Infrared ship image classification method, system and equipment based on feature reconstruction

Abstract

The invention discloses an infrared ship image classification method, system and equipment based on feature reconstruction. The method comprises the following steps: performing first feature reconstruction and second feature reconstruction on a feature matrix; based on the first feature importance score and the second feature importance score, screening features in the feature matrix to obtain a new feature matrix; fusing the extracted visual features with the new feature matrix to obtain a fused feature matrix; inputting the fusion feature matrix into a first classification network for classification evaluation; performing uncertainty evaluation on the sample features according to the classification evaluation result and the category labels of the samples to obtain an uncertainty evaluation result; and screening a target feature set according to an uncertainty evaluation result, and training a second classification network by using the target feature set so as to classify the infrared ship image through the trained second classification network. According to the method, the feature set with relatively high stability and robustness can be obtained, so that the infrared ship image classification accuracy is improved.

Description

Technical Field

[0001] The present application relates to the field of image processing technology, and in particular to a method, system and device for classifying infrared ship images based on feature reconstruction. Background Art

[0002] Feature validity evaluation is an important technology in the field of machine learning and pattern recognition. Its core goal is to screen out features with high discrimination and robustness for target tasks (such as classification, detection, and recognition, etc.) through quantitative analysis, while eliminating redundant or noisy features. In complex naval warfare environments, feature validity evaluation of infrared ship images is of vital importance.

[0003] Due to the complexity of the naval environment, including factors such as waves, sea fog, illumination changes, and low contrast between the target and the background, the feature representation of infrared ship targets is often unstable, with a large number of redundant features and poor anti-interference ability. These problems seriously affect the accuracy and reliability of the target recognition system. Therefore, developing an effective feature validity evaluation method is of great application value for improving the target recognition of infrared ship images in complex naval environments.

[0004] In summary, in complex naval combat environments, target features in infrared images often exhibit problems such as high noise, serious redundancy, and poor stability, which makes it difficult for traditional methods to fully explore the essential characteristics of the target for accurate classification. Summary of the invention

[0005] The present application aims to propose a method, system and device for infrared ship image classification based on feature reconstruction, which can obtain a feature set with high stability and robustness, thereby improving the accuracy of infrared ship image classification.

[0006] In a first aspect, an embodiment of the present application provides an infrared ship image classification method based on feature reconstruction, the method comprising: Collecting feature data of infrared ship images, and preprocessing the feature data to obtain a feature matrix, wherein a row of the feature matrix represents a sample, and a column represents a feature; Performing a first feature reconstruction on the feature matrix to obtain a first feature importance score; Performing a second feature reconstruction on the feature matrix to obtain a second feature importance score; Based on the first feature importance score and the second feature importance score, the features in the feature matrix are screened to obtain a new feature matrix; Extracting visual features from the infrared ship image, and fusing the visual features with the new feature matrix to obtain a fused feature matrix; Inputting the fused feature matrix into a first classification network for classification evaluation to obtain a classification evaluation result; According to the classification evaluation result and the category label of the sample, uncertainty evaluation is performed on the sample feature to obtain an uncertainty evaluation result; A target feature set is screened according to the uncertainty assessment result, and a second classification network is trained using the target feature set, so that the infrared ship image is classified through the trained second classification network.

[0007] Compared with the prior art, the first aspect of the present application has the following beneficial effects: The method collects feature data of infrared ship images and preprocesses the feature data to obtain a feature matrix, wherein one row of the feature matrix represents a sample and one column represents a feature; the feature matrix is ​​reconstructed for the first time to obtain a first feature importance score; the feature matrix is ​​reconstructed for the second time to obtain a second feature importance score; based on the first feature importance score and the second feature importance score, the features in the feature matrix are screened to obtain a new feature matrix; visual features in the infrared ship images are extracted, and the visual features are fused with the new feature matrix to obtain a fused feature matrix; the fused feature matrix is ​​input into a first classification network for classification evaluation to obtain a classification evaluation result; uncertainty evaluation is performed on the sample features according to the classification evaluation result and the category label of the sample to obtain an uncertainty evaluation result; a target feature set is screened according to the uncertainty evaluation result, and a second classification network is trained using the target feature set, so that the infrared ship images can be classified through the trained second classification network. In this way, feature redundancy and noise can be removed and feature robustness can be improved through two feature reconstructions. Feature information can be enriched by fusing visual features with the new feature matrix. Uncertainty assessment of sample features can measure sample uncertainty and screen out feature sets with higher stability and robustness. Finally, the second classification network is trained through the target feature set, so that the trained second classification network can accurately classify infrared ship images, thereby improving the accuracy of infrared ship image classification.

[0008] In some implementations, performing a first feature reconstruction on the feature matrix to obtain a first feature importance score includes: A low-rank matrix is ​​used to represent a feature reconstruction coefficient matrix, and a first feature reconstruction objective function is constructed by combining the feature reconstruction coefficient matrix and the feature matrix, wherein the first feature reconstruction objective function is used to perform a first feature reconstruction on the feature matrix; Solving the feature reconstruction coefficient matrix in the first feature reconstruction objective function by using alternating iteration; Based on the solved feature reconstruction coefficient matrix, the first feature importance score is calculated.

