Land damage and repair monitoring method based on differential feature enhancement and mining area detection

Abstract

The invention discloses a land damage and restoration monitoring method based on differential feature enhancement and mining area detection, and the method comprises the steps: S1, collecting remote sensing image data of a front time phase and a rear time phase, and cutting the remote sensing image data into image blocks; s2, the neural network model comprises a residual module, a change information enhancement module, a global difference fusion module, a strip mine detection module and a land damage and repair judgment module, the residual module performs feature extraction on the image blocks and inputs the features into the global difference fusion module to obtain difference features with local and global information; the change information enhancement module performs feature information enhancement processing; and S3, the strip mine detection module judges whether a strip mine area is included or not by using the difference features, and the land damage and restoration judgment module performs classification judgment by using the difference features and outputs a judgment result. According to the method, pixel-level or pixel-level land damage and restoration prediction can be realized, and land damage and restoration result distribution data of a research area according to pixels can be obtained through step-by-step statistics.

Description

Technical Field

[0001] The present invention relates to the field of mining area ecological monitoring, and in particular to a land damage and restoration monitoring method based on differential feature enhancement and mining area detection. Background Art

[0002] With the rapid development of the mining industry, open-pit mining activities are increasing. Although this has promoted economic growth, it has also caused serious damage to the ecological environment, including deforestation, soil erosion and biodiversity loss. Open-pit mine land damage and restoration monitoring can help timely discover and evaluate the damage caused by mining activities to the ecological environment, so as to take corresponding protection and restoration measures to reduce the negative impact on the ecological environment and ensure sustainable development. However, traditional monitoring methods usually use field surveys, which are time-consuming, labor-intensive and inefficient. With the development of remote sensing technology, although there are also methods based on remote sensing image data to monitor mining areas, such as membership functions, support vector machines and random forests, the huge amount of data and the complexity of mining area characteristics make it challenging to accurately assess the degree of land damage and restoration work. Existing technologies cannot realize the monitoring of open-pit mine land damage and restoration, resulting in a technical gap in open-pit mine land damage and restoration. Therefore, it is urgent to develop an intelligent and accurate open-pit mine land damage and restoration monitoring method. Summary of the invention

[0003] The purpose of the present invention is to solve the technical problems pointed out by the background technology, and to provide a land damage and restoration monitoring method based on differential feature enhancement and mining area detection. The global differential fusion module effectively captures the land change characteristics of the open-pit mining area, and the change information enhancement module enhances the feature information to improve the accuracy and reliability of monitoring; the open-pit mining detection module performs mining area feature detection, and the land damage and restoration judgment module uses differential features to perform classification judgment and outputs the judgment results of land damage and restoration monitoring, thereby achieving the purpose of classified monitoring of land damage and restoration in open-pit mines, and providing a new method model for the ecological environment protection and monitoring of open-pit mining areas, which has important practical significance and application value.

[0004] The purpose of the present invention is achieved through the following technical solutions:

[0005] A land damage and restoration monitoring method based on differential feature enhancement and mining area detection, the method comprising:

[0006] S1, collecting remote sensing image data of two phases, and cutting the remote sensing image data into image blocks P1 and P2 of the same area in the two phases;

[0007] S2. Construct a neural network model and use the sample data set to train the model. The neural network model includes a residual module, a change information enhancement module, a global difference fusion module, an open-pit mine detection module, and a land damage and restoration judgment module. The residual module extracts features from the image blocks P1 and P2 and inputs them into the global difference fusion module to obtain differential features with both local and global information. Then, the change information enhancement module uses the differential features to enhance the features extracted by the residual module;

[0008] S3. The open-pit mine detection module uses differential features to determine whether an open-pit mine area is included. The land damage and restoration judgment module uses differential features to perform classification judgment and output judgment results, which include land damage and land restoration.

[0009] In order to better implement the land damage and restoration monitoring method of the present invention, in step S3, the open-pit mine detection module detects whether the differential features corresponding to the image blocks P1 and P2 include features of the open-pit mine area. Before determining whether the open-pit mine area is included, it also includes a detection information enhancement module, which performs information enhancement processing on the differential features.

