Coal mine deformation boundary identification method, device, equipment and medium

Abstract

The invention discloses a coal mine deformation boundary identification method, device and equipment and a medium, and the method comprises the steps: generating a differential interferogram based on the multi-temporal synthetic aperture radar data of a target coal mine region; based on the differential interferogram, generating a model training data set, and training a coal mine subsidence area segmentation model through the model training data set; inputting the to-be-identified differential interferogram into a coal mine settlement area segmentation model to carry out settlement area segmentation operation to obtain a coal mine settlement area binary image; and determining a coal mine deformation boundary based on the coal mine settlement area binary image. By constructing the coal mine subsidence area segmentation model, pixel-level automatic segmentation of the deformation boundary caused by mining is realized, so that the large-range coal mine deformation boundary can be quickly and accurately obtained.

Description

Technical Field

[0001] This invention relates to the field of surface deformation monitoring technology, and in particular to a method, device, equipment and medium for identifying deformation boundaries in coal mines. Background Technology

[0002] Currently, surface deformation monitoring in coal mining areas typically employs time-series analysis methods from synthetic aperture radar interferometry. This method involves averaging multi-temporal interferograms to suppress noise and extract large-scale, gradually varying average surface deformation rate information. However, the output of this method is only the average deformation rate field over the entire time period, resulting in low spatial resolution and making it difficult to clearly identify the specific shape and location of subsidence boundaries. Summary of the Invention

[0003] In view of this, the present invention provides a method, device, electronic device and medium for identifying coal mine deformation boundaries, in order to solve the technical problem that the time-series analysis method in synthetic aperture radar interferometry is difficult to clearly identify the specific shape and location of settlement boundaries.

[0004] Firstly, a method for identifying deformation boundaries in coal mines is provided, the method comprising:

[0005] Differential interferograms are generated based on multi-temporal synthetic aperture radar data of the target coal mining area;

[0006] Based on the differential interferogram, a model training dataset is generated, and a coal mine subsidence area segmentation model is trained using the model training dataset.

[0007] The differential interferogram to be identified is input into the coal mine subsidence area segmentation model to perform subsidence area segmentation operation, and a binary image of the coal mine subsidence area is obtained.

[0008] Determine the deformation boundary of the coal mine based on the binary image of the coal mine subsidence area.

[0009] Secondly, a coal mine deformation boundary identification device is provided, the device comprising:

[0010] The first generation module is used to generate differential interferograms based on multi-temporal synthetic aperture radar data of the target coal mine area.

[0011] The second generation module is used to generate the model training dataset based on the differential interferogram;

[0012] The training module is used to train a coal mine subsidence area segmentation model using a model training dataset.

[0013] The third generation module is used to input the differential interferogram to be identified into the coal mine settlement area segmentation model to perform settlement area segmentation operation and obtain a binary image of the coal mine settlement area.

[0014] The determination module is used to determine the deformation boundary of a coal mine based on a binary image of the coal mine subsidence area.

[0015] Thirdly, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described coal mine deformation boundary identification method.

[0016] Fourthly, a computer-readable storage medium is provided, which stores a computer program that, when executed by a processor, implements the steps of the above-described coal mine deformation boundary identification method.

[0017] In the aforementioned scheme implemented by the coal mine deformation boundary identification method, device, electronic equipment, and storage medium, firstly, high-quality deformation images are generated by performing coherence optimization and denoising processing on multi-temporal synthetic aperture radar data of the coal mine area. Subsequently, a coal mine subsidence area segmentation model is constructed based on the YOLOv11 algorithm with an attention mechanism, achieving pixel-level automatic segmentation of the deformation boundaries caused by mining, thereby enabling rapid and accurate acquisition of large-scale coal mine deformation boundaries. Attached Figure Description

[0018] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of this application. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0019] Figure 1 This is a flowchart illustrating a method for identifying deformation boundaries in coal mines according to an embodiment of the present invention.

[0020] Figure 2 This is a schematic diagram illustrating the process of processing the original differential interferogram using a coal mine subsidence area segmentation model according to an exemplary embodiment;

[0021] Figure 3 This is a schematic diagram showing the overlay of coal mine subsidence area segmentation model results and Stacking-InSAR results according to an exemplary embodiment;

[0022] Figure 4 This is a schematic diagram comparing the temporal performance of a coal mine subsidence area segmentation model result with Stacking-InSAR results, based on an exemplary embodiment.

[0023] Figure 5 This is a schematic diagram of the structure of a coal mine deformation boundary identification device in one embodiment of the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the accompanying drawings in the present invention are only for illustrative and descriptive purposes and are not intended to limit the scope of protection of the present invention.