[0009] In some implementations, combining the feature reconstruction coefficient matrix and the feature matrix to construct a first feature reconstruction objective function includes: ; in, Represents the matrix variables in the feature reconstruction coefficient matrix and matrix variables , represents the total number of features in the feature matrix, Indicates Features, represents the regularization hyperparameter, represents the data reconstruction error term, Indicates based on The sparse regularization term of the norm, represents an adjustable hyperparameter, represents the orthogonality constraint, Indicates transpose.

[0010] In some implementations, performing a second feature reconstruction on the feature matrix to obtain a second feature importance score includes: Based on the sample weights, the relationship matrix and the feature matrix, construct a second feature reconstruction objective function, wherein the second feature reconstruction objective function is used to perform a second feature reconstruction on the feature matrix; Using alternating iteration to solve the sample weights and relationship matrix in the second feature reconstruction objective function; Based on the solved sample weight, the second feature importance score is calculated.

[0011] In some implementations, constructing a second feature reconstruction objective function based on the sample weights, the relationship matrix, and the feature matrix includes: ; in, represents the total number of samples in the feature matrix, Indicates The sample weights, represents the data label matrix, Indicates The class labels corresponding to the samples are Indicates samples, represents the relationship matrix, represents the regularization hyperparameter, represents an adjustable hyperparameter, Indicates using data to fit category labels. Indicates based on The sparse regularization term of the norm, represents the soft regularization term of self-paced learning, Represents the self-paced learning step size parameter.

[0012] In some embodiments, performing uncertainty assessment on sample features according to the classification assessment result and the category label of the sample to obtain the uncertainty assessment result includes: Dirichlet distribution is used to construct uncertainty loss function; According to the classification evaluation result and the category label of the sample, the uncertainty loss function is used to perform uncertainty evaluation on the sample feature to obtain an uncertainty evaluation result.

[0013] In some embodiments, constructing an uncertainty loss function using Dirichlet distribution includes: ; in, represents the entropy of the Dirichlet distribution, represents the class probability vector, represents the normalization constant, represents the total number of categories, Indicates The predicted confidence of the category.

[0014] In a second aspect, an embodiment of the present application further provides an infrared ship image classification system based on feature reconstruction, the system comprising: A data processing unit, used for collecting feature data of infrared ship images, and preprocessing the feature data to obtain a feature matrix, wherein a row of the feature matrix represents a sample, and a column represents a feature; A first reconstruction unit, used for performing a first feature reconstruction on the feature matrix to obtain a first feature importance score; A second reconstruction unit is used to perform a second feature reconstruction on the feature matrix to obtain a second feature importance score; A feature screening unit, configured to screen features in a feature matrix based on the first feature importance score and the second feature importance score to obtain a new feature matrix; A feature extraction unit, used for extracting visual features from the infrared ship image, and fusing the visual features with the new feature matrix to obtain a fused feature matrix; A classification evaluation unit, used for inputting the fused feature matrix into a first classification network for classification evaluation to obtain a classification evaluation result; A sample evaluation unit, used to perform uncertainty evaluation on sample features according to the classification evaluation result and the category label of the sample to obtain an uncertainty evaluation result; The image classification unit is used to screen a target feature set according to the uncertainty assessment result, and use the target feature set to train a second classification network, so as to classify the infrared ship image through the trained second classification network.

[0015] In a third aspect, an embodiment of the present application also provides an electronic device, comprising at least one control processor and a memory for communicating with the at least one control processor; the memory stores instructions that can be executed by the at least one control processor, and the instructions are executed by the at least one control processor so that the at least one control processor can execute the infrared ship image classification method based on feature reconstruction as described above.

[0016] In a fourth aspect, an embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are used to enable a computer to execute the infrared ship image classification method based on feature reconstruction as described above.

[0017] It can be understood that the beneficial effects of the second to fourth aspects compared with the related art are the same as the beneficial effects of the first aspect compared with the related art. Please refer to the relevant description in the first aspect, and no further details will be given here. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which: Figure 1 It is a flowchart of an embodiment of an infrared ship image classification method based on feature reconstruction provided by the present application; Figure 2 It is a schematic diagram of the overall method flow in the best embodiment of the infrared ship image classification method based on feature reconstruction provided by the present application; Figure 3 It is a structural schematic diagram of an embodiment of an infrared ship image classification system based on feature reconstruction provided by the present application; Figure 4 It is a schematic diagram of the structure of an embodiment of the electronic device provided by the present application. DETAILED DESCRIPTION

[0019] The embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present application, and cannot be understood as limiting the present application.

[0020] In the description of this application, if there is a description of first, second, etc., it is only for the purpose of distinguishing technical features, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features or implicitly indicating the order of the indicated technical features.

[0021] In the description of the present application, it should be understood that the descriptions involving orientation, such as the orientation or positional relationship indicated as up, down, etc., are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as a limitation on the present application.

[0022] In the description of this application, it should be noted that, unless otherwise clearly defined, terms such as setting, installing, connecting, etc. should be understood in a broad sense, and technicians in the relevant technical field can reasonably determine the specific meaning of the above terms in this application based on the specific content of the technical solution.

[0023] In complex naval combat environments, target features in infrared images often exhibit problems such as high noise, serious redundancy, and poor stability, making it difficult for traditional methods to fully explore the essential characteristics of the target for accurate classification.

