An Infrared Ship Image Classification Method, System and Device Based on Feature Reconstruction

Abstract

The present application discloses an infrared ship image classification method, system and device based on feature reconstruction. The method performs a first feature reconstruction and a second feature reconstruction on a feature matrix; based on a first feature importance score and a second feature importance score, filters the features in the feature matrix to obtain a new feature matrix; fuses the extracted visual features with the new feature matrix to obtain a fused feature matrix; inputs the fused feature matrix into a first classification network for classification evaluation; according to the classification evaluation result and the class label of the sample, performs uncertainty evaluation on the sample features to obtain an uncertainty evaluation result; filters a target feature set according to the uncertainty evaluation result, and uses the target feature set to train a second classification network, so as to classify infrared ship images through the trained second classification network. The present application can obtain a feature set with higher stability and robustness, thereby improving the classification accuracy of infrared ship images.

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:

[0007] 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;

[0008] Performing a first feature reconstruction on the feature matrix to obtain a first feature importance score;

[0009] Performing a second feature reconstruction on the feature matrix to obtain a second feature importance score;

[0010] 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;

[0011] Extract the visual features in the infrared ship image, and fuse the visual features with the new feature matrix to obtain a fused feature matrix;

[0012] Input the fused feature matrix into the first classification network for classification evaluation to obtain a classification evaluation result;

[0013] According to the classification evaluation result and the class label of the sample, perform uncertainty evaluation on the sample features to obtain an uncertainty evaluation result;

[0014] Screen the target feature set according to the uncertainty evaluation 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.

[0015] Compared with the prior art, the first aspect of the present application has the following beneficial effects:

[0016] This method collects the feature data of infrared ship images, preprocesses the feature data to obtain a feature matrix, where one row of the feature matrix represents a sample and one column represents a feature; performs the first feature reconstruction on the feature matrix to obtain the first feature importance score; performs the second feature reconstruction on the feature matrix to obtain the second feature importance score; based on the first feature importance score and the second feature importance score, screen the features in the feature matrix to obtain a new feature matrix; extract the visual features in the infrared ship image, and fuse the visual features with the new feature matrix to obtain a fused feature matrix; input the fused feature matrix into the first classification network for classification evaluation to obtain a classification evaluation result; according to the classification evaluation result and the class label of the sample, perform uncertainty evaluation on the sample features to obtain an uncertainty evaluation result; screen the target feature set according to the uncertainty evaluation 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. In this way, feature redundancy and noise can be removed and feature robustness can be improved through two feature reconstructions. By fusing the visual features with the new feature matrix, the feature information can be enriched. By performing uncertainty evaluation on the sample features, the uncertainty of the sample can be measured, and a feature set with higher stability and robustness can be screened out. Finally, the second classification network is trained with the target feature set, so that the trained second classification network can accurately classify the infrared ship image, improving the accuracy of infrared ship image classification.

[0017] In some embodiments, the performing the first feature reconstruction on the feature matrix to obtain the first feature importance score includes:

[0018] The feature reconstruction coefficient matrix is represented by a low-rank 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 the first feature reconstruction on the feature matrix;

[0019] The feature reconstruction coefficient matrix in the first feature reconstruction objective function is solved by using alternating iteration;

[0020] Based on the solved feature reconstruction coefficient matrix, a first feature importance score is calculated.

[0021] In some embodiments, the constructing the first feature reconstruction objective function by combining the feature reconstruction coefficient matrix and the feature matrix includes:

[0022] ;

[0023] where represents the matrix variable in the feature reconstruction coefficient matrix and the matrix variable , represents the total number of features in the feature matrix, represents the th feature, represents the regularization hyperparameter, represents the data reconstruction error term, represents based on the norm of the sparse regularization term, represents the tunable hyperparameter, represents the orthogonality constraint condition, represents the transpose.

[0024] In some embodiments, the performing the second feature reconstruction on the feature matrix to obtain a second feature importance score includes:

[0025] Based on the sample weights, the relationship matrix, and the feature matrix, a second feature reconstruction objective function is constructed. The second feature reconstruction objective function is used to perform the second feature reconstruction on the feature matrix;

[0026] The sample weights and the relationship matrix in the second feature reconstruction objective function are solved by using alternating iteration;

[0027] Based on the solved sample weights, a second feature importance score is calculated.

[0028] In some embodiments, the constructing the second feature reconstruction objective function based on the sample weights, the relationship matrix, and the feature matrix includes:

[0029] ;

[0030] Among them, represents the total number of samples in the feature matrix, represents the th sample weight, represents the data label matrix, represents the th class label corresponding to the sample, represents the th sample, represents the relationship matrix, represents the regularization hyperparameter, represents the tunable hyperparameter, represents fitting the class label with data, represents based on norm sparse regularization term, represents the soft regularization term of self-paced learning, represents the self-paced learning step size parameter.

[0031] In some embodiments, the uncertainty evaluation of the sample features is performed according to the classification evaluation result and the class label of the sample to obtain an uncertainty evaluation result, including:

[0032] Construct an uncertainty loss function using the Dirichlet distribution;

[0033] According to the classification evaluation result and the class label of the sample, use the uncertainty loss function to perform uncertainty evaluation on the sample features to obtain an uncertainty evaluation result.