[0025] Furthermore, it should be understood that the schematic drawings are not drawn to scale. The flowcharts used in this invention illustrate operations implemented according to some embodiments of the invention. It should be understood that the operations in the flowcharts may not be implemented in sequence, and steps without logical contextual relationships may be reversed or performed simultaneously. Moreover, those skilled in the art, guided by the content of this invention, may add one or more other operations to the flowcharts, or remove one or more operations from the flowcharts.

[0026] Furthermore, the embodiments described herein are merely some, not all, of the embodiments of the invention. The components of the embodiments of the invention described and illustrated herein can typically be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0027] It should be noted that the term "comprising" will be used in the embodiments of the present invention to indicate the presence of a feature subsequently declared, but does not exclude the addition of other features. It should also be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0028] The following is a detailed description of this case, in conjunction with the relevant accompanying drawings in the instruction manual.

[0029] Please see Figure 1 This description and embodiment provide a method for identifying deformation boundaries in coal mines, specifically including the following steps:

[0030] S10: Generate differential interferograms based on multi-temporal synthetic aperture radar data of the target coal mine area.

[0031] It is understood that the executing entity of this invention can be a coal mine deformation boundary identification device, a terminal, or a server; no specific limitation is made here. This embodiment of the invention will be described using a server as the executing entity as an example.

[0032] In this step, the target coal mining area is the selected area of ​​coal mining activities. Multiple radar satellite images of the target coal mining area taken at different times are processed using differential radar interferometry to generate an image that clearly shows when and where minute movements or deformations occurred on the surface of the target coal mining area—this is the differential interferogram. The aim is to extract surface deformation information mainly caused by activities such as coal mining from complex raw satellite radar data and visualize this information as an image containing interference fringes, providing a reliable data foundation for subsequent automatic identification and precise segmentation analysis of subsidence areas.

[0033] In one embodiment of this application, a specific differential interferogram generation scheme is provided. In S10, that is, generating a differential interferogram based on multi-temporal synthetic aperture radar data of the target coal mine area, the following steps S11-S13 are specifically included:

[0034] S11: Acquire multi-temporal synthetic aperture radar data of the target coal mine area.

[0035] In this step, multi-temporal radar images of the target coal mine area are acquired by synthetic aperture radar (SAR) satellites at different times, i.e., multi-temporal synthetic aperture radar data.

[0036] In practical applications, synthetic aperture radar is used to coherently process two complex numerical image data of the same area to obtain surface elevation information.

[0037] S12: Preprocess the multi-temporal synthetic aperture radar data, including registration, de-flattening, and filtering.

[0038] In this step, before generating the differential interferogram, the multi-temporal synthetic aperture radar (SAR) data needs to be corrected and optimized, specifically including registration, de-flattening, and filtering. Registration involves pixel-level alignment of SAR images from different time phases within the same area, eliminating spatial positional deviations caused by differences in satellite orbits and viewing angles, ensuring strict spatial consistency between images from different periods. De-flattening utilizes a regional digital elevation model (DEM) to remove phase components caused by terrain undulations from the interferometric phase, making the subsequently generated interferogram more focused on reflecting surface deformation information and reducing interference from terrain factors. Filtering involves introducing adaptive filtering algorithms (such as Goldstein filtering) to suppress noise in the interferogram, improve the signal-to-noise ratio, reduce interference from phase-disrupted speckle noise on deformation information extraction, and enhance the reliability of the effective signal.

[0039] The above method transforms the raw SAR data into high-quality data that can be used for high-precision differential interferometry, laying a reliable foundation for subsequent deformation information extraction and settlement identification.

[0040] S13: Based on the preprocessed multi-temporal synthetic aperture radar data, a series of differential interferometry diagrams are generated using differential interferometry.

[0041] In this step, based on the preprocessed multi-temporal synthetic aperture radar data, differential interferometry (D-InSAR) technology is used to calculate the phase difference between SAR images acquired at different times, and separate the main phase components caused by surface deformation, thereby forming a series of interferometric fringe images that intuitively show the minute surface movements of the target coal mine area within the corresponding time period.

[0042] In practical applications, the acquired SAR data is preprocessed to generate SLC files. Then, a suitable spatiotemporal baseline is selected to construct differential interferograms, and terrain phase is eliminated using precise orbital data and DEM data. The Goldstein adaptive filtering method is applied to suppress coherent noise, yielding filtered differential interferometric results. For example, Sentinel-1A SAR data (orbit number 11_121) of the target coal mine area from March 12, 2017 to December 18, 2023 is selected. After differential radar interferometry processing, including terrain phase removal and filtering, 171 InSAR differential interferograms with a 12-day time baseline are generated. The subsidence funnel feature is quite obvious in the InSAR interferograms, typically appearing as a series of approximately circular or elliptical interference fringe patterns.

