A method and system for remote sensing extraction of linear fault structures from a continuous perspective

Abstract

This invention provides a remote sensing method and system for extracting linear fault structures from a continuous perspective, relating to the field of fault structure interpretation technology. The method includes: preprocessing the labeled data and remote sensing image data in the dataset; inputting the preprocessed dataset into a pre-constructed fault structure probability map generation model; training and validating the model using K-fold cross-validation; obtaining K trained fault structure probability map generation models based on the optimal results of the validation set for each fold; inputting second continuous remote sensing image data covering a second target area and its surroundings into the K trained fault structure probability map generation models to obtain K probability distribution result maps; processing the K probability distribution maps using a post-processing module to finally form a fault structure map. This invention improves the continuity of fault structure interpretation and extraction accuracy.

Description

Technical Field

[0001] This invention relates to the field of fracture structure interpretation technology, and more specifically, to a method and system for remote sensing extraction of linear fracture structures from a continuous perspective. Background Technology

[0002] Fault structures are a phenomenon in which rocks lose their continuity and integrity due to mechanical fracturing along a certain direction caused by tectonic stress. Fault structures represent important geological structures in the Earth's crust, and their interpretation plays a crucial role in earthquake research, resource exploration, and disaster prevention.

[0003] Traditional fault structure investigation and interpretation mainly rely on field surveys, but this method is costly and inefficient, especially in remote areas with poor transportation where field surveys are extremely difficult. In such cases, using remote sensing data for human-computer interactive interpretation has become a more economical and efficient option. However, human-computer interactive interpretation relies on manually set parameters, making the interpretation process highly dependent on human intervention. This dependence may introduce subjective factors, which could lead to inaccurate interpretation results. Furthermore, the interpretation results may also show poor continuity of fault structures, which is not conducive to observation and research. Summary of the Invention

[0004] The present invention aims to address the problems in the prior art of interpreting remote sensing data, which are highly dependent on human intervention, potentially leading to inaccurate interpretation results and poor continuity of fracture structures in the interpretation results.

[0005] To address the aforementioned problems, in a first aspect, the present invention provides a remote sensing extraction method for linear fault structures from a continuous perspective, comprising:

[0006] Construct a dataset for the same region, wherein the dataset includes first continuous remote sensing image data of the first target region and its surrounding area, and label data of the fracture structure surface of the first target region;

[0007] The labeled data and remote sensing image data in the dataset are preprocessed. Based on the preprocessed dataset, the pre-constructed fault structure probability map generation model is trained and validated through multi-fold cross-training to obtain multiple trained fault structure probability map generation models.

[0008] The second continuous remote sensing image data covering the second target area and its surroundings are input into the trained multiple fault structure probability map generation models to obtain multiple probability distribution result maps. Among them, the multiple probability distribution result maps only include the probability distribution result map of the second target area. Each probability distribution result map is composed of multiple pixels, and each pixel has a corresponding probability of the existence of a fault structure.

[0009] Based on the probability, the pixels in multiple probability distribution result maps are filtered to obtain a result map of the fracture structure surface containing high-probability fracture structure points. The result map of the fracture structure surface is then filtered and refined to obtain a result map of the fracture structure line.

[0010] Optionally, the pre-built probability map generation model for fracture structures includes a multi-scale attention semantic segmentation model.

[0011] Optionally, the preprocessing of the label data and remote sensing image data in the dataset includes:

[0012] The labeled data and remote sensing image data in the dataset are uniformly segmented.

[0013] The dataset, after being uniformly divided, is randomly divided into a training validation set and a test set that meet a preset ratio.

[0014] The step of performing multi-fold cross-training and validation on the pre-constructed fracture structure probability map generation model to obtain multiple trained fracture structure probability map generation models includes:

[0015] The first continuous remote sensing image data in the training and validation set is divided into K equal subsets, and each subset is used as a validation set, and the remaining K-1 subsets are used as training sets. The pre-constructed fault structure probability map generation model is trained and validated K times to obtain K fault structure probability map generation models, where K is a positive integer greater than 1.

[0016] Optionally, the step of inputting the second continuous remote sensing image data covering the second target area and its surroundings into the trained multiple fault structure probability map generation models to obtain multiple probability distribution result maps includes: inputting the target remote sensing image block of the second target area and multiple surrounding remote sensing image blocks of the target remote sensing image block into the trained multiple fault structure probability map generation models to obtain multiple probability distribution result maps, wherein the second continuous remote sensing image data includes the target remote sensing image block and multiple surrounding remote sensing image blocks, and the target remote sensing image block and multiple surrounding remote sensing image blocks correspond to multiple continuous locations.

[0017] Optionally, the multi-scale attention semantic segmentation model includes an encoder, a decoder, and a discarding structure. The encoder is used to extract abstract features and spatial orientation of fracture structures in continuous remote sensing image data under receptive fields of different sizes, so as to obtain a result feature map that characterizes the direction-aware and position-sensitive fracture structures. The decoder is used to restore the result feature map to its original size to obtain the probability distribution result map. The discarding structure is used to retain and output the result feature map associated with the target remote sensing image patch, while discarding the result feature maps of the surrounding remote sensing image patches. The continuous remote sensing image data includes first continuous remote sensing image data and second continuous remote sensing image data.

[0018] Optionally, the step of filtering the pixels in multiple probability distribution result maps based on the probability to obtain a result map containing the fracture structure surface with high probability fracture structure points includes:

[0019] For each identical pixel in multiple probability distribution result maps, the average probability is determined, and pixels whose average probability is greater than or equal to the machine learning classification probability are retained as key points where fracture structures occur. The high-probability fracture structure points include the key points.

[0020] For each identical pixel in multiple probability distribution result maps, determine the maximum value of the probability, and retain the pixels whose maximum value is greater than or equal to the machine learning classification probability as possible surfaces where fracture structures may occur;

[0021] Select the pixel corresponding to the possible surface containing the key point as the result image of the fracture structure surface.

[0022] Optionally, the step of filtering and refining the resulting image of the fracture structure surface to obtain the resulting image of the fracture structure line includes:

[0023] The entire result image of the fracture structure is converted into vector shape data. The non-connected parts of the vector shape data are then split to form multiple independent continuous vector data.