[0034] In some embodiments, the construction of the uncertainty loss function using the Dirichlet distribution includes:

[0035] ;

[0036] Among them, represents the entropy of the Dirichlet distribution, represents the class probability vector, represents the normalization constant, represents the total number of classes, represents the th class prediction confidence.

[0037] In a second aspect, an infrared ship image classification system based on feature reconstruction is further provided in an embodiment of the present application. The system includes:

[0038] A data processing unit, configured to collect feature data of infrared ship images and preprocess the feature data to obtain a feature matrix, where one row of the feature matrix represents a sample and one column represents a feature;

[0039] The first reconstruction unit is used to perform the first feature reconstruction on the feature matrix to obtain the first feature importance score;

[0040] The second reconstruction unit is used to perform the second feature reconstruction on the feature matrix to obtain the second feature importance score;

[0041] The feature screening unit 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;

[0042] The feature extraction unit is used to extract the visual features in the infrared ship image and fuse the visual features with the new feature matrix to obtain a fused feature matrix;

[0043] The classification evaluation unit is used to input the fused feature matrix into the first classification network for classification evaluation to obtain a classification evaluation result;

[0044] The sample evaluation unit 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;

[0045] The image classification unit is used to screen the target feature set according to the uncertainty evaluation result and train the second classification network with the target feature set, so as to classify the infrared ship image through the trained second classification network.

[0046] In a third aspect, an embodiment of the present application further provides an electronic device, including at least one control processor and a memory communicatively connected to the at least one control processor; the memory stores instructions executable 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 a method for classifying infrared ship images based on feature reconstruction as described above.

[0047] In a fourth aspect, an embodiment of the present application further provides a computer-readable storage medium, and the computer-readable storage medium stores computer-executable instructions for causing a computer to execute a method for classifying infrared ship images based on feature reconstruction as described above.

[0048] It can be understood that the beneficial effects of the above second aspect to the fourth aspect compared with the related art are the same as those of the first aspect compared with the related art, and reference can be made to the relevant descriptions in the first aspect, which will not be elaborated here. Description of the Drawings

[0049] The above and / or additional aspects and advantages of the present application will become apparent and be readily understood from the description of the embodiments in conjunction with the following drawings, where:

[0050] Figure 1 is a schematic flowchart of an embodiment of the infrared ship image classification method based on feature reconstruction provided by the present application;

[0051] Figure 2 is a schematic overall method flowchart of the best embodiment of the infrared ship image classification method based on feature reconstruction provided by the present application;

[0052] Figure 3 is a schematic structural diagram of an embodiment of the infrared ship image classification system based on feature reconstruction provided by the present application;

[0053] Figure 4 is a schematic structural diagram of an embodiment of the electronic device provided by the present application. Detailed Embodiments

[0054] The embodiments of the present application will be described in detail below. Examples of the embodiments are shown in the drawings, where the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described by referring to the drawings are exemplary only for explaining the present application and should not be construed as limiting the present application.

[0055] In the description of the present application, if the first, second, etc. are described only for the purpose of distinguishing technical features, they should not be construed as indicating or implying relative importance or implicitly indicating the quantity of the indicated technical features or implicitly indicating the sequence of the indicated technical features.

[0056] In the description of the present application, it should be understood that for the orientation description, such as up, down, etc., the orientation or positional relationship indicated is based on the orientation or positional relationship shown in the drawings. It is only for the convenience of describing the present application and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus should not be construed as limiting the present application.

[0057] In the description of the present application, it should be noted that unless otherwise clearly defined, words such as setting, installing, connecting, etc. should be understood in a broad sense, and those skilled in the art can reasonably determine the specific meanings of the above words in the present application in combination with the specific content of the technical solution.

[0058] In a complex naval battle environment, the target features in infrared images often exhibit problems such as a lot of noise, serious redundancy, and poor stability, resulting in difficulties for traditional methods to fully extract the essential features of the target for accurate classification.

[0059] To solve the problem that it is difficult for traditional methods to fully extract the essential features of targets for accurate classification, this application proposes an infrared ship image classification method, system, and device based on feature reconstruction.

[0060] Referring to Figure 1 , an embodiment of this application provides an infrared ship image classification method based on feature reconstruction. The method includes the following steps:

[0061] Step S100: Collect the feature data of infrared ship images and preprocess the feature data to obtain a feature matrix. One row of the feature matrix represents a sample, and one column represents a feature;

[0062] Step S200: Perform the first feature reconstruction on the feature matrix to obtain the first feature importance score;

[0063] Step S300: Perform the second feature reconstruction on the feature matrix to obtain the second feature importance score;

[0064] Step S400: Based on the first feature importance score and the second feature importance score, screen the features in the feature matrix to obtain a new feature matrix;

[0065] Step S500: Extract the visual features of the infrared ship images and fuse the visual features with the new feature matrix to obtain a fused feature matrix;

[0066] Step S600: Input the fused feature matrix into the first classification network for classification evaluation to obtain a classification evaluation result;

[0067] Step S700: According to the classification evaluation result and the class label of the sample, perform uncertainty evaluation on the sample features to obtain an uncertainty evaluation result;

[0068] Step S800: Screen the target feature set according to the uncertainty evaluation result and train the second classification network using the target feature set, so as to classify the infrared ship images through the trained second classification network.

