Multi-source high-resolution remote sensing image vegetation extraction method based on multi-feature comprehensive perception

Abstract

The present application belongs to the technical field of photogrammetry data processing, and particularly relates to a multi-source high-resolution remote sensing image vegetation extraction method based on multi-feature comprehensive perception. Firstly, the present application performs feature optimization on vegetation indexes through a random forest model, filters out indexes capable of improving the inter-class difference indexes of vegetation and other ground objects, and constructs a multi-feature comprehensive perception convolution network. The network is constructed through a double-branch parallel network of spectral features and vegetation index features, and on the basis of simplifying the dense connection module, the network strengthens the multi-scale feature information extraction capability while reducing the loss of detailed features. In addition, in order to promote the global information interaction between the original spectral information and the vegetation index features, a double-path multi-head cross-attention fusion module is designed to expand the difference between vegetation and other ground objects and the consistency of vegetation, improve the generalization performance of the network, and realize vegetation extraction under multi-source high-resolution remote sensing data.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of photogrammetry data processing, and particularly relates to a multi-source high-resolution remote sensing image vegetation extraction method based on multi-feature comprehensive perception. BACKGROUND

[0002] Vegetation is an important part of the earth's ecological system and is of great significance to ecological environment protection and construction. Based on the spectral, texture, spatial, and temporal characteristics of vegetation on high-resolution remote sensing images, vegetation extraction is performed, and the spatial and temporal distribution information of vegetation is extracted by complementing multi-source data, which is of great significance for resource survey, urban planning, land survey, and forest fire monitoring. However, due to the adverse effects of background noise during remote sensing imaging, a series of problems such as false extraction, internal fragmentation, unclear boundaries, and decreased model generalization ability often occur in the recognition results.

[0003] In recent years, remote sensing technology has become an important technical means for vegetation monitoring due to its wide range and timeliness. The current methods for studying vegetation extraction mainly include threshold segmentation based on vegetation index. This method constructs a spectral index to extract vegetation by the difference in reflectivity of the red and near-infrared bands. This method can extract vegetation from images, but it is difficult to identify vegetation under the shadow of tall buildings. In sparse vegetation areas, the vegetation index method may also fail to accurately estimate the vegetation coverage. Many machine learning methods have also been used for vegetation extraction, such as maximum likelihood classification (MLC), support vector machine (SVM), and random forest (RF). These machine learning methods can achieve high accuracy in images, but the feature and threshold conditions need to be designed manually, making it difficult to achieve automatic extraction of wide-area and multi-source data.

[0004] With the improvement of image spatial resolution and the diversification of ground information, the semantic information of the scene is more complex, and deep learning methods have gradually become a research hotspot in artificial intelligence. Currently, it has been widely used in image recognition, image segmentation, target detection, and other tasks. In the field of vegetation remote sensing, deep learning methods are mainly divided into semantic segmentation-based methods and pixel-based methods, which learn related data features independently in an end-to-end manner and can meet the growing demand for vegetation evaluation and monitoring under increasing remote sensing data. However, although semantic segmentation methods can achieve good accuracy in vegetation extraction, they essentially aggregate homogeneous pixels and often ignore some detailed features, making it difficult to consider the internal details of vegetation features and easily confused with other coexisting features. Moreover, semantic segmentation methods require accurate boundary information when labeling, and it is difficult to distinguish mixed pixels at the boundaries of vegetation and other coexisting features, making it difficult to label samples on multi-source remote sensing images, which further leads to the problem of semantic segmentation model convergence difficulty during training.

[0005] Based on pixel classification, only some labeled pixels need to be added in the sample image of a specific scene, which can quickly and easily adapt to new scenes. However, most remote sensing vegetation extraction methods only extract vegetation from a single data source according to the spatial spectral characteristics of vegetation, or combine vegetation index characteristics to input the network in a spliced manner for training to extract vegetation. Using only spectral characteristics for vegetation extraction, when the spectral values differ greatly due to differences in time phase, radiation, etc., it is difficult to achieve high-precision vegetation extraction across data sources. The difference between spectral characteristic values and index characteristic values is huge, and the introduction of vegetation index characteristics in a spliced manner often ignores the index characteristics, making it difficult to achieve complementary correction between multiple characteristics, which can lead to missed extraction and false extraction of vegetation in different data sources. In addition, the selection of vegetation index characteristics relies solely on manual experience for feature selection, lacking a quantitative selection method, making it difficult to scientifically and reasonably verify the effectiveness of the index.

