Method and system for detecting surface subsidence of underground mine

Abstract

The present application relates to the technical field of geological monitoring, and discloses a kind of underground mine surface subsidence detection method and system.The method first obtains mining area interference phase diagram, then the feature extraction and phase recovery processing of interference phase diagram are carried out by the preset target phase unwrapping model to obtain the unwrapped absolute phase diagram, finally the deformation information extraction of absolute phase diagram obtains the detection result of mining area surface subsidence;Wherein, the target phase unwrapping model is based on the encoder-decoder architecture of Segformer-B0 model, after four Transform modules included in the encoder, respectively add hollow space pyramid pooling module and efficient channel attention module obtained, hollow space pyramid pooling module is used for multi-scale feature extraction to the input feature map;Efficient channel attention module is used for channel attention weighting to the input feature map, and the weighted feature map is output to the decoder;In this way, the subsidence detection accuracy of mining area surface can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of geological monitoring, and relates to a method and system for detecting surface subsidence of an underground mine. BACKGROUND

[0002] Continuous mining of an underground mine can easily cause surface subsidence, which in turn can cause disasters such as cracks in houses, deformation of roads, and water accumulation. Currently, the surface subsidence of an underground mine can be monitored by using an interferometric synthetic aperture radar (InSAR) technology. The InSAR data processing directly affects the accuracy of surface deformation monitoring. Among them, phase unwrapping is a core step of InSAR data processing, and the goal is to recover continuous absolute phase from wrapped mod 2pi phase, and then accurately quantify the surface deformation.

[0003] However, the traditional phase unwrapping method is mostly based on the continuity assumption of phase gradient, and is suitable for scenarios with gentle deformation and low noise intensity. Therefore, when the surface deformation of an underground mine is severe or there is strong noise interference, the traditional method is prone to phase unwrapping errors, resulting in poor accuracy of the detection result. SUMMARY

[0004] Therefore, the embodiments of the present application provide a method and system for detecting surface subsidence of an underground mine, which can improve the accuracy of the detection result of surface subsidence.

[0005] The technical scheme of the embodiments of the present application is as follows:

[0006] In a first aspect, a method for detecting surface subsidence of an underground mine is provided, comprising:

[0007] obtaining an interference phase image of a mining area;

[0008] performing feature extraction and phase recovery processing on the interference phase image by using a preset target phase unwrapping model to obtain an unwrapped absolute phase image;

[0009] extracting deformation information from the absolute phase image to obtain a detection result of surface subsidence of the mining area;

[0010] The target phase unwrapping model is obtained by adding a cavity spatial pyramid pooling module and an efficient channel attention module after four Transformer modules included in the encoder of a Segformer-B0 model based on the encoder-decoder architecture of the Segformer-B0 model. The cavity spatial pyramid pooling module is used for multi-scale feature extraction of the input feature map. The efficient channel attention module is used for channel attention weighting of the input feature map, and the weighted feature map is output to the decoder.

[0011] In some embodiments, the encoder is configured to perform the following steps:

[0012] The first feature map is obtained by modeling the global dependency relationship of the interferometric phase map using the Transformer module.

[0013] The first feature map is extracted using the hollow space pyramid pooling module to obtain the second feature map.

[0014] The first feature map and the second feature map are concatenated along the channel dimension and then convolved to obtain a fused feature map.

[0015] The efficient channel attention module performs channel attention weighting on the fused feature map to obtain a third feature map, which is then output to the decoder.

[0016] In some embodiments, the dilated spatial pyramid pooling module includes a convolutional branch, multiple dilated convolutional branches, a global average pooling branch, a channel splicing layer, and a convolutional compression layer; wherein the multiple dilated convolutional branches correspond to different dilation rates.

[0017] The void space pyramid pooling module is specifically used to perform the following steps:

[0018] Receive the first feature map, extract local features from the first feature map through the convolution branch and adjust the channel dimension to obtain a local feature map;

[0019] A multi-scale context feature map is obtained by performing multi-scale dilated convolution operations on the first feature map through multiple dilated convolution branches.

