Methods, models, systems, and computer-readable storage media for processing remote sensing images of geological and mineral resources

Abstract

This application relates to the field of remote sensing image processing technology, and more particularly to a method, model, system, and computer-readable storage medium for processing geological and mining remote sensing images. The method involves performing shared-weighted downsampling convolutions layer by layer on two input geological and mining remote sensing images taken at two different times. A first temporal attention feature is extracted from the first geological and mining remote sensing feature image using a channel attention mechanism, and a second temporal attention feature is extracted from the second geological and mining remote sensing feature image. Further processing using a spatial attention mechanism yields a first spatial attention feature and a second spatial attention feature. The first and second spatial attention features are then concatenated through skip connections to obtain a fused feature map. A convolution operation is performed on the fused feature map to generate a change probability map. Regions in the change probability map exceeding a preset threshold are identified as changed regions, and the rest are identified as unchanged regions. This aims to solve the problem of improving the accuracy of change detection in geological and mining remote sensing images.

Description

Technical Field

[0001] This application relates to the field of remote sensing image processing technology, and in particular to a method, model, system, and computer-readable storage medium for processing geological and mining remote sensing images. Background Technology

[0002] Change recognition in remote sensing images refers to the analysis of remote sensing images of the same area acquired at different times to detect changes in surface features. Currently, common change recognition methods are object-based change detection methods—that is, first segmenting the image into objects, and then comparing the object features at different times to detect changes in the remote sensing image. However, this method typically only focuses on pixel or object changes between two remote sensing images themselves, and cannot identify changes in a sequence of remote sensing images acquired over a period of time, making it difficult to accurately identify changes in remote sensing images over long periods.

[0003] Currently, some studies have combined deep learning with the recognition of changes in remote sensing images. For example, Chinese patent application number 202411466327.8 discloses a method, device, equipment and storage medium for detecting changes in remote sensing images. The paper mentions using recurrent neural networks (RNNs) in deep learning networks to capture the change information of images over a long time series.

[0004] However, during the process of conceiving and implementing this application, the inventors discovered that in the field of geology and mining, methods such as mining and blasting directly alter the surface morphology and geological structure, leading to the collapse of open-pit mines or underground mining voids. This can easily cause obvious local changes in remote sensing images within a short period of time. Relying solely on traditional RNN network models for change recognition in remote sensing images makes it difficult to effectively identify and locate sudden, small-scale, or rare changes in geology and mining remote sensing images. Furthermore, it is impossible to accurately locate the boundaries and shapes of the changed areas, resulting in recognition accuracy that cannot meet industrial needs. Summary of the Invention

[0005] The main objective of this application is to provide a method for processing geological and mineral remote sensing images, aiming to solve the problem of how to improve the accuracy of change detection in geological and mineral remote sensing images.

[0006] To achieve the above objectives, this application provides a method for processing geological and mineral remote sensing images, the method comprising:

[0007] S10, perform shared-weight downsampling convolution on the two different time-time geological and mineral remote sensing images layer by layer to obtain the first geological and mineral remote sensing feature map and the second geological and mineral remote sensing feature map;

[0008] S20, a channel attention mechanism is used to extract the first temporal attention feature from the first geological and mineral remote sensing feature map, and to extract the second temporal attention feature from the second geological and mineral remote sensing feature map;

[0009] S30, the first temporal attention feature is processed using a spatial attention mechanism to obtain the first spatial attention feature, and the second temporal attention feature is processed to obtain the second spatial attention feature;

[0010] S40, the first spatial attention feature and the second spatial attention feature are concatenated through a skip connection to obtain a fused feature map;

[0011] S50, perform a convolution operation on the fused feature map to generate a change probability map, and determine the regions in the change probability map that are greater than a preset threshold as change regions, and the rest as non-change regions.

[0012] Optionally, after S50, the following steps are also included:

[0013] S60, calculate the binary cross-entropy corresponding to the change probability map, and the overlap loss corresponding to the change region;

[0014] S70, based on the binary cross-entropy and the overlap loss, a loss function is determined, thereby adjusting the hyperparameters of the deep learning model with the goal of minimizing the loss function.

[0015] Optionally, in step S60, the expression for calculating the binary cross-entropy is:

[0016]

[0017] In the formula, For binary cross-entropy, It is the i-th real label. This is the probability map of the i-th change predicted by the model. Total number of pixels;

[0018] The expression for calculating the overlap loss is as follows:

[0019]

[0020] In the formula, For overlap loss, Indicates the predicted area of ​​change The intersection with the actual change region z.

