Remote sensing image multi-class ground feature recognition method based on self-supervised learning

Abstract

The invention discloses a remote sensing image multi-class ground feature recognition method based on self-supervised learning. The method comprises the following steps: S1, generating standardized remote sensing image data; s2, performing feature pre-training on the standardized remote sensing image data based on self-supervised learning, and outputting depth feature representation; s3, inputting the depth feature representation into a gradient enhancement decision tree classification module, and outputting a preliminary classification result; s4, updating a gradient enhancement decision tree classification module; s5, converting the depth feature representation into low-dimensional feature representation through a feature distillation module, and outputting an optimization classification result; and S6, implementing space consistency optimization, classification boundary refinement and category redistribution on the optimized classification result, and generating a final remote sensing image multi-category ground feature recognition result. According to the method, the classification precision and the category distinguishing capability of complex ground feature categories are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of remote sensing images, and in particular to a remote sensing image multi-category object recognition method based on self-supervised learning. Background Art

[0002] With the rapid development of remote sensing technology and artificial intelligence, the automated processing of remote sensing images plays an increasingly important role in land and resources management, ecological environment monitoring, agricultural surveying and mapping, and urban planning. Among them, multi-category object recognition is one of the important tasks of remote sensing image analysis, which aims to automatically extract different types of object information from high-resolution, multispectral or hyperspectral images to achieve accurate classification of objects.

[0003] At present, multi-category object recognition in remote sensing images mainly relies on traditional supervised learning methods, that is, training deep learning models or machine learning classifiers by manually annotating a large amount of remote sensing image data. The existing methods have improved the automation level of object classification to a certain extent, but the following main problems still exist: First, the high reliance on labeled data limits the adaptability of the model. Remote sensing images usually cover vast geographical areas and a variety of land object types. There are significant differences in the spectral characteristics of land objects in different regions. Therefore, in order to train a classification model with generalization capabilities, large-scale and diverse high-quality labeled data is required. However, the cost of manual labeling of remote sensing image data is extremely high. The precise labeling of multi-category complex land objects requires the participation of domain experts, which leads to high data acquisition costs and makes it difficult to cover all possible land object categories and environmental changes. In addition, manually labeled data is easily affected by subjective factors and contains certain noise, which further reduces the reliability of the classification model.

[0004] Secondly, existing classification algorithms find it difficult to efficiently model the spatial and spectral characteristics of complex remote sensing data. Remote sensing images have multi-dimensional characteristics, including spatial distribution, spectral information, and texture features, while traditional classification methods, such as random forests and support vector machines, often find it difficult to fully model the nonlinear relationships between these complex features. Although deep learning methods have been widely used in remote sensing image analysis in recent years, existing methods usually require huge amounts of training data and are prone to overfitting in small sample environments, resulting in poor adaptability of the model to different data sets.

[0005] In addition, current classification methods still face great challenges in dealing with multi-category imbalance problems. The distribution of object categories in remote sensing images is often extremely uneven. For example, buildings and roads in urban areas may account for a large proportion, while the amount of data for vegetation and wetland categories is relatively small. Traditional classification methods usually tend to favor high-frequency categories, resulting in low recognition accuracy for frequency categories below the threshold. Although existing methods such as category resampling and cost-sensitive learning can partially alleviate this problem, there is still a problem of reduced classification performance when the data distribution is highly skewed. The classifier has a high misjudgment rate when the boundaries of object categories are blurred and the spectral characteristics between categories are similar.

[0006] Therefore, there is an urgent need for a new method to break the reliance of traditional supervised learning on labeled data, while combining an efficient classification mechanism to improve the accuracy and adaptability of multi-category object recognition in remote sensing images. Summary of the invention

[0007] One purpose of the present invention is to propose a multi-category object recognition method for remote sensing images based on self-supervised learning, which significantly improves the classification accuracy and category distinction ability of complex object categories.

[0008] A method for multi-category object recognition in remote sensing images based on self-supervised learning according to an embodiment of the present invention comprises the following steps: S1. Obtain remote sensing image data and preprocess the remote sensing image data, including radiation correction, geometric correction, noise removal, cloud occlusion area filling and image enhancement processing, to generate standardized remote sensing image data; S2. performing feature pre-training on the standardized remote sensing image data based on self-supervised learning, wherein the self-supervised learning module automatically extracts deep feature representations with discriminative capabilities from the standardized remote sensing image data by constructing multi-view comparison tasks, spatial consistency tasks, and spectral domain transformation tasks, and outputs the deep feature representations; S3. inputting the deep feature representation into a gradient enhanced decision tree classification module, the gradient enhanced decision tree classification module is composed of a plurality of weak classifiers, performing preliminary multi-category classification of the deep feature representation through iterative optimization, and outputting preliminary classification results; S4. Calculate the classification error based on the preliminary classification results and dynamically adjust the weights of the weak classifiers based on the gradient feedback mechanism, optimize the decision boundaries of each weak classifier, and update the gradient enhanced decision tree classification module; S5. The deep feature representation is converted into a low-dimensional feature representation through the feature distillation module, and is fused with the intermediate classification features after weight adjustment of the gradient boosting decision tree classification module to generate a fused optimized feature representation, which is further input into the gradient boosting decision tree classification module for fine classification training, and outputs the optimized classification result; S6. Implement spatial consistency optimization, classification boundary refinement, and category reallocation on the optimized classification results to generate a final remote sensing image multi-category object recognition result.