[0069] In this embodiment, by collecting the feature data of infrared ship images and preprocessing the feature data, a feature matrix is obtained. One row of the feature matrix represents a sample, and one column represents a feature; the first feature reconstruction is performed on the feature matrix to obtain the first feature importance score; the second feature reconstruction is performed on the feature matrix to obtain the 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; the visual features in the infrared ship images are extracted and fused with the new feature matrix to obtain a fused feature matrix; the fused feature matrix is input into the first classification network for classification evaluation to obtain a classification evaluation result; according to the classification evaluation result and the class label of the sample, the uncertainty of the sample features is evaluated to obtain an uncertainty evaluation result; according to the uncertainty evaluation result, the target feature set is screened, and the second classification network is trained with the target feature set, so as to classify the infrared ship images through the trained second classification network. In this way, through two feature reconstructions, feature redundancy and noise can be removed, and feature robustness can be improved. By fusing the visual features with the new feature matrix, feature information can be enriched. By evaluating the uncertainty of the sample features, the uncertainty of the samples can be measured, and a feature set with higher stability and robustness can be screened out. Finally, the second classification network is trained with the target feature set, so that the trained second classification network can accurately classify the infrared ship images, improving the accuracy of infrared ship image classification.

[0070] The above preprocessing of the feature data can be mechanical energy preprocessing such as normalizing the feature data, processing missing values, and aligning the standard dimensions.

[0071] The above first feature reconstruction of the feature matrix to obtain the first feature importance score can be to perform the first feature reconstruction on the feature matrix using the data self-expression technique to obtain the first feature importance score.

[0072] The above second feature reconstruction of the feature matrix to obtain the second feature importance score can be to perform the second feature reconstruction on the feature matrix using the self-paced learning strategy to obtain the second feature importance score.

[0073] The above screening of the features in the feature matrix based on the first feature importance score and the second feature importance score can be to add the first feature importance score and the second feature importance score to obtain a sum of scores result, and screen multiple features with the top scores in the feature matrix according to the sum of scores result, and construct a new matrix with the screened features.

[0074] The above-mentioned visual features extracted from the infrared ship images can be the visual features extracted from the infrared ship images by using a visual feature extraction network. The visual feature extraction network can be a visual feature extraction network well-known to those skilled in the art, and no specific description or limitation is made in this embodiment.

[0075] The category labels of the above samples can be pre-labeled category labels.

[0076] The above first classification network and second classification network can be the same classification network. The classification network can be a classification network well-known to those skilled in the art. For example, a multi-layer perceptron (MLP) can be used as the classification network, and no specific description or limitation is made in this embodiment.

[0077] In some embodiments, performing a first feature reconstruction on the feature matrix to obtain a first feature importance score includes:

[0078] Representing the feature reconstruction coefficient matrix by a low-rank matrix, and combining the feature reconstruction coefficient matrix and the feature matrix to construct a first feature reconstruction objective function, where the first feature reconstruction objective function is used to perform the first feature reconstruction on the feature matrix;

[0079] Using alternating iteration to solve the feature reconstruction coefficient matrix in the first feature reconstruction objective function;

[0080] Based on the solved feature reconstruction coefficient matrix, calculating the first feature importance score.

[0081] In this embodiment, by representing the feature reconstruction coefficient matrix by a low-rank matrix, and combining the feature reconstruction coefficient matrix and the feature matrix to construct a first feature reconstruction objective function, where the first feature reconstruction objective function is used to perform the 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, calculating the first feature importance score. In this way, representing the feature reconstruction coefficient matrix by a low-rank matrix to achieve dimensionality reduction of the feature matrix and removing noise and redundant information in the feature data.

[0082] The above-mentioned calculating the first feature importance score based on the solved feature reconstruction coefficient matrix can be based on the solved feature reconstruction coefficient matrix, normalizing the features in the matrix that preserves the effective information of the features, and then calculating the feature importance score for the normalized features to obtain the first feature importance score.

[0083] In some embodiments, combining the feature reconstruction coefficient matrix and the feature matrix to construct a first feature reconstruction objective function includes:

[0084] ;

[0085] Among them, represents the matrix variable in the feature reconstruction coefficient matrix and the matrix variable , represents the total number of features in the feature matrix, represents the th feature, represents the regularization hyperparameter, represents the data reconstruction error term, represents based on the sparse regularization term of the norm, represents the adjustable hyperparameter, represents the orthogonality constraint condition, represents the transpose.