[0010] Preferably, in step S2, the residual module includes several residual blocks, each residual block includes a Conv block module and an Identity block module, the Conv block module includes a convolution processing main path and a jump connection layer connected to the input and output ends of the convolution processing main path, the convolution processing main path includes 3 convolution layers, and the jump connection layer includes a convolution layer; the Identity block module includes a convolution processing main path and a jump connection layer connected to the input and output ends of the convolution processing main path, the convolution processing main path includes 3 convolution layers; the residual module extracts several layers of features for image blocks P1 and P2 respectively.

[0011] Preferably, the global differential fusion module extracts features of two phases from the residual module through a perceptron, uses two branches to obtain the query Q, key K and value V corresponding to the two phases, and after the two branches generate the query Q, performs a differential operation on the query Q of the two branches to obtain a differential query Q′; uses the differential query Q′ to perform a matrix multiplication operation with the key K and value V of each branch, and performs an absolute difference operation on the matrix multiplication results obtained from the two branches to obtain differential features.

[0012] Preferably, the processing method of the change information enhancement module is as follows: extracting initial features from the residual module through a perceptron and generating a matrix of query Q, extracting differential features through a perceptron to calculate the matrices of key K and value V; multiplying the matrix of query Q with the matrix of key K to obtain a similarity matrix; performing Softmax normalization processing on the similarity matrix to obtain an attention weight matrix, and multiplying the attention weight matrix with the matrix of value V to generate enhanced differential features.

[0013] Preferably, the open-pit mine detection module detection method includes: dividing the input image into several small blocks, the input image is a differential feature, all the small blocks are linearly projected and flattened into a feature vector of fixed length; using the Transformer encoder module to capture the relationship between each small block through the self-attention mechanism to obtain a feature vector containing global context information and local details, determine whether it contains open-pit mine area features, and obtain detection features for enhancing differential features.

[0014] Preferably, the processing method of the detection information enhancement module is as follows: the differential features are processed by the perceptron to obtain the query Q, the detection features are processed by the perceptron to obtain the key K and the value V, the matrix of the query Q is multiplied by the matrix of the key K to obtain the similarity matrix; the similarity matrix is ​​subjected to Softmax normalization processing to obtain the attention weight matrix, and the attention weight matrix is ​​multiplied by the matrix of the value V to generate the enhanced differential features.

[0015] Preferably, the model training method of the neural network model is as follows:

[0016] A1. Prepare a sample data set. Collect remote sensing image data of the previous and next two time phases, annotate them for damage and repair, and convert them into raster images. Cut the remote sensing image data of the previous and next two time phases and the raster images with annotation information into several image blocks of the same area of ​​the previous and next two time phases and use them as sample data.

[0017] A2. Divide the sample data of the sample data set into a training set and a validation set in a ratio of 7:3. The training set is used to train the neural network model, and the validation set is used to test the accuracy of the neural network model.

[0018] A land damage and restoration monitoring system based on differential feature enhancement and mining area detection comprises a data acquisition module, a sample data set and a neural network model, wherein the data acquisition module is used to collect remote sensing image data of two phases before and after, and cut the remote sensing image data into image blocks P1 and P2 of the same area in the two phases before and after; the neural network model comprises a residual module, a change information enhancement module, a global differential fusion module, an open-pit mine detection module and a land damage and restoration judgment module, wherein the residual module extracts features from the image blocks P1 and P2 and inputs the features into the global differential fusion module to obtain differential features with both local and global information, and then the change information enhancement module uses the differential features to perform feature information enhancement processing on the features extracted by the residual module; the open-pit mine detection module uses the differential features to judge whether an open-pit mine area is included, and the land damage and restoration judgment module uses the differential features to perform classification judgment and output the judgment result, and the judgment result includes land damage and land restoration; the neural network model uses the sample data set to perform model training.

[0019] The land damage and restoration monitoring system of the present invention further includes a detection information enhancement module, and the open-pit mine detection module detects whether the differential features corresponding to the image blocks P1 and P2 include features of the open-pit mine area; before determining whether the open-pit mine area is included, the detection information enhancement module performs information enhancement processing on the differential features; the model training method of the neural network model is as follows:

[0020] A1. Prepare a sample data set. Collect remote sensing image data of the previous and next two time phases, annotate them for damage and repair, and convert them into raster images. Cut the remote sensing image data of the previous and next two time phases and the raster images with annotation information into several image blocks of the same area of ​​the previous and next two time phases and use them as sample data.