[0009] Optionally, the S1 includes the following steps: S11. Collect remote sensing image data, the resolution of the remote sensing image data is R, the number of bands is B, and the image size is ; S12. performing radiation correction, geometric correction and noise removal on the remote sensing image data; S13. Fill the cloud-occluded area of ​​the remote sensing image data after radiation correction, geometric correction and noise removal. Based on the principle of local similarity, the pixel values ​​of the cloud-occluded area are estimated using the surrounding pixel information, and the nearest neighbor unoccluded pixel set is calculated and filled using the weighted interpolation method; S14. Adopting adaptive histogram equalization method to optimize the contrast of remote sensing image data, and outputting standardized remote sensing image data D: ; in, is the Nth standardized remote sensing image, and N is the total number of remote sensing image data.

[0010] Optionally, S2 includes the following steps: S21. Use self-supervised learning method to pre-train the features of standardized remote sensing image data D and construct a cross-scale spatial-spectral fusion self-supervised feature extraction model F: ; in, To standardize remote sensing images After feature extraction, the feature representation is , d is the feature dimension; S22. Construct a cross-scale contrast learning mechanism, perform multi-scale decomposition on standardized remote sensing image data, define the scale transformation set T, and construct multi-scale enhanced samples and , calculate their eigenvectors respectively ; Define scale contrast loss: ; in, is the cosine similarity of the feature vector, is the temperature parameter, j represents the jth scale-transformed sample in the scale-contrast learning mechanism; S23. Design object-background segmentation constraints and construct object masks , define the ground feature pixel set and background pixel set , and calculate the mean of ground objects and background features:

[0011] ; Among them, M is the number of ground object pixels, K is the number of background pixels, and Represents ground feature pixels or background pixels Representation vector in deep feature space; Define the object discrimination loss to make the objects distinguishable from the background features: ; S24. Using the spectral-spatial fusion enhancement mechanism, construct spectral perturbation samples: ; in, is Gaussian noise, is the disturbance coefficient, Calculate the spectral invariance loss for the remote sensing image after spectral perturbation: ; in, is the remote sensing image after spectral perturbation Through the cross-scale spatial-spectral fusion self-supervised feature extraction model, For remote sensing images Features processed by the self-supervised feature extraction model through cross-scale spatial-spectral fusion; And integrate spatial information to define local consistency constraints: ; in, To standardize remote sensing images The mth pixel point, Pixel The surrounding neighborhood window contains pixels close to its spatial position. Represents the pixel point in the remote sensing image Feature representation after processing by the cross-scale spatial-spectral fusion self-supervised feature extraction model, Represents standardized remote sensing images A pixel in is the number of pixels in the neighborhood window; S25. Calculate the final improved self-supervised learning loss, combining cross-scale contrastive learning, object-background segmentation constraints, and spectral-spatial fusion mechanism: ; in, is the weight parameter; And output the set of deep feature representations extracted by self-supervised learning: ; in, is the deep feature representation of the Nth remote sensing image.

[0012] Optionally, S3 includes the following steps: S31. Use the gradient enhanced decision tree classification module to perform preliminary classification on the deep feature representation, build the gradient enhanced decision tree classification module G, and define the mapping: ; in, To predict the class label, is the depth feature representation of the i-th remote sensing image, and its value range is , C is the total number of land feature categories; S32. Use the decision tree model based on gradient boosting for classification training and construct the loss function: ; in, is the true category label, is the true probability of the cth class, The class probabilities predicted by the gradient boosted decision tree classification module; S33. Using the weighted gradient update strategy, calculate the negative gradient of the tth iteration: ; Based on negative gradient Construct the tth decision tree , update the classification model: ; in, is the learning rate, which controls the contribution of each decision tree; S34. Use a deep feature importance calculation mechanism to evaluate the contribution of different feature dimensions and define the feature importance score: ; in, is the jth feature, T is the total number of decision trees, is the number of nodes in the t-th decision tree, The loss reduction caused by node splitting is calculated, the feature importance scores are sorted, and the most important feature dimensions are selected; S35. Calculate the balance coefficient of category distribution using the category balance adjustment strategy: ; in, is the number of samples in the cth class, based on the equilibrium coefficient Adjust the loss function: ; S36. Iteratively optimize the gradient boosting decision tree classification module until the loss function converges or reaches the preset number of iterations , output preliminary multi-category classification results: ; in, is the preliminary classification label set for all remote sensing images, is the predicted category label of the Nth remote sensing image.