[0086] In some embodiments, a second feature reconstruction is performed on the feature matrix to obtain a second feature importance score, including:

[0087] Based on the sample weights, the relationship matrix, and the feature matrix, construct a second feature reconstruction objective function, which is used to perform a second feature reconstruction on the feature matrix;

[0088] Use alternating iteration to solve for the sample weights and the relationship matrix in the second feature reconstruction objective function;

[0089] Based on the solved sample weights, calculate the second feature importance score.

[0090] In this embodiment, by constructing a second feature reconstruction objective function based on the sample weights, the relationship matrix, and the feature matrix, which is used to perform a second feature reconstruction on the feature matrix; using alternating iteration to solve for the sample weights and the relationship matrix in the second feature reconstruction objective function; and calculating the second feature importance score based on the solved sample weights. In this way, by imposing dynamic weight constraints on the samples and adopting a "from easy to difficult" strategy to learn the feature data, the noise resistance and robustness of the model are enhanced.

[0091] The above-mentioned calculation of the second feature importance score based on the solved sample weights can be to normalize the solved sample weights to obtain the second feature importance score.

[0092] In some embodiments, constructing a second feature reconstruction objective function based on the sample weights, the relationship matrix, and the feature matrix includes:

[0093] ;

[0094] Among them, represents the total number of samples in the feature matrix, represents the a sample weight, denotes the data label matrix, denotes the category label corresponding to the denotes the th sample, denotes the relationship matrix, denotes the regularization hyperparameter, denotes the tunable hyperparameter, denotes fitting the category labels with the data, denotes based on the sparse regularization term of the denotes the soft regularization term of self-paced learning, denotes the self-paced learning step size parameter.

[0095] In this embodiment, by performing feature reconstruction evaluation through the self-paced learning strategy, the anti-noise ability and robustness of the model can be enhanced, and the robustness of the features can be improved.

[0096] The above-mentioned fitting of the category labels with the data can learn the relationship matrix.

[0097] The above-mentioned sparse regularization term is used to learn the effectiveness score of the features.

[0098] The above-mentioned soft regularization term of self-paced learning can learn the importance of the samples. That is, key samples are assigned larger weights, while outliers or noise samples are assigned smaller weights, so that the model can obtain a robust effect.

[0099] In some embodiments, according to the classification evaluation result and the category label of the sample, the uncertainty of the sample features is evaluated to obtain the uncertainty evaluation result, including:

[0100] Construct an uncertainty loss function using the Dirichlet distribution;

[0101] According to the classification evaluation result and the category label of the sample, use the uncertainty loss function to evaluate the uncertainty of the sample features to obtain the uncertainty evaluation result.

[0102] In this embodiment, an uncertainty loss function is constructed using the Dirichlet distribution; according to the classification evaluation result and the category label of the sample, the uncertainty of the sample features is evaluated using the uncertainty loss function to obtain the uncertainty evaluation result. In this way, the uncertainty of the sample features is evaluated through the uncertainty loss function to measure the uncertainty of the sample features and screen out a feature set with higher stability and robustness.

[0103] In some embodiments, constructing an uncertainty loss function using the Dirichlet distribution includes:

[0104] ;

[0105] wherein, represents the entropy of the Dirichlet distribution, represents the class probability vector, represents the normalization constant, represents the total number of classes, represents the predicted confidence of the

[0106] For the convenience of those skilled in the art to understand, the following provides a set of best embodiments:

[0107] Traditional feature effectiveness evaluation methods include three types: filter-based, embedded, and wrapper-based. Among them, the wrapper-based method regards feature effectiveness evaluation as a subset search problem, directly evaluates the performance of each subset by training models on different feature subsets (such as calculating metrics like accuracy and F1-score), and finally selects the feature combination with the best performance. This method can capture the interaction between features, but due to the need to train models for each subset, its computational cost is high. The filter-based method relies on the statistical relationship between features and the target variable. First, calculate the correlation between each feature and the target variable (such as Pearson correlation coefficient and mutual information), and then rank the features according to statistical metrics, and select the features with higher rankings as the finally selected features. Its computational efficiency is high, but it may ignore the complex relationships between features. The embedded method integrates feature effectiveness evaluation into the model training process, enabling the model to automatically identify and adjust the importance of each feature during learning, and select the most effective features. Typical methods include those based on regularization (such as L1 regularization making some feature weights zero) and those based on tree models (such as decision trees and random forests). This method can both capture the interaction between features and automatically adjust feature weights, but may be affected by the preferences of the model itself. Generally speaking, the selection of feature effectiveness 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.

[0108] In a complex naval battle environment, the target features in infrared images often exhibit problems such as a lot of noise, serious redundancy, and poor stability, resulting in traditional methods being difficult to fully mine the essential features of the target for accurate classification. This embodiment proposes a feature effectiveness evaluation method based on reconstruction and representation learning, which can not only fully mine and utilize the internal relationships between features, reduce redundant features, but also improve the robustness of feature evaluation by adjusting dynamic sample weights, providing more stable and reliable feature support for subsequent infrared ship image classification, thereby improving classification accuracy and anti-interference ability.

[0109] Refer to Figure 2, the 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, many redundant features, and poor anti-interference ability of infrared ship targets in complex naval warfare environments. This method combines 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. The specific steps are as follows:

[0110] 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 features from different sources into a feature matrix of the same dimension for model input.