[0021] A2. Divide the sample data of the sample data set into a training set and a validation set in a ratio of 7:3. The training set is used to train the neural network model, and the validation set is used to test the accuracy of the neural network model.

[0022] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0023] (1) The present invention creatively introduces a global differential fusion module and effectively captures the land change characteristics of the open-pit mining area. The change information enhancement module plays a role in enhancing the characteristic information and improving the accuracy and reliability of monitoring. The open-pit mining detection module performs mining area characteristic detection, and the land damage and restoration judgment module uses differential characteristics to perform classification judgment and output the judgment results of land damage and restoration monitoring, thereby achieving the purpose of classified monitoring of land damage and restoration in open-pit mining areas, providing a new method model for the ecological environment protection and monitoring of open-pit mining areas, which has important practical significance and application value.

[0024] (2) The residual module, change information enhancement module and global differential fusion module of the present invention constitute a twin residual enhancement and differential fusion structure. The features extracted from the remote sensing images of the previous and next phases are calculated by the global differential fusion module to obtain differential features. The change information enhancement module enhances the features and differential features extracted by the residual block, thereby enhancing the efficiency and accuracy of differential fusion. The open-pit mine detection module is used to identify and judge the open-pit mine. The land damage and restoration judgment module classifies and judges the differential features containing the open-pit mine, and finally obtains the land damage and restoration results of the open-pit mine.

[0025] (3) The present invention can realize the prediction of land damage and restoration at the pixel or element level, cut the remote sensing image data of the study area into image blocks for orderly processing, and gradually obtain the distribution data of land damage and restoration results in the study area according to pixels. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 A method flow chart of the land damage and restoration monitoring method of the present invention;

[0027] Figure 2 Schematic diagram of the principle of the land damage and restoration monitoring method in the embodiment;

[0028] Figure 3 : is a principle structure block diagram of a Conv block module of a residual module in an embodiment;

[0029] Figure 4 : is a principle structure block diagram of the Identity block module of the residual module in the embodiment;

[0030] Figure 5 It is a principle structure block diagram of the global difference fusion module in the embodiment;

[0031] Figure 6 It is a principle structure block diagram of the change information enhancement module in the embodiment;

[0032] Figure 7 It is a principle structure block diagram of the open-pit mine detection module in the embodiment;

[0033] Figure 8 It is a principle structure block diagram of the detection information enhancement module in the embodiment. DETAILED DESCRIPTION

[0034] The present invention is further described in detail below in conjunction with embodiments:

[0035] Example

[0036] like Figure 1 As shown, a land damage and restoration monitoring method based on differential feature enhancement and mining area detection includes:

[0037] S1. Collect remote sensing image data (preferably high-resolution remote sensing image data) of two phases, and cut the remote sensing image data into image blocks P1 and P2 of the same area in two phases.

[0038] S2. Build a neural network model and use the sample data set to train the model. The model training method of the neural network model is as follows:

[0039] A1. Make a sample data set, collect remote sensing image data (preferably high-resolution remote sensing image data) of the two phases before and after (such as the two phases T1 and T2 of the study mining area), and perform damage and repair annotation processing to convert it into a raster image. In this embodiment, high-resolution remote sensing images of the two phases T1 and T2 of the study mining area are collected and downloaded. The preferred technical solution requires preprocessing of the remote sensing image data, which includes radiation calibration, atmospheric correction, orthorectification, and image fusion. The damaged and repaired areas in the open-pit mine in the remote sensing image data of the two phases T1 and T2 of the study mining area are respectively annotated and converted into a raster image.

[0040] The remote sensing image data of the two phases and the raster images with annotation information are cropped into several image blocks of the same area in the two phases (the image block size is preferably C×H×W, where C is the number of image bands, M is the number of image rows, and N is the number of image columns) and used as sample data.

[0041] A2, the sample data of the sample data set is divided into a training set (only the training set is subjected to data enhancement processing, specifically including flipping, translation, scale change, contrast change and Gaussian noise, etc., to enhance the generalization ability of the model) and a validation set according to a ratio of 7:3. The training set is used to train the neural network model, and the validation set is used to test the accuracy of the neural network model. The loss function used in the model training process of the present invention is the cross entropy loss function (Cross Entropy Loss, L CE ), the cross entropy loss function consists of two parts, one is the loss value of change detection (i.e., differential feature recognition detection), and the other is the loss value of open-pit mine search. The calculation method is:

[0042]

[0043] ω1 is the loss function of change detection (i.e. differential feature recognition detection) The weight of ω2 is the loss function of the open-pit mine detection task. The weight of , L is the total loss value of the network model, which is used to indicate the error size of the model.