[0021] Optionally, S40 specifically includes:

[0022] S41, using dilated convolutions with different dilation rates to generate the first spatial attention features. Second spatial attention features Corresponding multi-scale feature sets:

[0023]

[0024]

[0025] In the formula, This represents the multi-scale feature set corresponding to the first spatial attention feature. Multiscale feature factors with index n in the middle. This represents the multi-scale feature set corresponding to the second spatial attention features. In the multi-scale feature factor with index n, where n represents the number of features, ... and Multiscale feature factors with the same ordinal number have the same inflation rate;

[0026] S42, forming feature pairs from multi-scale feature factors with the same expansion rate. Design a dynamic weight generator for features at each scale. :

[0027]

[0028] In the formula, For spatially adaptive weighted graphs, ;

[0029] S43, weighted fusion of feature pairs yields initial fused features at multiple scales:

[0030]

[0031] S44, calculate the cross-temporal attention map corresponding to the initial fusion features at each scale, and then... With initial fusion features Multiplication:

[0032]

[0033] Obtain the feature set after multi-scale interaction ;

[0034] S45, the feature groups after multi-scale interaction By upsampling to align the resolution, the fused feature map is obtained by sequentially stitching and compressing the data. :

[0035]

[0036] Furthermore, to achieve the above objectives, this application also provides a geological and mineral remote sensing image processing model, characterized in that the geological and mineral remote sensing image processing model includes:

[0037] The convolutional downsampling module is used to perform shared-weight downsampling convolution on the two different time-time geological and mineral remote sensing images layer by layer to obtain the first geological and mineral remote sensing feature map and the second geological and mineral remote sensing feature map.

[0038] The convolutional block attention module is used to extract first temporal attention features from the first geological and mineral remote sensing feature map and second temporal attention features from the second geological and mineral remote sensing feature map using a channel attention mechanism; and to process the first temporal attention features using a spatial attention mechanism to obtain first spatial attention features and second temporal attention features to obtain second spatial attention features.

[0039] A multi-scale feature fusion module is used to concatenate the first spatial attention feature and the second spatial attention feature through a skip connection to obtain a fused feature map;

[0040] The change detection head network is used to perform convolution operations on the fused feature map to generate a change probability map. Regions in the change probability map that are greater than a preset threshold are identified as change regions, and the rest are identified as non-change regions.

[0041] Optionally, the geological and mining remote sensing image processing model includes at least two convolutional layers, and the convolutional block attention module is inserted between every two convolutional layers of the geological and mining remote sensing image processing model to realize channel attention mechanism extraction and spatial attention mechanism extraction through the channel attention module and spatial attention module in the convolutional block attention module.

[0042] Optionally, the convolutional downsampling module includes a Siamese structure, which assigns shared weights to the convolutional layers when the model implements downsampling convolution.

[0043] Optionally, the multi-scale feature fusion module includes:

[0044] A multi-scale spatial attention feature extraction unit is used to generate the first spatial attention feature using dilated convolutions with different dilation rates. Second spatial attention features Corresponding multi-scale feature sets:

[0045]

[0046]

[0047] In the formula, This represents the multi-scale feature set corresponding to the first spatial attention feature. Multiscale feature factors with index n in the middle. This represents the multi-scale feature set corresponding to the second spatial attention features. In the multi-scale feature factor with index n, where n represents the number of features, ... and Multiscale feature factors with the same ordinal number have the same inflation rate;

[0048] The cross-temporal dynamic weight generation unit is used to construct feature pairs from multi-scale feature factors with the same expansion rate. Design a dynamic weight generator for features at each scale. :

[0049]

[0050] In the formula, For spatially adaptive weighted graphs, ;

[0051] And the fused features are obtained by weighting and fusing the various feature pairs:

[0052]

[0053] A multi-scale feature interaction enhancement unit is used to calculate the cross-temporal attention map corresponding to the fused features at each scale, and to process the cross-temporal attention map. With fusion features Multiplication:

[0054]

[0055] Obtain the feature set after multi-scale interaction ;

[0056] Progressive feature aggregation unit, used to aggregate feature groups after multi-scale interaction By upsampling to align the resolution, the fused feature map is obtained by sequentially stitching and compressing the data. :

[0057]

[0058] In addition, to achieve the above objectives, this application also provides a geological and mineral remote sensing image processing system, which includes: a memory, a processor, and a geological and mineral remote sensing image processing program stored in the memory and executable on the processor. When the geological and mineral remote sensing image processing program is executed by the processor, it implements the steps of the geological and mineral remote sensing image processing method as described in any of the above claims.

[0059] In addition, to achieve the above objectives, this application also provides a computer-readable storage medium storing a geological and mineral remote sensing image processing program, which, when executed by a processor, implements the steps of the geological and mineral remote sensing image processing method as described in any of the preceding claims.