[0024] The continuous vector data corresponding to the possible surfaces containing the key points are retained and rasterized to form the raster data of the fracture structure surface;

[0025] The raster data of the fracture structure surface is refined to obtain the resulting image of the fracture structure line.

[0026] Optionally, the dataset constructed from the same region includes:

[0027] Acquire remote sensing image data of the same area at a preset resolution, and fault structure vector line data of a geological map at a preset scale;

[0028] The fracture structure vector line data is buffered to form the tag data.

[0029] Optionally, the preprocessing of the label data and remote sensing image data in the dataset includes:

[0030] The labeled data and remote sensing image data in the dataset are uniformly segmented.

[0031] The dataset, after being uniformly divided, is randomly divided into a training validation set and a test set that meet a preset ratio;

[0032] The dataset, after being uniformly divided, is randomly divided into a training validation set and a test set that meet a preset ratio.

[0033] The step of performing multi-fold cross-training and validation on the pre-constructed fracture structure probability map generation model to obtain multiple trained fracture structure probability map generation models includes:

[0034] The first continuous remote sensing image data in the training and validation set is divided into K equal subsets, and each subset is used as a validation set, and the remaining K-1 subsets are used as training sets. The pre-constructed fault structure probability map generation model is trained and validated K times to obtain K fault structure probability map generation models, where K is a positive integer greater than 1.

[0035] This invention facilitates the training of a fault structure probability map generation model and improves the accuracy of subsequent extraction by constructing and preprocessing a dataset including remote sensing image data and labeled data of fault structures. The pre-constructed fault structure probability map generation model undergoes multi-fold cross-training and validation to obtain multiple trained fault structure probability map generation models. The input for training the model is a first continuous remote sensing image dataset, covering a first target area and multiple surrounding continuous locations. Therefore, the multiple first continuous remote sensing images used for training ultimately appear as multiple overlapping continuous image patches, simulating the observation continuity from the perspective of fault structure interpretation by geologists, and indirectly expanding the interpretation observation field, thus contributing to improved interpretation accuracy. It is also important to emphasize that the labels during model training only include the labeled raster data of the first target area, excluding the labeled data of multiple surrounding continuous locations. Each raster image of the target area is spatially non-overlapping, so the output of the trained model only includes the interpretation results of the target remote sensing image patches. Similarly, to obtain the interpreted fracture structure results, multiple fracture structure probability map generation models trained using continuous image input (these models are generated from a single training and validation dataset and a multi-fold cross-validation method) are used to obtain multiple probability distribution result maps of the second target region corresponding to the second continuous remote sensing image data (each input of the second continuous remote sensing image data can correspond to multiple probability distribution result maps of the second target region). The probability of fracture structure existence under each pixel in the probability distribution result map of the target region is interpreted, and not just one probability distribution result map is generated. Based on the probability, the pixels in the generated multiple probability distribution result maps are filtered to obtain a result map containing fracture structure surfaces with high probability fracture structure points. This filtering design fully considers the linear continuity characteristics of fracture structures, that is, a fracture structure surface is likely to exist near a high probability fracture structure point. Therefore, the resulting image of the fracture structure surface containing high-probability fracture structure points interprets the distribution of the fracture structure surface more accurately. Subsequently, the resulting image of the fracture structure surface is filtered and refined to obtain the result image of the fracture structure line, which improves the continuity of fracture structure interpretation. Compared with the processing method of setting thresholds often used in the fracture structure processing of the prior art, the processing method of filtering fracture structure surfaces based on high-probability fracture structure points of the present invention reduces or avoids the introduction of human subjective factors and improves the accuracy of fracture structure line extraction.

[0036] Secondly, the present invention also provides a remote sensing extraction system for linear fracture structures from a continuous perspective, comprising:

[0037] A dataset construction module is used to construct a dataset for the same region, wherein the dataset includes first continuous remote sensing image data of the first target region and its surrounding area, and label data of the fracture structure surface of the first target region;

[0038] The model optimization module is used to preprocess the label data and remote sensing image data in the dataset, and based on the preprocessed dataset, to perform multi-fold cross-training and validation on the pre-built fault structure probability map generation model to obtain multiple trained fault structure probability map generation models.

[0039] The probability distribution result map generation module is used to input the second continuous remote sensing image data covering the second target area and its surroundings into the trained multiple fault structure probability map generation models to obtain multiple probability distribution result maps. The multiple probability distribution result maps only include the probability distribution result map of the second target area. Each probability distribution result map is composed of multiple pixels, and each pixel has a corresponding probability of the existence of a fault structure.

[0040] The fracture structure post-processing module is used to filter the pixels in multiple probability distribution result maps based on the probability to obtain a result map of the fracture structure surface containing high-probability fracture structure points, and to filter and refine the result map of the fracture structure surface to obtain a result map of the fracture structure line.

[0041] Thirdly, the present invention provides an electronic device, including a memory and a processor;

[0042] The memory is used to store computer programs;

[0043] The processor is configured to, when executing the computer program, implement the remote sensing extraction method for linear fracture structures from a continuous viewpoint as described in the first aspect.

[0044] Fourthly, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the remote sensing extraction method for linear fracture structures from a continuous viewpoint as described in the first aspect.

[0045] The linear fault structure remote sensing extraction system, electronic device, and computer-readable storage medium provided by this invention have the same beneficial effects as the linear fault structure remote sensing extraction method under continuous perspective compared to the prior art, and will not be repeated here. Attached Figure Description

[0046] Figure 1 A flowchart illustrating a method for remote sensing extraction of linear fracture structures from a continuous perspective, according to an embodiment of the present invention, is shown.

[0047] Figure 2 This diagram illustrates the extraction of fractured structural lines after inputting continuous remote sensing image data from multiple consecutive locations in an embodiment of the present invention.

[0048] Figure 3 This diagram illustrates the model training and post-processing in an embodiment of the present invention.