[0111] Step 2: Represent the preprocessed feature matrix as , where 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 kind of feature.

[0112] Step 3: Use data self-expression technology and low-rank technology for feature reconstruction to solve the problems of feature redundancy and noise. The specific implementation is shown in Steps 4 to 6.

[0113] Step 4: Use low-rank matrices and to represent the feature reconstruction coefficient matrix to reduce the dimension of the feature matrix and remove the noise and redundant information in the data, where is the feature sparse matrix, is the orthogonal matrix, is the rank of the matrix. The specific optimization objective function (i.e., the first feature reconstruction objective function) is shown in Formula (1):

[0114] (1);

[0115] The first term of the objective function is the data reconstruction error term, where represents the th feature, and there are features in total. This term approximates the original data through a linear combination of feature subsets, aiming to minimize the reconstruction error, so as to screen out the most representative feature subset during the feature evaluation process. The second term of the objective function is the sparse regularization term, which represents adding regularization to encourage obtaining a small and informative feature subset, where is a regularization hyperparameter, and is an orthogonality constraint condition, which is used to ensure the independence and low correlation between high-score features, thereby reducing redundancy and collinearity, and improving the generalization ability, robustness, and interpretability of the model.

[0116] Step 5: Solve the matrix variables and by using the alternating iterative optimization method. First, update the variable while fixing the variable : Take the derivative of Equation (1) with respect to and set its derivative to zero, then the analytical solution of can be obtained, as shown in Equation (2), where is a diagonal matrix, and its element ;

[0117] (2);

[0118] Next, update the variable while fixing the variable : After the matrix trace transformation of Equation (1), it can be transformed into the optimization formula (3). Take the derivative of Equation (3) with respect to to obtain the expression . Since the solution of is subject to orthogonality constraints, it can be regarded as an orthogonal Procrustes problem, and directly perform singular value decomposition on , that is , represents the left singular vector matrix, represents the singular value diagonal matrix, represents the right singular vector matrix, and thus the optimal solution of can be obtained as , that is, finally optimize the variable according to Equation (4);

[0119] (3);

[0120] (4).

[0121] Step 6: After the iteration ends, obtain the final matrix variables and , where the matrix stores the valid information of the features. After normalizing the matrix , calculate the feature importance score (i.e., the first feature importance score) to quantify the contribution degree of each feature.

[0122] Step 7: Further impose dynamic weight constraints on the samples, and adopt a "from easy to difficult" strategy to learn the feature data, so as to enhance the anti-noise ability and robustness of the model. This method is based on the self-paced learning strategy to evaluate feature reconstruction, and the specific implementation is shown in Steps 8 to 10.

[0123] Step 8: First, construct the following objective function (i.e., the second feature reconstruction objective function):

[0124] (5);

[0125] In the above formula, the purpose of the first term is to fit the class labels with the data, so as to learn the relationship matrix, where represents the weight of the th sample, represents the data label matrix (i.e., the class labels of all samples), is the th sample, and there are samples in total, is the relationship matrix. The second term is a sparse regularization term based on the norm, which is used to learn the effectiveness score of the features, where is the regularization hyperparameter. The third term is the soft regularization term of self-paced learning, which can learn the importance of the samples. That is, the key samples are given larger weights, while the outliers or noise samples are assigned smaller weights, so that the model can obtain a robust effect, where is the adjustable hyperparameter, is the self-paced learning step size parameter.

[0126] Step 9: Adopt an alternating iteration strategy for optimization. First, when fixing the variable , update the variable : For the convenience of solution, use the matrix trace transformation to transform formula (5) into formula (6), where , , , is a diagonal matrix, is used to construct the diagonal matrix, and the diagonal elements are , ;

[0127] (6);

[0128] Subsequently, take the derivative of formula (6) with respect to , and finally obtain the solution of as shown in formula (7);

[0129] (7);

[0130] Next, with the fixed variable held constant, update the variable : Let , transform formula (5) into formula (8), and thus obtain The solution of is shown in formula (9);

[0131] (8);

[0132] (9).

[0133] Step 10. After the iteration ends, obtain the final variables and , where stores the validity information of the features. After normalizing it, the feature importance score (i.e., the second feature importance score) is obtained .

[0134] Step 11. Combine and to obtain the comprehensive score . To reduce the computational amount, select the top features ranked by score to form a new feature matrix for uncertainty quantification analysis;

[0135] Step 12. First, use the visual feature extraction network to extract the multi-scale feature vectors (i.e., visual features) of the infrared ship image and fuse them with the new feature matrix to embed high-dimensional visual features. During the fusion process, to ensure the semantic consistency of the feature representation, the mean squared error loss function ( ) is used as the feature alignment loss to optimize the feature distribution, so that the feature vectors are aligned within the same feature space.

[0136] Step 13. Input the concatenated new feature matrix into the classification network (i.e., the first classification network) for training and classification evaluation, and generate the class output score of the sample (i.e., the classification evaluation result ). Subsequently, combine with the class label of the sample and use the uncertainty loss function for optimization to measure the uncertainty of the sample and select a feature set with high stability and robustness. The specific implementation is shown in Steps 14 to 16.