[0006] Therefore, the present application provides a multi-source high-resolution remote sensing image vegetation extraction method based on multi-feature comprehensive perception. SUMMARY

[0007] The present application aims to overcome the above-mentioned problems in the prior art and provide a multi-source high-resolution remote sensing image vegetation extraction method based on multi-feature comprehensive perception.

[0008] To achieve the above technical purposes and effects, the present application is implemented by the following technical solutions:

[0009] The present application provides a multi-source high-resolution remote sensing image vegetation extraction method based on multi-feature comprehensive perception, comprising the following steps:

[0010] Step 1: Sample set preparation

[0011] Step 2: Vegetation index feature optimization

[0012] Step 3: Network model structure design

[0013] Step 4: Multi-scale dense connection module

[0014] Step 5: Dual-path multi-head cross-attention feature fusion module

[0015] Step 6: Model extraction result accuracy evaluation

[0016] Further, the sub-steps of step 1 are as follows:

[0017] 1) Obtain high-resolution remote sensing image data of the target area

[0018] 2) To better represent vegetation features, remote sensing images were displayed in Arcgis 10.7 software using the 4-3-2 band, with vegetation features colored red. Vegetation was delineated through manual visual interpretation.

[0019] 3) The preliminary delineation results are repeatedly revised, and samples are collected in the interior and edge areas of the delineated plot to obtain vegetation labels and background labels, and obtain sample coordinate points;

[0020] 4) Considering the spatial correlation between adjacent pixels, a 5×5 neighborhood window centered on the sampling point is expanded outward as the training sample to obtain more spatial and spectral information. Combining various data sources, the vegetation contours are outlined through manual visual interpretation to obtain ground truth label data as verification data.

[0021] Furthermore, the sub-steps of step 2 are as follows:

[0022] 1) Using vegetation index features as independent variables and sample labels as dependent variables, a random forest regression model was used to fit the nonlinear relationship between the two and perform feature optimization. The vegetation indices selected were as follows: NDVI, GNDVI, EVI, OSAVI, RVI, DVI, TVI, GVI, GI, NDGI, MCARI, and TCARI. The number of decision trees in the model was set to 20, and the depth was set to 4.

[0023] 2) According to the fitting results of the random forest model, the values ​​are sorted. The larger the value, the more important the vegetation index feature is to the accuracy of the model prediction. The top four vegetation index features are selected.

[0024] Furthermore, in step three, the input data of the model are remote sensing image spectral data and vegetation index. The cropped 5×5 samples are input into the parallel two-branch network respectively. The two networks contain three parallel sub-networks with scales of 1×1, 3×3, and 5×5 respectively; a simplified dense connection module is introduced to learn spectral and index features of different scales, and downsampling dense connections are used to perform interactive correction between various features.

[0025] Furthermore, in step three, based on the self-attention of the visual transformer to obtain the maximum receptive field, a two-way cross-attention feature fusion module is designed between the 5×5 and 3×3 branches; the shallow initial features themselves are cross-fused with the enhanced index features to establish multi-feature comprehensive perception to achieve global information complementarity, and the hybrid channel output is used to enhance the features.

[0026] Further, in step three, a parallel double-branch multi-scale network is designed, i.e., one spectral multi-scale network branch and one exponential multi-scale network branch; coordinate convolution is introduced into the 5x5 and 3x3 branch networks respectively, the convolution layer is used to optimize the extraction of spatial feature position information, and a skip connection is designed in the branch network for extracting features of different scales to enhance the fusion of vegetation index features at different scales and make the vegetation boundary extraction more complete; after continuous convolution of the double-branch parallel sub-network, 1x1 spectral features and 1x1 index features are output respectively;

[0027] Then, the spectral features and index features generated by parallel splicing of the network at different scales are merged; a deep separable convolution layer DWCConv 3×3 The features are extracted in detail, and the parameter calculation amount is reduced;

[0028] Finally, the vegetation classification results are output through two fully connected layers and a Softmax classifier.