[0049] Figure 4 A schematic diagram of the structure of the multi-scale attention module in an embodiment of the present invention is shown;

[0050] Figure 5 A visual comparison diagram of key points and possible surfaces in an embodiment of the present invention is shown;

[0051] Figure 6 A schematic diagram of the structure of the remote sensing extraction system for linear fracture structures in an embodiment of the present invention is shown;

[0052] Figure 7 A schematic diagram of the structure of an electronic device according to an embodiment of the present invention is shown. Detailed Implementation

[0053] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0054] It should be noted that relational terms such as "first" and "second" in this invention are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0055] In the description of this specification, references to terms such as "embodiment," "one embodiment," and "one implementation" indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or implementation is included in at least one embodiment or illustrative implementation of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or implementation. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or implementations.

[0056] Reference Figure 1 As shown in the figure, this invention proposes a remote sensing extraction method for linear fracture structures from a continuous perspective.

[0057] The method for remote sensing extraction of linear fracture structures from continuous perspectives includes:

[0058] S100: Construct a dataset for the same region, wherein the dataset includes first continuous remote sensing image data of the first target region and its surrounding area, and label data of the fracture structure surface of the first target region.

[0059] Specifically, after determining the study area, remote sensing image data of that area is acquired, such as various types of remote sensing image data, including very high resolution (VHR), hyperspectral (HSI), and synthetic aperture radar (SAR). The remote sensing image data serves as sample data in the dataset, corresponding to the label data. The label data is the output result corresponding to the sample data, also known as the target value. The labels are used to guide the model's learning and help the model identify patterns and relationships in the data.

[0060] S200: Preprocess the label data and remote sensing image data in the dataset, and based on the preprocessed dataset, perform multi-fold cross-training and validation on the pre-constructed fault structure probability map generation model to obtain multiple trained fault structure probability map generation models.

[0061] Specifically, the processing of the dataset before model training may include dataset partitioning, block processing of remote sensing image data, and data cleaning to ensure that it meets the requirements of model training and validation. The constructed model can be a deep learning model. The constructed model undergoes multi-fold cross-training and validation. For each fold, optimal model parameters are selected to obtain multiple corresponding fault structure probability map generation models, so as to output multiple probability distribution result maps. Unlike the prior art that selects a single model under the cross-validation framework, this invention obtains multiple models under each validation and training result.

[0062] S300: Input the second continuous remote sensing image data covering the second target area and its surroundings into the trained multiple fault structure probability map generation models to obtain multiple probability distribution result maps. The multiple probability distribution result maps only include the probability distribution result map of the second target area. Each probability distribution result map is composed of multiple pixels, and each pixel has a corresponding probability of the existence of a fault structure.

[0063] A pixel is a fundamental concept in Geographic Information Systems (GIS), referring to the smallest resolvable unit in a map or image.

[0064] Specifically, the relationship between the second target area and the first target area is as follows: the second target area is generally a region not far from the first target area, and the two have the same or similar tectonic background. The first and second continuous remote sensing image data are input in the form of target remote sensing image blocks and multiple surrounding remote sensing image blocks, which enhances the observation continuity from the perspective of fault structure interpretation and indirectly expands the observation field of interpretation. After inputting the second continuous remote sensing image data into the fault structure probability map generation model, multiple probability distribution result maps are obtained, which enhances the observation continuity from the perspective of fault structure interpretation and generates K probability result maps instead of just one. This can improve the reliability of the results of the continuity post-processing related to fault structures (in the next step) and the ability to assess the uncertainty of the results.

[0065] S400: Based on the probability, the pixels in multiple probability distribution result maps are filtered to obtain a result map of the fracture structure surface containing high-probability fracture structure points. The result map of the fracture structure surface is then filtered and refined to obtain a result map of the fracture structure line.

[0066] A comprehensive analysis of multiple probability distribution maps is performed. If most of the K result maps show the presence of a fault structure in the area, the confidence level is relatively high. If the probability values ​​of the same pixel are generally high across the K result maps, the area where that pixel is located is highly likely to be a "high-probability fault structure point." If a pixel in the K result maps has a higher probability value than others, the coverage area corresponding to that pixel is highly likely to be the potential fault structure occurrence surface, i.e., the fault structure surface. Therefore, in summary, the fault structure surface containing high-probability fault structure points can be used as the final fault structure surface. Since the fault structure surface may contain some noise, noise removal can be performed during the refinement process. Thus, potential fault structures can be identified from multiple probability distribution map results, and a refined fault structure line result map can be obtained, providing important reference information for subsequent geological research and risk assessment. High-probability fault structure points can be considered as fault structure points that are determined to be highly likely to experience fault formation.

[0067] In practical application, this embodiment utilizes the construction and preprocessing of a dataset including remote sensing image data and label data of fault structures to facilitate the training of the fault structure probability map generation model and improve the accuracy of subsequent extraction. The pre-constructed fault structure probability map generation model undergoes multi-fold cross-training and validation to obtain multiple trained fault structure probability map generation models. The input for model training is continuous remote sensing imagery covering the first target area and multiple surrounding continuous locations. Therefore, the multiple first continuous remote sensing image data used for training ultimately appear as multiple overlapping continuous image patches, simulating the continuity of observation from the perspective of fault structure interpretation by geologists, and indirectly expanding the interpretation observation field, thus contributing to improved interpretation accuracy. It is also important to emphasize that the labels during model training only include the label raster data of the first target area, excluding the label data of multiple surrounding continuous locations. All target area raster data are spatially non-overlapping, so the output of the trained model only includes the interpretation results of the target image patch locations (multiple probability distribution result maps only include the probability distribution result map of the second target area). Similarly, to obtain the interpreted fault structure results, multiple fault structure probability map generation models trained using continuous image input (these models are generated from a single training and validation dataset and a multi-fold cross-validation method) are used to obtain multiple probability distribution result maps of the second target region corresponding to the second continuous remote sensing image data (each input of the second continuous remote sensing image data can correspond to multiple probability distribution result maps of the second target region). The probability of a fault structure existing under each pixel in the probability distribution map of the second target region is interpreted. Furthermore, not only one probability distribution result map is generated, but based on the probability, a filtering design is applied to the pixels in the generated multiple probability distribution result maps to obtain a result map containing the fault structure surface that contains high-probability fault structure points. This filtering design fully considers the linear continuity characteristics of fault structures, meaning that a fault structure is highly likely to exist near a high-probability fault structure point. Therefore, the resulting image of the fracture structure surface containing high-probability fracture structure points interprets the distribution of the fracture structure surface more accurately. Subsequently, the resulting image of the fracture structure surface is filtered and refined to obtain the result image of the fracture structure line, which improves the continuity of fracture structure interpretation. Compared with the processing method of setting thresholds often used in the fracture structure processing of the prior art, the processing method of filtering fracture structure surfaces based on high-probability fracture structure points of the present invention reduces or avoids the introduction of human subjective factors and improves the accuracy of fracture structure line extraction.