[0013] Optionally, S4 includes the following steps: S41. By the loss function Find the gradient of the preliminary classification results and obtain the gradient feedback value: ; S42. For multiple weak classifiers of the gradient enhanced decision tree classification module, let the weight of the tth weak classifier be ,in, , T is the total number of weak classifiers, according to the gradient feedback value of each sample i Calculate the weight adjustment of the weak classifier and define the adjustment as: ; in, represents the weight adjustment of the t-th weak classifier under the guidance of the gradient for the i-th sample; S43. Implement dynamic weight update for each weak classifier: ; in, is the updated weight of the tth weak classifier, is the learning rate parameter, which controls the update step size; S44. Using updated weights Re-optimize the decision boundaries of each weak classifier and combine it with deep feature representation Fine-tune or retrain the gradient boosted decision tree classification module as a whole to construct an updated classification module G'; S45. Output the classification result obtained after the updated gradient enhanced decision tree classification module G' predicts the deep feature representation Z: ; in, is the predicted category probability vector of the Nth remote sensing image after dynamic adjustment.

[0014] Optionally, S5 includes the following steps: S51. The feature distillation module is used to reduce the dimension of the deep feature representation. The feature distillation module optimizes the redundant features while maintaining the original feature space information by constructing a lightweight representation learning network, and generates a low-dimensional feature representation after dimensionality reduction; S52. Use information entropy optimization strategy to screen the low-dimensional feature representation after dimensionality reduction, calculate the information contribution of each feature dimension based on the category discrimination ability, retain the most discriminative features and eliminate low-relevance features; S53. A feature fusion strategy is used to fuse the low-dimensional feature representation output by the feature distillation module with the intermediate classification features after weight adjustment by the optimized gradient boosting decision tree classification module. During the fusion process, dynamic weighting parameters are set for different feature sources so that the deep features extracted by self-supervised learning and the high-discriminative features generated by the gradient boosting decision tree classification module work together to form a fused optimized feature representation. S54. The gradient boosting decision tree classification module is retrained using fusion feature representation. A hierarchical learning strategy is introduced in the training process. The easy-to-classify samples are trained first, and then gradually transitioned to the boundary fuzzy and easily confused categories. S55. Use a fine classification training strategy to perform local iterative optimization on the fusion feature representation, dynamically adjust the category discrimination boundary during the training process, increase the decision tree depth for the confusion categories above the threshold, and use a sample balancing strategy for the frequency categories below the threshold; S56. Outputting an optimized multi-category ground object recognition and classification result, wherein the optimized multi-category ground object recognition and classification result is generated based on the fusion feature representation and undergoes multi-level optimization and adjustment.

[0015] The beneficial effects of the present invention are: (1) The present invention adopts a self-supervised learning strategy of cross-scale spatial-spectral fusion, combined with multi-view contrast learning, spatial consistency learning and spectral transformation enhancement, to achieve high-quality feature extraction of multi-category objects in remote sensing images without the need for a large amount of manually labeled data. By introducing object-background segmentation constraints and spectral invariance regularization, robust characterization features can be extracted under different object categories, lighting conditions and spatiotemporal scales. When processing hyperspectral and multispectral data, the intrinsic correlation of spectral information can be fully utilized, so that object classification can still maintain a high resolution when facing spectral mixing effects.

[0016] (2) The present invention combines feature distillation with a gradient boosted decision tree classification module in the remote sensing image classification process. The feature distillation module is used to reduce the dimensionality of high-dimensional deep features, remove redundant information, extract the most discriminative low-dimensional feature representation, and fuse it with the intermediate features of the gradient boosted decision tree classifier, thereby improving the classification accuracy and reducing the computational complexity. By constructing a lightweight feature distillation network, the feature input dimension is reduced by 35%-50%, but the information retention rate can still reach more than 95%. In addition, the feature fusion mechanism enables GBDT to utilize the discriminative features extracted by deep learning while maintaining adaptability to small sample data.

[0017] (3) The present invention introduces a multi-stage dynamic weight adjustment strategy based on gradient feedback in the training process of the gradient enhanced decision tree classification module, performs gradient analysis on the classification errors of different categories, and adjusts the decision weights of the weak classifiers in each round of iteration. By constructing an adaptive category balance loss function, the classification accuracy of frequency categories below the threshold is improved. In addition, by utilizing the local feature reweighting mechanism, the classifier can dynamically adjust the decision boundary for the confusing categories above the threshold, thereby improving the classification model's ability to distinguish in complex terrain environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings: Figure 1 The present invention provides a flowchart of a method for multi-category object recognition in remote sensing images based on self-supervised learning. DETAILED DESCRIPTION