[0024] In order to solve the problem that the above-mentioned traditional methods are difficult to fully explore the essential characteristics of the target for accurate classification, the present application proposes an infrared ship image classification method, system and device based on feature reconstruction.

[0025] Reference Figure 1 The present application embodiment provides an infrared ship image classification method based on feature reconstruction, the method comprising the following steps: Step S100, collecting feature data of infrared ship images, and preprocessing the feature data to obtain a feature matrix, where a row of the feature matrix represents a sample, and a column represents a feature; Step S200, performing a first feature reconstruction on the feature matrix to obtain a first feature importance score; Step S300, performing a second feature reconstruction on the feature matrix to obtain a second feature importance score; Step S400: Screening the features in the feature matrix based on the first feature importance score and the second feature importance score to obtain a new feature matrix; Step S500, extracting visual features from the infrared ship image, and fusing the visual features with the new feature matrix to obtain a fused feature matrix; Step S600: input the fused feature matrix into the first classification network for classification evaluation to obtain a classification evaluation result; Step S700: perform uncertainty assessment on sample features according to the classification assessment result and the category label of the sample to obtain an uncertainty assessment result; Step S800: Filter the target feature set according to the uncertainty assessment result, and use the target feature set to train the second classification network, so as to classify the infrared ship image through the trained second classification network.

[0026] In this embodiment, feature data of infrared ship images are collected and preprocessed to obtain a feature matrix, wherein one row of the feature matrix represents a sample and one column represents a feature; a first feature reconstruction is performed on the feature matrix to obtain a first feature importance score; a second feature reconstruction is performed on the feature matrix to obtain a second feature importance score; based on the first feature importance score and the second feature importance score, features in the feature matrix are screened to obtain a new feature matrix; visual features in the infrared ship images are extracted, and the visual features are fused with the new feature matrix to obtain a fused feature matrix; the fused feature matrix is ​​input into a first classification network for classification evaluation to obtain a classification evaluation result; uncertainty evaluation is performed on the sample features according to the classification evaluation result and the category label of the sample to obtain an uncertainty evaluation result; a target feature set is screened according to the uncertainty evaluation result, and the target feature set is used to train a second classification network, so that the infrared ship image can be classified through the trained second classification network. In this way, feature redundancy and noise can be removed and feature robustness can be improved through two feature reconstructions. Feature information can be enriched by fusing visual features with the new feature matrix. Uncertainty assessment of sample features can measure sample uncertainty and screen out feature sets with higher stability and robustness. Finally, the second classification network is trained through the target feature set, so that the trained second classification network can accurately classify infrared ship images, thereby improving the accuracy of infrared ship image classification.

[0027] The above-mentioned preprocessing of the feature data may be mechanical energy preprocessing such as normalization, missing value processing, and standard dimension alignment of the feature data.

[0028] The first feature reconstruction of the feature matrix to obtain the first feature importance score may be performed by using a data self-expression technology to perform the first feature reconstruction of the feature matrix to obtain the first feature importance score.

[0029] The second feature reconstruction of the feature matrix is ​​performed for the second time to obtain the second feature importance score, and the second feature importance score can be obtained by using a self-paced learning strategy to perform the second feature reconstruction of the feature matrix for the second time.

[0030] The above-mentioned screening of features in the feature matrix based on the first feature importance score and the second feature importance score can be performed by adding the first feature importance score and the second feature importance score to obtain a score addition result, screening multiple features with high scores in the feature matrix according to the score addition result, and constructing a new matrix through the screened features.

[0031] The above-mentioned extraction of visual features from the infrared ship image may be performed by using a visual feature extraction network to extract visual features from the infrared ship image. The visual feature extraction network may be a visual feature extraction network known to those skilled in the art, and this embodiment does not make any specific description or limitation.

[0032] The category labels of the above samples may be pre-labeled category labels.

[0033] The first classification network and the second classification network may be the same classification network. The classification network may be a classification network known to those skilled in the art, such as a multi-layer perceptron (MLP) as a classification network, which is not specifically described or limited in this embodiment.

[0034] In some implementations, performing a first feature reconstruction on the feature matrix to obtain a first feature importance score includes: A low-rank matrix is ​​used to represent a feature reconstruction coefficient matrix, and a first feature reconstruction objective function is constructed by combining the feature reconstruction coefficient matrix and the feature matrix. The first feature reconstruction objective function is used to perform a first feature reconstruction on the feature matrix. Using alternating iteration to solve the feature reconstruction coefficient matrix in the first feature reconstruction objective function; Based on the solved feature reconstruction coefficient matrix, the first feature importance score is calculated.

[0035] In this embodiment, a low-rank matrix is ​​used to represent the feature reconstruction coefficient matrix, and the feature reconstruction coefficient matrix and the feature matrix are combined to construct a first feature reconstruction objective function, which is used to perform the first feature reconstruction on the feature matrix; the feature reconstruction coefficient matrix in the first feature reconstruction objective function is solved by alternating iteration; based on the solved feature reconstruction coefficient matrix, the first feature importance score is calculated. In this way, a low-rank matrix is ​​used to represent the feature reconstruction coefficient matrix to achieve dimensionality reduction of the feature matrix and remove noise and redundant information in the feature data.