[0068] like Figure 2 When extracting fracture structural lines after inputting continuous remote sensing image data from multiple consecutive locations, the ratio of the number of the first target area and the multiple surrounding remote sensing influence blocks in the first continuous remote sensing image data includes 1:8.

[0069] The second continuous remote sensing image data includes the target remote sensing image block and multiple surrounding remote sensing image blocks in a ratio of 1:8.

[0070] For example, the input is set to one target remote sensing image block and eight surrounding remote sensing image blocks, that is, a total of nine image blocks are input into the model for learning and extraction.

[0071] As an optional embodiment of the present invention, the step of inputting the second continuous remote sensing image data covering the second target area and its surroundings into the trained multiple fault structure probability map generation models to obtain multiple probability distribution result maps includes:

[0072] The target remote sensing image patch of the second target region and multiple surrounding remote sensing image patches are input into a pre-trained probability map generation model for multiple fracture structures to obtain multiple probability distribution result maps. The second continuous remote sensing image data includes the target remote sensing image patch and multiple surrounding remote sensing image patches, which correspond to multiple consecutive locations. The multiple probability distribution result maps only include the probability distribution result map of the second target region (the probability distribution result maps of the surrounding regions are discarded). Each probability distribution result map is composed of multiple pixels, and each pixel has a corresponding probability of the presence of a fracture structure.

[0073] By inputting the target area and its surrounding images into multiple models, and using information from the surrounding areas to assist in the prediction of the target area, the prediction accuracy of the models can be improved.

[0074] By inputting continuous remote sensing image data into the trained fault structure probability map generation model, the connection between the target remote sensing image patch and the surrounding remote sensing image patches is strengthened, and the observation perspective is further magnified, enhancing the continuity of the fault structure interpretation network's perspective. Secondly, using, for example, eight image patches instead of more to enhance the continuity of structure interpretation is to avoid increasing the number of network parameters too much. Thirdly, multi-scale attention modules under different levels of features are designed to mine the features of fault image patches at different sizes and perspectives, in order to better obtain their accurate spatial location information.

[0075] As an optional embodiment of the present invention, the pre-constructed fracture structure probability map generation model includes a multi-scale attention semantic segmentation model.

[0076] Specifically, the multi-scale attention semantic segmentation model performs semantic segmentation tasks under the deep learning model (U-Net-like model), assigning each pixel in the image to a specific category. The multi-scale attention mechanism can help the model better capture information at different scales in the image, thereby improving the accuracy of semantic segmentation and determining the probability of a fracture structure in each pixel of the target remote sensing image block in the second continuous remote sensing image data.

[0077] like Figure 3 and Figure 4 As shown, the trained fault structure probability map generation model includes an encoder, a decoder, and a discarding structure. During model training, the encoder extracts abstract features and spatial orientation of fault structures in continuous remote sensing image data under receptive fields of different sizes to obtain a result feature map characterizing the direction-awareness and location sensitivity of the fault structures. The decoder restores the result feature map to its original size to obtain the probability distribution result map. The discarding structure retains and outputs the result feature map associated with the target remote sensing image patch, while discarding the result feature maps of surrounding remote sensing image patches. This is because when calculating the loss function, only the labels corresponding to the target remote sensing image patch are interpreted. Therefore, only the output feature values ​​associated with the target remote sensing image patch need to be selectively retained, while the output feature values ​​of surrounding remote sensing image patches are directly discarded, in order to perform the loss calculation. Thus, the result feature maps of these surrounding remote sensing image patches are discarded.

[0078] Specifically, the encoder includes four downsampling modules and two first convolutional modules. The receptive fields corresponding to the four downsampling modules decrease sequentially. Each downsampling module includes two second convolutional modules and a max pooling layer with a kernel size of 2×2. The first convolutional module includes a convolutional layer with a kernel size of 1×1 and a stride of 1, a Bn layer, and a ReLU layer. The second convolutional module includes a convolutional layer with a kernel size of 3×3 and a stride of 1, a Bn layer, and a ReLU layer. The downsampling modules also have multi-scale attention modules added (which can be determined based on the pixel block size of the continuous remote sensing image data. For example, in practice, the input is a 64×64 image block. The receptive fields corresponding to the four downsampling modules decrease sequentially. Therefore, the feature map obtained by the fourth sampling module is too small and it is not suitable to add a multi-scale attention module. Therefore, in this case, at least the first three sampling modules have multi-scale attention modules added. With a suitable size set, all four sampling modules have multi-scale attention modules added).

[0079] For this encoder, corresponding to the decoder structure, the input can be nine images, including the training target and its surrounding images, effectively increasing observation continuity. In terms of model structure, as mentioned earlier, the multi-attention module, namely the "ASPP+Coordinate Attention" module, learns the features of observed fracture structures from receptive fields of different sizes, while the Coordinate Attention module helps learn the spatial directionality of fracture structures from the results of receptive fields of different sizes, obtaining a direction-aware and position-sensitive feature map, expressed by the following formula:

[0080] Out=ASPPCA(Maxpooling(Conv(Conv(x))));

[0081] Where Out represents the output of a downsampling module, x represents the input data of the downsampling module, ASPPCA represents the multi-scale attention module, Conv represents the convolution module, and Maxpooling represents the max pooling layer.

