Method and device for extracting coal mining subsidence area based on HRNet model
Through the HRNet model combined with SBAS-InSAR and D-InSAR technology, coal subsidence areas are automatically extracted, solving the high cost and low efficiency problems of traditional surveying and mapping methods in the deformation monitoring of mining areas, and achieving more accurate and efficient subsidence areas.
Patent Information
- Application Number
- CN202411096041.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-12
- Publication Date
- 2024-12-03
AI Technical Summary
Traditional surveying and mapping methods have problems such as high cost, low efficiency and inability to fully reveal the overall deformation laws of the mining area in the mining area, and the monitoring points are easily damaged, making it difficult to achieve systematic deformation monitoring.
Using a method based on the HRNet model, combined with SBAS-InSAR and D-InSAR technology, SAR images are processed, deformation information is obtained and training sample data is obtained through visual interpretation, and the HRNet network model is trained to automatically extract the coal mining subsidence area.
It improves the identification accuracy and efficiency of coal mining subsidence areas, reduces the phenomenon of misclassification of leakage classification, provides more complete details of subsidence areas, and reduces the uncertainty of manual interpretation.
Smart Images

Figure CN119068350A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of radar remote sensing processing and application technology, and in particular to a method and device for extracting coal mining subsidence areas based on a HRNet model. Background Art
[0002] With the large-scale mining of coal, ground subsidence and infrastructure damage may occur, which threaten the local natural environment and property safety. Therefore, it is of great significance to monitor the deformation of the mining area and extract the scope of the mining area subsidence. Although traditional surveying and mapping methods such as leveling and GPS measurement have high accuracy in mining area deformation monitoring, they also have obvious limitations and shortcomings. Specifically, these methods require a lot of manpower and material resources, resulting in high operating costs, long measurement cycles, and low monitoring efficiency. In addition, the problem that monitoring points are easily damaged cannot be ignored. More importantly, these traditional surveying and mapping methods can only provide measurement results of discrete points, and cannot comprehensively and systematically reveal the overall deformation and mining subsidence laws of the mining area, which to a certain extent limits their application in mining area deformation monitoring. Summary of the invention
[0003] In view of this, the embodiments of the present disclosure provide a method and device for extracting coal mining subsidence areas based on the HRNet model, which at least partially solve the problems existing in the prior art.
[0004] In a first aspect, an embodiment of the present disclosure provides a method for extracting coal mining subsidence areas based on a HRNet model, comprising:
[0005] Using SBAS-InSAR technology, the SAR image of the target area is processed to obtain the deformation information of the target area;
[0006] Using D-InSAR technology, deformation monitoring is performed on the area containing the mine to obtain a phase unwrapping result map reflecting the surface deformation of the area;
[0007] Visually interpreting the phase unwrapping result image to obtain image blocks containing coal mining subsidence areas as training sample data;
[0008] Using the training sample data to train the HRNet network model to obtain a trained HRNet network model; and
[0009] The deformation information of the target area is input into the trained HRNet network model to obtain the coal mining subsidence area.
[0010] In a specific embodiment, the method further includes manually interpreting the deformation information of the target area, and comparing the manually interpreted deformation information with the coal mining subsidence area obtained through the HRNet network model.
[0011] In a specific embodiment, interference image pairs collected in autumn and winter are selected to monitor deformation of an area including a mine.
[0012] In a specific embodiment, the step of using D-InSAR technology to monitor the deformation of the area including the mine to obtain a phase unwrapping result map reflecting the surface deformation of the area includes:
[0013] Select a main image from the SAR images;
[0014] Performing image registration on the SAR images of other time phases and the main image;
[0015] Generate interferogram based on external DEM data and registered SAR image data;
[0016] A differential interference image is generated based on the interference image, the differential interference image is filtered, and phase unwrapping is performed on the filtered differential interference image to obtain a phase unwrapping result image.
[0017] In a specific embodiment, the step of performing phase unwrapping on the filtered differential interferogram to obtain a phase unwrapping result graph includes:
[0018] Identify phase changes in the filtered differential interferogram due to surface deformation;
[0019] Applying a phase unwrapping algorithm to estimate the phase hidden in the phase variation; and
[0020] The estimated hidden phase is added to the original phase to restore the phase integrity.
[0021] In a specific implementation, a two-track method is used to obtain the differential interference pattern.
[0022] In a specific embodiment, visually interpreting the phase unwrapping result image to obtain an image block containing a coal mining subsidence area as training sample data includes:
[0023] ArcGIS software is used to visually interpret the phase unwrapping result map;
[0024] Manually circle the coal mining subsidence area in the phase unwrapping result diagram;
[0025] The phase unwrapping result image is cropped according to a specification of 512×512;
[0026] Image blocks containing coal mining subsidence areas are selected from the cropped image blocks as training sample data.
[0027] In a specific embodiment, the segmentation performance of the obtained HRNet network model is also evaluated by accuracy, precision and average intersection-over-union.