[0029] Further, in step four, in order to realize the exchange of feature information between 5x5, 3x3 and 1x1, a multi-scale dense connection module is introduced between the parallel double-branch network of spectral features and index features; in the multi-scale dense connection module, each sub-network can receive feature information from other parallel networks, so that the network can exchange information at different scales and enhance the feature expression ability; in the dense connection block used, the features of the previous layer are transmitted to all the following layers by stacking;

[0030] In the 5x5 and 3x3 branch structure, vertical direction downsampling is performed to expand the width of the network; through the dense connection mode, the features of the 5x5 branch are superimposed into the 3x3 and 1x1 branch networks, and then the features of the 3x3 branch are superimposed into the 1x1 branch network;

[0031] In order to retain the feature information at different scales and realize the fusion of features, the branch network of each layer is spliced with other branch networks in the channel dimension. This method not only effectively alleviates the problem of gradient disappearance, but also reduces the parameter amount, thereby enhancing the ability of feature extraction and reuse. That is, in the L-layer deep network structure, the network of the embodiment has Lx(L+1) / 2 connections.

[0032] Further, in step five, in this module, the input feature F(H, W, C) is flattened to F(N, C), and then linear embedding is used to generate K (key), V (value) and Q (query) for each attention head, each head is calculated by a multispectral branch and a vegetation index branch Attention score G MS and G VI ;

[0033] Then, the attention scores are normalized by Softmax, and the vegetation index feature G VI is cross-multiplied with the spectral feature Q MS , so that the vegetation index branch allocates attention to the features of the spectral branch; meanwhile, the spectral feature G MS is cross-multiplied with the vegetation index feature Q VI , so that the spectral feature branch allocates attention to the features of the vegetation index branch.

[0034] In the two sequence branches, branch complementary interaction is performed to achieve feature fusion; the calculation formula is as follows:

[0035]

[0036] Further, in step six, four evaluation indexes are used: overall accuracy OA, F1 score, intersection over union (IoU), and accuracy for precision evaluation.

[0037] Overall accuracy OA refers to the ratio of the sum of the number of pixels correctly identified by the model in the test image to all identified pixels of the class.

[0038] F1 score is the harmonic mean of precision and recall, used to evaluate the ability of the model to handle imbalanced data sets or sensitivity to misclassified classes.

[0039] IoU is the degree of overlap between the model's prediction and the true labeled area, used to evaluate the accuracy and robustness of the model.

[0040] The Kappa coefficient (Kappa) of remote sensing interpretation accuracy evaluation estimates the consistency of prediction and ground reference.

[0041] The formulas of the above evaluation indexes are as follows:

[0042]

[0043] Where TP represents true positive, FP represents false positive, TN represents true negative, FN represents false negative, and N represents the total number of samples; false positive indicates the number of non-field pixels incorrectly classified as positive, and false negative indicates the number of positive pixels incorrectly classified as negative.

[0044] The beneficial effects of the present application are:

[0045] 1. The method of the present invention is scientifically and rationally designed. First, a random forest model is used to optimize the features of the vegetation index, screening out an index that can enhance the inter-class difference between vegetation and other land objects. On this basis, a multi-feature comprehensive perception convolutional network (MSCIN) is constructed with multi-scale feature information enhancement, global information interaction, and feature cross-fusion. By constructing a dual-branch parallel network of spectral features and vegetation index features, this network strengthens the multi-scale feature information extraction capability while reducing the loss of detailed features on the basis of simplifying the dense connection module. In addition, to promote global information interaction between the original spectral information and vegetation index features, a dual-path multi-head cross-attention fusion module is designed to expand the difference between vegetation and other land objects and the consistency of vegetation, improve the generalization performance of the network, and realize vegetation extraction under multi-source high-resolution remote sensing data.

[0046] 2. The extraction results of the method of the present invention are robust in accuracy under different sensors, overcoming the problem of rapid accuracy degradation of other methods under different sensors, and at the same time solving the problems of internal fragmentation, false extraction, and missed extraction caused by sample generalization and image diversity.