[0036] The first feature importance score is calculated based on the solved feature reconstruction coefficient matrix. The first feature importance score can be obtained by normalizing the features in the matrix that stores the validity information of the features based on the solved feature reconstruction coefficient matrix, and then calculating the feature importance scores of the normalized features to obtain the first feature importance score.

[0037] In some embodiments, the first feature reconstruction objective function is constructed by combining the feature reconstruction coefficient matrix and the feature matrix, including: ; in, Represents the matrix variable in the feature reconstruction coefficient matrix and matrix variables , represents the total number of features in the feature matrix, Indicates Features, represents the regularization hyperparameter, represents the data reconstruction error term, Indicates based on The sparse regularization term of the norm, represents an adjustable hyperparameter, represents the orthogonality constraint, Indicates transpose.

[0038] In some implementations, performing a second feature reconstruction on the feature matrix to obtain a second feature importance score includes: Based on the sample weights, the relationship matrix and the feature matrix, a second feature reconstruction objective function is constructed, and the second feature reconstruction objective function is used to perform a second feature reconstruction on the feature matrix; Alternating iteration is used to solve the sample weight and relationship matrix in the second feature reconstruction objective function; Based on the solved sample weight, the second feature importance score is calculated.

[0039] In this embodiment, a second feature reconstruction objective function is constructed based on sample weights, relationship matrix and feature matrix, and the second feature reconstruction objective function is used to perform a second feature reconstruction on the feature matrix; the sample weights and relationship matrix in the second feature reconstruction objective function are solved by alternating iteration; based on the solved sample weights, the second feature importance score is calculated. In this way, by applying dynamic weight constraints to the samples, the feature data is learned using a "first easy then difficult" strategy to enhance the noise resistance and robustness of the model.

[0040] The second feature importance score is calculated based on the solved sample weights, and the solved sample weights may be normalized to obtain the second feature importance score.

[0041] In some implementations, constructing a second feature reconstruction objective function based on the sample weights, the relationship matrix, and the feature matrix includes: ; in, represents the total number of samples in the feature matrix, Indicates The sample weights, represents the data label matrix, Indicates The class labels corresponding to the samples are Indicates samples, represents the relationship matrix, represents the regularization hyperparameter, represents an adjustable hyperparameter, Indicates using data to fit category labels. Indicates based on The sparse regularization term of the norm, represents the soft regularization term of self-paced learning, Represents the self-paced learning step size parameter.

[0042] In this embodiment, the feature reconstruction evaluation is performed through a self-paced learning strategy, which can enhance the noise resistance and robustness of the model and improve the robustness of the features.

[0043] The above method uses data to fit category labels and learn the relationship matrix.

[0044] The above sparse regularization term is used to learn the effectiveness score of the feature.

[0045] The soft regularization term of the self-paced learning can learn the importance of samples. That is, key samples are given larger weights, while outliers or noise samples are assigned smaller weights, so that the model can achieve a robust effect.

[0046] In some embodiments, uncertainty assessment is performed on sample features according to the classification assessment result and the category label of the sample to obtain an uncertainty assessment result, including: Dirichlet distribution is used to construct uncertainty loss function; According to the classification evaluation results and the category labels of the samples, the uncertainty loss function is used to perform uncertainty evaluation on the sample features to obtain the uncertainty evaluation results.

[0047] In this embodiment, the Dirichlet distribution is used to construct an uncertainty loss function; according to the classification evaluation result and the category label of the sample, the uncertainty loss function is used to perform uncertainty evaluation on the sample features to obtain an uncertainty evaluation result. In this way, the uncertainty of the sample features is evaluated by the uncertainty loss function to measure the uncertainty of the sample features and screen out a feature set with higher stability and robustness.

[0048] In some embodiments, the uncertainty loss function is constructed using Dirichlet distribution, including: ; in, represents the entropy of the Dirichlet distribution, represents the class probability vector, represents the normalization constant, represents the total number of categories, Indicates The predicted confidence of the category.

[0049] To facilitate understanding by those skilled in the art, a set of best embodiments is provided below: Traditional feature validity evaluation methods include filtering, embedded, and wrapping. The wrapping method regards feature validity evaluation as a subset search problem. It directly evaluates the performance of each subset by training the model on different feature subsets (such as calculating indicators such as accuracy and F1 score), and finally selects the feature combination with the best performance. This method can capture the interaction between features, but its computational cost is high because it needs to train the model for each subset. The filtering method relies on the statistical relationship between features and target variables. It first calculates the correlation between each feature and the target variable (such as Pearson correlation coefficient and mutual information), and then sorts the features according to statistical indicators. The top-ranked features are selected as the final selected features. Its computational efficiency is high, but it may ignore the complex relationship between features. The embedded method integrates feature validity evaluation into the model training process, so that the model can automatically identify and adjust the importance of each feature during learning and select the most effective features. Typical methods include methods based on regularization (such as L1 regularization to zero the weights of some features) and tree models (such as decision trees and random forests). This method can capture the interaction between features and automatically adjust feature weights, but may be affected by the model's own preferences. In general, the selection of feature validity evaluation methods should comprehensively consider the nature of the problem, data characteristics, and computing resources. In different scenarios, each method has its unique advantages and limitations.