[0060] This application has at least the following beneficial effects:

[0061] 1. Acquire dual-temporal remote sensing images of geological and mineral resources, and through channel-spatial attention cascade, first extract important channel features in the temporal dimension, then focus on salient areas in the space through spatial attention, and finally achieve alignment and information complementarity between dual-temporal features (i.e., feature maps of geological and mineral resources remote sensing images at two different times) through skip connections.

[0062] 2. The accuracy and region consistency of pixel-level classification are supervised by the loss function, and the accuracy and robustness of change detection are balanced by adjusting the hyperparameters with dynamic weights, thereby enhancing the model's generalization ability to multi-scale and multi-type changes.

[0063] 3. The spatial attention features of the two temporal phases of the input are decomposed using a multi-scale pyramid, and dilated convolution (DC) with different dilation rates is used to capture the changing regions at different scales in the geological and mineral remote sensing images.

[0064] 4. Design a dynamic weight generator (DWG) to generate dynamic weights across time phases, thereby suppressing the generation of non-changing noise such as light intensity changes in geological and mineral remote sensing images;

[0065] 5. Input the fusion features of each scale into the cross-attention module. The cross-attention module calculates the cross-temporal attention map corresponding to the initial fusion features of each scale. Multiply the cross-temporal attention map with the initial fusion features. The resulting multi-scale interactive feature group is upsampled to align the resolution, and then stitched and compressed step by step to obtain the final required fusion feature map, thus realizing multi-scale feature interaction enhancement. Attached Figure Description

[0066] Figure 1 This is a flowchart illustrating the first embodiment of the geological and mineral remote sensing image processing method of this application.

[0067] Figure 2 This is a schematic diagram of the architecture of the geological and mineral remote sensing image processing model involved in the embodiments of this application;

[0068] Figure 3 This is a schematic diagram of the architecture of the scale feature fusion module involved in the embodiments of this application;

[0069] Figure 4 This is a schematic diagram of the architecture of the geological and mining remote sensing image processing system involved in the embodiments of this application.

[0070] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0071] To better understand the above technical solutions, exemplary embodiments of this disclosure will be described in more detail below with reference to the accompanying drawings. While exemplary embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of this disclosure to those skilled in the art.

[0072] First Embodiment

[0073] Reference Figure 1 This embodiment provides a method for processing geological and mineral remote sensing images, the method comprising the following steps:

[0074] S10, perform shared-weight downsampling convolution on the two different time-time geological and mineral remote sensing images layer by layer to obtain the first geological and mineral remote sensing feature map and the second geological and mineral remote sensing feature map;

[0075] In this embodiment, the encoder uses multi-layer convolution and pooling operations to downsample the input geological remote sensing images at different times layer by layer, and assigns shared weights to the convolutional layers in the encoder during the downsampling process.

[0076] The different times mentioned include a first time and a second time, and the geological and mineral remote sensing feature map corresponding to the first time is the first geological and mineral remote sensing feature map, and the geological and mineral remote sensing feature map corresponding to the second time is the second geological and mineral remote sensing feature map.

[0077] It should be noted that in this embodiment, shared weights refer to the repeated use of the same set of weight parameters in different modules or locations, sharing the weights at different time steps in the recurrent neural network, so that the model can learn the temporal dependencies in the sequence.

[0078] S20, a channel attention mechanism is used to extract the first temporal attention feature from the first geological and mineral remote sensing feature map, and to extract the second temporal attention feature from the second geological and mineral remote sensing feature map;

[0079] In this embodiment, a temporal attention module is introduced between different levels of feature extraction to calculate attention weights between corresponding feature maps. The purpose of calculating attention weights is to emphasize the feature regions that require special attention in the geological and mineral remote sensing feature maps. In this embodiment, the regions that change between two geological and mineral remote sensing feature maps are used as the feature regions.

[0080] Optionally, the feature region can be a region in the geological and mineral remote sensing image where the spectral characteristics change significantly.

[0081] For example, let the first geological and mineral remote sensing feature map be... The second geological and mineral remote sensing feature map is After performing shared-weight downsampling convolution, the first geological and mineral remote sensing feature map is obtained. and the second geological and mineral remote sensing feature map Weights are generated through channel attention mechanism. and Its expression is:

[0082]

[0083]

[0084] in:

[0085]

[0086]

[0087] In the formula, This represents the learnable temporal attention weight matrix, with dimension 1. ; softmax represents the function used to normalize the weights generated in the time dimension.