[0047] Of course, any product implementing the present invention does not necessarily need to achieve all of the above advantages at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0049] Fig. 1 This is a diagram of the overall model structure of an embodiment of the present invention;

[0050] Fig. 2 This is a structural diagram of a dual-path cross-attention feature fusion module according to an embodiment of the present invention;

[0051] Fig. 3 Graphs showing vegetation extraction results for different models according to an embodiment of the present invention. DETAILED DESCRIPTION

[0052] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0053] like Figs. 1-3As shown, the embodiment provides a multi-source high-resolution remote sensing image vegetation extraction method based on multi-feature comprehensive perception, including the following steps:

[0054] Step one, sample set making;

[0055] Step two, vegetation index feature optimization;

[0056] Step three, network model structure design;

[0057] Step four, multi-scale dense connection module;

[0058] Step five, double-path multi-head cross attention feature fusion module;

[0059] Step six, model extraction result precision evaluation.

[0060] Step one specifically includes the following content:

[0061] 1) Obtain target area high-resolution remote sensing image data;

[0062] 2) To better represent the characteristics of vegetation, display the remote sensing image in 4-3-2 bands in Arcgis 10.7 software, and the characteristics of vegetation are represented as red, and the vegetation is outlined by manual visual interpretation;

[0063] 3) Repeat the correction of the preliminary outline result, sample collection in the interior and edge area of the outlined plot, obtain vegetation labels and background labels respectively, and obtain sample coordinate points;

[0064] 4) Considering the spatial correlation between adjacent pixels, taking the pixel point of the sample point as the center, a 5x5 neighborhood window is expanded outward as the training sample to obtain more spatial information and spectral information; combined with various data sources, the vegetation contour is outlined by manual visual interpretation to obtain ground true label data as verification data.

[0065] Step two specifically includes the following content:

[0066] 1) Take the vegetation index feature as the independent variable and the sample label as the dependent variable, use the random forest regression model to fit the nonlinear relationship between the two, and optimize the features; the vegetation index is selected as follows: NDVI, GNDVI, EVI, OSAVI, RVI, DVI, TVI, GVI, GI, NDGI, MCARI, TCARI; the number of decision trees in the model is set to 20, and the depth is set to 4;

[0067] 2) According to the fitting result of the random forest model, sort the numerical value, the larger the numerical value, the more important the vegetation index feature to the accuracy of model prediction, select the top four vegetation index features.

[0068] Step 3 specifically includes the following:

[0069] The input data of the model are remote sensing image spectral data and vegetation index. The cropped 5×5 samples are input into the parallel two-branch network respectively. The two networks contain three parallel sub-networks with scales of 1×1, 3×3, and 5×5 respectively. A simplified dense connection module is introduced to learn spectral and index features of different scales, and downsampling dense connections are used to perform interactive correction between features.

[0070] Furthermore, based on the self-attention of the visual transformer to obtain the maximum receptive field, a two-way cross-attention feature fusion module is designed between the 5×5 and 3×3 branches; the shallow initial features themselves are cross-fused with the enhanced exponential features to establish multi-feature comprehensive perception to achieve global information complementarity, and the hybrid channel output is used to enhance the features.

[0071] Furthermore, a parallel dual-branch multi-scale network is designed, namely a spectral multi-scale network branch and an exponential multi-scale network branch; coordinate convolution is introduced into the 5×5 and 3×3 branch networks respectively, and the convolution layer is used to optimize the extraction of spatial feature position information. At the same time, jump connections are designed within the branch networks for extracting index features at different scales to enhance the fusion of vegetation index features at different scales and make the vegetation boundary extraction more complete; after continuous convolution, the dual-branch parallel sub-network outputs 1×1 spectral features and 1×1 exponential features respectively; then, the spectral features and exponential features generated by parallel splicing of networks of different scales are merged; a depthwise separable convolutional layer DWCConv is designed. 3×3 The features are extracted finely to reduce the amount of parameter calculation; finally, after two fully connected layers and a Softmax classifier, the vegetation classification results are output.

[0072] Step 4 specifically includes the following:

[0073] In order to realize the feature information exchange between 5×5, 3×3 and 1×1, a multi-scale dense connection module is introduced between the spectral feature and exponential feature parallel dual-branch networks; in the multi-scale dense connection module, each sub-network can receive feature information from other parallel networks, enabling the network to exchange information at different scales, thereby enhancing the feature expression capability.