[0043] S20: Based on the differential interferogram, generate a model training dataset, and train the coal mine subsidence area segmentation model using the model training dataset.

[0044] In this step, a model training dataset is constructed using the generated differential interferogram, and the coal mine subsidence area segmentation model is trained using the model training dataset, so that the model can identify and segment the corresponding coal mine subsidence area based on the input differential interferogram.

[0045] In one embodiment of this application, a specific model training scheme is provided. In S20, a model training dataset is generated based on the differential interferogram, and a coal mine subsidence area segmentation model is trained using the model training dataset. This specifically includes the following steps S21-S24:

[0046] S21: Crop the differential interferogram according to the preset size to generate multiple image blocks.

[0047] In this step, a large-format differential interferogram is traversed through the entire image using a fixed-size window and cut into multiple regularly sized image blocks. These blocks serve as the basic data units for model training, thereby adapting to the input requirements of deep learning models and significantly improving data utilization efficiency and processing performance.

[0048] Optionally, the preset size can be 320×320 pixels.

[0049] S22: Label each image block to generate labeled data with coordinate information for the boundary labels of the settlement area.

[0050] In this step, the outline of the settlement area is drawn on each cropped image patch, and structured annotation data containing geometric coordinates is generated.

[0051] In practical applications, professionals use annotation tools (such as LabelMe) to accurately delineate the contours of the subsidence area in the segmented image blocks based on the characteristics of interference fringes, and generate 1,000 sets of annotation data (such as JSON or XML format) containing the pixel coordinates of the boundary of the subsidence area. Common formats include polygon point sets or binary masks.

[0052] S23: Generate a model training dataset based on labeled data.

[0053] In this step, labeled data is paired and associated with corresponding image patches to form "image-label" data pairs. Then, the labeled data is standardized, converting it into a specific format required for model training. Subsequently, the processed data is divided into training, validation, and test sets according to a preset ratio (e.g., 8:1:1), collectively forming a standardized dataset that meets the requirements for training machine learning models.

[0054] S24: Train a coal mine subsidence area segmentation model based on the model training dataset.

[0055] In this step, using the established training dataset, the model learns to automatically identify and segment the characteristic patterns of subsidence areas from differential interferogram image patches through deep learning algorithms, and finally obtains a coal mine subsidence area segmentation model with predictive capabilities.

[0056] In one embodiment of this application, a specific model training scheme is provided. In S24, that is, based on the model training dataset, a coal mine subsidence area segmentation model is trained, which specifically includes the following steps S241-S243:

[0057] S241: Construct a YOLOv11 model based on the attention mechanism.

[0058] In this step, an attention module is embedded in the YOLOv11 architecture as the backbone network to construct an attention-based YOLOv11 model. By introducing the attention mechanism, the expressive power of subsidence area features is enhanced, thereby improving the accuracy and robustness of identifying coal mine deformation areas.

[0059] In one embodiment of this application, the YOLOv11 model includes an input layer, a feature extraction layer, and a segmentation head, wherein the input layer is a Backbone network;

[0060] The steps for building a YOLOv11 model based on an attention mechanism include:

[0061] The DB-SimAM attention mechanism is added to the tail of the backbone network and the head of the segmentation head of the YOLOv11 model to form an attention-based YOLOv11 model.

[0062] In this embodiment, YOLOv11, as the latest generation of single-stage object detection framework in the YOLO series, significantly improves detection accuracy while maintaining real-time performance through its architecture. This model is built upon a deep convolutional neural network, and its core architecture mainly includes three stages:

[0063] (1) Input layer: The input layer is a Backbone network, which achieves efficient feature extraction through depthwise separable convolution and cross-stage partial connections, while reducing the amount of computation and preserving multi-scale feature information;

[0064] (2) Feature extraction layer: ResNet or Darknet are usually used as the basic backbone network. These networks gradually extract high-level features in the image through multi-layer convolution operations, improving the detection capability of small targets and large-scale sinking areas.

[0065] (3) Segmentation Head: A decoupled prediction head design is adopted, which includes a detection branch, an instance segmentation branch and a semantic segmentation branch to generate the prediction results of the target.

[0066] Based on the output of the feature map, the network generates bounding boxes, target categories, and confidence values ​​for each grid cell. Each predicted box contains a box location coordinate and a corresponding target category.