[0088] It should be noted that the purpose of the channel attention mechanism is to enable the model to capture long-term series changes. This mechanism highlights important change information between different time points by calculating the attention weights of multi-temporal images. It not only focuses on changes in a single time phase, but can also handle complex change patterns in long-term series. In other words, it enables the model to not only focus on the feature changes between two geological and mineral remote sensing feature maps at two time points, but also on the feature changes between multiple geological and mineral remote sensing feature maps collected over a period of time.

[0089] S30, the first temporal attention feature is processed using a spatial attention mechanism to obtain the first spatial attention feature, and the second temporal attention feature is processed to obtain the second spatial attention feature;

[0090] In this embodiment, a spatial attention module is applied to the temporal attention feature map to enhance the feature representation of key areas and suppress the interference of irrelevant areas in the geological and mineral remote sensing image on the key change areas of interest.

[0091] For example, the first spatial attention feature after applying the spatial attention mechanism. Second spatial attention features Represented as:

[0092]

[0093]

[0094] in,

[0095]

[0096]

[0097] In the formula, Sigmoid() represents the activation function. ( ) represents the convolution function. This represents the average pooling function. () represents the max pooling function.

[0098] It should be noted that in this step, a spatial attention mechanism is applied to the feature map after temporal attention processing. This mechanism combines the weight calculation of max pooling and average pooling to improve the sensitivity of change detection to spatial features, further strengthen the feature representation of key regions, suppress the influence of irrelevant regions, and thus enhance the detection accuracy of change regions.

[0099] S40, the first spatial attention feature and the second spatial attention feature are concatenated through a skip connection to obtain a fused feature map;

[0100] Furthermore, in the decoder section, the first spatial attention features and the second spatial attention features obtained from the encoder are fused and spliced ​​together by skip connections to obtain a fused feature map.

[0101] In this embodiment, the purpose of introducing skip connections is to preserve the detailed information of low-level features during the upsampling process. Skip connections are extended from the traditional inter-layer transfer within the same network to the fusion of cross-temporal features, solving the problem of alignment and information complementarity between dual-temporal features (i.e., feature maps of geological and mineral remote sensing images at two different times).

[0102] The fused feature map is represented as , can be expressed as:

[0103]

[0104] In the formula, This indicates a jump connection operation.

[0105] S50, perform a convolution operation on the fused feature map to generate a change probability map, and determine the regions in the change probability map that are greater than a preset threshold as change regions, and the rest as non-change regions.

[0106] After obtaining the fused feature map, a convolutional layer is used to locally enhance the fused features, highlighting the areas of change and making the features focus more on areas of significant change. Finally, a change probability map and a binarized change map are output, thereby achieving accurate localization of the change areas.

[0107] For example, let the fused feature map A probability map is generated through convolution operations. Its expression is:

[0108]

[0109] In the formula, This is a probability plot, where R is a real number. It's about height. It is wide; Let be the Sigmoid activation function representing the probability of change.

[0110] By setting a threshold The probability change map is binarized to generate the final change map. :

[0111]

[0112] The regions marked as 1 are considered as changing regions, and the regions marked as 0 are considered as non-changing regions.

[0113] In the technical solution provided in this embodiment, dual-temporal remote sensing images of geological and mineral resources are acquired. Through channel-spatial attention cascade, important channel features in the temporal dimension are first extracted, and then spatial attention is used to focus on significant areas in space. Finally, skip connections are used to achieve alignment and information complementarity between dual-temporal features (i.e., feature maps of geological and mineral remote sensing images at two different times).

[0114] Second Embodiment

[0115] Based on the first embodiment, this embodiment introduces a multi-task joint loss function to supervise the accuracy of the aforementioned method, ensuring the accuracy of pixel-level classification and region consistency. Specifically, after step S50, the method further includes:

[0116] S60, calculate the binary cross-entropy corresponding to the change probability map, and the overlap loss corresponding to the change region;

[0117] Specifically, the expression for calculating the binary cross-entropy is:

[0118]

[0119] In the formula, For binary cross-entropy, It is the i-th real label. This is the probability map of the i-th change predicted by the model. Total number of pixels;

[0120] Specifically, the expression for calculating the overlap loss is as follows:

[0121]

[0122] In the formula, For overlap loss, Indicates the predicted area of ​​change The intersection with the actual change region z.

[0123] S70, based on the binary cross-entropy and the overlap loss, a loss function is determined, thereby adjusting the hyperparameters of the deep learning model with the goal of minimizing the loss function.

[0124] Specifically, the expression for the loss function is:

[0125]

[0126] In the formula The hyperparameter is used to adjust the balance between the two losses.