[0082] Reference Figure 4 The multi-scale attention module includes a multi-view learning module (ASPP) and an attention module (Coordinate Attention);

[0083] The multi-view learning module includes a regular convolutional module, two dilated convolutional modules, and a global pooling module (four modules). Each of the regular and dilated convolutional modules includes a 3×3 kernel convolutional layer, a Bn layer, and a ReLU layer. In the two dilated convolutional modules, the first dilated convolutional layer has a dilation size of 2 and a patch size of 2, while the second dilated convolutional layer has a dilation size of 4 and a patch size of 4. The global pooling module includes an adaptive average pooling function, a third regular convolutional layer, and a bilinear interpolation function. The third regular convolutional layer includes a 1×1 kernel convolutional layer, a Bn layer, and a ReLU layer. Bilinear interpolation is then used to restore the feature map size of the original input ASPP layer. Next, the results from the above four modules are concatenated along the channels, and then passed through a 3×3 kernel convolutional layer, a Bn layer, a ReLU layer, and a Dropout layer, where the Dropout layer parameter is 0.5.

[0084] The multi-view learning module is used to generate fracture structure features under receptive fields of different sizes.

[0085] The Coordinate Attention module is used to learn the spatial directionality of fracture structures at different scales after the multi-scale fracture structure features are obtained from the ASPP module. This results in a feature map that is both direction-aware and position-sensitive.

[0086] The attention module includes a dimensional average pooling function, a third convolution module, and a normalization module. The dimensional average pooling function is used to perform average pooling on the multi-scale fracture structure features in both horizontal and vertical spatial dimensions. The results of the two average pooling are then merged in the third dimension and processed by a convolutional layer with a kernel size of 1×1 and a stride of 1, a Bn layer, and an h_swish layer to obtain new weight values ​​in the two spatial dimensions. Based on the new weight values, the weight values ​​of the two spatial dimensions are restored. After passing through a convolutional layer with a kernel size of 1×1 and a stride of 1 and a Sigmoid normalization, the final positional weight values ​​in the two spatial dimensions are obtained. The two obtained positional weight values ​​are multiplied by the input fracture structure features to obtain the feature map of the position-sensitive result.

[0087] More specifically, the decoder's main goal is to restore the original image size using the abstract features extracted during the encoding process. The decoder consists of four upsampling modules, each containing a 2×2 deconvolutional layer with a stride of 2, a skip connection, and two convolutional blocks. These convolutional blocks consist of a 3×3 convolutional layer with a stride of 1, a Bn layer, and a ReLU layer. The skip connection between the decoder and encoder connects features of the same resolution from different levels of the encoder, thus mitigating the spatial information loss caused by downsampling. Finally, a 1×1 convolutional layer is applied for dimensionality reduction, reducing the number of channels to the number of classes, thereby obtaining a feature map of the target remote sensing image patch plus surrounding remote sensing image patches.

[0088] In practical application, this embodiment uses the composition of the neural network structure under the fault structure probability map generation model to input the first continuous remote sensing image data into the model for training. This can effectively increase the continuity of observation and learn the spatial directionality of fault structures from the results of receptive fields of different sizes. Thus, the final result feature map representing direction perception and position sensitivity is obtained, and the feature map of the target remote sensing image block plus the surrounding remote sensing image blocks is output. Only the output feature map associated with the target remote sensing image block is selectively retained, while the output feature map of the surrounding remote sensing image blocks is directly discarded, resulting in multiple probability score result maps.

[0089] After actual comparison, it was found that the multi-scale attention semantic segmentation model obtained after training and validation in this application is superior to classic popular or newer semantic segmentation models. Specifically, at the pixel scale, the model's Precision, Recall, and F1 scores are improved by 28.98%, 32.83%, and 31.77% respectively compared to the suboptimal model. These data fully demonstrate the key role of the proposed background image input and discard structure, multi-scale attention module, and post-processing module based on key points and possible surfaces in accurately predicting fracture structures.

[0090] The metrics are as follows: ① Precision: This metric measures the proportion of pixels predicted as fracture structures that are actually correct fracture structures (Precision = Number of correctly predicted fracture structure pixels / Total number of pixels predicted as fracture structures). ② Recall: This metric represents the percentage of pixels correctly predicted as fracture structures among those labeled as fracture structures (Recall = Number of correctly predicted fracture structure pixels / Total number of pixels labeled as fracture structures). ③ F1 Score: This is a combined evaluation metric of precision and recall, used to balance the performance of the two (F1 = 2 × Precision × Recall / (Precision + Recall)). A lower recall indicates fewer fracture structures in the true labels covered in the prediction results; conversely, a lower precision indicates that the prediction results include more fracture structures not in the true labels.

[0091] As an optional embodiment of the present invention, the preprocessing of the label data and remote sensing image data in the dataset includes:

[0092] The labeled data and remote sensing image data in the dataset are uniformly segmented.

[0093] Specifically, the label data of the fault structure surface in the 1:50,000 scale geological map and the high-resolution image of the area are divided into 64×64 (pixels) sizes.

[0094] The dataset, after being uniformly divided, is randomly divided into a training validation set and a test set that meet a preset ratio;

[0095] Specifically, Gaussian distributed random numbers are used to divide the training and validation sets and the test sets according to a set ratio, or a random function is used to automatically select a fixed number of random numbers from the dataset to meet the preset ratio, such as 50% training and validation set and 50% test set.

[0096] like Figure 3As shown, in an optional embodiment of the present invention, the step of performing multi-fold cross-training and validation on the pre-constructed fracture structure probability map generation model to obtain multiple trained fracture structure probability map generation models includes:

[0097] The first continuous remote sensing image data in the training and validation set is divided into K equal subsets, and each subset is used as a validation set, and the remaining K-1 subsets are used as training sets. The pre-constructed fault structure probability map generation model is trained and validated K times to obtain K fault structure probability map generation models, where K is a positive integer greater than 1.