[0067] Furthermore, in the task of segmenting subsidence basins in mining areas, the morphology of subsidence areas typically exhibits a funnel shape with certain regularity. These areas differ significantly from their surrounding environment, and the segmentation effect is easily affected, especially under complex terrain or noise interference. To improve the accuracy of subsidence basin segmentation and avoid environmental interference and missed detections, the DB-SimAM attention mechanism is introduced into the YOLOv11 architecture at the Backbone tail and Head head. This mechanism integrates the attention module into the convolutional neural network (CNN) structure, enabling the model to focus on key image regions, enhancing feature extraction capabilities and overall model performance, thereby allowing the model to extract more effective and comprehensive feature representations.

[0068] Specifically, SimAM is a parameter-free attention module based on neuroscience theory that effectively enhances the network's feature representation capabilities by defining an energy function. This method boasts the advantages of being lightweight and highly efficient. SimAM further enhances the model's accuracy in detecting settlement funnels of different sizes by improving the network's adaptive learning ability towards targets of different scales.

[0069] Furthermore, to capture finer spatial details and improve sensitivity to local structures, a novel local statistical branch is employed to enhance the network's feature extraction capabilities. This branch effectively captures fine-grained spatial information while retaining the global statistical branch based on the original SimAM module. This dual-branch structure enables the network to more accurately focus on the target region and suppress irrelevant background interference. Through...

[0070] The mechanism introduced in YOLOv11 significantly reduces the impact of background noise and improves the accuracy of target recognition and segmentation in complex mineral subsidence scenarios, thereby ensuring high-quality and high-precision segmentation results.

[0071] Optionally, the performance evaluation of deep learning models typically relies on mean precision (mAP). This metric measures the combined performance of different trained models in terms of accuracy and recall. mAP is the arithmetic mean of the mean precision (AP) for each class, and the mean precision (AP) is obtained by calculating the area under the precision-recall curve, using an interpolated AP method during evaluation. For example, AP50 represents the mean precision with an intersection-over-union (IoU) threshold of 0.5, and AP50-95 represents the average AP for all IoU thresholds from 0.5 to 0.9 (step size 0.05).

[0072] The formula for calculating average accuracy is:

[0073]

[0074] Where AP is the average precision; n is the number of equal divisions of the recall interval [0,1]; r is the recall rate; and P(r) is the precision value when the recall rate is r.

[0075] The formula for calculating the mean precision is:

[0076]

[0077] Where mAP is the mean precision; N is the total number of categories; k is the category index; AP k represents the average precision of the k-th category.

[0078] Furthermore, in order to comprehensively evaluate the model's performance in different aspects, auxiliary evaluation metrics such as precision, recall, and F1 score are used to supplement the evaluation capability of mAP from different dimensions, providing a more comprehensive and detailed model performance analysis.

[0079] The formula for calculating accuracy is:

[0080]

[0081] Wherein, TP is the number of samples whose true class is positive and which the model correctly predicts as positive; FP is the number of samples whose true class is negative but which the model incorrectly predicts as positive; TP+FP is the total number of samples predicted as positive by the model (including both correct and incorrect predictions).

[0082] For example, if the model correctly detects 80 settlement areas (TP) and misidentifies 20 non-settlement areas as settlement in the coal mine area, then Precision = 0.8, indicating that 80% of the settlement areas predicted by the model are actual settlements.

[0083] The formula for calculating recall rate is:

[0084]

[0085] Where FN is the number of samples that are actually positive but were incorrectly predicted as negative by the model.

[0086] For example, assuming there are 100 actual settlement areas in the coal mining area, and the model correctly detects 90 settlement areas (TP) and misses 90 settlement areas (FN), then the Recall is 0.9, indicating that the model has identified 90% of the actual settlement areas.

[0087] The formula for calculating the F1 score is:

[0088]

[0089] For example, in the model's detection of coal mine subsidence, if the precision is 0.8 (meaning 80% of the predicted subsidence areas are real) and the recall is 0.9 (meaning 90% of the real subsidence areas are detected), then the F1 score is 0.847. A high F1 score indicates that the model possesses both high reliability (low false positives) and high coverage (low false negatives), demonstrating excellent overall performance; a low F1 score indicates that the model has not achieved a balance between precision and recall (e.g., more false positives or more false negatives).

[0090] S242: Input the model training dataset into the attention-based YOLOv11 model, and perform backpropagation and gradient calculation on the model parameters based on the binary cross-entropy loss function. Use the Adam optimizer to iteratively update the model parameters of the attention-based YOLOv11 model based on the gradient calculation results.