[0050] In complex naval warfare environments, target features in infrared images often show problems such as high noise, serious redundancy and poor stability, which makes it difficult for traditional methods to fully explore the essential features of the target for accurate classification. This embodiment proposes a feature validity evaluation method based on reconstruction and representation learning, which can fully explore and utilize the intrinsic connection between features, reduce redundant features, and use dynamic sample weight adjustment to improve the robustness of feature evaluation, providing more stable and reliable feature support for subsequent infrared ship image classification, thereby improving classification accuracy and anti-interference ability.

[0051] Reference Figure 2The infrared ship image classification method based on feature reconstruction and uncertainty quantification proposed in this embodiment aims to solve the problems of unstable feature representation of infrared ship targets, many redundant features and poor anti-interference ability in complex naval combat environments. This method integrates a linear reconstruction method based on data self-expression, a feature reconstruction evaluation method based on self-paced learning and uncertainty quantification technology to optimize feature representation and enhance classification performance. By screening and optimizing multi-dimensional features, the accuracy and robustness of infrared ship image classification in complex environments are improved. Specifically, it includes the following steps: Step 1: Collect the original feature data of infrared ship images, and perform preprocessing such as normalization, missing value processing, and standard dimension alignment on the original feature data to unify the features from different sources into a feature matrix of the same dimension. , for model input.

[0052] Step 2: Express the preprocessed feature matrix as ,in Represents the number of samples, that is, each row in the feature matrix represents a sample, Represents the feature dimension, that is, each column in the feature matrix represents a feature.

[0053] Step 3: Use data self-expression technology and low-rank technology to reconstruct features to solve feature redundancy and noise problems. For specific implementation, see steps 4 to 6.

[0054] Step 4: Use low-rank matrix and Represents the feature reconstruction coefficient matrix to achieve the feature matrix Perform dimensionality reduction to remove noise and redundant information in the data. is the feature sparse matrix, is an orthogonal matrix, is the rank of the matrix, and the specific optimization objective function (i.e., the first feature reconstruction objective function) is shown in formula (1): (1); The first term of the objective function Reconstruct the error term for the data, where Indicates Features, total This term approximates the original data through a linear combination of feature subsets, aiming to minimize the reconstruction error, thereby selecting the most representative feature subset during the feature evaluation process. The second term of the objective function is the sparse regularization term, which means that by adding Regularization is used to encourage the acquisition of a small subset of informative features, where is the regularization hyperparameter, is a tunable hyperparameter. It is an orthogonality constraint, which is used to ensure the independence and low correlation between high-scoring features, thereby reducing redundancy and collinearity and improving the generalization ability, robustness and interpretability of the model.

[0055] Step 5: Solve the matrix variables using alternating iterative optimization and First, fix the variable In the case of :For formula (1) Taking the derivative and setting it to zero, we get The analytical solution is shown in formula (2), where is a diagonal matrix whose elements ; (2); Next, in the fixed variable In the case of :After matrix trace transformation, formula (1) can be transformed into the optimized formula (3). Derivative, we get the expression . Because of the The solution of is orthogonal, so it can be regarded as an orthogonal Prucker problem. Perform singular value decomposition, that is , represents the left singular vector matrix, represents a diagonal matrix of singular values, represents the right singular vector matrix, from which we can get The optimal solution is , that is, finally optimize the variables according to formula (4) ; (3); (4).

[0056] Step 6: After the iteration, the final matrix variable is obtained and , where the matrix Saves the validity information of the feature. After normalization, calculate the feature importance score (i.e. the first feature importance score) , to quantify the contribution of each feature.

[0057] Step 7: Further apply dynamic weight constraints to the samples, and adopt the strategy of "easy first, difficult later" to learn feature data to enhance the model's noise resistance and robustness. This method is based on the self-paced learning strategy for feature reconstruction evaluation, and the specific implementation is shown in steps 8 to 10.

[0058] Step 8: First construct the following objective function (i.e., the second feature reconstruction objective function): (5); In the above formula, the purpose of the first term is to use data to fit the category label, so as to learn the relationship matrix, where Indicates The weight of the samples, represents the data label matrix (i.e., contains the category labels of all samples), For the samples, a total of samples, is a relationship matrix. The second term is a norm, used to learn the effectiveness score of the feature, where is a regularization hyperparameter. The third term is a soft regularization term for self-paced learning, which can learn the importance of samples. That is, key samples are given larger weights, while outliers or noise samples are assigned smaller weights, so that the model can achieve a robust effect. is a tunable hyperparameter, is the step size parameter for self-learning.

[0059] Step 9: Use the alternating iteration strategy for optimization. First, fix the variable In the case of : To facilitate the solution, we use matrix trace transformation to transform formula (5) into formula (6), where , , , is a diagonal matrix, Used to construct a diagonal matrix with diagonal elements , ; (6); Then, the formula (6) is Derivation, finally get The solution is shown in formula (7); (7); Next, in the fixed variable In the case of :make , transform formula (5) into formula (8), and then we can get The solution is shown in formula (9); (8); (9).

[0060] Step 10: After the iteration, the final variable is obtained and ,in The validity information of the feature is saved, and after normalization, the feature importance score (i.e. the second feature importance score) is obtained. .