[0020] Global feature extraction is performed on the first feature map using the global average pooling branch to obtain a global feature map;

[0021] The local feature map, the multi-scale context feature map, and the global feature map are concatenated and convolved by the channel concatenation layer and the convolutional compression layer to obtain the second feature map.

[0022] In some embodiments, the efficient channel attention module includes a global average pooling layer, a first convolutional layer, and an activation function layer;

[0023] The high-efficiency channel attention module is specifically used to perform the following steps:

[0024] The fused feature map is received, and global average pooling is performed on the fused feature map through the global average pooling layer to obtain a global feature vector;

[0025] The global feature vector is subjected to cross-channel information interaction through the first convolutional layer and the activation function layer to generate channel attention weights.

[0026] The channel attention weights are multiplied channel by channel by the fused feature map to obtain a third feature map, which is then output to the decoder.

[0027] In some embodiments, the decoder is configured to perform the following operations:

[0028] The third feature maps output by the four efficient channel attention modules are received, and the four third feature maps are projected onto the same embedding dimension to obtain four projected feature maps.

[0029] The four projected feature maps are upsampled to the same spatial resolution to obtain four upsampled feature maps.

[0030] The four upsampled feature maps are concatenated along the channel dimension to obtain a concatenated feature map. The concatenated feature map is then fused and regressed to obtain the absolute phase map.

[0031] In some embodiments, the method further includes:

[0032] A training sample set is generated by simulating the deformation field and interference phase noise field of the mining area; wherein, the training sample set includes a wrapped phase map and the true absolute phase map corresponding to the wrapped phase map;

[0033] Using the wrapped phase map as input and the true absolute phase map as supervision label, the initial phase unwrapping model is iteratively trained until the initial phase unwrapping model converges, thus obtaining the target phase unwrapping model.

[0034] In some embodiments, generating a training sample set by simulating the deformation field and interference phase noise of the mining area includes:

[0035] Based on the preset geometric features of the mining area settling funnel, a simulated settling funnel is obtained by superimposing multiple two-dimensional Gaussian functions. The simulated settling funnel is then randomly superimposed on a real digital elevation model to obtain a simulated settling model. The geometric features include at least the width, stripe density, shape, and distribution pattern of the mining area settling funnel.

[0036] Based on preset radar system parameters and the simulated subsidence model, the deformation field of the mining area and the interferometric phase noise field are generated; wherein, the interferometric phase noise field includes geometric decoherence noise, temporal decoherence noise and thermal decoherence noise;

[0037] The deformation field of the mining area is superimposed with the interference phase noise field to obtain a continuous interference phase field. The continuous interference phase field is then subjected to phase wrapping to obtain the wrapped phase map. The deformation field of the mining area is used as the true absolute phase map.

[0038] A second aspect of this application provides a surface settlement detection system for underground mines, comprising:

[0039] The acquisition module is used to acquire an interference phase map of the mining area, wherein the interference phase map contains phase entanglement information caused by mining disturbance;

[0040] The phase unwrapping module is used to perform feature extraction and phase recovery processing on the interference phase map through a preset target phase unwrapping model to obtain the unwrapped absolute phase map; wherein, the target phase unwrapping model is based on the encoder-decoder architecture of the Segformer-B0 model, and a hole space pyramid pooling module and an efficient channel attention module are added after the four Transformer modules included in the encoder.

[0041] The settlement detection module is used to extract deformation information from the absolute phase map to obtain the detection results of surface settlement in the mining area.