[0137] Step 14. Use the distribution to model the class probability vector of the sample, assuming that it follows a distribution with a concentration parameter of , that is . The corresponding probability density function is . Among them is the normalization constant, called the multivariate Beta function, is the total number of categories, represents the probability of category , represents the predicted confidence of category , indicating the certainty of the model's prediction for this category.

[0138] Step 15: Determine the prediction probability of the model for each category by calculating the expected value of the category probability.

[0139] Step 16: Use the entropy of the distribution to measure the total uncertainty. The larger the entropy value, the higher the uncertainty of the model's prediction result. Among them is the derivative of the function with respect to the parameter , indicating the contribution of the predicted confidence of each category to the model's uncertainty.

[0140] Step 17: Use the filtered stable and robust feature set to train the classification network (i.e., the second classification network) to optimize the model performance and obtain a highly accurate and stable classification result.

[0141] Compared with the existing technical solution, this embodiment has the following advantages:

[0142] This embodiment includes four stages: feature evaluation, feature fusion, feature selection, and classification. By utilizing feature reconstruction techniques and uncertainty quantification techniques based on data self-expression and self-paced learning, a stable and robust feature set is screened out to improve classification performance and anti-interference ability. Specifically, in the feature evaluation stage, to reduce feature redundancy and solve the noise problem, this embodiment proposes regarding the feature vector of the ship target as data points, and by learning the linear combination coefficients, expressing some features as linear combinations of other features, thereby achieving feature redundancy reduction. On this basis, this embodiment further introduces orthogonal constraints to improve the stability of data expression. In addition, to improve the robustness of the model, this embodiment imposes dynamic weight constraints on the samples and adopts a "from easy to difficult" strategy to learn feature data, making the proposed method have the ability to resist noise. In the feature fusion stage, to ensure the semantic consistency of feature expression, this embodiment proposes a feature alignment loss function to guide model training and optimize the feature distribution, so that the feature vectors are aligned within the same feature space. In the feature selection stage, this embodiment uses a distribution-based uncertainty quantification mechanism to evaluate the features and screen out a feature set with high stability and robustness to enhance the effectiveness of the features in the classification task. In the classification stage, this embodiment uses the screened high-quality feature set to train a classification model, thereby achieving accurate classification of infrared ship images.

[0143] Referring to Figure 3 , this embodiment of the present application also provides an infrared ship image classification system based on feature reconstruction. The system 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, where:

[0144] The data processing unit 100 is configured to collect feature data of infrared ship images and preprocess the feature data to obtain a feature matrix, where one row of the feature matrix represents a sample and one column represents a feature;

[0145] The first reconstruction unit 200 is configured to perform a first feature reconstruction on the feature matrix to obtain a first feature importance score;

[0146] The second reconstruction unit 300 is configured to perform a second feature reconstruction on the feature matrix to obtain a second feature importance score;

[0147] The feature screening unit 400 is configured 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;

[0148] The feature extraction unit 500 is configured 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;

[0149] The classification and evaluation unit 600 is configured to input the fused feature matrix into the first classification network for classification and evaluation to obtain a classification and evaluation result;

[0150] The sample evaluation unit 700 is configured to perform uncertainty evaluation on the sample features according to the classification and evaluation result and the class label of the sample to obtain an uncertainty evaluation result;

[0151] The image classification unit 800 is configured to screen the target feature set according to the uncertainty evaluation result and train the second classification network with the target feature set, so as to classify the infrared ship image through the trained second classification network.

[0152] In some embodiments, the first reconstruction unit 200 may specifically be configured to:

[0153] Represent the feature reconstruction coefficient matrix with a low-rank matrix, and combine the feature reconstruction coefficient matrix and the feature matrix to construct a first feature reconstruction objective function, which is used to perform the first feature reconstruction on the feature matrix;

[0154] Solve the feature reconstruction coefficient matrix in the first feature reconstruction objective function by using alternating iteration;

[0155] Calculate the first feature importance score based on the solved feature reconstruction coefficient matrix.

[0156] In some embodiments, the first reconstruction unit 200 may specifically be configured to:

[0157] ;

[0158] Wherein, represents the matrix variable in the feature reconstruction coefficient matrix and the matrix variable , represents the total number of features in the feature matrix, represents the th feature, represents the regularization hyperparameter, represents the data reconstruction error term, represents based on the sparse regularization term of the norm, represents the tunable hyperparameter, represents the orthogonality constraint condition, represents the transpose.

[0159] In some embodiments, the second reconstruction unit 300 may be specifically configured to:

[0160] Construct a second feature reconstruction objective function based on the sample weights, the relationship matrix, and the feature matrix, where the second feature reconstruction objective function is used to perform a second feature reconstruction on the feature matrix;

[0161] Solve for the sample weights and the relationship matrix in the second feature reconstruction objective function by alternating iteration;

[0162] Calculate a second feature importance score based on the solved sample weights.