[0061] Step 11: Combine and The calculation results give a comprehensive score To reduce the amount of calculation, select the top scorers. The features of compose a new feature matrix , conduct uncertainty quantification analysis; Step 12: First, use the visual feature extraction network to extract the multi-scale feature vector (i.e., visual feature) of the infrared ship image and compare it with the new feature matrix Fusion to embed high-dimensional visual features. In the fusion process, in order to ensure the semantic consistency of feature expression, the mean square error loss function ( ) as the feature alignment loss to optimize the feature distribution so that the feature vectors are aligned in the same feature space.

[0062] Step 13: The concatenated new feature matrix Input the classification network (i.e. the first classification network), perform training and classification evaluation, and generate the category output score of the sample (i.e. the classification evaluation result ). Then, combine The uncertainty loss function is used to optimize the class labels of the samples to measure the uncertainty of the samples and select the feature sets with higher stability and robustness. For specific implementation, see steps 14 to 16.

[0063] Step 14: Use Distribution of class probability vectors for samples Modeling is performed, assuming that it obeys the concentration parameter of Distribution, that is The corresponding probability density function is .in is a normalizing constant, called the multivariate Beta function, is the total number of categories, Indicates category The probability of Representative Category The prediction confidence of indicates the certainty of the model's prediction for this category.

[0064] Step 15: Calculate the expected value of the class probability , determine the model's predicted probability for each category.

[0065] Step 16: Use Entropy of distribution It measures the total uncertainty. The larger the entropy value, the higher the uncertainty of the model in predicting the results. for Function pair parameters The derivative of represents the contribution of each category’s prediction confidence to the model uncertainty.

[0066] Step 17: Use the screened stable and robust feature set to train the classification network (i.e., the second classification network) to optimize the model performance and obtain high-accuracy and stable classification results.

[0067] Compared with the prior art solutions, this embodiment has the following advantages: This embodiment includes four stages: feature evaluation, feature fusion, feature selection and classification. By using feature reconstruction technology based on data self-expression and self-paced learning and uncertainty quantification technology, a stable and robust feature set is selected to improve classification performance and anti-interference ability. Specifically, in the feature evaluation stage, in order to reduce feature redundancy and solve the noise problem, this embodiment proposes to regard the feature vector of the ship target as a data point, and by learning the linear combination coefficient, some features are expressed as a linear combination of other features, thereby realizing feature redundancy removal. On this basis, this embodiment further introduces orthogonal restrictions to improve the stability of data expression. In addition, in order to improve the robustness of the model, this embodiment imposes dynamic weight constraints on the samples and adopts a "first easy and then difficult" strategy to learn feature data, so that the proposed method has the ability to resist noise. In the feature fusion stage, in order to ensure the semantic consistency of feature expression, this embodiment proposes a feature alignment loss function to guide model training and optimize feature distribution, so that feature vectors are aligned in the same feature space. In the feature selection stage, this embodiment uses a distribution-based uncertainty quantification mechanism to evaluate features and select feature sets with higher stability and robustness to enhance the effectiveness of features in classification tasks. In the classification stage, this embodiment uses the screened high-quality feature set to train the classification model, thereby achieving accurate classification of infrared ship images.

[0068] Reference Figure 3The embodiment of the present application also provides an infrared ship image classification system based on feature reconstruction, which includes a data processing unit 100, a first reconstruction unit 200, a second reconstruction unit 300, a feature screening unit 400, a feature extraction unit 500, a classification evaluation unit 600, a sample evaluation unit 700 and an image classification unit 800, wherein: The data processing unit 100 is used to collect feature data of the infrared ship image and pre-process the feature data to obtain a feature matrix, where a row of the feature matrix represents a sample and a column represents a feature; A first reconstruction unit 200 is used to perform a first feature reconstruction on the feature matrix to obtain a first feature importance score; A second reconstruction unit 300 is used to perform a second feature reconstruction on the feature matrix to obtain a second feature importance score; A feature screening unit 400 is used to screen the features in the feature matrix based on the first feature importance score and the second feature importance score to obtain a new feature matrix; The feature extraction unit 500 is used to extract visual features from the infrared ship image and fuse the visual features with the new feature matrix to obtain a fused feature matrix; A classification evaluation unit 600 is used to input the fused feature matrix into the first classification network for classification evaluation to obtain a classification evaluation result; The sample evaluation unit 700 is used to perform uncertainty evaluation on the sample features according to the classification evaluation result and the class label of the sample to obtain an uncertainty evaluation result; The image classification unit 800 is used to screen the target feature set according to the uncertainty assessment result, and use the target feature set to train the second classification network, so as to classify the infrared ship image through the trained second classification network.

[0069] In some implementations, the first reconstruction unit 200 may be specifically configured to: A low-rank matrix is ​​used to represent a feature reconstruction coefficient matrix, and a first feature reconstruction objective function is constructed by combining the feature reconstruction coefficient matrix and the feature matrix. The first feature reconstruction objective function is used to perform a first feature reconstruction on the feature matrix. Using alternating iteration to solve the feature reconstruction coefficient matrix in the first feature reconstruction objective function; Based on the solved feature reconstruction coefficient matrix, the first feature importance score is calculated.

[0070] In some implementations, the first reconstruction unit 200 may be specifically configured to: ; in, Represents the matrix variable in the feature reconstruction coefficient matrix and matrix variables , represents the total number of features in the feature matrix, Indicates Features, represents the regularization hyperparameter, represents the data reconstruction error term, Indicates based on The sparse regularization term of the norm, represents an adjustable hyperparameter, represents the orthogonality constraint, Indicates transpose.

