an adaptive feature projector for enhancing feature generalization and plasticity, and optimization of only the feature projector to mitigate forgetting. This structural design addresses the problems of feature-label misalignment and mismatch with the prototype caused by unstable feature representations in existing methods.

[0026] Simultaneously, an optimized feature extractor is used to extract features from the second dataset, and an extended layer is used to generate optimized extended features. Using these optimized features, the model can adjust and learn for new data while preserving the prototypes of new classes. This incremental learning approach allows the model to continuously adapt to new image data without losing existing knowledge, effectively mitigating the forgetting problem.

[0027] Step 110: In the forgetting compensation and testing module, copy the linear layer and autocorrelation matrix in the analysis classifier, calculate the prototype features of each new class using the feature backbone, generate pseudo features of the old class based on the similarity between the new class and the old class, pass the pseudo features through the optimized feature extractor and then through the expansion layer to obtain the expanded pseudo features; perform label alignment on the expanded pseudo features according to the parameters of the second linear layer to obtain the trained material microscopy image prediction model.

[0028] In each incremental stage, the linear layer (Replicated AnalyticClassifier) ​​and autocorrelation matrix of the analytical classifier are first replicated. Then, the feature backbone (f) is utilized. FB )Calculate each new class C n The prototype, the formula is in W FB These are the core feature parameters. The prototypes are saved to a prototype set, and the similarity between the old and new classes is calculated using cosine similarity, generating pseudo-features for the old class. If the old class is assumed to be most similar to the new class, then pseudo-features exist.

[0029] The generated pseudo-features are then passed through a trained feature projector (i.e., f). FP To obtain features with better generalization ability, let's call them... Then, after passing through the expansion layer, the expanded pseudo-features are obtained.

[0030] Align the expanded pseudo-features with the replicated analysis classifier, and let the weights of the replicated linear layer be... The autocorrelation matrix is ​​R′ k The model is aligned and then tested using the replicated analysis classifier to obtain the test results for the current stage. After testing, relevant intermediate results, such as expanded pseudo-features, replicated classifier weight matrices, and autocorrelation matrices, are discarded to prepare for the next incremental stage. Model training is performed sequentially until a pre-set threshold is reached, at which point model training is complete. The new class refers to a symptom in a different part than the old class. The forgetting compensation and testing module replicates the second linear layer and autocorrelation matrix of the analysis classifier. By preserving the prototype, it calculates the relationship between the old and new prototypes to generate pseudo-features. These pseudo-features are then passed through a feature projector to obtain more generalizable features and aligned with the replicated classifier, compensating for forgetting and avoiding overfitting. This unique pseudo-feature generation and alignment mechanism is a key innovation in solving the forgetting problem, effectively improving the model's performance in sample-less incremental learning. In specific applications, by replicating the linear layer and autocorrelation matrix in the analysis classifier, a balance between old and new knowledge can be achieved. By generating pseudo-features of the old class and comparing them with the new class, the model can dynamically adjust the weights between learning new knowledge and retaining old knowledge. This mechanism ensures that the model can still effectively identify old data while continuously learning new data, thereby improving overall prediction accuracy. By establishing an autocorrelation matrix and prototype features, the model can not only provide results when making predictions, but also provide interpretability for these results.

[0031] Step 112: Use the trained material microscopy image prediction model to predict the material microscopy image to be predicted, and obtain the material category prediction result.