[0163] In some embodiments, the second reconstruction unit 300 may be specifically configured to:

[0164] ;

[0165] Wherein, represents the total number of samples in the feature matrix, represents the th sample weight, represents the data label matrix, represents the th class label corresponding to the sample, represents the th sample, represents the relationship matrix, represents the regularization hyperparameter, represents the tunable hyperparameter, represents fitting the data to the class labels, represents the sparse regularization term based on the norm, represents the soft regularization term of self-paced learning, represents the self-paced learning step size parameter.

[0166] In some embodiments, the sample evaluation unit 700 may be specifically configured to:

[0167] Construct an uncertainty loss function using the Dirichlet distribution;

[0168] Perform uncertainty evaluation on the sample features using the uncertainty loss function according to the classification evaluation result and the class label of the sample to obtain an uncertainty evaluation result.

[0169] In some embodiments, the sample evaluation unit 700 may be specifically configured to:

[0170] ;

[0171] Wherein, represents the entropy of the Dirichlet distribution, represents a category probability vector, represents a normalization constant, represents the total number of categories, represents the predicted confidence of the category.

[0172] It should be noted that since an 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 content in the method embodiment also applies to the system embodiment of the present application and will not be elaborated here.

[0173] Referring to Figure 4 , an embodiment of the present application also provides an electronic device, which includes:

[0174] at least one memory;

[0175] at least one processor;

[0176] at least one program;

[0177] The program is 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.

[0178] The electronic device can be any intelligent terminal including a mobile phone, a tablet computer, a personal digital assistant (PDA), an in-vehicle computer, etc.

[0179] The electronic device in the embodiment of the present application will be introduced in detail below.

[0180] The processor 1600 can be implemented by using a general-purpose central processing unit (CPU), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, etc., and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present disclosure;

[0181] 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), etc. The memory 1700 can store an operating system and other application programs. When implementing the technical solutions provided in the embodiments of this specification through software or firmware, the relevant program codes are stored in the memory 1700 and are called by the processor 1600 to execute the infrared ship image classification method based on feature reconstruction according to the embodiments of the present disclosure.

[0182] The input / output interface 1800 is used to implement information input and output;

[0183] The communication interface 1900 is used to implement communication and interaction between this device and other devices. It can achieve communication through wired means (such as USB, network cable, etc.) or through wireless means (such as mobile network, WIFI, Bluetooth, etc.);

[0184] The bus 2000 transmits information between various components of the device (such as the processor 1600, the memory 1700, the input / output interface 1800, and the communication interface 1900);

[0185] Among them, the processor 1600, the memory 1700, the input / output interface 1800, and the communication interface 1900 are communicatively connected to each other inside the device through the bus 2000.

[0186] The embodiments of the present disclosure also provide 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 cause a computer to execute the above-mentioned infrared ship image classification method based on feature reconstruction.

[0187] As a non-transitory computer-readable storage medium, the memory can be used to store non-transitory software programs and non-transitory computer-executable programs. In addition, the memory can include high-speed random access memory, and can also include non-transitory memory, such as at least one magnetic disk storage device, a flash memory device, or other non-transitory solid-state storage devices. In some embodiments, the memory optionally includes a memory remotely set relative to the processor, and these remote memories can be connected to the processor through a network. Examples of the above-mentioned network include but are not limited to the Internet, an enterprise intranet, a local area network, a mobile communication network, and combinations thereof.

[0188] The embodiments described in the embodiments of the present disclosure are 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 know 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 equally applicable to similar technical problems.

[0189] Those skilled in the art can understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the present disclosure, and may include more or fewer steps than those shown in the figures, or combine certain steps, or different steps.

[0190] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, that is, they may be located in one place, or may be distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0191] Those of ordinary skill in the art can understand that all or some of the steps in the methods disclosed above, and the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, and their appropriate combinations.

[0192] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of this application and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily need to be used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments of the present application described here can be implemented in an order other than those illustrated or described here. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or are inherent to these processes, methods, products, or devices.

[0193] It should be understood that in this application, "at least one (item)" means one or more, and "a plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that there can be three relationships. For example, "A and / or B" can mean: only A exists, only B exists, and both A and B exist at the same time. Among them, A and B can be singular or plural. The character " / " generally represents an "or" relationship between the associated objects before and after. "At least one (item) of the following" or its similar expressions refer to any combination of these items, including any combination of single item (item) or plural items (items). For example, at least one (item) 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.

[0194] In several embodiments provided in this 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 illustrative. For example, the division of units is only a logical function division. In actual implementation, there can be other division methods. For example, 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 displayed or discussed coupling or direct coupling or communication connection between each other can be through some interfaces. The indirect coupling or communication connection of devices or units can be in electrical, mechanical or other forms.

[0195] The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they can be located in one place, or they can be distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0196] In addition, each functional unit in each embodiment of this application can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above integrated units can be implemented in the form of hardware or in the form of software functional units.