[0071] In some implementations, the second reconstruction unit 300 may be specifically configured to: Based on the sample weights, the relationship matrix and the feature matrix, a second feature reconstruction objective function is constructed, and the second feature reconstruction objective function is used to perform a second feature reconstruction on the feature matrix; Alternating iteration is used to solve the sample weight and relationship matrix in the second feature reconstruction objective function; Based on the solved sample weight, the second feature importance score is calculated.

[0072] In some implementations, the second reconstruction unit 300 may be specifically configured to: ; in, represents the total number of samples in the feature matrix, Indicates The sample weights, represents the data label matrix, Indicates The class labels corresponding to the samples are Indicates samples, represents the relationship matrix, represents the regularization hyperparameter, represents an adjustable hyperparameter, Indicates using data to fit category labels. Indicates based on The sparse regularization term of the norm, represents the soft regularization term of self-paced learning, Represents the self-paced learning step size parameter.

[0073] In some implementations, the sample evaluation unit 700 may be specifically configured to: Dirichlet distribution is used to construct uncertainty loss function; According to the classification evaluation results and the category labels of the samples, the uncertainty loss function is used to perform uncertainty evaluation on the sample features to obtain the uncertainty evaluation results.

[0074] In some implementations, the sample evaluation unit 700 may be specifically configured to: ; in, represents the entropy of the Dirichlet distribution, represents the class probability vector, represents the normalization constant, represents the total number of categories, Indicates The predicted confidence of the category.

[0075] It should be noted that since the infrared ship image classification system based on feature reconstruction in this embodiment and the above-mentioned infrared ship image classification method based on feature reconstruction are based on the same inventive concept, the corresponding contents in the method embodiment are also applicable to the system embodiment and will not be described in detail here.

[0076] Reference Figure 4 The present application also provides an electronic device, which includes: at least one memory; at least one processor; at least one program; The programs are stored in the memory, and the processor executes at least one program to implement the infrared ship image classification method based on feature reconstruction described above in the present disclosure.

[0077] The electronic device may be any intelligent terminal including a mobile phone, a tablet computer, a personal digital assistant (PDA), a vehicle-mounted computer, etc.

[0078] The electronic device according to the embodiment of the present application is described in detail below.

[0079] The processor 1600 may be implemented by a general-purpose central processing unit (CPU), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present disclosure; The memory 1700 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 1700 can store an operating system and other application programs. When the technical solution provided in the embodiment of this specification is implemented by software or firmware, the relevant program code is stored in the memory 1700, and the processor 1600 calls and executes the infrared ship image classification method based on feature reconstruction in the embodiment of the present disclosure.

[0080] Input / output interface 1800, used to implement information input and output; Communication interface 1900, used to realize communication interaction between the device and other devices, which can be realized through wired mode (such as USB, network cable, etc.) or wireless mode (such as mobile network, WIFI, Bluetooth, etc.); Bus 2000 , which transmits information between various components of the device (e.g., processor 1600 , memory 1700 , input / output interface 1800 , and communication interface 1900 ); The processor 1600 , the memory 1700 , the input / output interface 1800 , and the communication interface 1900 are connected to each other in communication within the device via the bus 2000 .

[0081] The embodiment of the present disclosure also provides a storage medium, which is a computer-readable storage medium. The computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are used to enable a computer to execute the above-mentioned infrared ship image classification method based on feature reconstruction.

[0082] The memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may optionally include a memory remotely disposed relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0083] The embodiments described in the embodiments of the present disclosure are intended to more clearly illustrate the technical solutions of the embodiments of the present disclosure and do not constitute a limitation on the technical solutions provided by the embodiments of the present disclosure. Those skilled in the art will appreciate that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of the present disclosure are also applicable to similar technical problems.

[0084] Those skilled in the art will appreciate that the technical solutions shown in the figures do not limit the embodiments of the present disclosure and may include more or fewer steps than shown in the figures, or a combination of certain steps, or different steps.

[0085] The device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separated, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0086] Those skilled in the art will appreciate that all or some of the steps in the methods disclosed above, and the functional modules / units in the systems and devices may be implemented as software, firmware, hardware, or a suitable combination thereof.

[0087] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0088] It should be understood that in the present application, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0089] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0090] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0091] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0092] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including multiple instructions to enable an electronic device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, ROM), random access memory (Random Access Memory, RAM), disk or optical disk and other media that can store programs. The above is a detailed description of the embodiments of the present application in conjunction with the accompanying drawings, but the present application is not limited to the above embodiments. Within the scope of knowledge possessed by ordinary technicians in the relevant technical field, various changes can be made without departing from the purpose of the present application.

[0093] The embodiments of the present application are described in detail above in conjunction with the accompanying drawings, but the present application is not limited to the above embodiments. Various changes can be made within the knowledge scope of ordinary technicians in the relevant technical field without departing from the purpose of the present application.