[0028] In a specific embodiment, the deformation of the area including the mine is obtained by the following formula:
[0029]
[0030] In the formula, ΔR is the deformation, λ is the wavelength, is the deformation phase.
[0031] In a second aspect, the embodiment of the present disclosure provides a coal mining subsidence area extraction device based on the HRNet model, comprising:
[0032] The target area deformation information acquisition device uses SBAS-InSAR technology to process the SAR image of the target area to obtain the deformation information of the target area;
[0033] A phase unwrapping result map acquisition device uses D-InSAR technology to monitor the deformation of the area containing the mine to obtain a phase unwrapping result map reflecting the surface deformation of the area;
[0034] A training sample data acquisition device is used to visually interpret the phase unwrapping result image to obtain an image block containing a coal mining subsidence area as training sample data;
[0035] A HRNet network model training device, which trains the HRNet network model using the training sample data to obtain a trained HRNet network model; and
[0036] The coal mining subsidence area acquisition device inputs the deformation information of the target area into the trained HRNet network model to obtain the coal mining subsidence area.
[0037] The coal mining subsidence area extraction method based on the HRNet model in the embodiment of the present disclosure includes using SBAS-InSAR technology to process the SAR image of the target area to obtain the deformation information of the target area; using D-InSAR technology to monitor the deformation of the area containing the mine to obtain a phase unwrapping result map reflecting the surface deformation of the area; visually interpreting the phase unwrapping result map to obtain image blocks containing coal mining subsidence areas as training sample data; using the training sample data to train the HRNet network model to obtain a trained HRNet network model; and inputting the deformation information of the target area into the trained HRNet network model to obtain the coal mining subsidence area. Through the processing scheme disclosed in the present disclosure, high-resolution features can be maintained without loss, and the detailed information of the subsidence area can be more completely retained and utilized, thereby improving the accuracy of the prediction results and reducing the phenomenon of missed classification and misclassification of the subsidence area. BRIEF DESCRIPTION OF THE DRAWINGS
[0038] In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.
[0039] Figure 1 A schematic diagram of a process for extracting coal mining subsidence areas based on the HRNet model provided in an embodiment of the present disclosure;
[0040] Figure 2 A diagram of the experimental area provided for the embodiments of the present disclosure;
[0041] Figure 3 A diagram of an image data set selected for an embodiment of the present disclosure;
[0042] Figure 4 is a deformation rate result diagram of an embodiment of the present disclosure;
[0043] Figure 5 The D-InSAR technical principle of the embodiment of the present disclosure;
[0044] Figure 6 a disentanglement graph generated for an embodiment of the present disclosure;
[0045] Figure 7 A graph showing sample labels of a training data set generated according to an embodiment of the present disclosure. DETAILED DESCRIPTION
[0046] The embodiments of the present disclosure are described in detail below with reference to the accompanying drawings.
[0047] The following describes the embodiments of the present disclosure through specific examples, and those skilled in the art can easily understand other advantages and effects of the present disclosure from the contents disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of the present disclosure, rather than all of the embodiments. The present disclosure can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present disclosure. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present disclosure.
[0048] It should be noted that various aspects of the embodiments within the scope of the appended claims are described below. It should be apparent that the aspects described herein may be embodied in a wide variety of forms, and any specific structure and / or function described herein is merely illustrative. Based on the present disclosure, it should be understood by those skilled in the art that an aspect described herein may be implemented independently of any other aspect, and two or more of these aspects may be combined in various ways. For example, any number of aspects described herein may be used to implement the device and / or practice the method. In addition, other structures and / or functionalities other than one or more of the aspects described herein may be used to implement this device and / or practice this method.
[0049] It should also be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present disclosure. The drawings only show components related to the present disclosure rather than being drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component may be changed arbitrarily, and the component layout may also be more complicated.
[0050] Additionally, in the following description, specific details are provided to facilitate a thorough understanding of the examples. However, it will be understood by those skilled in the art that the aspects described may be practiced without these specific details.
[0051] For the extraction of the scope of coal mining subsidence areas, the Interferometric Synthetic Aperture Radar (InSAR) technology, with its unique all-day and all-weather monitoring capabilities, provides an efficient and accurate deformation monitoring solution for coal mining areas, which can effectively make up for the shortcomings of traditional surveying and mapping methods in dealing with large-scale deformation monitoring.
[0052] Small Baseline Subset InSAR (SBAS-InSAR) is an improved InSAR technology that combines the small baseline set method and differential interferometry. By processing multiple synthetic aperture radar SAR images, this technology can provide high-precision surface deformation monitoring and geological change analysis over a long span. When extracting coal mining subsidence areas, SBAS-InSAR technology usually relies on manual visual analysis and interpretation of the interferometric result images. However, due to the lack of unified standards for visual interpretation, different interpreters may obtain different extraction results, which to a certain extent increases the uncertainty and variability of the results of extracting coal mining subsidence areas.