[0127] In the technical solution provided in this embodiment, a loss function is formed by combining binary cross-entropy loss and overlap loss. The loss function is used to supervise the accuracy of pixel-level classification and the consistency of regions. The hyperparameters are adjusted by dynamic weights to balance the accuracy and robustness of change detection, thereby enhancing the model's ability to generalize to changes of multiple scales and types.

[0128] Third Embodiment

[0129] Based on any embodiment, in this embodiment, in the detection of changes in dual-temporal remote sensing images, traditional skip connections typically directly stitch together or add spatial attention features from different times. However, the contributions of features at different times to the changed region are different. If features at a single scale are used, it is difficult to capture subtle changes between features in dual-temporal remote sensing images and structural information of complex regions. In addition, the light intensity changes in geological and mining remote sensing images are significant, easily generating a lot of non-changing noise, which can cause the model to ignore the correlation between features of the two real images. Therefore, this embodiment further proposes a dynamic multi-scale interactive skip fusion module, the specific steps of which are as follows:

[0130] S41, using dilated convolutions with different dilation rates to generate the first spatial attention features. Second spatial attention features Corresponding multi-scale feature sets:

[0131]

[0132]

[0133] In the formula, This represents the multi-scale feature set corresponding to the first spatial attention feature. Multiscale feature factors with index n in the middle. This represents the multi-scale feature set corresponding to the second spatial attention features. In the multi-scale feature factor with index n, where n represents the number of features, ... and Multiscale feature factors with the same ordinal number have the same inflation rate;

[0134] First, the spatial attention features of the two temporal phases of the input are decomposed using a multi-scale pyramid, and dilated convolution (DC) with different dilation rates is used to capture the changing regions at different scales in the geological and mineral remote sensing images.

[0135] S42, forming feature pairs from multi-scale feature factors with the same expansion rate. Design a dynamic weight generator for features at each scale. :

[0136]

[0137] In the formula, For spatially adaptive weighted graphs, ;

[0138] S43, weighted fusion of feature pairs yields initial fused features at multiple scales:

[0139]

[0140] In steps S42 and S43, a dynamic weight generator (DWG) is designed to generate dynamic weights across time phases, thereby suppressing the generation of non-changing noise such as light intensity changes in geological and mineral remote sensing images.

[0141] S44, calculate the cross-temporal attention map corresponding to the initial fusion features at each scale, and then... With initial fusion features Multiplication:

[0142]

[0143] Obtain the feature set after multi-scale interaction ;

[0144] S45, the feature groups after multi-scale interaction By upsampling to align the resolution, the fused feature map is obtained by sequentially stitching and compressing the data. :

[0145]

[0146] Finally, the fusion features at each scale { The input is the Cross-Attention Module (CAM), which calculates the cross-temporal attention map corresponding to the initial fusion features at each scale. The cross-temporal attention map is then processed. With initial fusion features The resulting multi-scale interactive feature sets are upsampled to align the resolution, then concatenated and compressed step by step to obtain the final fused feature map. This enables multi-scale feature interaction enhancement.

[0147] Fourth embodiment

[0148] Based on any embodiment, as an implementation scheme, refer to Figure 2 The diagram shown illustrates the architecture of a geological and mineral remote sensing image processing model, which includes:

[0149] The convolutional downsampling module 100 is used to perform shared-weight downsampling convolution on the two different time-time input geological and mineral remote sensing images layer by layer to obtain the first geological and mineral remote sensing feature map and the second geological and mineral remote sensing feature map.

[0150] The convolutional block attention module 200 is used to extract first temporal attention features from the first geological and mineral remote sensing feature map and second temporal attention features from the second geological and mineral remote sensing feature map using a channel attention mechanism; and to process the first temporal attention features to obtain first spatial attention features and the second temporal attention features to obtain second spatial attention features using a spatial attention mechanism.

[0151] The multi-scale feature fusion module 300 is used to concatenate the first spatial attention feature and the second spatial attention feature through a skip connection to obtain a fused feature map;

[0152] The change detection head network 400 is used to perform convolution operations on the fused feature map to generate a change probability map, and to determine the regions in the change probability map that are greater than a preset threshold as changed regions, and the rest as non-changed regions.

[0153] Optionally, the geological and mining remote sensing image processing model includes at least two convolutional layers, and the convolutional block attention module is inserted between every two convolutional layers of the geological and mining remote sensing image processing model to realize channel attention mechanism extraction and spatial attention mechanism extraction through the channel attention module and spatial attention module in the convolutional block attention module.

[0154] Optionally, the convolutional downsampling module includes a Siamese structure, which assigns shared weights to the convolutional layers when the model implements downsampling convolution.