[0091] In this step, a training dataset containing image patches and their corresponding labels is input into a YOLOv11 model with an embedded attention mechanism. The binary cross-entropy loss function is used to calculate the difference between the model's predicted segmentation results and the true labels, quantifying the model's current performance error. Using the chain rule, the loss value is backpropagated to each layer of the network, calculating the gradient of the loss function with respect to each model parameter, representing the direction and intensity of parameter adjustment. The Adam optimizer is used to dynamically adjust the learning rate based on the gradient calculation results and iteratively update the model parameters, gradually minimizing the loss function to accurately identify coal mine subsidence areas.

[0092] S243: When the change in the loss function is less than the preset threshold or the number of iterations is equal to the maximum number of iterations, output the YOLOv11 model based on the attention mechanism as the coal mine subsidence area segmentation model.

[0093] In this step, when the decrease in the loss function value is lower than the set threshold, it indicates that the model parameters have stabilized. The final version of the model weights and structural parameters are saved to generate a deployable model with the ability to segment coal mine subsidence areas, which serves as the coal mine subsidence area segmentation model.

[0094] S30: Input the differential interferogram to be identified into the coal mine subsidence area segmentation model to perform subsidence area segmentation operation and obtain a binary image of the coal mine subsidence area.

[0095] In this step, the differential interferogram to be analyzed is input into the pre-trained coal mine subsidence area segmentation model, which enables the model to automatically analyze and identify the corresponding subsidence areas and finally output a binary image with only black and white colors. This image can clearly show the specific location and range of the coal mine subsidence area.

[0096] By employing the above methods, deep learning is used to automatically extract mining deformation regions from InSAR deformation rate maps, overcoming the limitations of traditional manual thresholding methods that are sensitive to noise and prone to false positives and false negatives, thus significantly improving segmentation accuracy and processing efficiency. Simultaneously, deep networks can learn deformation characteristics under different terrain conditions, seasonal variations, and radar perspectives, exhibiting strong tolerance to interference from InSAR phase noise, multi-view geometric distortion, and atmospheric delay, thereby significantly improving the stability of deformation field detection in complex mining environments.

[0097] In practical application scenarios, such as Figure 2The diagram shown illustrates the process of processing the original differential interferogram using a coal mine subsidence area segmentation model. Specifically, Figure 2 In Figures (a), (c), and (e), the original differential interferograms (DII) obtained after differential interferometry (DI) processing, before deformation information extraction, contain the deformation signals (interference fringes) desired by researchers, but are also mixed with a large amount of noise, atmospheric delay phase, and other interference information, making it difficult to accurately determine the boundary of the subsidence area with the naked eye. The DB-SimAM-YOLOv11 model receives the original interferograms (a), (c), and (e) as input and outputs segmentation results (b), (d), and (f). The figures clearly show that the model effectively filters noise from the original interferograms, accurately identifies the interference fringe features representing mining deformation, and outputs clear and accurate subsidence area boundaries. Similarly, (g), (i), and (k) in the figures represent another set of original DIIs. Inputting (g), (i), and (k) into the DB-SimAM-YOLOv11 model for segmentation yields segmentation results (h), (j), and (l).

[0098] S40: Determine the deformation boundary of the coal mine based on the binary image of the coal mine subsidence area.

[0099] In this step, based on the binary image of the settlement area generated by the model, image processing techniques are used to extract accurate, coherent, and smooth contour lines to determine the spatial range and boundary location of the coal mine deformation area.

[0100] In one embodiment of this application, a specific scheme for determining the deformation boundary of a coal mine is provided. In S40, that is, determining the deformation boundary of the coal mine based on the binary image of the coal mine subsidence area, specifically includes the following steps S41-S43:

[0101] S41: Denoise reduction processing is performed on the binary image of the coal mine subsidence area.

[0102] In this step, image processing techniques are used to eliminate noise interference in the binarized segmentation results, improve the accuracy and continuity of the settlement area contour, and provide a high-quality data foundation for subsequent boundary extraction and quantitative analysis.

[0103] S42: Perform morphological optimization on the denoised binary image of the coal mine subsidence area.

[0104] In this step, mathematical morphology methods are used to refine the contour of the settlement area to improve regional connectivity, smooth the boundaries, and preserve geometric structural features, providing an optimized basis for subsequent accurate boundary extraction.

[0105] S43: Extract the contour information of the optimized binary image of the coal mine subsidence area to obtain the coal mine deformation boundary.