[0053] D-InSAR (Differential Interferometric Synthetic Aperture Radar) is a differential interferometric synthetic aperture radar technology that measures tiny deformations of the surface by analyzing the phase difference between two or more SAR images. This technology can detect millimeter-level deformations and is therefore widely used in earthquakes, volcanic activity, ground subsidence, crustal movement, and large-scale structural health monitoring.
[0054] The basic principle of D-InSAR technology is to use SAR images acquired twice or more times at different times to extract the deformation information of the ground surface in the direction of the radar line of sight (LOS) through interferometric measurement technology. The specific steps include image registration, interferogram generation, noise filtering, baseline estimation, flat ground effect elimination, phase unwrapping, geocoding, etc.
[0055] In addition, deep learning technology has developed rapidly in recent years. Through training, this technology can automatically learn the intrinsic characteristics of sample images and encode these characteristics into external feature models. It can be efficiently applied to sample images of the same type to quickly and accurately realize feature extraction, image recognition and classification.
[0056] The mining subsidence area shows small and scattered spatial features in the deformation result image. At the same time, the atmospheric residual error contained in the deformation result image will interfere with the identification of the subsidence area. When using traditional convolutional neural networks to identify mining subsidence areas, there are problems such as information loss, low edge extraction accuracy, and unsatisfactory classification results. It is easy to misclassify or miss the area of interest. Since the mining subsidence area has obvious block characteristics and obvious contextual semantic characteristics, its identification requires special attention to boundary information.
[0057] The HRNet (High-Resolution Network) network model is a deep learning network model designed for visual tasks that require high-resolution representation, such as human pose estimation, semantic segmentation, and target detection. HRNet always maintains high-resolution feature maps during the forward propagation of the entire network. The core idea is to connect high-resolution subnetworks and low-resolution subnetworks in parallel, which is in sharp contrast to the high-to-bottom series method mainly used by traditional networks. This design enables the HRNet network to continuously maintain high-resolution feature representation without the need to design additional complex upsampling modules from low to high, while also maintaining the accuracy of classification tasks. When processing features of different scales and resolutions, HRNet cleverly implements information exchange, allowing the network model to make full use of spatial location information. This feature is particularly critical in preserving the edge details of the segmented area, especially for coal mining shafts with scattered distribution and variable shapes. The HRNet model's feature extraction capability is significantly improved, providing strong support for fine coal mine monitoring and mining area boundary identification. Therefore, it is of great significance to invent a complete method to automatically extract the boundaries of mining subsidence areas from SBAS-InSAR monitoring results using the HRNet model, which can not only solve the problem of inconsistent visual interpretation standards in mining subsidence areas, but also improve the accuracy and recognition efficiency.
[0058] In view of the problem that it is difficult to accurately extract the boundaries of the mining subsidence area mentioned above, the present invention proposes a method for extracting the surface deformation range of the mining area from the SBAS-InSAR image using the HRNet model. In addition, the applicant also compares the results extracted by the method of this application with the surface deformation results monitored by SBAS-InSAR technology to prove the feasibility of the method. It provides a data basis for the rapid and accurate delineation of the scope of mining subsidence and provides a reference for risk prevention and subsequent management of subsidence areas.
[0059] The idea of the present invention is to use satellite data for differential interferometric measurement (Differential Intereferometric Synthetic Aperture Radar, D-InSAR) processing to obtain a large number of subsidence area samples and select the HRNet network model for training.
[0060] Next, with reference to the accompanying drawings, the coal mining subsidence area extraction method based on the HRNet model of the present invention is described in detail. Figure 1 , describing the method of the present invention, it should be noted that, in Figure 1 The left side is the deformation information of the coal mining subsidence area obtained by SBAS-InSAR technology, and Figure 1The right side of is the process of training the HRNet model using the image data obtained by D-InSAR technology. In the present invention, the deformation information of the coal mining subsidence area obtained by SBAS-InSAR technology is used to evaluate the deformation information output by the HRNet model. This is because the deformation information of the coal mining subsidence area obtained by SBAS-InSAR technology is highly accurate and can effectively judge the accuracy of the HRNet model.
[0061] The coal mining subsidence area extraction method based on the HRNet model of the present invention comprises:
[0062] S100: Obtaining a deformation rate result map (deformation information) of the coal mining subsidence area. Specifically, in the present invention, SBAS-InSAR technology is used to perform fine interference processing on SAR images to obtain a deformation rate result map of the area. This result map is the data basis for understanding the surface deformation characteristics.
[0063] In a specific embodiment, for example, Datong Coal Mine in Shanxi Province can be selected as the research object. Figure 2 As shown in Figure 2, multi-view time-series SAR image data of Datong coal mining area were taken by satellite (such as 4D Gaojing-2 satellite) to conduct surface deformation monitoring research. The selected image data set is as follows: Figure 3 The SBAS-InSAR technology was used to obtain the deformation rate result diagram of the surface in the radar line of sight (LOS). The deformation rate result diagram is shown in Figure 4 shown.