[0032] The aforementioned improved sample-free incremental learning-based material microscopy image prediction method first constructs a material microscopy image prediction model. This model includes a feature extractor, a feature and label alignment module, a feature optimization and alignment module, and a forgetting compensation and testing module. Addressing the limitations of existing technologies in sample feature generalization and insufficient mitigation of the forgetting problem, this application introduces self-supervised label enhancement and knowledge distillation techniques into the feature extractor for feature optimization. This enables the model to learn more generalizable features, overcoming the limited feature generalization problem caused by feature extractor freezing in existing technologies. Simultaneously, knowledge distillation constrains the feature optimization process, ensuring that the model does not excessively deviate from learned knowledge when learning new data features, effectively mitigating the forgetting problem. Secondly, to address the issues of insufficient backbone network utilization and limited mitigation of sample imbalance, the feature extractor architecture of this application consists of an invariant feature backbone and an adaptive feature projector. In the feature and label alignment module, the initial features are expanded using the extended layer of the analysis classifier, and the parameters of the linear layer are determined. The linear layer with determined parameters then performs initial alignment between the expanded features and labels to preserve the feature backbone and important information of the feature extractor. The unchanged feature backbone preserves the original and generates pseudo-features, providing a stable foundation for the model and fully utilizing the learning ability of the backbone network. The forgetting compensation and testing module further improves the model's ability to handle different categories of data by generating pseudo-features and performing alignment tests, more effectively alleviating the sample imbalance problem. The feature optimization and alignment module optimizes the feature extractor, enhancing the generalization and plasticity of the features, enabling better fitting of data with complex nonlinear relationships and effectively solving the underfitting problem. Simultaneously, the forgetting compensation and testing module dynamically adjusts the balance between learning new knowledge and retaining old knowledge by replicating the analysis classifier and performing pseudo-feature alignment, more accurately balancing stability and plasticity, improving the overall performance and generalization ability of the model, and effectively improving the prediction accuracy of material microscopy images. This also ensures that the model can not only process new data, but also maintain stable performance and efficient learning ability in the continuous emergence of material microscopic images, thus having important value in material design and application practice.

[0033] In one embodiment, features are extracted from the first dataset using a feature extractor to obtain initial features, including:

[0034] The feature extractor uses self-supervised label enhancement technology and a fully connected classifier to extract features from the first dataset, obtaining initial features.

[0035] In a specific embodiment, the feature extractor uses self-supervised label enhancement (SSLA) technology and a fully connected classifier (FCN) on the first dataset. Training is performed on the training samples. Self-supervised label augmentation technology generates additional supervised information by rotating the training samples, specifically rotating the sample x by 90 degrees, 180 degrees, and 270 degrees, and expanding the label y. This can help the feature extractor learn features with better generalization ability.

[0036] In one embodiment, the feature and label alignment module expands the initial features using an extension layer of the analysis classifier and determines the parameters of the linear layer, including:

[0037] The initial features are expanded using the expansion layer of the analysis classifier to obtain the expanded features.

[0038]

[0039] Among them, f act f represents the activation function. E Indicates an extension layer, f flat Represents the flattening operator. Indicates initial features;

[0040] The problem of optimizing the parameters of a linear layer is as follows:

[0041]

[0042] in, The parameters representing the linear layer, Indicates extended features, || || F Let Frobenius norm be denoted, and γ be the regularization hyperparameter;

[0043] The optimal solution for optimizing the parameters of the linear layer is obtained by solving the linear layer parameter optimization problem using the least squares method, i.e., the parameters of the linear layer are:

[0044] Where I is the identity matrix, (.) T It is a matrix transpose operation.

[0045] In a specific embodiment, the trained feature extractor transforms the input data into features, which are then aligned with labels using an analytical classifier. This analytical classifier comprises a feature projector and a linear classifier. Let the input be X, then the feature extractor f... FE (Parameter is W) FE ) to obtain feature X FE =f FE (X;W FEThen, an analytical classifier is used to align the features and labels. This analytical classifier contains an extension layer f. E (shape is d) FE ×d E ) and a linear layer f L Feature X FE After the extended layer f E Obtain extended features This is f act It is an activation function (such as ReLU), f flat It is a flattening operator. Then, the linear layer f is determined by solving an optimization problem. L The parameters are estimated using the least squares method to obtain the optimal solution. After alignment, certain parts of the feature extractor (such as the feature backbone f) are frozen. FB It also saves some important information, such as the autocorrelation matrix R0 and the prototype of the initial class (Save Prototype Set).

[0046] In one embodiment, the feature extractor is optimized, including:

[0047] The feature extractor is optimized by combining self-supervised label augmentation and knowledge distillation techniques with a fully connected classifier and the second dataset of the current stage, by minimizing the current feature extractor f. FE；k and the previous stage feature extractor f FE；k-1 The optimization is constrained by the distance between the extracted new data features, as shown in the formula. The total loss is L total =L KD +L CE L CE It is cross-entropy loss.