[0044] The change detection loss function and the open-pit mine detection task loss function both use the following loss function:

[0045]

[0046] In the formula, is the true category, y i is the result predicted by the model, and N is the total number of pixels or pixels or samples.

[0047] Set hyperparameters such as the number of iterative training times, learning rate, batch size, optimizer, and perform multiple iterative training. In each iterative training, use the gradient descent algorithm to reduce the model loss value, and optimize and update the model parameters (the parameters of the neural network model are the weight values ​​of the connections between layers in the neural network model).

[0048] This example uses Shenhua Coal as a research mining area and an implementation case, collects GF-6 remote sensing images of the area in 2019 and 2020, performs data preprocessing, and produces a sample data set for open-pit mine land damage and restoration monitoring. The specific method is as follows:

[0049] A1. Collect or download the GF-6 remote sensing images of the Shenhua mining area in 2019 and 2020, and perform image preprocessing including radiation calibration, atmospheric correction, orthorectification, and image fusion.

[0050] A2. Use geographic information software such as ArcMap to mark the land use change areas of open-pit mines in the remote sensing images of the area in 2019 and 2020, that is, create line vector files, and convert the marked vector files into raster images through vector-to-raster conversion.

[0051] A3. Use ArcMap to crop the remote sensing image data of the Shenhua mining area in 2019 and 2020 and the sample raster images obtained in step A3 into several 256 pixel × 256 pixel image blocks of the same area, and divide the cropped data into a training set and a validation set in a ratio of 7:3.

[0052] A4. Perform data augmentation only on the training set: flipping, translation, scale change, contrast change, and Gaussian noise to enhance the generalization ability of the model.

[0053] The neural network model of the present invention includes a residual module, a change information enhancement module, a global differential fusion module, an open-pit mine detection module and a land damage and restoration judgment module. The residual module extracts features from the image blocks P1 and P2 and inputs them into the global differential fusion module to obtain differential features with both local and global information. Then, the change information enhancement module uses the differential features to enhance the features extracted by the residual module. In the example case using Shenhua Coal as the research mining area, the neural network model is built using the pytorch deep learning framework, and the network parameters of the neural network model are set as shown in the following table:

[0054]

[0055] The server configuration of the neural network model is as follows:

[0056]

[0057]

[0058] The neural network model is trained using the training set, and the model accuracy after each iteration of training is tested using the validation set. The accuracy of the neural network model is tested using three classification evaluation indicators: Precision, Recall, and F1-score. Set the number of iterations of training epochs and perform multiple iterations of training. In each iteration of training, the gradient descent algorithm is used to reduce the model loss value, while optimizing and updating the model parameters (the model parameters refer to the weight values ​​of the connections between layers in the neural network).

[0059] In some embodiments, Figure 2 As shown, the residual module in step S2 includes several residual blocks (four residual blocks are used as an example in this embodiment), each residual block includes a Conv block module and an Identity block module, such as Figure 3 As shown in the figure, the Conv block module includes a convolution processing main path and a jump connection layer connected to the input and output ends of the convolution processing main path. The convolution processing main path includes 3 convolution layers (the convolution kernels are 1×1, 3×3, and 1×1 respectively), and the jump connection layer includes a convolution layer (the convolution kernel size is 1×1). The Conv block module extracts features while expanding the number of feature map channels, which can extract more deep features and reduce the size of the feature map, alleviating the calculation pressure. Figure 4 As shown in the figure, the Identity block module includes a convolution processing main path and a jump connection layer connected to the input and output ends of the convolution processing main path. The convolution processing main path contains 3 convolution layers. The Identity block module is used to extract features while maintaining the size and number of channels of the feature map, and to directly propagate low-level features to high-level features, so that information can be transmitted more smoothly. The residual module extracts several layers of features from image blocks P1 and P2 respectively. Figure 2 The residual module extracts multiple layers of features from the previous and next phase images, and the size of the multiple layers of features is 64×256 respectively. 2 , 96×128 2 , 128×64 2 , 256×32 2 , respectively marked as {E t1-1 , E t1-2 , E t1-3 , Et1-4} and {E t2-1 , E t2-2 , E t2-3 , E t2-4}.