[0064] Specifically, the steps of processing the SAR image of the coal mining subsidence area (target area) by using the SBAS-InSAR technology to obtain the deformation information of the coal mining subsidence area include:
[0065] S101: Acquire SAR images of coal mining areas. SAR images, namely Synthetic Aperture Radar images, are a technology that uses a radar system to transmit and receive microwave signals to acquire surface images. The acquisition of SAR images usually involves SAR equipment carried by satellites or aircraft, which transmits and receives radar signals, records the time delay and intensity of echo signals, and then generates SAR images through signal processing. In the present invention, SAR images can be acquired by the multi-view four-dimensional Gaojing-2 satellite.
[0066] S102: Select reference images from the acquired SAR images, and perform registration and resampling. In SBAS-InSAR technology, the selection of SAR images, determination of reference images, registration and resampling are key steps: First, select a series of SAR images with small baselines to ensure that the deformation difference between adjacent images is small, thereby reducing the influence of atmospheric and orbital errors; then, determine one or more reference images that have high coherence throughout the time series as a benchmark to reduce the phase difference between other images and them; then, perform image registration to ensure that all images are accurately aligned in space, which is usually achieved by finding matching points between the reference image and other images to compensate for possible geometric deviations; finally, resample the images to unify the spatial resolution and pixel size of all images to ensure the accuracy of interferometric measurement. These steps together lay a solid foundation for subsequent interferogram generation, phase unwrapping and deformation analysis.
[0067] S103: Forming interferometric image pairs based on baseline thresholds. In SBAS-InSAR technology, the step of forming interferometric image pairs based on baseline thresholds is a key step in ensuring data quality. First, the baseline distance between each pair of SAR images is calculated, which is achieved by determining the position change of the sensor relative to the surface during two imagings; then, a baseline length threshold is set, which is usually a function of the image acquisition time and spatial baseline to ensure that the time delay and spatial change between interferometric image pairs are within an acceptable range; then, based on the baseline length and time interval, image pairs that meet the threshold conditions are selected for interferometric processing. The baselines of these image pairs are shorter, which can ensure the coherence and accuracy of interferometric measurement; finally, the interferometric image pairs formed in this way can reduce the impact of atmospheric and orbital errors and improve the reliability of deformation monitoring. This process is the basis for achieving high-precision surface deformation monitoring in SBAS-InSAR technology.
[0068] S104: Generate interferogram. In SBAS-InSAR technology, after forming an interferogram pair based on the baseline threshold, the steps to generate the interferogram are as follows: First, perform image registration on the selected interferogram pair to ensure that the two images are accurately aligned at the pixel level; then, use the registered image to calculate the phase difference between the two SAR images to obtain the initial interferogram, which contains the terrain phase and possible deformation information; then, use the known DEM data to remove the terrain effect. This step is called deflating, which is achieved by subtracting the simulated terrain phase from the interferogram; then, perform multi-view processing and filtering to improve the signal-to-noise ratio and spatial resolution of the interferogram; finally, generate the coherence map of the interferogram to evaluate the quality of the interferometric data and ensure the reliability of subsequent analysis.
[0069] S104: Generate a correlation coefficient map and a differential interferogram based on the interferogram and external DEM data. In the SBAS-InSAR technology, the steps of generating a correlation coefficient map and a differential interferogram based on the interferogram and external DEM data are: first, use the external DEM data to correct the terrain phase in the interferogram, simulate the phase change caused by the terrain through the DEM data, and subtract this part of the simulated phase from the original interferogram to eliminate the terrain effect, and obtain an interferogram that mainly reflects the surface deformation and the atmospheric delay effect; then, perform a coherence analysis on the corrected interferogram, calculate the coherence coefficient of each pixel point, and generate a coherence coefficient map, which reflects the signal consistency of the pixel point between two imagings. A high coherence coefficient indicates that the radar signal is stable and suitable for deformation analysis; then, use the coherence coefficient map to identify the pixels in the high coherence area, perform differential processing on these pixels, extract the surface deformation information, and generate a differential interferogram.
[0070] In addition, it also involves phase unwrapping of differential interferograms. First, identify the phase changes caused by surface deformation in the differential interferogram, which may exceed the observable range of the phase (i.e., the period of 2π), resulting in phase wrapping; then, apply phase unwrapping algorithms, such as minimum cost flow, genetic algorithm, or multi-baseline method, to estimate the integer multiples of 2π hidden in the phase change. This process needs to consider the phase continuity and consistency in the entire interferogram; then, add the estimated integer multiples of 2π to the original phase to restore the integrity of the phase and solve the phase wrapping problem; then, use the unwrapped phase to calculate the exact value of the surface deformation, which usually involves converting the phase change into an actual displacement or deformation rate; finally, generate a differential interferogram after phase unwrapping, which provides an accurate data basis for further analysis and interpretation of surface deformation.