[0106] In this step, computer vision technology is used to automatically identify and vectorize the edge contours of the settlement area from the processed binary mask, transforming them into spatial boundary data with a clear mathematical expression.

[0107] In practical application scenarios, such as Figure 3 As shown, the results of the coal mine subsidence area segmentation model are compared with...

[0108] A schematic diagram of the stacking-InSAR results. Specifically, the detection boundaries identified by the coal mine subsidence area segmentation model are superimposed onto the average surface deformation rate map obtained after processing synthetic aperture radar interferometry (InSAR) data using stacking technology. The coal mine subsidence area segmentation model of this application not only completely characterizes the boundaries of the aforementioned deformation field, but also...

[0109] The consistency of the spatial location of the deformation field in the Stacking-InSAR results also indicates good agreement between the two methods. Furthermore, the coal mine subsidence area segmentation model provides a broader coverage of coal mine subsidence area, making it more advantageous in coal mine subsidence analysis because it offers a more comprehensive understanding of the impact zone's extent. Further, such as Figure 4 The diagram shows a comparison of the temporal performance of the coal mine subsidence area segmentation model results and the Stacking-InSAR results. Specifically, to analyze the model's temporal performance in depth, a typical mining area was selected and its coal mine boundary was extracted. Subsequently, the model output was validated based on the established InSAR time series results. Figure 5 (a), (b), and (c) in the middle represent the detection boundary results of Stacking-InSAR under different time series; Figure 5 (d), (e), and (f) are respectively the same sequence as those mentioned above.

[0110] The detection boundary results of the DB-SimAM-YOLOv11 model, where the coal mine detection boundary is marked with a black outline. The overlay results of the boundaries derived from the DB-SimAM-YOLOv11 model and the Stacking-InSAR monitoring boundaries show a highly consistent spatial alignment, fully demonstrating...

[0111] The output results of the DB-SimAM-YOLOv11 model are in significant agreement with the Stacking-InSAR monitoring data.

[0112] In one embodiment of this application, a specific spatiotemporal evolution map generation scheme is provided, which, after determining the deformation boundary of the coal mine based on the binary image of the coal mine subsidence area, further includes the following steps:

[0113] Align the coal mine deformation boundaries at different time points according to the observation time sequence;

[0114] Acquire spatial variation data of coal mine deformation boundary between two adjacent time points, wherein the spatial variation data includes displacement variation data, area variation data and expansion direction data;

[0115] Based on continuous temporal spatial variation data, a spatiotemporal evolution map of coal mine deformation boundaries is generated.

[0116] In this embodiment, the subsidence area boundaries extracted from multi-temporal remote sensing data are sorted and spatially registered according to their corresponding observation times to establish a boundary sequence with a unified spatiotemporal reference system. Subsequently, by comparing and analyzing the differences in position and morphology of subsidence boundaries between two consecutive periods, the dynamic evolution characteristics of the subsidence area in the temporal dimension are quantified. Specifically, this includes three types of spatial change data: displacement change data, area change data, and expansion direction data. Displacement change is calculated by calculating the coordinate offset of the centroid point of the subsidence area; area change is calculated by calculating the area difference between the two periods of subsidence; and the expansion direction is determined by identifying the main expansion direction through boundary overlay analysis. Then, the spatial change data of the multi-period coal mine subsidence boundaries are integrated with the temporal dimension, and a spatiotemporal evolution map that comprehensively reflects the spatiotemporal dynamic evolution law of the subsidence area is constructed using visualization technology.

[0117] By employing the above methods, spatiotemporal evolution maps are constructed to reveal the dynamic evolution of subsidence in coal mining areas. This enables continuous tracking and dynamic visualization of subsidence regions, accurately identifying subsidence development trends and spatial expansion characteristics at different time points, and quantitatively assessing the deformation diffusion rate and direction. This provides a scientific basis for geological hazard assessment, mining activity impact analysis, and prevention and control decisions. Furthermore, the automated spatiotemporal boundary tracking method can process multi-temporal data across mining areas of tens to hundreds of square kilometers in batches.

[0118] InSAR data enables large-scale, near real-time boundary identification and updates, providing reliable technical support for rapid early warning, emergency response, and long-term monitoring of mining areas.

[0119] As can be seen, in the above scheme, firstly, high-quality deformation images are generated by performing coherence optimization and denoising processing on multi-temporal synthetic aperture radar data of the coal mining area. Subsequently, a coal mine subsidence area segmentation model is constructed based on the YOLOv11 algorithm with an attention mechanism, which achieves pixel-level automatic segmentation of deformation boundaries caused by mining, thereby enabling rapid and accurate acquisition of large-scale coal mine deformation boundaries.