[0098] In the experiments, the PyTorch 1.7 deep learning framework was used to build and train the model. All experiments were conducted on an Ubuntu 20.04.3 operating system environment, using a GeForce RTX 3080 graphics card. Initially, the images were normalized using the channel mean and standard deviation. The batch size was set to 10, the number of training iterations to 200, and the learning rate to 0.001, with the Adam optimizer used for learning rate optimization. During training, a five-fold cross-validation method was used to train the model, with cross-entropy loss as the loss function. In the multi-scale attention semantic segmentation model, the optimal parameters for each fold were determined by the maximum MIOU value obtained on the validation set and applied to the test set to generate the corresponding probability score result map. The final result map generated by the fracture construction post-processing module 400 was then used for accuracy calculation. In contrast, traditional semantic segmentation models do not have a fracture construction post-processing module 400. Therefore, in this application, the accuracy calculation directly uses the model parameters corresponding to the maximum MIOU value in the five-fold cross-validation method to generate the final fracture surface result map for accuracy calculation.

[0099] Specifically, by using different subsets of data for training and testing, the model's performance on different datasets can be evaluated more accurately, avoiding overfitting of the training data. Applying cross-validation, based on the first continuous remote sensing image data, can yield more stable model evaluation results. The choice of K value depends on the amount of data; generally, K values ​​of 5 or 10 are common, and the larger the K value, the more stable the evaluation results.

[0100] In practical applications, this embodiment divides the first continuous remote sensing image data into K subsets and trains and validates them in turn, which can yield more reliable model evaluation results, improve the model's generalization ability, and prevent overfitting.

[0101] The fault structure remote sensing extraction model constructed in this invention can obtain a superior fault structure extraction result map by using background image input and discarding structures, multi-scale attention module and post-processing module, providing technical support for geological surveys in hard-to-reach areas.

[0102] like Figure 3 and Figure 5 As shown, in an optional embodiment of the present invention, the step of filtering the pixels in multiple probability distribution result maps based on the probability to obtain a result map containing the fracture structure surface with high probability fracture structure points includes:

[0103] For each identical pixel in multiple probability distribution result maps, the average probability is determined, and pixels whose average probability is greater than or equal to the machine learning classification probability are retained as key points where fracture structures occur. The high-probability fracture structure points include the key points.

[0104] A machine learning classification probability of 0.5 indicates a high probability of the existence of fracture structures.

[0105] Specifically, expressed by a formula, K l =If[(K1+K2+……+K n ) / n>=0.5];

[0106] Among them, K l This indicates that a single pixel is identified as a key point. n refers to the number of probability maps obtained by the K-fold cross-validation method. Kn refers to the probability value of a fracture structure at a certain position in the nth probability map. n is a positive integer greater than or equal to 1.

[0107] For each identical pixel in multiple probability distribution result maps, determine the maximum value of the probability, and retain the pixels whose maximum value is greater than or equal to the machine learning classification probability as possible surfaces where fracture structures may occur;

[0108] Specifically, expressed by a formula, K m =If[max(K1,K2,……,K n [)>=0.5];

[0109] Among them, K m This indicates that a single cell is considered a possible surface, where m is less than or equal to n.

[0110] Select the pixel corresponding to the possible surface containing the key point as the result image of the fracture structure surface.

[0111] Specifically, firstly, for each 64×64 (pixel) output image, all output images are stitched together to generate K complete result images. The average value of the K stitched images at the same location is taken, and positions with a probability value ≥ 0.5 are retained, representing high-probability areas of fracture structures, called key points. Simultaneously, the maximum value is taken at the same location in the K stitched images, again retaining positions with a probability value ≥ 0.5, indicating potential fracture structures, called probabilities. Probability surfaces are then selected using key points. Given the linear nature of fracture structures, a high-probability point is likely to be near a fracture structure; therefore, the probabilities surface containing key points is selected as the final fracture structure surface. The determination of machine learning classification probabilities avoids manually setting thresholds that could influence the final interpretation of fracture structure lines.

[0112] First, a visual explanation of the key points and potential aspects, such as... Figure 5 As shown in Figure A, five probability maps were obtained using the K-fold (5-fold cross-validation method was used in this example). After averaging these maps, regions with a probability greater than or equal to 0.5 were selected as key points in the map. Then, the maximum value of the five probability maps was calculated, and the possible surfaces in the map were determined based on a probability greater than or equal to 0.5. It can be observed that there are many possible surfaces in the model's prediction results, but not all possible surfaces represent a high probability of fault formation. Therefore, high-probability key points are used to select the range of possible surfaces. This is because fault formations are often linearly continuous, and high-probability fault formations are also likely to exist around them. This approach considers the continuity of the structural lines, avoids selecting too many low-probability structural surfaces, and significantly reduces subjective dependence. The results of comparing the selected possible surface results with the fault structural line labels are shown below. Figure 5 As shown in B. Finally, the final fracture structure line result diagram is obtained through a refinement algorithm ( Figure 5 C).

[0113] In summary, by using both possible surfaces and key points for screening, the fracture structure surface can be accurately determined, thus allowing the subsequent extracted fracture structure lines to be obtained precisely.

[0114] like Figure 3 (see Figure 3 As shown in the lower half of the diagram, as an optional embodiment of the present invention, the step of filtering and refining the result diagram of the fracture structure surface to obtain the result diagram of the fracture structure line includes:

[0115] The resulting image of the fracture structure (initially in raster format) is converted into vector shape data. The non-connected parts of the vector shape data are then split to form multiple independent continuous vector data.

[0116] Shape data refers to a format in which geographic information is stored as vector data in a shapefile.

[0117] The continuous vector data corresponding to the possible surfaces containing the key points are retained and rasterized to form the raster data of the fracture structure surface;

[0118] Specifically, after converting the raster data of possible surfaces and key points (results of fracture structure surfaces) into vector shape data, the spatially unconnected vector data are split to form multiple independent continuous vector data. Then, the vector data of key points are traversed, and the continuous vector data corresponding to the possible surfaces containing the key points are retained. The vector data of possible surfaces containing key points are retained. The retained vector data of possible surfaces are then rasterized to form the final raster data of possible surfaces.

[0119] The raster data of the fracture structure surface is refined to obtain the resulting image of the fracture structure line.

[0120] Specifically, the model output is a planar distribution. In order to maintain consistency with the linear representation on the geological map, the Zhang-Suen refinement algorithm was further used to refine the structural surface, resulting in the final interpretation result of the fault structural lines. The fault structural lines are the final interpretation result of the target remote sensing image block in the second continuous remote sensing image data.