[0155] Fifth embodiment

[0156] Based on the fourth embodiment, as one implementation scheme, refer to Figure 3 The scale feature fusion module of the dynamic multi-scale interactive skip fusion architecture shown includes:

[0157] Multi-scale spatial attention feature extraction unit 301 is used to generate first spatial attention features using dilated convolutions with different dilation rates. Second spatial attention features Corresponding multi-scale feature sets:

[0158]

[0159]

[0160] In the formula, This represents the multi-scale feature set corresponding to the first spatial attention feature. Multiscale feature factors with index n in the middle. This represents the multi-scale feature set corresponding to the second spatial attention features. In the multi-scale feature factor with index n, where n represents the number of features, ... and Multiscale feature factors with the same ordinal number have the same inflation rate;

[0161] The cross-temporal dynamic weight generation unit 302 is used to form feature pairs from multi-scale feature factors with the same expansion rate. Design a dynamic weight generator for features at each scale. :

[0162]

[0163] In the formula, For spatially adaptive weighted graphs, ;

[0164] And the fused features are obtained by weighting and fusing the various feature pairs:

[0165]

[0166] The multi-scale feature interaction enhancement unit 303 is used to calculate the cross-temporal attention map corresponding to the fused features at each scale, and to apply the cross-temporal attention map... With fusion features Multiplication:

[0167]

[0168] Obtain the feature set after multi-scale interaction ;

[0169] Progressive feature aggregation unit 304 is used to aggregate feature groups after multi-scale interaction. By upsampling to align the resolution, the fused feature map is obtained by sequentially stitching and compressing the data. :

[0170]

[0171] Sixth Embodiment

[0172] Based on any embodiment, using the test set data of the open-source dataset LEVIR-CD as an example, see the performance comparison table of the various models shown below:

[0173] Table 1. Performance Comparison of Various Models

[0174]

[0175] As can be seen, the algorithm model proposed in this application has higher performance in all evaluation metrics than the current algorithm models.

[0176] As one implementation scheme, Figure 4 This is a schematic diagram of the hardware operating environment of the geological and mining remote sensing image processing system involved in the embodiments of this application.

[0177] like Figure 4 As shown, the geological and mining remote sensing image processing system may include: a processor 1001, such as a CPU; a memory 1005; a user interface 1003; a network interface 1004; and a communication bus 1002. The communication bus 1002 is used to enable communication between these components. The user interface 1003 may include a display screen or an input unit such as a keyboard; optionally, the user interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface). The memory 1005 may be a high-speed RAM or a stable, non-volatile memory, such as a disk drive. Optionally, the memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0178] Those skilled in the art will understand that Figure 4The architecture of the geological and mineral remote sensing image processing system shown does not constitute a limitation on the geological and mineral remote sensing image processing system. It may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0179] like Figure 4 As shown, the memory 1005, which serves as a storage medium, may include an operating system, a network communication module, a user interface module, and a geological and mining remote sensing image processing program. The operating system is a program that manages and controls the hardware and software resources of the geological and mining remote sensing image processing system, as well as the operation of the geological and mining remote sensing image processing program and other software or programs.

[0180] exist Figure 4 In the shown geological and mining remote sensing image processing system, the user interface 1003 is mainly used to connect to the terminal and communicate data with the terminal; the network interface 1004 is mainly used to communicate data with the back-end server; and the processor 1001 can be used to call the geological and mining remote sensing image processing program stored in the memory 1005.

[0181] In this embodiment, the geological and mining remote sensing image processing system includes: a memory 1005, a processor 1001, and a geological and mining remote sensing image processing program stored in the memory and executable on the processor, wherein:

[0182] When processor 1001 calls the geological and mineral remote sensing image processing program stored in memory 1005, it performs the following operations:

[0183] S10, perform shared-weight downsampling convolution on the two different time-time geological and mineral remote sensing images layer by layer to obtain the first geological and mineral remote sensing feature map and the second geological and mineral remote sensing feature map;

[0184] S20, a channel attention mechanism is used to extract the first temporal attention feature from the first geological and mineral remote sensing feature map, and to extract the second temporal attention feature from the second geological and mineral remote sensing feature map;

[0185] S30, the first temporal attention feature is processed using a spatial attention mechanism to obtain the first spatial attention feature, and the second temporal attention feature is processed to obtain the second spatial attention feature;

[0186] S40, the first spatial attention feature and the second spatial attention feature are concatenated through a skip connection to obtain a fused feature map;

[0187] S50, perform a convolution operation on the fused feature map to generate a change probability map, and determine the regions in the change probability map that are greater than a preset threshold as change regions, and the rest as non-change regions.