[0048] In a specific implementation, a self-supervised label enhancement (SSLA) technique is employed, such as rotating the training samples to expand the labels. This is combined with a fully connected classifier and cross-entropy loss to learn generalizable features at all stages. This is one of the key innovations that distinguishes it from existing analytical learning (AL) methods, which often focus only on the initial stage during feature extractor training, neglecting the feature generalization requirements in the incremental stage. Knowledge distillation (KD) is used to constrain the feature optimization process, limiting the distance between the current feature extractor and the new data features extracted by the previous stage, thus avoiding the forgetting problem caused by over-optimization. This combination of self-supervised learning and knowledge distillation provides an effective approach to feature optimization, enabling the model to retain learned knowledge while improving feature generalization ability.

[0049] In one embodiment, the parameters of the second linear layer are determined using the optimized extended features, including:

[0050] The parameters of the second linear layer are determined using the optimized extended features.

[0051]

[0052] Among them, R k Represents the autocorrelation matrix. This represents the optimized diffusion characteristics. This represents the label matrix of the training data for stage K. This represents the optimal estimate of the linear layer weight matrix of the previous stage k-1.

[0053] In one embodiment, the prototype features of each new class are calculated using the feature backbone, including:

[0054] The prototype features of each new class are calculated using the feature backbone, using the following formula:

[0055]

[0056]

[0057] Among them, C n To represent a new class, f FB W represents the main feature. FB X represents the main feature parameters. Cn Represents the new class C n The sample.

[0058] In one embodiment, generating pseudo-features of the old class based on the similarity between the new class and the old class further includes:

[0059] Based on the similarity between the new and old classes, pseudo-features of the old class are generated.

[0060]

[0061] in, Represents the new class C n The features after invariant feature backbone (FB) processing, μ(C) n ) represents the new class C n The prototype is obtained by calculating sample features. The mean value is obtained, where μ(C0) represents the prototype of the old class C0, obtained by calculating sample features. The mean value is obtained from the average.

[0062] In specific embodiments, the invariant feature backbone can extract features with a certain degree of stability and representativeness. These features serve as the basis for generating pseudo-features and, combined with the new class prototype, provide information related to the new class for the generation of old class pseudo-features. This helps the model establish connections between feature spaces of different classes and reduces forgetting. New class prototype μ(C) n ) and new class features Through their combined action, the new class's feature information is converted into pseudo-features of the old class by calculating the difference with the old class prototype μ(C0). This allows the model to use the information of the new class to compensate for the forgetting of the old class, and also reflects the correlation and transformation of features between different categories.

[0063] In one embodiment, the pseudo-features are passed through an optimized feature extractor and then through an expansion layer to obtain expanded pseudo-features, including:

[0064] The pseudo-features are then passed through an optimized feature extractor and an expansion layer to obtain the expanded pseudo-features.

[0065]

[0066] Among them, f act f represents the activation function. E Indicates an extension layer, f flat Represents the flattening operator. This represents the features obtained after the pseudo-features have passed through the optimized feature extractor.

[0067] It should be understood that, although Figure 1 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figure 1 At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but may be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but may be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0068] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 2As shown, the computer device includes a processor, memory, network interface, display screen, and input devices connected via a system bus. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used to communicate with external terminals via a network connection. When executed by the processor, the computer program implements an improved sample-free incremental learning-based method for predicting materials microscopic images. The display screen can be a liquid crystal display (LCD) or an e-ink display. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the computer device casing, or an external keyboard, touchpad, or mouse.