[0121] The conversion between raster data and vector data (shape data) involved in this embodiment is a common conversion format in the prior art. By vectorizing, splitting, filtering and rasterizing the result map of the fracture structure surface, the final accurate result map of the fracture structure line is obtained, which is convenient for observation from a continuous perspective.

[0122] As an optional embodiment of the present invention, the construction of the dataset for the same region includes:

[0123] Acquire remote sensing image data of the same area at a preset resolution, and fault structure vector line data of a geological map at a preset scale;

[0124] Specifically, after selecting the study area, high-resolution images of that area are collected, including multispectral remote sensing images, high-precision DEM data, and SAR data. The downloaded SAR, multispectral, and other remote sensing images are merged into a single image containing data from multiple channels. Then, the images are cropped to an A×A size (which can be set according to actual needs, such as 64×64). The cropped image ranges do not overlap, and the label data is also cropped at the same ratio.

[0125] Collect fault structure vector line data from publicly published 1:50,000 scale geological maps. The fault structure vector line data is line data of type SHP (a format that stores geographic information as vector data in a SHP file).

[0126] The fracture structure vector line data is buffered to form the tag data.

[0127] Label data refers to target numerical values ​​or categories used to guide model training. For a fracture structure probabilistic map generation model, label data might be information about the presence or absence of fracture structures, the type of fracture structure, etc. Labels are typically categorical or numerical variables.

[0128] Specifically, ArcGIS software is used to process the line buffer at the xm specification, such as 200m, to form fracture structure data.

[0129] In summary, obtaining remote sensing image data and label data of the same region at a preset resolution can serve as the basis and guiding requirements for training subsequent models.

[0130] like Figure 6 As shown, on the other hand, the present invention also provides a remote sensing extraction system 200 for linear fault structures from a continuous perspective, which applies the remote sensing extraction method for linear fault structures from a continuous perspective as described in the above embodiments, including:

[0131] Dataset construction module 210 is used to construct a dataset of the same region, wherein the dataset includes first continuous remote sensing image data of the first target region and its surrounding area, and label data of the fracture structure surface of the first target region;

[0132] The model optimization module 220 is used to preprocess the label data and remote sensing image data in the dataset, and based on the preprocessed dataset, to perform multi-fold cross-training and validation on the pre-constructed fault structure probability map generation model to obtain multiple trained fault structure probability map generation models.

[0133] The probability distribution result map generation module 230 is used to input the second continuous remote sensing image data covering the second target area and its surroundings into the trained multiple fault structure probability map generation models to obtain multiple probability distribution result maps. The multiple probability distribution result maps only include the probability distribution result map of the second target area. Each probability distribution result map is composed of multiple pixels, and each pixel has a corresponding probability of the existence of a fault structure.

[0134] The fracture structure post-processing module 240 is used to filter the pixels in multiple probability distribution result maps based on the probability to obtain a result map of the fracture structure surface containing high-probability fracture structure points, and to filter and refine the result map of the fracture structure surface to obtain a result map of the fracture structure line.

[0135] The specific implementation method of this embodiment can be referred to the corresponding implementation method described above, and will not be described again here.

[0136] like Figure 7 As shown, an electronic device 300 provided in this embodiment of the invention includes a memory 310 and a processor 320; the memory 310 is used to store a computer program; the processor 320 is used to implement the linear fracture structure remote sensing extraction method under continuous view as described above when the computer program is executed.

[0137] Alternatively, an electronic device 300 includes a memory 310 and a processor 320 coupled to the memory 310; the memory 310 is configured to store a computer program; and the processor 320 is configured to perform the following operations when the computer program is executed:

[0138] Construct a dataset for the same region, wherein the dataset includes first continuous remote sensing image data of the first target region and its surrounding area, and label data of the fracture structure surface of the first target region;

[0139] The labeled data and remote sensing image data in the dataset are preprocessed. Based on the preprocessed dataset, the pre-constructed fault structure probability map generation model is trained and validated through multi-fold cross-training to obtain multiple trained fault structure probability map generation models.

[0140] The second continuous remote sensing image data covering the second target area and its surroundings are input into the trained multiple fault structure probability map generation models to obtain multiple probability distribution result maps. Among them, the multiple probability distribution result maps only include the probability distribution result map of the second target area. Each probability distribution result map is composed of multiple pixels, and each pixel has a corresponding probability of the existence of a fault structure.

[0141] Based on the probability, the pixels in multiple probability distribution result maps are filtered to obtain a result map of the fracture structure surface containing high-probability fracture structure points. The result map of the fracture structure surface is then filtered and refined to obtain a result map of the fracture structure line.

[0142] This invention provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the remote sensing extraction method for linear fracture structures from a continuous viewpoint as described above.

[0143] Alternatively, a non-volatile computer-readable storage medium storing a computer program that, when executed by a processor, causes the processor to perform the following operations:

[0144] Construct a dataset for the same region, wherein the dataset includes first continuous remote sensing image data of the first target region and its surrounding area, and label data of the fracture structure surface of the first target region;

[0145] The labeled data and remote sensing image data in the dataset are preprocessed. Based on the preprocessed dataset, the pre-constructed fault structure probability map generation model is trained and validated through multi-fold cross-training to obtain multiple trained fault structure probability map generation models.

[0146] The second continuous remote sensing image data covering the second target area and its surroundings are input into the trained multiple fault structure probability map generation models to obtain multiple probability distribution result maps. Among them, the multiple probability distribution result maps only include the probability distribution result map of the second target area. Each probability distribution result map is composed of multiple pixels, and each pixel has a corresponding probability of the existence of a fault structure.

[0147] Based on the probability, the pixels in multiple probability distribution result maps are filtered to obtain a result map of the fracture structure surface containing high-probability fracture structure points. The result map of the fracture structure surface is then filtered and refined to obtain a result map of the fracture structure line.

[0148] 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 program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided by this invention can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0149] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

[0150] While the present invention has been disclosed above, its scope of protection is not limited thereto. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the present invention, and all such changes and modifications will fall within the scope of protection of the present invention.