[0074] In the densely connected block used, the features of the previous layer are passed to all subsequent layers in a stacked manner; in the 5×5 and 3×3 branch structures, vertical downsampling is performed to expand the width of the network; through dense connections, the features of the 5×5 branch are superimposed on the 3×3 and 1×1 branch networks, and then the features of the 3×3 branch are superimposed on the 1×1 branch network. In order to retain feature information at different scales and achieve feature fusion, the branch networks of each layer are spliced ​​with other branch networks in the channel dimension. This method not only effectively alleviates the problem of gradient disappearance, but also reduces the number of parameters, thereby enhancing the ability of feature extraction and reuse. That is to say, in the deep network structure of L layers, the network of this embodiment has L×(L+1) / 2 connections.

[0075] Step 5 specifically includes the following:

[0076] In this module, the input features F(H, W, C) are flattened to F(N, C), and then linear embedding is used to generate K(key), V(value), and Q(query) for each attention head. Each head calculates the attention score G of K and V by the multispectral branch and the vegetation index branch. MS and G VI ;

[0077] Then, Softmax is used to normalize the attention score and the vegetation index feature G VI With spectral characteristics Q MS Cross-multiplication makes the vegetation index branch focus on the characteristics of the spectral branch; at the same time, the spectral feature G MS and vegetation index characteristic Q VI Cross-multiplication makes the spectral feature branch allocate attention to the features of the vegetation index branch;

[0078] In the two sequence branches, branch complementary interaction is performed to achieve feature fusion; the calculation formula is as follows:

[0079]

[0080] Step 6 specifically includes the following:

[0081] In order to verify the effectiveness of this method in extracting vegetation, four evaluation indicators were used for accuracy evaluation: overall accuracy (OA), F1 score, intersection over union (IoU) and precision; overall accuracy (OA) refers to the ratio of the sum of the number of pixels in the category correctly identified by the model in the test image to the pixels of all identified categories; F1 score is the harmonic mean of precision and recall, which is used to evaluate the model's ability to handle unbalanced data sets or sensitivity to misclassified categories; IoU is the degree of overlap between the model's prediction results and the true annotated area, which is used to evaluate the accuracy and robustness of the model; the Kappa coefficient (Kappa) of remote sensing interpretation accuracy evaluation estimates the consistency between the prediction and the ground reference.

[0082] The formula for the above evaluation index is as follows:

[0083]

[0084]

[0085] Among them, TP stands for true positive, FP stands for false positive, TN stands for true negative, FN stands for false negative, and N is the total number of samples; false positive refers to the number of non-field pixels that are mistakenly classified as positive, and false negative refers to positive pixels that are mistakenly classified as negative.

[0086] A specific application of this embodiment is:

[0087] Two GF-2 image data of Ma'anshan City, Anhui Province in October 2022 were obtained, and they were manually outlined and sampled to obtain 266,547 vegetation pixels and 435,205 background pixels, respectively. They were made into 5×5 standard sample data and input into the model. This embodiment was experimented with the tensflower2.4.0 framework in the Python3.7 environment. All experiments were performed on a device equipped with a 1080Ti graphics card. The Adam optimizer was used to optimize the network. During the training process, all experiments used the same hyperparameters, including the number of training rounds (40 rounds), batch size (512), and initial learning rate (0.001).

[0088] As shown in Table 1 and Fig. 3 As shown in the figure, using the Gaofen-2 remote sensing image as the data source, it can be seen from the figure that the four models are all very effective in vegetation extraction, and can extract vegetation relatively completely in mountainous areas. However, in some complex urban areas A(4)-(7), the mixture of urban roads, buildings and vegetation is prone to misclassification. Compared with other methods, the method of this embodiment has fewer cases of misclassification and omission, and the boundary extraction is more complete, reaching 86.46%, 92.85% and 85.58% in IoU, OA and Kappa coefficient respectively.

[0089] The method of this embodiment can extract vegetation relatively completely on both GF7 and Planet images, with fewer omissions and errors.

[0090] In the GF7 image, the method of this embodiment achieved 84.21%, 92.03%, and 86.81% in IoU, OA, and Kappa coefficient, respectively.