[0071] S105: SVD decomposition is used to obtain linear deformation based on high coherence points and phase unwrapping. Specifically, first, high coherence points are screened out from the coherence coefficient map. These points usually correspond to areas with stable surface and consistent radar scattering characteristics, such as building corners or rock surfaces; then, these high coherence points are phase unwrapped to obtain continuous and unwrapped deformation time series; then, the phase change data of these points in multiple interference cycles are collected and the deformation matrix is constructed; then, the deformation matrix is decomposed by applying SVD decomposition technology to extract the main deformation modes and corresponding singular values, which represent the linear components of surface deformation; then, through the SVD decomposition results, the signals related to linear deformation are identified and separated, and nonlinear or random deformation noise is excluded; finally, these linear deformation components are used to estimate the deformation trend and rate of the entire monitoring area, providing a scientific basis for surface deformation analysis. This method can effectively extract stable linear deformation features from complex deformation time series and improve the accuracy and reliability of deformation monitoring.
[0072] S106: Remove the deformation model and obtain the residual phase. In SBAS-InSAR technology, after the linear deformation model is obtained by singular value decomposition (SVD), the steps of removing the model and obtaining the residual phase are as follows: First, a mathematical model describing the linear deformation trend of the entire monitoring area is constructed based on the linear deformation parameters obtained by SVD decomposition; then, this linear deformation model is applied to the unwrapped phase data to predict the deformation phase of each pixel during the monitoring period; then, the predicted linear deformation phase is subtracted from the actual observed phase data to obtain the residual phase, which represents the surface deformation information after removing the influence of linear deformation, and can be the result of crustal movement, surface subsidence or other nonlinear deformation processes; finally, the residual phase is further analyzed to identify and study local deformation features or abnormal deformation events in the region.
[0073] S107: Perform time and space filtering to obtain atmospheric delay phase and nonlinear deformation. In SBAS-InSAR technology, after removing the deformation model to obtain the residual phase, the steps of performing time and space filtering to obtain the atmospheric delay phase and nonlinear deformation are as follows: First, perform time series analysis on the residual phase data to identify and separate the phase delay associated with changes in atmospheric conditions, which usually involves modeling and filtering periodic or non-periodic signals in the time series; then, apply spatial filtering techniques, such as Gaussian filtering or median filtering, to smooth spatially discontinuous noise and outliers while retaining useful information of surface deformation; then, extract the atmospheric delay phase through the combination of time and space filtering, which helps to further correct the atmospheric effects in the residual phase and improve the accuracy of deformation measurement; then, analyze the filtered residual phase to identify nonlinear deformation signals, which may be related to local geological activities, building construction or other non-systematic factors on the surface; finally, use these filtered residual phase data to conduct a more in-depth analysis of surface deformation to provide accurate deformation information for geological disaster assessment, infrastructure monitoring and environmental change research.
[0074] After the nonlinear deformation is obtained, the deformation information can be manually extracted to serve as a comparison basis for the deformation information extracted by HRNet.
[0075] The above describes the process of obtaining deformation information of coal mining subsidence area by using SBAS-InSAR technology. Next, the steps of constructing the HRNet model of the present invention are described.
[0076] S200: In the coal mining areas in Shanxi Province and its surrounding areas, a large number of interferometric image pairs collected in autumn and winter were selected, because the data collected during this period are less affected by weather and vegetation and have higher coherence. Deformation monitoring of the mining area was carried out using D-InSAR technology. After data processing and analysis, phase unwrapping result maps reflecting the surface deformation of these mining areas were obtained.
[0077] Specifically, the steps of using D-InSAR technology to monitor the deformation of the mining area and obtain the phase unwrapping result map reflecting the surface deformation of the mining area include:
[0078] S201: Select the main image. In D-InSAR technology, the steps of selecting the main image from SAR images are: first, collect a series of temporally continuous SAR image data, which cover the same area but have different imaging times; then, select one or more images with high coherence as the main image based on the quality and coherence of the image, and the selection of the main image takes into account its coherence with other images in the entire time series, as well as its position in the time series, to ensure that the key period of surface deformation can be captured; then, preprocess the selected main image, including image focusing, denoising and possible geometric correction, to ensure its accuracy as a reference; in addition, the selection of the main image should also take into account its representativeness in the time series, so as to facilitate effective interference comparison with other images and capture the characteristics of surface deformation; finally, the selected main image will be used as a benchmark and paired with other images to generate multiple interference maps to provide basic data for further deformation analysis and monitoring.