[0060] In some embodiments, Figure 5 As shown, the global differential fusion module (capable of differential fusion of features of each dimension) extracts features of two time phases from the residual module through the perceptron, and uses two branches to obtain the query Q, key K and value V corresponding to the two time phases. After the two branches generate the query Q, the query Q of the two branches is differentially operated to obtain a differential query Q′; the differential query Q′ is used to perform matrix multiplication operation with the key K and value V of each branch, and the matrix multiplication results obtained from the two branches are absolutely differentially operated to obtain differential features. Through interactive feature fusion, the feature information at different time points is integrated to obtain a feature expression with more temporal consistency; the two obtained feature features are further absolutely differentially operated to obtain the final global differential feature. The present invention creatively uses query Q′ as a shared query vector, so that the two feature branches can be compared and analyzed in a common space, thereby extracting global change features. The global differential fusion module of the present invention can effectively reduce the interference of background noise and focus on real scene change information. The features of each layer are fused through the global differential fusion module to obtain multi-layer differential features with both local and global information. The size is 64×2562, 96×1282, 128×64 2 , 256×32 2 , marked as {D t-1 , D t-2 , D t-3 , D t-4}.

[0061] In some embodiments, Figure 6 As shown, the processing method of the change information enhancement module is as follows: extract the initial features from the residual module through the perceptron and generate the matrix of the query Q, extract the differential features through the perceptron to calculate the matrix of the key K and the value V; multiply the matrix of the query Q with the matrix of the key K to obtain the similarity matrix; perform Softmax normalization on the similarity matrix to obtain the attention weight matrix, and multiply the attention weight matrix with the matrix of the value V to generate the enhanced differential features; the change information enhancement module of the present invention can enhance the recognition and differential fusion of fine features, and significantly improve the performance of the neural network model in complex scenes.

[0062] In some embodiments, Figure 7As shown, the open-pit mine detection module (used to determine whether there is an open-pit mine) detection method includes: dividing the input image into several small blocks, the input image is a differential feature, all the small blocks are linearly projected and flattened into a feature vector of fixed length; using the Transformer encoder module to capture the relationship between each small block through the self-attention mechanism to obtain a feature vector containing global context information and local details (through the self-attention mechanism of the Transformer, the model's understanding of the image content is effectively improved, and it can effectively distinguish complex visual feature recognition), determine whether it contains open-pit mine area features, and obtain detection features for open-pit mine detection (enhanced differential features). In the training stage of the model, regardless of whether the open-pit mine detection module determines whether there is an open-pit mine, the neural network model must perform subsequent land damage and restoration judgment module processing; in the testing stage of the model or the application stage after training, if the open-pit mine detection module determines that there is no open-pit mine, the neural network model terminates subsequent processing. The neural network model of the present invention focuses on land damage and restoration monitoring in the research mining area.

[0063] S3. The open-pit mine detection module uses differential features to determine whether the open-pit mine area is included. The land damage and restoration judgment module uses differential features to perform classification judgment and output the judgment results, which include land damage and land restoration. After multiple iterations of training, the neural network model with the highest accuracy and the best visual effect is selected, and the remote sensing image (C×M×N) of the area to be monitored is input into the trained neural network model. If the result of the open-pit mine detection module is No, that is, there is no open-pit mine in the area, no subsequent processing is required. If the result of the open-pit mine detection module is Yes, subsequent land damage and restoration monitoring is required to obtain the land damage and restoration monitoring judgment results of the open-pit mine (the judgment results include land damage and land restoration).

[0064] In some embodiments, the open-pit mine detection module in step S3 detects whether the differential features corresponding to the image blocks P1 and P2 contain features of the open-pit mine area, and before determining whether the open-pit mine area is included, it also includes a detection information enhancement module, which performs information enhancement processing on the differential features. Figure 8 As shown in FIG, the processing method of the detection information enhancement module is as follows: the differential features are processed by the perceptron to obtain the query Q, the detection features are processed by the perceptron to obtain the key K and the value V, and the matrix of the query Q is multiplied by the matrix of the key K to obtain the similarity matrix; the similarity matrix is ​​subjected to Softmax normalization to obtain the attention weight matrix (on the basis of the original differential information features, the information of the detection features is further integrated to make the model more sensitive to the details of the changes), and the attention weight matrix is ​​multiplied by the matrix of the value V to generate the enhanced differential features.