[0019] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0020] refer to Figure 1 , a multi-category object recognition method for remote sensing images based on self-supervised learning, comprising the following steps: S1. Obtain remote sensing image data and preprocess the remote sensing image data, including radiation correction, geometric correction, noise removal, cloud occlusion area filling and image enhancement processing, to generate standardized remote sensing image data; S2. Based on self-supervised learning, the standardized remote sensing image data is pre-trained. The self-supervised learning module automatically extracts deep feature representations with discriminative capabilities from the standardized remote sensing image data by constructing multi-view comparison tasks, spatial consistency tasks, and spectral domain transformation tasks, and outputs deep feature representations; S3. Input the deep feature representation into the gradient boosting decision tree classification module, which is composed of multiple weak classifiers, performs preliminary multi-category classification of the deep feature representation through iterative optimization, and outputs preliminary classification results; S4. Calculate the classification error based on the preliminary classification results and dynamically adjust the weights of the weak classifiers based on the gradient feedback mechanism, optimize the decision boundaries of each weak classifier, and update the gradient enhanced decision tree classification module; S5. The deep feature representation is converted into a low-dimensional feature representation through the feature distillation module, and is fused with the intermediate classification features after weight adjustment of the gradient boosting decision tree classification module to generate a fused optimized feature representation, which is further input into the gradient boosting decision tree classification module for fine classification training, and outputs the optimized classification result; S6. Implement spatial consistency optimization, classification boundary refinement, and category redistribution on the optimized classification results to generate the final multi-category land object recognition results of remote sensing images.

[0021] In this implementation, S1 includes the following steps: S11. Collect remote sensing image data. The resolution of the remote sensing image data is R, the number of bands is B, and the image size is ; S12. Perform radiation correction, geometric correction and noise removal on remote sensing image data; S13. Fill the cloud-occluded area of ​​the remote sensing image data after radiation correction, geometric correction and noise removal. Based on the principle of local similarity, the pixel values ​​of the cloud-occluded area are estimated using the surrounding pixel information, and the nearest neighbor unoccluded pixel set is calculated and filled using the weighted interpolation method; S14. Adopting adaptive histogram equalization method to optimize the contrast of remote sensing image data, and outputting standardized remote sensing image data D: ; in, is the Nth standardized remote sensing image, and N is the total number of remote sensing image data.

[0022] In this implementation, S2 includes the following steps: S21. Use self-supervised learning method to pre-train the features of standardized remote sensing image data D and construct a cross-scale spatial-spectral fusion self-supervised feature extraction model F: ; in, To standardize remote sensing images After feature extraction, the feature representation is , d is the feature dimension; S22. Construct a cross-scale contrast learning mechanism, perform multi-scale decomposition on standardized remote sensing image data, define the scale transformation set T, and construct multi-scale enhanced samples and , calculate their eigenvectors respectively ; Define scale contrast loss: ; in, is the cosine similarity of the feature vector, is the temperature parameter, j represents the jth scale-transformed sample in the scale-contrast learning mechanism; S23. Design object-background segmentation constraints and construct object masks , define the ground feature pixel set and background pixel set , and calculate the mean of ground objects and background features:

[0023] ; Among them, M is the number of ground object pixels, K is the number of background pixels, and Represents ground feature pixels or background pixels Representation vector in deep feature space; Define the object discrimination loss to make the objects distinguishable from the background features: ; S24. Using the spectral-spatial fusion enhancement mechanism, construct spectral perturbation samples: ; in, is Gaussian noise, is the disturbance coefficient, Calculate the spectral invariance loss for the remote sensing image after spectral perturbation: ; in, is the remote sensing image after spectral perturbation Through the cross-scale spatial-spectral fusion self-supervised feature extraction model, For remote sensing images Features processed by the self-supervised feature extraction model through cross-scale spatial-spectral fusion; And integrate spatial information to define local consistency constraints: ; in, To standardize remote sensing images The mth pixel point, Pixel The surrounding neighborhood window contains pixels close to its spatial position. Represents the pixel point in the remote sensing image Feature representation after processing by the cross-scale spatial-spectral fusion self-supervised feature extraction model, Represents standardized remote sensing images A pixel in is the number of pixels in the neighborhood window; S25. Calculate the final improved self-supervised learning loss, combining cross-scale contrastive learning, object-background segmentation constraints, and spectral-spatial fusion mechanism: ; in, is the weight parameter; And output the set of deep feature representations extracted by self-supervised learning: ; in, is the deep feature representation of the Nth remote sensing image.