[0120] In one embodiment, a coal mine deformation boundary identification device is provided, which corresponds one-to-one with the coal mine deformation boundary identification method described in the above embodiments. For example... Figure 5As shown, the coal mine deformation boundary recognition device 100 includes: a first generation module 101, a second generation module 102, a training module 103, a third generation module 104, and a determination module 105. Detailed descriptions of each functional module are as follows:

[0121] The first generation module 101 is used to generate differential interferograms based on multi-temporal synthetic aperture radar data of the target coal mine area.

[0122] The second generation module 102 is used to generate a model training dataset based on the differential interferogram;

[0123] Training module 103 is used to train a coal mine subsidence area segmentation model using a model training dataset;

[0124] The third generation module 104 is used to input the differential interferogram to be identified into the coal mine settlement area segmentation model to perform settlement area segmentation operation and obtain a binary image of the coal mine settlement area.

[0125] The determination module 105 is used to determine the deformation boundary of the coal mine based on the binary image of the coal mine subsidence area.

[0126] In one embodiment, the first generation module 101 is specifically used for:

[0127] Acquire multi-temporal synthetic aperture radar data of the target coal mining area;

[0128] Preprocessing is performed on multi-temporal synthetic aperture radar data, which includes registration, de-flattening, and filtering.

[0129] Based on preprocessed multi-temporal synthetic aperture radar data, a series of differential interferograms are generated using differential interferometry.

[0130] In one embodiment, the training module 103 is specifically used for:

[0131] The differential interferogram is cropped according to a preset size to generate multiple image blocks;

[0132] Each image patch is labeled to generate labeled data of settlement area boundary labels with coordinate information;

[0133] Generate a model training dataset based on labeled data;

[0134] A coal mine subsidence area segmentation model was trained based on the model training dataset.

[0135] In one embodiment, the training module 103 is further configured to:

[0136] Construct a YOLOv11 model based on the attention mechanism;

[0137] The model training dataset is input into the attention-based YOLOv11 model, and the model parameters are backpropagated and gradients are calculated based on the binary cross-entropy loss function. The Adam optimizer is used to iteratively update the model parameters of the attention-based YOLOv11 model based on the gradient calculation results.

[0138] When the change in the loss function is less than a preset threshold or the number of iterations equals the maximum number of iterations, the YOLOv11 model based on the attention mechanism is output as a coal mine subsidence area segmentation model.

[0139] In one embodiment, the attention-based YOLOv11 model includes an input layer, a feature extraction layer, and a segmentation head, wherein the input layer is a Backbone network;

[0140] Training module 103 is also used for:

[0141] The DB-SimAM attention mechanism is added to the tail of the backbone network and the head of the segmentation head of the YOLOv11 model to form an attention-based YOLOv11 model.

[0142] In one embodiment, the determining module 105 is specifically used for:

[0143] Denoising processing is performed on binary images of coal mine subsidence areas;

[0144] Morphological optimization operations were performed on the denoised binary image of the coal mine subsidence area.

[0145] Contour information of the optimized binary image of the coal mine subsidence area is extracted to obtain the deformation boundary of the coal mine.

[0146] In one embodiment, the device further includes:

[0147] The alignment module is used to align the coal mine deformation boundaries at different time points according to the observation time sequence;

[0148] The acquisition module is used to acquire spatial change data of the coal mine deformation boundary between two adjacent time nodes. The spatial change data includes displacement change data, area change data, and expansion direction data.

[0149] The fourth generation module is used to generate a spatiotemporal evolution map of coal mine deformation boundaries based on continuous temporal spatial variation data.

[0150] This invention provides a coal mine deformation boundary identification device 100. First, high-quality deformation images are generated by performing coherence optimization and denoising processing on multi-temporal synthetic aperture radar data of the coal mine area. Subsequently, a coal mine subsidence area segmentation model is constructed based on the YOLOv11 algorithm with an attention mechanism, achieving pixel-level automatic segmentation of deformation boundaries caused by mining, thereby enabling rapid and accurate acquisition of large-scale coal mine deformation boundaries.

[0151] Specific limitations regarding the coal mine deformation boundary identification device can be found in the limitations of the coal mine deformation boundary identification method described above, and will not be repeated here. Each module in the aforementioned coal mine deformation boundary identification device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in the electronic device in hardware form, or stored in the memory of the electronic device in software form, so that the processor can call and execute the corresponding operations of each module.

[0152] In one embodiment, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described coal mine deformation boundary identification method.