[0065] A land damage and restoration monitoring system based on differential feature enhancement and mining area detection comprises a data acquisition module, a sample data set and a neural network model. The data acquisition module is used to collect remote sensing image data of two phases before and after, and cut the remote sensing image data into image blocks P1 and P2 of the same area in the two phases before and after; the neural network model comprises a residual module, a change information enhancement module, a global differential fusion module, an open-pit mine detection module and a land damage and restoration judgment module. The residual module extracts features from the image blocks P1 and P2 and inputs the features into the global differential fusion module to obtain differential features with both local and global information. Then, the change information enhancement module uses the differential features to perform feature information enhancement processing on the features extracted by the residual module; the open-pit mine detection module uses the differential features to judge whether the open-pit mine area is included, and the land damage and restoration judgment module uses the differential features to perform classification judgment and output the judgment results, and the judgment results include land damage and land restoration; the neural network model uses the sample data set to perform model training.

[0066] The land damage and restoration monitoring system of the present invention further includes a detection information enhancement module, and the open-pit mine detection module detects whether the differential features corresponding to the image blocks P1 and P2 include features of the open-pit mine area; before determining whether the open-pit mine area is included, the detection information enhancement module performs information enhancement processing on the differential features; the model training method of the neural network model is as follows:

[0067] A1. Prepare a sample data set. Collect remote sensing image data of the previous and next two time phases, annotate them for damage and repair, and convert them into raster images. Cut the remote sensing image data of the previous and next two time phases and the raster images with annotation information into several image blocks of the same area of ​​the previous and next two time phases and use them as sample data.

[0068] A2. Divide the sample data of the sample data set into a training set and a validation set in a ratio of 7:3. The training set is used to train the neural network model, and the validation set is used to test the accuracy of the neural network model.

[0069] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A land damage and restoration monitoring method based on differential feature enhancement and mining area detection, characterized by: The methods include: S1, collecting remote sensing image data of two phases, and cutting the remote sensing image data into image blocks P1 and P2 of the same area in the two phases; S2. Construct a neural network model and use the sample data set to train the model. The neural network model includes a residual module, a change information enhancement module, a global difference fusion module, an open-pit mine detection module, and a land damage and restoration judgment module. The residual module extracts features from the image blocks P1 and P2 and inputs them into the global difference fusion module to obtain differential features with both local and global information. Then, the change information enhancement module uses the differential features to enhance the features extracted by the residual module; S3. The open-pit mine detection module uses differential features to determine whether an open-pit mine area is included. The land damage and restoration judgment module uses differential features to perform classification judgment and output judgment results, which include land damage and land restoration.

2. The land damage and restoration monitoring method based on differential feature enhancement and mining area detection according to claim 1 is characterized by: In step S3, the open-pit mine detection module detects whether the differential features corresponding to the image blocks P1 and P2 include features of the open-pit mine area. Before determining whether the open-pit mine area is included, it also includes a detection information enhancement module, which performs information enhancement processing on the differential features.

3. The land damage and restoration monitoring method based on differential feature enhancement and mining area detection according to claim 1 is characterized by: In step S2, the residual module includes several residual blocks, each residual block includes a Conv block module and an Identity block module, the Conv block module includes a convolution processing main path and a jump connection layer connected to the input and output ends of the convolution processing main path, the convolution processing main path includes 3 convolution layers, and the jump connection layer includes a convolution layer; the Identity block module includes a convolution processing main path and a jump connection layer connected to the input and output ends of the convolution processing main path, the convolution processing main path includes 3 convolution layers; the residual module extracts several layers of features for image blocks P1 and P2 respectively.

4. The land damage and restoration monitoring method based on differential feature enhancement and mining area detection according to claim 1 is characterized by: The global difference fusion module extracts the features of the two time phases from the residual module through the perceptron, obtains the query Q, key K and value V corresponding to the two time phases by using two branches, and after the two branches generate the query Q, performs a difference operation on the query Q of the two branches to obtain a differential query Q′; Use the differential query Q′ to perform matrix multiplication operation with the key K and value V of each branch, and perform absolute difference operation on the matrix multiplication results obtained from the two branches to obtain differential features.