[0024] In this implementation, S3 includes the following steps: S31. Use the gradient enhanced decision tree classification module to perform preliminary classification on the deep feature representation, build the gradient enhanced decision tree classification module G, and define the mapping: ; in, To predict the class label, is the depth feature representation of the i-th remote sensing image, and its value range is , C is the total number of land feature categories; S32. Use the decision tree model based on gradient boosting for classification training and construct the loss function: ; in, is the true category label, is the true probability of the cth class, The class probabilities predicted by the gradient boosted decision tree classification module; S33. Using the weighted gradient update strategy, calculate the negative gradient of the tth iteration: ; Based on negative gradient Construct the tth decision tree , update the classification model: ; in, is the learning rate, which controls the contribution of each decision tree; S34. Use a deep feature importance calculation mechanism to evaluate the contribution of different feature dimensions and define the feature importance score: ; in, is the jth feature, T is the total number of decision trees, is the number of nodes in the t-th decision tree, The loss reduction caused by node splitting is calculated, the feature importance scores are sorted, and the most important feature dimensions are selected; S35. Calculate the balance coefficient of category distribution using the category balance adjustment strategy: ; in, is the number of samples in the cth class, based on the equilibrium coefficient Adjust the loss function: ; S36. Iteratively optimize the gradient boosting decision tree classification module until the loss function converges or reaches the preset number of iterations , output preliminary multi-category classification results: ; in, is the preliminary classification label set for all remote sensing images, is the predicted category label of the Nth remote sensing image.

[0025] In this implementation, S4 includes the following steps: S41. By the loss function Find the gradient of the preliminary classification results and obtain the gradient feedback value: ; S42. For multiple weak classifiers of the gradient enhanced decision tree classification module, let the weight of the tth weak classifier be ,in, , T is the total number of weak classifiers, according to the gradient feedback value of each sample i Calculate the weight adjustment of the weak classifier and define the adjustment as: ; in, represents the weight adjustment of the t-th weak classifier under the guidance of the gradient for the i-th sample; S43. Implement dynamic weight update for each weak classifier: ; in, is the updated weight of the tth weak classifier, is the learning rate parameter, which controls the update step size; S44. Using updated weights Re-optimize the decision boundaries of each weak classifier and combine it with deep feature representation Fine-tune or retrain the gradient boosted decision tree classification module as a whole to construct an updated classification module G'; S45. Output the classification result obtained after the updated gradient enhanced decision tree classification module G' predicts the deep feature representation Z: ; in, is the predicted category probability vector of the Nth remote sensing image after dynamic adjustment.

[0026] In this implementation, S5 includes the following steps: S51. The feature distillation module is used to reduce the dimension of the deep feature representation. The feature distillation module optimizes the redundant features while maintaining the original feature space information by constructing a lightweight representation learning network, and generates a low-dimensional feature representation after dimensionality reduction; S52. Use information entropy optimization strategy to screen the low-dimensional feature representation after dimensionality reduction, calculate the information contribution of each feature dimension based on the category discrimination ability, retain the most discriminative features and eliminate low-relevance features; S53. A feature fusion strategy is used to fuse the low-dimensional feature representation output by the feature distillation module with the intermediate classification features after weight adjustment by the optimized gradient boosting decision tree classification module. During the fusion process, dynamic weighting parameters are set for different feature sources so that the deep features extracted by self-supervised learning and the high-discriminative features generated by the gradient boosting decision tree classification module work together to form a fused optimized feature representation. S54. The gradient boosting decision tree classification module is retrained using fusion feature representation. A hierarchical learning strategy is introduced in the training process. The easy-to-classify samples are trained first, and then gradually transitioned to the boundary fuzzy and easily confused categories. S55. Use a fine classification training strategy to perform local iterative optimization on the fusion feature representation, dynamically adjust the category discrimination boundary during the training process, increase the decision tree depth for the confusion categories above the threshold, and use a sample balancing strategy for the frequency categories below the threshold; S56. Outputting optimized multi-category land object recognition and classification results, the optimized multi-category land object recognition and classification results are generated based on the fusion feature representation and are optimized and adjusted at multiple levels.

[0027] Example 1: In order to verify the effectiveness of the multi-category land object recognition method of remote sensing images based on gradient enhancement decision tree and self-supervised learning in complex land object classification tasks, this example selected multispectral remote sensing image data of City A and its surrounding areas, and conducted land object classification experiments in combination with the geographic information system. The area contains multi-category land objects such as urban buildings, roads, farmlands, forests, water bodies, and wetlands, which have complex spectral characteristics and highly mixed spatial distribution, providing a real experimental environment for remote sensing image classification tasks.

[0028] Traditional remote sensing object classification methods usually rely on a large amount of manually annotated data, using random forests, support vector machines or deep neural networks for classification. However, due to high annotation costs, data imbalance, and insufficient classification accuracy in complex environments, traditional methods have great limitations when dealing with multi-category, cross-scale, and hyperspectral object classification tasks. The present invention combines self-supervised learning with gradient-enhanced decision trees to automatically learn the deep features of remote sensing images without the need for large-scale manually annotated data, and combines decision tree models for efficient classification.

[0029] This experiment uses Sentinel-2 multispectral remote sensing images of City A and its surrounding areas from January to December 2023, with a spatial resolution of 10m, 12 spectral bands, and an image coverage area of ​​about 5,000 square kilometers. The experimental data includes 1,200 high-resolution images, 800 of which are used for training and 400 for testing. The object distribution data provided by the geographic database is used to annotate the object categories in order to evaluate the classification accuracy.