[0079] S202: Perform image registration. In D-InSAR technology, the steps of image registration after selecting the main image are as follows: first, the selected main image is accurately geocoded to ensure that it has an accurate geographic reference; then, the SAR images of other phases are registered with the main image. This process usually involves geometric and radiometric corrections between images to eliminate differences caused by terrain, sensor viewing angle or radar beam characteristics; then, image processing algorithms, such as cross-correlation or feature matching methods, are used to accurately calculate and compensate for the spatial deviation between the two images to ensure that each pixel point corresponds to the same surface position in different images; in addition, the pixel size and direction of the image may also need to be adjusted to achieve the best registration effect; finally, the registered image will be used to generate an interferometric pair to extract surface deformation information by calculating the phase difference between the main image and other images.
[0080] S203: Generate interferogram. In D-InSAR technology, the steps of generating interferogram after image registration are as follows: First, the phase difference between the main image and other phase images is calculated by precise interferometric processing algorithm using the registered SAR image data. These phase differences contain surface deformation and terrain information; then, the external DEM data is used to simulate and remove the terrain effect in the interferogram. This step is called terrain correction, which is achieved by subtracting the terrain phase simulated by DEM data from the phase difference; then, the corrected interferogram is multi-viewed and filtered to improve the phase stability and quality of the interferogram; then, the coherence map of the interferogram is generated to evaluate the quality and coherence level of the interferometric data to ensure that high coherence areas are selected for analysis; finally, the interferogram obtained through these steps reflects the deformation information of the surface in the radar line of sight (LOS), providing basic data for further deformation monitoring and analysis.
[0081] S104: Generate a differential interference pattern. The step of generating a differential interference pattern is similar to the method of generating a differential interference pattern in the above step S104, and will not be described again.
[0082] S105: Differential interferogram filtering. In D-InSAR technology, the steps of filtering the differential interferogram are as follows: first, identify and analyze the noise characteristics in the differential interferogram, including phase noise and possible speckle noise; then, select a suitable filtering algorithm, such as Gaussian filtering, median filtering or adaptive filtering, to smooth the noise and retain valuable deformation signals; then, apply the selected filtering algorithm to process the differential interferogram, adjust the filtering parameters to balance noise suppression and signal fidelity, and avoid excessive filtering leading to loss of deformation information; then, evaluate the filtering effect to ensure that the filtered interferogram can accurately reflect the surface deformation characteristics while removing noise; finally, use the filtered differential interferogram for subsequent phase unwrapping and deformation analysis to provide support for obtaining clearer and more reliable surface deformation information. This filtering step is crucial to improving the accuracy and reliability of deformation monitoring in D-InSAR technology.
[0083] S106: Phase unwrapping. In D-InSAR technology, the steps of phase unwrapping of filtered differential interferograms are as follows: first, identify the phase changes caused by surface deformation in the filtered differential interferograms, which may exceed the observable range of the phase (i.e., the period of 2π), resulting in phase wrapping; then, apply a phase unwrapping algorithm, such as the least squares method, the region growing method, or the genetic algorithm, to estimate the integer multiples of 2π hidden in the phase change. This process needs to consider the phase continuity and consistency in the entire interferogram; then, add the estimated integer multiples of 2π to the original phase to restore the integrity of the phase and solve the phase wrapping problem; then, verify the accuracy of the unwrapping result, usually by checking whether the unwrapped phase map is smooth and physically reasonable; finally, use the unwrapped phase to calculate the exact value of the surface deformation, which usually involves converting the phase change into an actual displacement or deformation rate. This step is crucial for accurately extracting surface deformation information from D-InSAR data.
[0084] The term "phase unwrapping" refers to a concept in signal processing and image processing, especially in synthetic aperture radar (SAR) imaging and interferometric SAR (InSAR). In SAR imaging, since the phase information of the radar signal is quantized into a discrete number of bits, phase wrapping occurs when the phase change exceeds a certain threshold (called phase wrapping or phase jump). The goal of phase unwrapping is to restore the original continuous phase information.
[0085] Specifically, the present invention focuses on the mining areas in Shanxi Province and its surrounding areas, and selects a large number of interferometric image data collected in autumn and winter with short time-space baselines. These data are relatively less affected by the atmosphere and vegetation, have short time-space baselines, and have high coherence, which provides favorable conditions for obtaining more accurate surface deformation information. The D-InSAR technology is used to generate an unwrapped map by processing the interferometric image data set through registration, differential interference, filtering, phase unwrapping, etc. The generated unwrapped map is as follows: Figure 6 shown.
[0086] In the present invention, the unwrapped result map is used as the original data when constructing the data set. This choice is based on a key consideration: the deformation area shows significant color information on the unwrapped image, and these color information forms a sharp contrast with the surrounding background, which is convenient for the recognition of the deep learning model. In addition, directly using the unwrapped image to extract the coal mining subsidence area can save part of the InSAR data processing process, save calculation and time costs, and make the entire analysis process more efficient. The differential interference signal extraction method adopted by the present invention is the two-track method, which mainly relies on two SAR images obtained before and after the deformation occurs and the external DEM data of the area. Differential interference processing can accurately obtain the deformation information of the ground in the line of sight direction LOS.