5. The land damage and restoration monitoring method based on differential feature enhancement and mining area detection according to claim 1 is characterized by: The processing method of the change information enhancement module is as follows: extract the initial features from the residual module through the perceptron and generate the matrix of the query Q, extract the differential features through the perceptron to calculate the matrix of the key K and the value V; multiply the matrix of the query Q with the matrix of the key K to obtain the similarity matrix; perform Softmax normalization on the similarity matrix to obtain the attention weight matrix, and multiply the attention weight matrix with the matrix of the value V to generate the enhanced differential features.

6. The land damage and restoration monitoring method based on differential feature enhancement and mining area detection according to claim 1 is characterized by: The open-pit mine detection module detection method includes: dividing an input image into a plurality of small blocks, the input image is a differential feature, linearly projecting all the small blocks and flattening them into a feature vector of a fixed length; using a Transformer encoder module to capture the relationship between each small block through a self-attention mechanism to obtain a feature vector containing global context information and local details, judging whether it contains open-pit mine area features, and obtaining detection features for enhancing differential features.

7. The land damage and restoration monitoring method based on differential feature enhancement and mining area detection according to claim 6 is characterized by: The detection information enhancement module processing method is as follows: the differential features are processed by a perceptron to obtain a query Q, the detection features are processed by a perceptron to obtain a key K and a value V, and the matrix of the query Q is multiplied by the matrix of the key K to obtain a similarity matrix; The similarity matrix is ​​normalized by Softmax to obtain the attention weight matrix, and the attention weight matrix is ​​multiplied by the matrix of value V to generate the enhanced differential features.

8. The land damage and restoration monitoring method based on differential feature enhancement and mining area detection according to claim 1 is characterized by: The model training method of the neural network model is as follows: A1. Prepare a sample data set. Collect remote sensing image data of the previous and next two time phases, annotate them for damage and repair, and convert them into raster images. Cut the remote sensing image data of the previous and next two time phases and the raster images with annotation information into several image blocks of the same area of ​​the previous and next two time phases and use them as sample data. A2. Divide the sample data of the sample data set into a training set and a validation set in a ratio of 7:

3. The training set is used to train the neural network model, and the validation set is used to test the accuracy of the neural network model.

9. A land damage and restoration monitoring system based on differential feature enhancement and mining area detection, characterized by: The invention comprises a data acquisition module, a sample data set and a neural network model. The data acquisition module is used to acquire remote sensing image data of two phases before and after, and to crop the remote sensing image data into image blocks P1 and P2 of the same area in the two phases before and after. The neural network model comprises a residual module, a change information enhancement module, a global differential fusion module, an open-pit mine detection module and a land damage and restoration judgment module. The residual module extracts features of the image blocks P1 and P2 and inputs the features into the global differential fusion module to obtain differential features with both local and global information. Then, the change information enhancement module uses the differential features to perform feature information enhancement processing on the features extracted by the residual module. The open-pit mine detection module uses the differential features to judge whether an open-pit mine area is included. The land damage and restoration judgment module uses the differential features to perform classification judgment and output the judgment results, and the judgment results include land damage and land restoration. The neural network model uses the sample data set to perform model training.

10. The land damage and restoration monitoring system based on differential feature enhancement and mining area detection according to claim 9 is characterized in that: It also includes a detection information enhancement module, and the open-pit mine detection module detects whether the differential features corresponding to the image blocks P1 and P2 contain the features of the open-pit mine area; before judging whether the open-pit mine area is included, the detection information enhancement module performs information enhancement processing on the differential features; the model training method of the neural network model is as follows: A1. Prepare a sample data set. Collect remote sensing image data of the previous and next two time phases, annotate them for damage and repair, and convert them into raster images. Cut the remote sensing image data of the previous and next two time phases and the raster images with annotation information into several image blocks of the same area of ​​the previous and next two time phases and use them as sample data. A2. Divide the sample data of the sample data set into a training set and a validation set in a ratio of 7:

3. The training set is used to train the neural network model, and the validation set is used to test the accuracy of the neural network model.

Images