[0030] The specific types of features and sample numbers are shown in Table 1 below: Table 1 Statistics of training and testing samples of ground object categories Feature Type Number of training samples Number of test samples Proportion (%) Urban Architecture 50,000 20,000 18.3 the way 40,000 15,000 14.6 farmland 70,000 25,000 21.1 forest 65,000 22,000 19.8 Water 35,000 12,000 10.5 Wetlands 25,000 10,000 7.9 other 20,000 8,000 7.8 During the experiment, we used traditional methods (CNN, random forest, SVM) and the method of the present invention (self-supervised learning + gradient boosting decision tree) to perform multi-category land object recognition on remote sensing images, and made a comprehensive comparison of classification accuracy, computational efficiency, and model stability.

[0031] Firstly, the remote sensing images are subjected to radiation correction, geometric correction, noise removal, cloud-occluded area filling and image enhancement to generate standardized remote sensing image data. The traditional method directly uses the data for training, while the method of the present invention further adopts self-supervised learning for feature pre-training to mine the spectral and spatial characteristics of the data.

[0032] The present invention adopts cross-scale spatial-spectral fusion self-supervised learning to perform multi-perspective comparative learning, object-background segmentation constraints, and spectral enhancement tasks on the training data, automatically learns the deep features of different categories of objects, and constructs an efficient feature extraction network through unsupervised feature learning, which reduces the dependence on manually labeled data and improves the separability and generalization ability of classification features.

[0033] In the classification stage, traditional methods directly train CNN, random forest and SVM based on labeled data, while the method of the present invention uses gradient enhanced decision trees for classification and combines feature distillation to reduce the dimensionality of high-dimensional features, reduce computational overhead, and improve the decision-making ability of the model. During the training process, we use multi-stage gradient feedback optimization to adjust the decision boundary to improve the ability to distinguish confused categories.

[0034] During the testing phase, we used 400 test images to compare the performance of the proposed method with traditional methods in terms of classification accuracy, computational efficiency, and category recognition ability.

[0035] The experimental results shown in Table 2 below show that the method of the present invention is significantly superior to the traditional method in terms of classification accuracy, and has better discrimination ability in categories with high spectral similarity (forest and wetland, road and building).

[0036] Table 2 Comparison of classification accuracy between the method of the present invention and the traditional method method Overall classification accuracy (OA, %) Kappa coefficient Urban buildings (%) the way(%) farmland(%) forest(%) Water body (%) Wetlands (%) Random Forest 79.6 0.75 85.2 78.1 80.4 76.3 82.5 68.7 SVM 82.3 0.78 86.7 80.5 82.8 78.9 84.2 71.2 CNN 85.4 0.81 89.1 84.2 86.7 82.1 87.5 74.5 Method of the present invention 92.6 0.89 94.3 90.8 93.1 89.7 92.4 85.2 The method of the present invention is superior to traditional methods in both training time and inference time, and can achieve efficient processing on large-scale data sets. The method of the present invention reduces the training time by 53.5% and increases the inference speed by 59.8%, and has a significant advantage in computing efficiency.

[0037] Traditional methods have large errors in the classification of frequency categories below the threshold. The method of the present invention improves the recognition ability of frequency categories below the threshold through feature distillation and gradient optimization, which increases the wetland recognition accuracy by 10.7% and significantly improves the classification accuracy of farmland and forest categories.

[0038] This embodiment shows that the method of the present invention can effectively solve the problems of traditional remote sensing object classification that is highly dependent on labeled data, has low computational efficiency, and insufficient category differentiation capability. It is superior to traditional methods in terms of classification accuracy, computational speed, and generalization capability, and can be widely used in the fields of automatic classification of remote sensing data, land use monitoring, and environmental assessment, providing an efficient and stable solution for intelligent analysis of remote sensing images.

[0039] The present invention adopts a self-supervised learning strategy of cross-scale spatial-spectral fusion, combined with multi-view contrast learning, spatial consistency learning and spectral transformation enhancement, to achieve high-quality feature extraction of multi-category objects in remote sensing images without the need for a large amount of manually labeled data. By introducing object-background segmentation constraints and spectral invariance regularization, robust characterization features can be extracted under different object categories, lighting conditions and spatiotemporal scales. When processing hyperspectral and multispectral data, the inherent correlation of spectral information can be fully utilized, so that object classification can still maintain a high resolution when facing spectral mixing effects.

[0040] The present invention combines feature distillation with a gradient boosted decision tree classification module in the remote sensing image classification process. The feature distillation module is used to reduce the dimensionality of high-dimensional deep features, remove redundant information, extract the most discriminative low-dimensional feature representation, and fuse it with the intermediate features of the gradient boosted decision tree classifier, thereby improving the classification accuracy and reducing the computational complexity. By constructing a lightweight feature distillation network, the feature input dimension is reduced by 35%-50%, but the information retention rate can still reach more than 95%. In addition, the feature fusion mechanism enables GBDT to utilize the discriminative features extracted by deep learning while maintaining adaptability to small sample data.