[0087] The principle of D-InSAR technology is as follows Figure 5 As shown in the figure, S1 and S2 represent the satellite sensors that take the primary and secondary images respectively. Assuming that the ground deforms upward in the LOS direction when the satellite passes twice, the surface deformation is expressed as:
[0088]
[0089] In the formula, ΔR is the surface deformation, λ is the wavelength, is the deformation phase.
[0090] S300: Perform detailed visual interpretation of the phase unwrapping result images obtained in step S200 one by one, aiming to accurately determine the specific subsidence areas of each coal mining area and mark them. The image blocks that have been delineated to contain the subsidence boundaries are used as data sets to produce training samples for deep learning, which are used to train the model to improve the recognition ability of mining area subsidence.
[0091] In the present invention, ArcGIS software can be used to visually interpret the unwrapping result map obtained in step S200, and the coal mining subsidence area is manually circled in the unwrapping map. In order to improve the rapid acquisition of sample features by the model, a unit of 512×512 specifications is selected as the training sample size for deep learning and the unwrapping map is cropped according to this specification. The image blocks containing the coal mining subsidence area are selected from the cropped image blocks. The sample label sample of the generated training data set is as follows: Figure 7 shown.
[0092] S400: After the data set is ready, the HRNet network model is used for training. By continuously adjusting the parameters and structure of the model, the model can gradually learn the rules and features in the data set, so that the boundaries of the deformation area of the mining area can be accurately identified. After the model training is completed, the accuracy of the produced samples needs to be evaluated. A part of the independent test set is used to evaluate the prediction accuracy of the model and calculate various evaluation indicators.
[0093] When training the HRNet model, the prepared data set needs to be divided into a training set, a validation set, and a test set. The training set is used for learning and optimizing the model, the validation set is used to adjust hyperparameters and prevent overfitting, and the test set is used to evaluate the generalization ability of the model. Through multiple iterations of training, the model will gradually learn the rules and features in the data, so that it can make accurate predictions on new samples. Specifically, when the HRNet model is used to extract the scope of mining subsidence, samples of the subsidence area are used to establish a sample set and train the HRNet network to obtain the optimal model weights, and then the HRNet model is used to extract features and segment the image. HRNet can capture deformation information of different scales through multi-scale feature fusion and parallel convolutional network design, thereby achieving accurate segmentation of the subsidence area. The following indicators are used to evaluate the segmentation performance of the network model, namely accuracy, precision, and average intersection-over-union ratio.
[0094]
[0095]
[0096]
[0097] In the formula, TP (True Positive) is the true positive, which refers to the number of pixels that are marked as deformation areas in the label and are correctly identified as deformation areas by the classifier's detection results. TN (True Negative) is the true negative, which refers to the number of pixels that are marked as non-deformation areas in the label and are also correctly identified as non-deformation areas by the classifier. FP (False Positive) is the false positive, which refers to the number of pixels that are actually non-deformation areas in the label but are mistakenly identified as deformation areas by the classifier. FN (False Negative) is the false negative, which refers to the number of pixels that are clearly marked as deformation areas in the label but the classifier fails to correctly identify them and mistakenly judges them as non-deformation areas.
[0098] S500: Manual visual interpretation of the deformation rate map generated by SBAS-InSAR technology in step S100 to extract the results of coal mining subsidence area. Then, the specific boundary information of the coal mining subsidence area extracted by the HRNet model is superimposed on the manually interpreted deformation rate map, and the result is compared with the surface deformation result of visual interpretation to prove the feasibility of the method of the present invention and to more comprehensively analyze and understand the distribution of coal mining subsidence area.
[0099] In addition, the present invention also provides a coal mining subsidence area extraction device based on the HRNet model, the device comprising:
[0100] The target area deformation information acquisition device uses SBAS-InSAR technology to process the SAR image of the target area to obtain the deformation information of the target area;
[0101] A phase unwrapping result map acquisition device uses D-InSAR technology to monitor the deformation of the area containing the mine to obtain a phase unwrapping result map reflecting the surface deformation of the area;
[0102] A training sample data acquisition device is used to visually interpret the phase unwrapping result image to obtain an image block containing a coal mining subsidence area as training sample data;
[0103] A HRNet network model training device, which trains the HRNet network model using the training sample data to obtain a trained HRNet network model; and
[0104] The coal mining subsidence area acquisition device inputs the deformation information of the target area into the trained HRNet network model to obtain the coal mining subsidence area.
[0105] The various components and functions of the coal mining subsidence area extraction device based on the HRNet model of the present invention have been described above with reference to the accompanying drawings and will not be repeated here.
[0106] The present invention uses the deep learning method of the HRNet model to automatically extract the coal mining subsidence area on the deformation rate result image obtained by using the SBAS-InSAR technology, which can effectively solve the problem of inconsistent visual interpretation standards in mining subsidence areas.