[0091] On the Planet image, the method in this embodiment achieved 86.96%, 91.82%, and 83.28% in IoU, OA, and Kappa coefficient, respectively.

[0092] This shows that MSICN still maintains high internal consistency and boundary accuracy for vegetation extraction across data sources and has high generalization ability.

[0093] Table 1 Comparison of accuracy verification results of different models

[0094]

[0095] In response to the current problems of missing and incorrect extraction in vegetation extraction from multiple data sources, a convolutional network (MSICN) for vegetation extraction that takes into account multi-feature comprehensive perception is constructed. This method first uses a random forest model to perform feature optimization on the vegetation index, and screens out an index that can enhance the inter-class difference between vegetation and other ground objects. On this basis, a multi-feature comprehensive perception convolutional network (MSCIN) with multi-scale feature information enhancement, global information interaction, and feature cross-fusion is constructed. This network strengthens the multi-scale feature information extraction capability while reducing the loss of detail features by constructing a dual-branch parallel network of spectral features and vegetation index features on the basis of simplifying the dense connection module. In addition, in order to promote global information interaction between the original spectral information and vegetation index features, a dual-path multi-head cross-attention fusion module is designed to expand the difference between vegetation and other ground objects and the consistency of vegetation, improve the generalization performance of the network, and realize vegetation extraction under multi-source high-resolution remote sensing data. The extraction results of the method of the present invention are robust in accuracy under different sensors, overcoming the problem of rapid accuracy decay of other methods under different sensors, and solving the problems of internal fragmentation, incorrect extraction, and missing extraction caused by sample generalization and image diversity.

[0096] The preferred embodiments of the present invention disclosed above are intended only to help illustrate the present invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the present invention to specific embodiments. Obviously, many modifications and variations are possible based on the contents of this specification. These embodiments are selected and described in detail in this specification to better explain the principles and practical applications of the present invention, thereby enabling those skilled in the art to better understand and utilize the present invention. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A vegetation extraction method based on multi-source high-resolution remote sensing images based on multi-feature comprehensive perception, characterized by: The steps include: Step 1: Sample set preparation; Step 2: Optimization of vegetation index features; Step 3: Network model structure design; The model's input data consists of remote sensing image spectral data and vegetation indices. Cropped 5×5 samples are fed into a parallel two-branch network. Each network contains three parallel sub-networks with scales of 1×1, 3×3, and 5×5, respectively. A simplified dense connection module is introduced to learn spectral and index features of different scales, and downsampled dense connections are used to perform interactive corrections between features. Based on the self-attention of the visual transformer to obtain the maximum receptive field, a two-way cross-attention feature fusion module is designed between the 5×5 and 3×3 branches; Cross-modal feature fusion is performed by combining its own shallow initial features with enhanced index features to establish multi-feature comprehensive perception to achieve global information complementarity, and enhanced features are output using hybrid channels. Step 4: Multi-scale dense connection module; To enable feature information exchange between 5×5, 3×3, and 1×1 networks, a multi-scale dense connection module is introduced between the spectral feature and exponential feature parallel dual-branch networks. In this multi-scale dense connection module, each sub-network can receive feature information from other parallel networks, enabling the network to exchange information at different scales, thereby enhancing feature expression capabilities. In the dense connection block used, the features of the previous layer are passed to all subsequent layers in a stacked manner. In the 5×5 and 3×3 branch structures, vertical downsampling is performed to expand the width of the network. Through dense connections, the features of the 5×5 branch are superimposed on the 3×3 and 1×1 branch networks, and then the features of the 3×3 branch are superimposed on the 1×1 branch network. Each layer of the branch network is spliced ​​with other branch networks in the channel dimension; Step 5: Dual-path multi-head cross-attention feature fusion module; Step 6: Accuracy evaluation of model extraction results.