Claims

1. A method for classifying infrared ship images based on feature reconstruction, characterized in that: The method comprises: Collecting feature data of infrared ship images, and preprocessing the feature data to obtain a feature matrix, wherein a row of the feature matrix represents a sample, and a column represents a feature; Performing a first feature reconstruction on the feature matrix to obtain a first feature importance score; Performing a second feature reconstruction on the feature matrix to obtain a second feature importance score; Based on the first feature importance score and the second feature importance score, the features in the feature matrix are screened to obtain a new feature matrix; Extracting visual features from the infrared ship image, and fusing the visual features with the new feature matrix to obtain a fused feature matrix; Inputting the fused feature matrix into a first classification network for classification evaluation to obtain a classification evaluation result; According to the classification evaluation result and the category label of the sample, uncertainty evaluation is performed on the sample feature to obtain an uncertainty evaluation result; A target feature set is screened according to the uncertainty assessment result, and a second classification network is trained using the target feature set, so that the infrared ship image is classified through the trained second classification network.

2. The infrared ship image classification method based on feature reconstruction according to claim 1 is characterized in that: The first feature reconstruction of the feature matrix to obtain a first feature importance score includes: A low-rank matrix is ​​used to represent a feature reconstruction coefficient matrix, and a first feature reconstruction objective function is constructed by combining the feature reconstruction coefficient matrix and the feature matrix, wherein the first feature reconstruction objective function is used to perform a first feature reconstruction on the feature matrix; Solving the feature reconstruction coefficient matrix in the first feature reconstruction objective function by using alternating iteration; Based on the solved feature reconstruction coefficient matrix, the first feature importance score is calculated.

3. The infrared ship image classification method based on feature reconstruction according to claim 2 is characterized in that: The step of combining the feature reconstruction coefficient matrix and the feature matrix to construct a first feature reconstruction objective function includes: ； in, Represents the matrix variables in the feature reconstruction coefficient matrix and matrix variables , represents the total number of features in the feature matrix, Indicates Features, represents the regularization hyperparameter, represents the data reconstruction error term, Indicates based on The sparse regularization term of the norm, represents an adjustable hyperparameter, represents the orthogonality constraint, Indicates transpose.

4. The infrared ship image classification method based on feature reconstruction according to claim 1 is characterized in that: The performing a second feature reconstruction on the feature matrix to obtain a second feature importance score includes: Based on the sample weights, the relationship matrix and the feature matrix, construct a second feature reconstruction objective function, wherein the second feature reconstruction objective function is used to perform a second feature reconstruction on the feature matrix; Using alternating iteration to solve the sample weights and relationship matrix in the second feature reconstruction objective function; Based on the solved sample weight, the second feature importance score is calculated.

5. The infrared ship image classification method based on feature reconstruction according to claim 4 is characterized in that: The constructing a second feature reconstruction objective function based on the sample weights, the relationship matrix and the feature matrix includes: ； in, represents the total number of samples in the feature matrix, Indicates The sample weights, represents the data label matrix, Indicates The class labels corresponding to the samples are Indicates samples, represents the relationship matrix, represents the regularization hyperparameter, represents an adjustable hyperparameter, Indicates using data to fit category labels. Indicates based on The sparse regularization term of the norm, represents the soft regularization term of self-paced learning, Represents the self-paced learning step size parameter.

6. The infrared ship image classification method based on feature reconstruction according to claim 1 is characterized in that: The step of performing uncertainty assessment on sample features according to the classification assessment result and the category label of the sample to obtain an uncertainty assessment result includes: Dirichlet distribution is used to construct uncertainty loss function; According to the classification evaluation result and the category label of the sample, the uncertainty loss function is used to perform uncertainty evaluation on the sample feature to obtain an uncertainty evaluation result.

7. The infrared ship image classification method based on feature reconstruction according to claim 6 is characterized in that: The use of Dirichlet distribution to construct an uncertainty loss function includes: ； in, represents the entropy of the Dirichlet distribution, represents the class probability vector, represents the normalization constant, represents the total number of categories, Indicates The prediction confidence of the category, It means partial derivative.

8. An infrared ship image classification system based on feature reconstruction, characterized in that: The system comprises: A data processing unit, used for collecting feature data of infrared ship images, and preprocessing the feature data to obtain a feature matrix, wherein a row of the feature matrix represents a sample, and a column represents a feature; A first reconstruction unit, used for performing a first feature reconstruction on the feature matrix to obtain a first feature importance score; A second reconstruction unit is used to perform a second feature reconstruction on the feature matrix to obtain a second feature importance score; A feature screening unit, configured to screen features in a feature matrix based on the first feature importance score and the second feature importance score to obtain a new feature matrix; A feature extraction unit, used for extracting visual features from the infrared ship image, and fusing the visual features with the new feature matrix to obtain a fused feature matrix; A classification evaluation unit, used for inputting the fused feature matrix into a first classification network for classification evaluation to obtain a classification evaluation result; A sample evaluation unit, used to perform uncertainty evaluation on sample features according to the classification evaluation result and the category label of the sample to obtain an uncertainty evaluation result; The image classification unit is used to screen a target feature set according to the uncertainty assessment result, and use the target feature set to train a second classification network, so as to classify the infrared ship image through the trained second classification network.

9. An electronic device, characterized in that: It includes at least one control processor and a memory for communicating with the at least one control processor; the memory stores instructions that can be executed by the at least one control processor, and the instructions are executed by the at least one control processor so that the at least one control processor can execute the infrared ship image classification method based on feature reconstruction as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are used to enable a computer to execute the infrared ship image classification method based on feature reconstruction according to any one of claims 1 to 7.

Images