[0041] The present invention introduces a multi-stage dynamic weight adjustment strategy based on gradient feedback in the training process of the gradient enhanced decision tree classification module, performs gradient analysis on the classification errors of different categories, and adjusts the decision weights of weak classifiers in each round of iteration. By constructing an adaptive category balance loss function, the classification accuracy of frequency categories below the threshold is improved, and by utilizing the local feature reweighting mechanism, the classifier can dynamically adjust the decision boundary for the confusion categories above the threshold, thereby improving the discrimination ability of the classification model in complex terrain environments.

[0042] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A multi-category object recognition method for remote sensing images based on self-supervised learning, characterized in that: The steps include: S1. Obtain remote sensing image data and preprocess the remote sensing image data, including radiation correction, geometric correction, noise removal, cloud occlusion area filling and image enhancement processing, to generate standardized remote sensing image data; S2. performing feature pre-training on the standardized remote sensing image data based on self-supervised learning, wherein the self-supervised learning module automatically extracts deep feature representations with discriminative capabilities from the standardized remote sensing image data by constructing multi-view comparison tasks, spatial consistency tasks, and spectral domain transformation tasks, and outputs the deep feature representations; S3. inputting the deep feature representation into a gradient enhanced decision tree classification module, the gradient enhanced decision tree classification module is composed of a plurality of weak classifiers, performing preliminary multi-category classification of the deep feature representation through iterative optimization, and outputting preliminary classification results; S4. Calculate the classification error based on the preliminary classification results and dynamically adjust the weights of the weak classifiers based on the gradient feedback mechanism, optimize the decision boundaries of each weak classifier, and update the gradient enhanced decision tree classification module; S5. The deep feature representation is converted into a low-dimensional feature representation through the feature distillation module, and is fused with the intermediate classification features after weight adjustment of the gradient boosting decision tree classification module to generate a fused optimized feature representation, which is further input into the gradient boosting decision tree classification module for fine classification training, and outputs the optimized classification result; S6. Implement spatial consistency optimization, classification boundary refinement, and category reallocation on the optimized classification results to generate a final remote sensing image multi-category object recognition result.

2. The method for multi-category object recognition in remote sensing images based on self-supervised learning according to claim 1, characterized in that: The S1 comprises the following steps: S11. Collect remote sensing image data, the resolution of the remote sensing image data is R, the number of bands is B, and the image size is ; S12. performing radiation correction, geometric correction and noise removal on the remote sensing image data; S13. Fill the cloud-occluded area of ​​the remote sensing image data after radiation correction, geometric correction and noise removal. Based on the principle of local similarity, the pixel values ​​of the cloud-occluded area are estimated using the surrounding pixel information, and the nearest neighbor unoccluded pixel set is calculated and filled using the weighted interpolation method; S14. Adopting adaptive histogram equalization method to optimize the contrast of remote sensing image data, and outputting standardized remote sensing image data D: ； in, is the Nth standardized remote sensing image, and N is the total number of remote sensing image data.

3. The method for multi-category object recognition in remote sensing images based on self-supervised learning according to claim 1, characterized in that: The S2 comprises the following steps: S21. Use self-supervised learning method to pre-train the features of standardized remote sensing image data D and construct a cross-scale spatial-spectral fusion self-supervised feature extraction model F: ； in, To standardize remote sensing images After feature extraction, the feature representation is , d is the feature dimension; S22. Construct a cross-scale contrast learning mechanism, perform multi-scale decomposition on standardized remote sensing image data, define the scale transformation set T, and construct multi-scale enhanced samples and , calculate their eigenvectors respectively ; Define scale contrast loss: ； in, is the cosine similarity of the feature vector, is the temperature parameter, j represents the jth scale-transformed sample in the scale-contrast learning mechanism; S23. Design object-background segmentation constraints and construct object masks , define the ground feature pixel set and background pixel set , and calculate the mean of ground objects and background features: ； Among them, M is the number of ground object pixels, K is the number of background pixels, and Represents ground feature pixels or background pixels Representation vector in deep feature space; Define the object discrimination loss to make the objects distinguishable from the background features: ； S24. Using the spectral-spatial fusion enhancement mechanism, construct spectral perturbation samples: ； in, is Gaussian noise, is the disturbance coefficient, Calculate the spectral invariance loss for the remote sensing image after spectral perturbation: ； in, is the remote sensing image after spectral perturbation Through the cross-scale spatial-spectral fusion self-supervised feature extraction model, For remote sensing images Features processed by the self-supervised feature extraction model through cross-scale spatial-spectral fusion; And integrate spatial information to define local consistency constraints: ； in, To standardize remote sensing images The mth pixel point, Pixel The surrounding neighborhood window contains pixels close to its spatial position. Represents the pixel point in the remote sensing image Feature representation after processing by the cross-scale spatial-spectral fusion self-supervised feature extraction model, Represents standardized remote sensing images A pixel in is the number of pixels in the neighborhood window; S25. Calculate the final improved self-supervised learning loss, combining cross-scale contrastive learning, object-background segmentation constraints, and spectral-spatial fusion mechanism: ； in, is the weight parameter; And output the set of deep feature representations extracted by self-supervised learning: ； in, is the deep feature representation of the Nth remote sensing image.