[0153] In one embodiment, a computer-readable storage medium is provided, which stores a computer program that, when executed by a processor, implements the above-described coal mine deformation boundary identification method.

[0154] It should be noted that the functions or steps that can be implemented by the computer-readable storage medium or electronic device described above can be referred to the relevant descriptions on the server side and client side in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.

[0155] 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 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 in this application 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 memory bus dynamic RAM (RDRAM), etc.

[0156] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0157] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for identifying deformation boundaries in coal mines, characterized in that, include: Differential interferograms are generated based on multi-temporal synthetic aperture radar data of the target coal mining area; Based on the differential interferogram, a model training dataset is generated, and a coal mine subsidence area segmentation model is trained using the model training dataset. The differential interferogram to be identified is input into the coal mine subsidence area segmentation model to perform subsidence area segmentation operation, and a binary image of the coal mine subsidence area is obtained. Based on the binary image of the coal mine subsidence area, the deformation boundary of the coal mine is determined.

2. The method according to claim 1, characterized in that, The step of generating a differential interferogram based on multi-temporal synthetic aperture radar data of the target coal mine area specifically includes: Acquire the multi-temporal synthetic aperture radar data of the target coal mine area; The multi-temporal synthetic aperture radar data is preprocessed, wherein the preprocessing includes registration processing, terrain removal processing, and filtering processing; Based on the preprocessed multi-temporal synthetic aperture radar data, a series of differential interferometry patterns are generated using differential interferometry.

3. The method according to claim 1, characterized in that, The step of generating a model training dataset based on the differential interferogram and training a coal mine subsidence area segmentation model using the model training dataset specifically includes: The differential interferogram is cropped according to a preset size to generate multiple image blocks; Each image patch is labeled to generate labeled data of settlement area boundary labels with coordinate information; Based on the labeled data, the model training dataset is generated; The coal mine subsidence area segmentation model is trained based on the model training dataset.

4. The method according to claim 3, characterized in that, The step of training the coal mine subsidence area segmentation model based on the model training dataset specifically includes: Construct a YOLOv11 model based on the attention mechanism; The training dataset is input into the attention-based YOLOv11 model, and the model parameters are backpropagated and gradients are calculated based on the binary cross-entropy loss function. The Adam optimizer is used to iteratively update the model parameters of the attention-based YOLOv11 model based on the gradient calculation results. When the change in the loss function is less than a preset threshold or the number of iterations equals the maximum number of iterations, the YOLOv11 model based on the attention mechanism is output as the coal mine subsidence area segmentation model.

5. The method according to claim 4, characterized in that, The attention-based YOLOv11 model includes an input layer, a feature extraction layer, and a segmentation head, wherein the input layer is a Backbone network; The steps for constructing a YOLOv11 model based on an attention mechanism specifically include: The DB-SimAM attention mechanism is added to the tail of the Backbone network and the head of the segmentation head of the YOLOv11 model to form the YOLOv11 model based on the attention mechanism.

6. The method according to claim 1, characterized in that, The step of determining the deformation boundary of the coal mine based on the binary image of the coal mine subsidence area specifically includes: The binary image of the coal mine subsidence area is subjected to noise reduction processing; Morphological optimization operations were performed on the denoised binary image of the coal mine subsidence area. The contour information of the optimized binary image of the coal mine subsidence area is extracted to obtain the deformation boundary of the coal mine.

7. The method according to claim 1, characterized in that, After determining the deformation boundary of the coal mine based on the binary image of the coal mine subsidence area, the process further includes: Align the coal mine deformation boundaries at different time points according to the observation time sequence; Acquire spatial variation data of coal mine deformation boundary between two adjacent time points, wherein the spatial variation data includes displacement variation data, area variation data and expansion direction data; Based on continuous temporal spatial variation data, a spatiotemporal evolution map of coal mine deformation boundaries is generated.

8. A coal mine deformation boundary identification device, characterized in that, include: The first generation module is used to generate differential interferograms based on multi-temporal synthetic aperture radar data of the target coal mine area. The second generation module is used to generate a model training dataset based on the differential interferogram, and the training module is used to train a coal mine subsidence area segmentation model using the model training dataset. The third generation module is used to input the differential interferogram to be identified into the coal mine subsidence area segmentation model to perform subsidence area segmentation operation and obtain a binary image of the coal mine subsidence area. The determination module is used to determine the deformation boundary of the coal mine based on the binary image of the coal mine subsidence area.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the coal mine deformation boundary identification method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the coal mine deformation boundary identification method as described in any one of claims 1 to 7.

Images