[0188] When processor 1001 calls the geological and mineral remote sensing image processing program stored in memory 1005, it performs the following operations:

[0189] S60, calculate the binary cross-entropy corresponding to the change probability map, and the overlap loss corresponding to the change region;

[0190] S70, based on the binary cross-entropy and the overlap loss, a loss function is determined, thereby adjusting the hyperparameters of the deep learning model with the goal of minimizing the loss function.

[0191] When processor 1001 calls the geological and mineral remote sensing image processing program stored in memory 1005, it performs the following operations:

[0192] The expression for calculating the binary cross-entropy is:

[0193]

[0194] In the formula, For binary cross-entropy, It is the i-th real label. This is the probability map of the i-th change predicted by the model. Total number of pixels;

[0195] The expression for calculating the overlap loss is as follows:

[0196]

[0197] In the formula, For overlap loss, Indicates the predicted area of ​​change The intersection with the actual change region z.

[0198] When processor 1001 calls the geological and mineral remote sensing image processing program stored in memory 1005, it performs the following operations:

[0199] S41, using dilated convolutions with different dilation rates to generate the first spatial attention features. Second spatial attention features Corresponding multi-scale feature sets:

[0200]

[0201]

[0202] In the formula, This represents the multi-scale feature set corresponding to the first spatial attention feature. Multiscale feature factors with index n in the middle. This represents the multi-scale feature set corresponding to the second spatial attention features. In the multi-scale feature factor with index n, where n represents the number of features, ... and Multiscale feature factors with the same ordinal number have the same inflation rate;

[0203] S42, forming feature pairs from multi-scale feature factors with the same expansion rate. Design a dynamic weight generator for features at each scale. :

[0204]

[0205] In the formula, For spatially adaptive weighted graphs, ;

[0206] S43, weighted fusion of feature pairs yields initial fused features at multiple scales:

[0207]

[0208] S44, calculate the cross-temporal attention map corresponding to the initial fusion features at each scale, and then... With initial fusion features Multiplication:

[0209]

[0210] Obtain the feature set after multi-scale interaction ;

[0211] S45, the feature groups after multi-scale interaction By upsampling to align the resolution, the fused feature map is obtained by sequentially stitching and compressing the data. :

[0212]

[0213] Furthermore, those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program includes program instructions and can be stored in a storage medium, which is a computer-readable storage medium. The program instructions are executed by at least one processor in the geological remote sensing image processing system to implement the process steps of the embodiments of the above methods.

[0214] Therefore, this application also provides a computer-readable storage medium storing a geological and mineral remote sensing image processing program, which, when executed by a processor, implements the various steps of the geological and mineral remote sensing image processing method as described in the above embodiments.

[0215] The computer-readable storage medium can be any computer-readable storage medium capable of storing program code, such as a USB flash drive, portable hard drive, read-only memory (ROM), magnetic disk, or optical disk.

[0216] It should be noted that, since the storage medium provided in the embodiments of this application is the storage medium used to implement the methods of the embodiments of this application, those skilled in the art can understand the specific structure and variations of the storage medium based on the methods described in the embodiments of this application, and therefore will not be repeated here. All storage media used in the methods of the embodiments of this application fall within the scope of protection of this application.

[0217] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0218] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0219] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0220] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0221] It should be noted that any reference signs placed between parentheses in the claims should not be construed as limiting the claims. The word "comprising" does not exclude the presence of components or steps not listed in the claims. The word "a" or "an" preceding a component does not exclude the presence of a plurality of such components. This application can be implemented by means of hardware comprising several different components and by means of a suitably programmed computer. In a unit claim enumerating several means, several of these means may be embodied by the same item of hardware. The use of the words first, second, and third, etc., does not indicate any order. These words can be interpreted as names.

[0222] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0223] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for processing remote sensing images of geological and mineral resources, characterized in that, The method includes the following steps: S10, perform shared-weight downsampling convolution on the two different time-time geological and mineral remote sensing images layer by layer to obtain the first geological and mineral remote sensing feature map and the second geological and mineral remote sensing feature map; S20, a channel attention mechanism is used to extract the first temporal attention feature from the first geological and mineral remote sensing feature map, and to extract the second temporal attention feature from the second geological and mineral remote sensing feature map; S30, the first temporal attention feature is processed using a spatial attention mechanism to obtain the first spatial attention feature, and the second temporal attention feature is processed to obtain the second spatial attention feature; S40, the first spatial attention feature and the second spatial attention feature are concatenated through a skip connection to obtain a fused feature map; S50, perform a convolution operation on the fused feature map to generate a change probability map, and determine the regions in the change probability map that are greater than a preset threshold as change regions, and the rest as non-change regions; S40 specifically includes: S41, using dilated convolutions with different dilation rates to generate the first spatial attention features. Second spatial attention features Corresponding multi-scale feature sets: ； ； In the formula, This represents the multi-scale feature set corresponding to the first spatial attention feature. Multiscale feature factors with index n in the middle. This represents the multi-scale feature set corresponding to the second spatial attention features. In the multi-scale feature factor with index n, where n represents the number of features, ... and Multiscale feature factors with the same ordinal number have the same inflation rate; S42, forming feature pairs from multi-scale feature factors with the same expansion rate. Design a dynamic weight generator for features at each scale. : ； In the formula, A weighted graph that adapts to spatial conditions; S43, weighted fusion of feature pairs yields initial fused features at multiple scales: ； S44, calculate the cross-temporal attention map corresponding to the initial fusion features at each scale, and then... With initial fusion features Multiplication: ； Obtain the feature set after multi-scale interaction ; S45, the feature groups after multi-scale interaction By upsampling to align the resolution, the fused feature map is obtained by sequentially stitching and compressing the data. :