Claims

1. A remote sensing method for extracting linear fault structures from a continuous viewpoint, characterized in that, include: Construct a dataset for the same region, wherein the dataset includes first continuous remote sensing image data of the first target region and its surrounding area, and label data of the fracture structure surface of the first target region; The labeled data and remote sensing image data in the dataset are preprocessed. Based on the preprocessed dataset, the pre-constructed fault structure probability map generation model is trained and validated through multi-fold cross-training to obtain multiple trained fault structure probability map generation models. The second continuous remote sensing image data covering the second target area and its surroundings are input into the trained multiple fault structure probability map generation models to obtain multiple probability distribution result maps. Among them, the multiple probability distribution result maps only include the probability distribution result map of the second target area. Each probability distribution result map is composed of multiple pixels, and each pixel has a corresponding probability of the existence of a fault structure. Based on the probability, the pixels in multiple probability distribution result maps are filtered to obtain a result map of the fracture structure surface containing high probability fracture structure points. The result map of the fracture structure surface is then filtered and refined to obtain a result map of the fracture structure line. The step of filtering pixels from multiple probability distribution result maps based on the probability to obtain a result map containing the fracture structure surface with high probability fracture structure points includes: For each identical pixel in multiple probability distribution result maps, the average probability is determined, and pixels whose average probability is greater than or equal to the machine learning classification probability are retained as key points where fracture structures occur. The high-probability fracture structure points include the key points. For each identical pixel in multiple probability distribution result maps, determine the maximum value of the probability, and retain the pixels whose maximum value is greater than or equal to the machine learning classification probability as possible surfaces where fracture structures may occur; Select the pixel corresponding to the possible surface containing the key point as the result image of the fracture structure surface.

2. The remote sensing extraction method for linear fracture structures from a continuous perspective according to claim 1, characterized in that, The step of inputting the second continuous remote sensing image data covering the second target area and its surroundings into multiple trained fault structure probability map generation models to obtain multiple probability distribution result maps includes: The target remote sensing image block of the second target area and multiple surrounding remote sensing image blocks of the target remote sensing image block are input into multiple trained fault structure probability map generation models to obtain multiple probability distribution result maps. The second continuous remote sensing image data includes the target remote sensing image block and multiple surrounding remote sensing image blocks, and the target remote sensing image block and multiple surrounding remote sensing image blocks correspond to multiple continuous locations.

3. The remote sensing extraction method for linear fracture structures from a continuous perspective according to claim 2, characterized in that, The pre-built probability map generation model for fracture structures includes a multi-scale attention semantic segmentation model.

4. The remote sensing extraction method for linear fracture structures from a continuous perspective according to claim 2, characterized in that, The preprocessing of the label data and remote sensing image data in the dataset includes: The labeled data and remote sensing image data in the dataset are uniformly segmented. The dataset, after being uniformly divided, is randomly divided into a training validation set and a test set that meet a preset ratio.

5. The method for remote sensing extraction of linear fracture structures from a continuous perspective according to claim 4, characterized in that, The step of performing multi-fold cross-training and validation on the pre-constructed fracture structure probability map generation model to obtain multiple trained fracture structure probability map generation models includes: The first continuous remote sensing image data in the training and validation set is divided into K equal subsets, and each subset is used as a validation set, and the remaining K-1 subsets are used as training sets. The pre-constructed fault structure probability map generation model is trained and validated K times to obtain K fault structure probability map generation models, where K is a positive integer greater than 1.

6. The remote sensing extraction method for linear fracture structures from a continuous perspective according to claim 3, characterized in that, The multi-scale attention semantic segmentation model includes an encoder, a decoder, and a discarding structure. The encoder is used to extract abstract features and spatial directionality of fracture structures in continuous remote sensing image data under receptive fields of different sizes, so as to obtain a result feature map that characterizes the direction-aware and position-sensitive fracture structures. The decoder is used to restore the result feature map to its original size to obtain the probability distribution result map. The discarding structure is used to retain and output the result feature map associated with the target remote sensing image patch, while discarding the result feature maps of the surrounding remote sensing image patches. The continuous remote sensing image data includes first continuous remote sensing image data and second continuous remote sensing image data.

7. The remote sensing extraction method for linear fracture structures from a continuous perspective according to claim 1, characterized in that, The process of filtering and refining the resulting image of the fracture structure surface to obtain the resulting image of the fracture structure line includes: The entire result image of the fracture structure is converted into vector shape data. The non-connected parts of the vector shape data are then split to form multiple independent continuous vector data. The continuous vector data corresponding to the possible surfaces containing the key points are retained and rasterized to form the raster data of the fracture structure surface; The raster data of the fracture structure surface is refined to obtain the resulting image of the fracture structure line.

8. The remote sensing extraction method for linear fracture structures from a continuous viewing angle according to any one of claims 1-5, characterized in that, The datasets constructed for the same region include: Acquire remote sensing image data of the same area at a preset resolution, and fault structure vector line data of a geological map at a preset scale; The fracture structure vector line data is buffered to form the tag data.

9. A remote sensing extraction system for linear fracture structures from a continuous viewpoint, characterized in that, The method for remote sensing extraction of linear fault structures from a continuous viewpoint as described in any one of claims 1-8 includes: A dataset construction module is used to construct a dataset for the same region, wherein the dataset includes first continuous remote sensing image data of the first target region and its surrounding area, and label data of the fracture structure surface of the first target region; The model optimization module is used to preprocess the label data and remote sensing image data in the dataset, and based on the preprocessed dataset, to perform multi-fold cross-training and validation on the pre-built fault structure probability map generation model to obtain multiple trained fault structure probability map generation models. The probability distribution result map generation module is used to input the second continuous remote sensing image data covering the second target area and its surroundings into the trained multiple fault structure probability map generation models to obtain multiple probability distribution result maps. The multiple probability distribution result maps only include the probability distribution result map of the second target area. Each probability distribution result map is composed of multiple pixels, and each pixel has a corresponding probability of the existence of a fault structure. The fracture structure post-processing module is used to filter the pixels in multiple probability distribution result maps based on the probability to obtain a result map of the fracture structure surface containing high-probability fracture structure points, and to filter and refine the result map of the fracture structure surface to obtain a result map of the fracture structure line.

Images