[0107] The present invention uses the HRNet model to maintain high-resolution features without loss during the training process, and retains and utilizes the detailed information of the subsidence area more completely, thereby improving the accuracy of the prediction results and reducing the phenomenon of missed classification and misclassification of the subsidence area.
[0108] The present invention uses the HRNet model to efficiently identify and classify coal mining subsidence areas, saving time and labor costs.
[0109] The above is only a specific implementation of the present disclosure, but the protection scope of the present disclosure is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present disclosure should be included in the protection scope of the present disclosure. Therefore, the protection scope of the present disclosure should be based on the protection scope of the claims.
Claims
1. A method for extracting coal mining subsidence areas based on the HRNet model, characterized in that: include: Using SBAS-InSAR technology, the SAR image of the target area is processed to obtain the deformation information of the target area; Using D-InSAR technology, deformation monitoring is performed on the area containing the mine to obtain a phase unwrapping result map reflecting the surface deformation of the area; Visually interpreting the phase unwrapping result image to obtain image blocks containing coal mining subsidence areas as training sample data; Using the training sample data to train the HRNet network model to obtain a trained HRNet network model; as well as The deformation information of the target area is input into the trained HRNet network model to obtain the coal mining subsidence area.
2. The method for extracting coal mining subsidence area based on the HRNet model according to claim 1, characterized in that: The method also includes manually interpreting the deformation information of the target area, and comparing the manually interpreted deformation information with the coal mining subsidence area obtained through the HRNet network model.
3. The method for extracting coal mining subsidence area based on HRNet model according to claim 1, characterized in that: Interferometric image pairs collected in autumn and winter are selected to monitor deformation in the area containing the mine.
4. The method for extracting coal mining subsidence area based on HRNet model according to claim 1, characterized in that: The step of using the D-InSAR technology to monitor the deformation of the area including the mine to obtain a phase unwrapping result map reflecting the surface deformation of the area includes: Select a main image from the SAR images; Performing image registration on the SAR images of other time phases and the main image; Generate interferogram based on external DEM data and registered SAR image data; A differential interference image is generated based on the interference image, the differential interference image is filtered, and phase unwrapping is performed on the filtered differential interference image to obtain a phase unwrapping result image.
5. The method for extracting coal mining subsidence area based on the HRNet model according to claim 4 is characterized in that: The steps of performing phase unwrapping on the filtered differential interference image to obtain a phase unwrapping result image include: Identify phase changes in the filtered differential interferogram due to surface deformation; Applying a phase unwrapping algorithm to estimate the phase hidden in the phase variation; and The estimated hidden phase is added to the original phase to restore the phase integrity.
6. The method for extracting coal mining subsidence area based on HRNet model according to claim 4, characterized in that: The differential interference pattern is obtained using a two-track method.
7. The method for extracting coal mining subsidence area based on HRNet model according to claim 1, characterized in that: Visually interpreting the phase unwrapping result image to obtain image blocks containing coal mining subsidence areas as training sample data, including: ArcGIS software is used to visually interpret the phase unwrapping result map; Manually circle the coal mining subsidence area in the phase unwrapping result diagram; The phase unwrapping result image is cropped according to a specification of 512×512; Image blocks containing coal mining subsidence areas are selected from the cropped image blocks as training sample data.
8. The method for extracting coal mining subsidence area based on HRNet model according to claim 7, characterized in that: The segmentation performance of the obtained HRNet network model was also evaluated by accuracy, precision and average intersection-over-union.
9. The method for extracting coal mining subsidence area based on HRNet model according to claim 1, characterized in that: The deformation of the area containing the mine is obtained by the following formula: In the formula, ΔR is the deformation, λ is the wavelength, is the deformation phase.
10. A coal mining subsidence area extraction device based on HRNet model, characterized in that: include: The target area deformation information acquisition device uses SBAS-InSAR technology to process the SAR image of the target area to obtain the deformation information of the target area; A phase unwrapping result map acquisition device uses D-InSAR technology to monitor the deformation of the area containing the mine to obtain a phase unwrapping result map reflecting the surface deformation of the area; A training sample data acquisition device is used to visually interpret the phase unwrapping result image to obtain an image block containing a coal mining subsidence area as training sample data; A HRNet network model training device, which trains the HRNet network model using the training sample data to obtain a trained HRNet network model; as well as The coal mining subsidence area acquisition device inputs the deformation information of the target area into the trained HRNet network model to obtain the coal mining subsidence area.
Citation Information
Patent Citations
Mining subsidence area identification method and system, storage medium and electronic equipment
CN115331096A
Mining area semantic segmentation model construction method and device and mining area semantic segmentation method
CN115862014A
Intelligent identification method and device for underground mining subsidence area, electronic equipment and storage medium
CN115980745A