2. The method as described in claim 1, characterized in that, Following S50, the following is also included: S60, calculate the binary cross-entropy corresponding to the change probability map, and the overlap loss corresponding to the change region; S70, based on the binary cross-entropy and the overlap loss, a loss function is determined, thereby adjusting the hyperparameters of the deep learning model with the goal of minimizing the loss function.

3. The method as described in claim 2, characterized in that, In step S60, the expression for calculating the binary cross-entropy is: ； In the formula, For binary cross-entropy, It is the i-th real label. This is the probability map of the i-th change predicted by the model. Total number of pixels; The expression for calculating the overlap loss is as follows: ； In the formula, For overlap loss, Indicates the predicted area of ​​change The intersection with the actual change region z.

4. A geological and mining remote sensing image processing system, characterized in that, The geological and mineral remote sensing image processing system includes: The convolutional downsampling module is used to perform shared-weight downsampling convolution on the two different time-time geological and mineral remote sensing images layer by layer to obtain the first geological and mineral remote sensing feature map and the second geological and mineral remote sensing feature map. The convolutional block attention module is used to extract first temporal attention features from the first geological and mineral remote sensing feature map and second temporal attention features from the second geological and mineral remote sensing feature map using a channel attention mechanism; and to process the first temporal attention features using a spatial attention mechanism to obtain first spatial attention features and second temporal attention features to obtain second spatial attention features. A multi-scale feature fusion module is used to concatenate the first spatial attention feature and the second spatial attention feature through a skip connection to obtain a fused feature map; A change detection head network is used to perform convolution operations on the fused feature map to generate a change probability map. Regions in the change probability map that are greater than a preset threshold are identified as change regions, and the rest are identified as non-change regions. The multi-scale feature fusion module includes: A multi-scale spatial attention feature extraction unit is used to generate the first spatial attention feature using dilated convolutions with different dilation rates. Second spatial attention features Corresponding multi-scale feature sets: ； ； In the formula, This represents the multi-scale feature set corresponding to the first spatial attention feature. Multiscale feature factors with index n in the middle. This represents the multi-scale feature set corresponding to the second spatial attention features. In the multi-scale feature factor with index n, where n represents the number of features, ... and Multiscale feature factors with the same ordinal number have the same inflation rate; The cross-temporal dynamic weight generation unit is used to construct feature pairs from multi-scale feature factors with the same expansion rate. Design a dynamic weight generator for features at each scale. : ； In the formula, For spatially adaptive weighted graphs, ; And the fused features are obtained by weighting and fusing the various feature pairs: ； A multi-scale feature interaction enhancement unit is used to calculate the cross-temporal attention map corresponding to the fused features at each scale, and to process the cross-temporal attention map. With fusion features Multiplication: ； Obtain the feature set after multi-scale interaction ; Progressive feature aggregation unit, used to aggregate feature groups after multi-scale interaction By upsampling to align the resolution, the fused feature map is obtained by sequentially stitching and compressing the data. :

5. The geological and mining remote sensing image processing system as described in claim 4, characterized in that, The geological and mining remote sensing image processing system includes at least two convolutional layers. The convolutional block attention module is inserted between every two convolutional layers of the geological and mining remote sensing image processing model to realize channel attention mechanism extraction and spatial attention mechanism extraction through the channel attention module and spatial attention module in the convolutional block attention module.

6. The geological and mining remote sensing image processing system as described in claim 4, characterized in that, The convolutional downsampling module includes a Siamese structure, which assigns shared weights to the convolutional layers when the model implements downsampling convolution.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a geological and mineral remote sensing image processing program, which, when executed by a processor, implements the steps of the geological and mineral remote sensing image processing method as described in any one of claims 1 to 3.

Images