2. The method for extracting vegetation from multi-source high-resolution remote sensing images based on multi-feature integrated perception according to claim 1 is characterized in that: The sub-steps of step one are as follows: 1) Obtaining Gaofen-2 remote sensing image data of the target area; 2) To better represent vegetation features, remote sensing images were displayed in Arcgis 10.7 software using the 4-3-2 band, with vegetation features colored red. Vegetation was delineated through manual visual interpretation. 3) Repeatedly revise the preliminary delineation results, collect samples inside and on the edge of the delineated plot, obtain vegetation labels and background labels, and obtain sample coordinate points; 4) Considering the spatial correlation between adjacent pixels, a 5×5 neighborhood window is used to expand outward from the sampling point as the center as the training sample to obtain more spatial and spectral information; Combining various data sources, vegetation contours were outlined through manual visual interpretation to obtain ground truth label data as verification data.

3. The method for extracting vegetation from multi-source high-resolution remote sensing images based on multi-feature integrated perception according to claim 2 is characterized in that: The sub-steps of step 2 are as follows: 1) Using vegetation index features as independent variables and sample labels as dependent variables, a random forest regression model was used to fit the nonlinear relationship between the two and perform feature optimization. The vegetation indices selected were as follows: NDVI, GNDVI, EVI, OSAVI, RVI, DVI, TVI, GVI, GI, NDGI, MCARI, and TCARI. The number of decision trees in the model was set to 20, and the depth was set to 4. 2) According to the fitting results of the random forest model, the values ​​are sorted. The larger the value, the more important the vegetation index feature is to the accuracy of the model prediction. The top four vegetation index features are selected.

4. The method for extracting vegetation from multi-source high-resolution remote sensing images based on multi-feature integrated perception according to claim 3 is characterized in that: In step three, a parallel two-branch multi-scale network is designed, namely, a spectral multi-scale network branch and an exponential multi-scale network branch. Coordinate convolution is introduced into the 5×5 and 3×3 branch networks respectively, and the convolution layer is used to optimize the extraction of spatial feature position information. At the same time, jump connections are designed within the branch networks for extracting index features at different scales to enhance the fusion of vegetation index features at different scales and make vegetation boundary extraction more complete. After continuous convolution, the two-branch parallel sub-network outputs 1×1 spectral features and 1×1 exponential features respectively. Then, the spectral features and exponential features generated by parallel splicing of networks of different scales are merged; a depth-wise separable convolutional layer DWCConv is designed. 3×3 Finely extract features to reduce parameter calculations; Finally, after two fully connected layers and a Softmax classifier, the vegetation classification result is output.

5. The method for extracting vegetation from multi-source high-resolution remote sensing images based on multi-feature integrated perception according to claim 4 is characterized in that: In step 5, the input features F(H, W, C) are flattened to F(N, C) in this module, and then linear embedding is used to generate K, V, Q for each attention head. Each head calculates the attention score G of K and V by the multispectral branch and the vegetation index branch. MS and G VI ; Then, Softmax is used to normalize the attention score and the vegetation index feature G VI With spectral characteristics Q MS Cross-multiplication makes the vegetation index branch focus on the characteristics of the spectral branch; at the same time, the spectral feature G MS and vegetation index characteristic Q VI Cross-multiplication makes the spectral feature branch allocate attention to the features of the vegetation index branch; In the two sequence branches, branch complementary interaction is performed to achieve feature fusion; The calculation formula is as follows: （1） （2） （3） （4）。 6. The method for extracting vegetation from multi-source high-resolution remote sensing images based on multi-feature integrated perception according to claim 5, characterized in that: In step 6, four evaluation indicators are used to evaluate the accuracy: overall accuracy OA, F1 score, intersection over union ratio and accuracy rate; Overall accuracy OA refers to the ratio of the sum of the number of pixels in the category correctly identified by the model to the pixels of all identified categories in the test image; The F1 score is the harmonic mean of precision and recall, and is used to evaluate the model's ability to handle imbalanced datasets or sensitivity to misclassified categories; IoU is the degree of overlap between the model prediction result and the true annotation area, which is used to evaluate the accuracy and robustness of the model; The Kappa coefficient for remote sensing interpretation accuracy assessment estimates the consistency between the prediction and the ground reference; The formula for the above evaluation index is as follows: （5） （6） （7） （8） （9） （10） in, TP represents a true positive, FP represents a false positive, TN represents a true negative, FN represents false negatives, N is the total number of samples; false positives represent the number of non-field pixels that are misclassified as positive, and false negatives represent positive pixels that are misclassified as negative.

Images