4. The method for multi-category object recognition in remote sensing images based on self-supervised learning according to claim 1, characterized in that: The S3 comprises the following steps: S31. Use the gradient enhanced decision tree classification module to perform preliminary classification on the deep feature representation, build the gradient enhanced decision tree classification module G, and define the mapping: ； in, To predict the class label, is the depth feature representation of the i-th remote sensing image, and its value range is , C is the total number of land feature categories; S32. Use the decision tree model based on gradient boosting for classification training and construct the loss function: ； in, is the true category label, is the true probability of the cth class, The class probabilities predicted by the gradient boosted decision tree classification module; S33. Using the weighted gradient update strategy, calculate the negative gradient of the tth iteration: ； Based on negative gradient Construct the tth decision tree , update the classification model: ； in, is the learning rate, which controls the contribution of each decision tree; S34. Use a deep feature importance calculation mechanism to evaluate the contribution of different feature dimensions and define feature importance scores: ； in, is the jth feature, T is the total number of decision trees, is the number of nodes in the t-th decision tree, The loss reduction caused by node splitting is calculated, the feature importance scores are sorted, and the most important feature dimensions are selected; S35. Calculate the balance coefficient of category distribution using the category balance adjustment strategy: ； in, is the number of samples in the cth class, based on the equilibrium coefficient Adjust the loss function: ； S36. Iteratively optimize the gradient boosting decision tree classification module until the loss function converges or reaches the preset number of iterations , output preliminary multi-category classification results: ； in, is the preliminary classification label set for all remote sensing images, is the predicted category label of the Nth remote sensing image.

5. The method for multi-category object recognition in remote sensing images based on self-supervised learning according to claim 1, characterized in that: The S4 comprises the following steps: S41. By the loss function Find the gradient of the preliminary classification results and obtain the gradient feedback value: ； S42. For multiple weak classifiers of the gradient enhanced decision tree classification module, let the weight of the tth weak classifier be ,in, , T is the total number of weak classifiers, according to the gradient feedback value of each sample i Calculate the weight adjustment of the weak classifier and define the adjustment as: ； in, represents the weight adjustment of the t-th weak classifier under the guidance of the gradient for the i-th sample; S43. Implement dynamic weight update for each weak classifier: ； in, is the updated weight of the tth weak classifier, is the learning rate parameter, which controls the update step size; S44. Using updated weights Re-optimize the decision boundaries of each weak classifier and combine it with deep feature representation Fine-tune or retrain the gradient boosted decision tree classification module as a whole to construct an updated classification module G'; S45. Output the classification result obtained after the updated gradient enhanced decision tree classification module G' predicts the deep feature representation Z: ； in, is the predicted category probability vector of the Nth remote sensing image after dynamic adjustment.

6. The method for multi-category object recognition in remote sensing images based on self-supervised learning according to claim 1, characterized in that: The S5 comprises the following steps: S51. The feature distillation module is used to reduce the dimension of the deep feature representation. The feature distillation module optimizes the redundant features while maintaining the original feature space information by constructing a lightweight representation learning network, and generates a low-dimensional feature representation after dimensionality reduction; S52. Use information entropy optimization strategy to screen the low-dimensional feature representation after dimensionality reduction, calculate the information contribution of each feature dimension based on the category discrimination ability, retain the most discriminative features and eliminate low-relevance features; S53. A feature fusion strategy is used to fuse the low-dimensional feature representation output by the feature distillation module with the intermediate classification features after weight adjustment by the optimized gradient boosting decision tree classification module. During the fusion process, dynamic weighting parameters are set for different feature sources so that the deep features extracted by self-supervised learning and the high-discriminative features generated by the gradient boosting decision tree classification module work together to form a fused optimized feature representation. S54. The gradient boosting decision tree classification module is retrained using fusion feature representation. A hierarchical learning strategy is introduced in the training process. The easy-to-classify samples are trained first, and then gradually transitioned to the boundary fuzzy and easily confused categories. S55. Use a fine classification training strategy to perform local iterative optimization on the fusion feature representation, dynamically adjust the category discrimination boundary during the training process, increase the decision tree depth for the confusion categories above the threshold, and use a sample balancing strategy for the frequency categories below the threshold; S56. Outputting an optimized multi-category ground object recognition and classification result, wherein the optimized multi-category ground object recognition and classification result is generated based on the fusion feature representation and undergoes multi-level optimization and adjustment.

Images