[0069] Those skilled in the art will understand that Figure 2 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0070] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0071] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0072] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for predicting material microscopic images based on improved sample-free class incremental learning, characterized in that, The method includes: Obtain a set of labeled material microscopic images of common materials; the set of material microscopic images includes a first dataset and a second dataset; construct a material microscopic image prediction model; the material microscopic image prediction model includes a feature extractor, a feature and label alignment module, a feature optimization and alignment module, and a forgetting compensation and testing module; The material microscopy image prediction model is trained based on the set of material microscopy images, and the feature extractor extracts features from the first dataset to obtain initial features; In the feature and label alignment module, the initial features are expanded using the expansion layer of the analysis classifier, and the parameters of the linear layer are determined. The expanded features and labels are then aligned in the initial stage using the linear layer after the parameters are determined to preserve the feature backbone and important information of the feature extractor. The important information includes the autocorrelation matrix and the prototype of the initial class. The feature extractor is optimized in the feature optimization and alignment module. Based on the optimized feature extractor and the feature backbone, features are extracted from the second dataset and passed through the extension layer to obtain optimized extended features. The parameters of the second linear layer are determined using the optimized extended features. Based on the parameters of the second linear layer, the optimized extended features and labels are aligned in an incremental stage and the prototype of the new class is saved. The forgetting compensation and testing module replicates the linear layer and autocorrelation matrix in the analysis classifier, calculates the prototype features of each new class using the feature backbone, generates pseudo features of the old class based on the similarity between the new class and the old class, passes the pseudo features through the optimized feature extractor and then through the expansion layer to obtain expanded pseudo features; and performs label alignment on the expanded pseudo features according to the parameters of the second linear layer to obtain the trained material microscopy image prediction model. The trained material microscopy image prediction model is used to predict the material microscopy image to be predicted, and the predicted material category corresponding to the material microscopy image is obtained. The first dataset is subjected to feature extraction using the feature extractor to obtain initial features, including: The feature extractor uses self-supervised label enhancement technology and a fully connected classifier to extract features from the first dataset to obtain initial features; Optimize the feature extractor, including: The feature extractor is optimized by combining a fully connected classifier and a second dataset from the current stage using self-supervised label augmentation and knowledge distillation techniques, thereby minimizing the current feature extractor. and the previous stage feature extractor The optimization is constrained by the distance between the extracted new data features, as shown in the formula. The overall loss is ,in It is cross-entropy loss.

2. The method according to claim 1, characterized in that, The feature and label alignment module expands the initial features using the extended layer of the analysis classifier and determines the parameters of the linear layer, including: The initial features are expanded using the extension layer of the analysis classifier to obtain the expanded features. in, This represents the activation function. Indicates an extension layer. Represents the flattening operator. Indicates initial features; The problem of optimizing the parameters of a linear layer is as follows: in, The parameters representing the linear layer, Indicates extended features, Denotes the Frobenius norm. It is a regularization hyperparameter; The optimal solution for optimizing the parameters of the linear layer is obtained by solving the least squares method, i.e., the parameters of the linear layer are: ,in It is the identity matrix. It is a matrix transpose operation.

3. The method according to claim 1, characterized in that, The parameters of the second linear layer are determined using the optimized extended features, including: The parameters of the second linear layer are determined using the optimized extended features. in, Represents the autocorrelation matrix. This represents the optimized diffusion characteristics. This represents the label matrix of the training data for stage K. Indicates the previous stage k The optimal estimate of the linear layer weight matrix of -1.

4. The method according to claim 1, characterized in that, The prototype features of each new class are calculated using the aforementioned feature backbone, including: The prototype features of each new class are calculated using the aforementioned feature backbone, using the following formula: in, Indicates a new class, Indicates the main features, Indicates the main feature parameters. Represents a new class The sample.

5. The method according to claim 1, characterized in that, Generating pseudo-features for the old class based on the similarity between the new and old classes also includes: Based on the similarity between the new and old classes, pseudo-features of the old class are generated. in, Represents a new class Passing through the invariant feature trunk Processed features Represents a new class The prototype is obtained by calculating sample features. The mean was obtained. Represents the old class The prototype is obtained by calculating sample features. The mean value is obtained from the average.

6. The method according to claim 5, characterized in that, The pseudo-features are then passed through an optimized feature extractor and an expansion layer to obtain expanded pseudo-features, including: The pseudo-features are then passed through an optimized feature extractor and an expansion layer to obtain the expanded pseudo-features. in, This represents the activation function. Indicates an extension layer. Represents the flattening operator. This represents the features obtained after the pseudo-features have passed through the optimized feature extractor.

Images