[0197] When an 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 this application, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions for causing an electronic device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods of various embodiments of this application. The aforementioned storage medium includes: various media that can store programs, such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs. The embodiments of this application have been described in detail above in conjunction with the accompanying drawings, but this application is not limited to the above embodiments. Within the scope of knowledge possessed by those of ordinary skill in the art, various changes can also be made without departing from the gist of this application.

[0198] The embodiments of this application have been described in detail above in conjunction with the accompanying drawings, but this application is not limited to the above embodiments. Within the scope of knowledge possessed by those of ordinary skill in the art, various changes can also be made without departing from the gist of this application.

Claims

1. An infrared ship image classification method based on feature reconstruction, characterized in that The method includes: Collecting the feature data of infrared ship images, and preprocessing the feature data to obtain a feature matrix, where one row of the feature matrix represents a sample and one column represents a kind of feature; Performing a first feature reconstruction on the feature matrix to obtain a first feature importance score, specifically including: Using a low-rank matrix to represent the feature reconstruction coefficient matrix, and combining the feature reconstruction coefficient matrix and the feature matrix to construct a first feature reconstruction objective function, which is used to perform the first feature reconstruction on the feature matrix; Using alternating iteration to solve the feature reconstruction coefficient matrix in the first feature reconstruction objective function; Calculating a first feature importance score based on the solved feature reconstruction coefficient matrix; Performing a second feature reconstruction on the feature matrix to obtain a second feature importance score, specifically including: Based on the sample weights, the relationship matrix, and the feature matrix, constructing a second feature reconstruction objective function, which is used to perform the second feature reconstruction on the feature matrix; Using alternating iteration to solve the sample weights and the relationship matrix in the second feature reconstruction objective function; Calculating a second feature importance score based on the solved sample weights; Based on the first feature importance score and the second feature importance score, screening the features in the feature matrix to obtain a new feature matrix; Extracting the visual features in 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 class label of the sample, performing uncertainty evaluation on the sample features to obtain an uncertainty evaluation result; Screening a target feature set according to the uncertainty evaluation result, and training a second classification network using the target feature set, so as to classify the infrared ship image through the trained second classification network.

2. The infrared ship image classification method based on feature reconstruction according to claim 1, wherein, The combining the feature reconstruction coefficient matrix and the feature matrix to construct a first feature reconstruction objective function includes: ； Among them, represents the matrix variable in the feature reconstruction coefficient matrix and the matrix variable , represents the total number of features in the feature matrix, represents the th feature, represents the regularization hyperparameter, represents the data reconstruction error term, represents based on the sparse regularization term of the norm, represents the adjustable hyperparameter, represents the orthogonality constraint condition, represents the transpose.

3. The infrared ship image classification method based on feature reconstruction according to claim 1, 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.

4. The infrared ship image classification method based on feature reconstruction according to claim 1, characterized in that The performing 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 includes: Constructing an uncertainty loss function using the Dirichlet distribution; According to the classification evaluation result and the class label of the sample, performing uncertainty evaluation on the sample features using the uncertainty loss function to obtain an uncertainty evaluation result.

5. The infrared ship image classification method based on feature reconstruction according to claim 4, characterized in that The constructing an uncertainty loss function using the 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 prediction confidence of the category, It means partial derivative.

6. An infrared ship image classification system based on feature reconstruction, characterized in that, The system includes: A data processing unit, configured to collect the feature data of infrared ship images, and preprocess the feature data to obtain a feature matrix, where one row of the feature matrix represents a sample and one column represents a kind of feature; A first reconstruction unit, configured to perform a first feature reconstruction on the feature matrix to obtain a first feature importance score, specifically including: The feature reconstruction coefficient matrix is represented by a low-rank matrix, and the 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 the first feature reconstruction on the feature matrix; The feature reconstruction coefficient matrix in the first feature reconstruction objective function is solved by using alternating iteration; Based on the solved feature reconstruction coefficient matrix, the first feature importance score is calculated; A second reconstruction unit is configured to perform a second feature reconstruction on the feature matrix to obtain a second feature importance score, specifically including: Based on the sample weights, the relationship matrix, and the feature matrix, a second feature reconstruction objective function is constructed. The second feature reconstruction objective function is used to perform the second feature reconstruction on the feature matrix; The sample weights and the relationship matrix in the second feature reconstruction objective function are solved by using alternating iteration; Based on the solved sample weights, the second feature importance score is calculated; A feature screening unit is configured 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; A feature extraction unit is configured to extract the visual features in the infrared ship image and fuse the visual features with the new feature matrix to obtain a fused feature matrix; A classification evaluation unit is configured to input the fused feature matrix into a first classification network for classification evaluation to obtain a classification evaluation result; A sample evaluation unit is configured 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; An image classification unit is configured to screen a target feature set according to the uncertainty evaluation result and train a second classification network by using the target feature set, so as to classify the infrared ship image by using the trained second classification network.

7. An electronic device, characterized in that, It includes at least one control processor and a memory for communicatively connecting with the at least one control processor; the memory stores instructions executable by the at least one control processor. 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 according to any one of claims 1 to 5.

8. 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 cause a computer to execute the infrared ship image classification method based on feature reconstruction according to any one of claims 1 to 5.

Images