[0042] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following:

[0043] In this embodiment of the invention, an interferometric phase map of the mining area is first acquired. Then, a preset target phase unwrapping model is used to perform feature extraction and phase recovery processing on the interferometric phase map to obtain an unwrapped absolute phase map. Finally, deformation information is extracted from the absolute phase map to obtain the detection result of surface subsidence in the mining area. The target phase unwrapping model is based on the encoder-decoder architecture of the Segformer-B0 model. A hollow spatial pyramid pooling module and an efficient channel attention module are added after the four Transformer modules included in the encoder. In this way, a multi-layer Transformer structure (i.e., four-stage Transformer modules) is used to model the global dependency of the image. A hollow spatial pyramid pooling module is introduced to extract multi-scale features and enhance the local structural expression ability. An efficient channel attention mechanism (ECA) is introduced in the feature fusion stage to guide the target phase unwrapping model to focus on key features and suppress redundant information, which can improve the accuracy of surface subsidence detection in the mining area. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort, wherein:

[0045] Figure 1 A schematic flowchart of the underground mine surface subsidence detection method provided in an embodiment of the present invention;

[0046] Figure 2 This is a schematic diagram of the target phase unwrapping model provided in an embodiment of the present invention;

[0047] Figure 3 This is a schematic diagram of the structure of the hollow space pyramid pooling module provided in an embodiment of the present invention;

[0048] Figure 4 This is a schematic diagram of the structure of the high-efficiency channel attention module provided in an embodiment of the present invention;

[0049] Figure 5 A schematic diagram illustrating the comparative experimental results provided in an embodiment of the present invention;

[0050] Figure 6 This is a schematic diagram of the structure of the underground mine surface subsidence detection system provided in an embodiment of the present invention. Detailed Implementation

[0051] 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. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. The following embodiments are used to illustrate the present invention, but are not intended to limit the scope of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0052] In the following description, references are made to “some embodiments,” which describe a subset of all possible embodiments. However, it is understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0053] It should be noted that the terms "first, second, and third" used in the embodiments of the present invention are only used to distinguish similar objects and do not represent a specific ordering of objects. It is understood that "first, second, and third" can be interchanged in a specific order or sequence where permitted, so that the embodiments of the present invention described herein can be implemented in an order other than that illustrated or described herein.

[0054] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which these embodiments of the invention pertain. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art and should not be interpreted in an idealized or overly formal sense unless specifically defined as herein.

[0055] The Segformer-B0 model is a lightweight real-time semantic segmentation model based on the Transformer architecture. It adopts an encoder-decoder architecture, combining hierarchical feature representation and lightweight design. It extracts multi-scale features through the Transformer encoder and generates semantic segmentation results through the segmentation head decoding.

[0056] Figure 1 This is a schematic flowchart illustrating a method for detecting surface subsidence in underground mines, provided by an embodiment of the present invention. The method can be executed via a control device, which may include at least one of a personal computer, laptop computer, smartphone, tablet computer, and portable wearable device; however, this embodiment does not limit the specific device used.

[0057] like Figure 1 As shown, the underground mine surface subsidence detection method provided in this embodiment of the invention may include steps S101-S103.

[0058] S101. Obtain the interferometric phase diagram of the mining area.

[0059] In some embodiments, the interferometric phase map includes phase entanglement information caused by mining disturbances.

[0060] For example, interferometric phase maps of mining areas can be acquired using interferometric synthetic aperture radar technology.

[0061] S102. The interference phase map is subjected to feature extraction and phase recovery processing through a preset target phase unwrapping model to obtain the unwrapped absolute phase map.

[0062] In some embodiments, the target phase unwrapping model is based on the encoder-decoder architecture of the Segformer-B0 model, with a hollow spatial pyramid pooling module and an efficient channel attention module added after the four Transformer modules in the encoder. Specifically, the Transformer modules are used to model the global dependencies of the interferometric phase map using a self-attention mechanism to obtain the first feature map; the hollow spatial pyramid pooling module is used to extract multi-scale features from the input feature map; and the efficient channel attention module is used to apply channel attention weighting to the input feature map and output the weighted feature map to the decoder.

[0063] In some embodiments, the encoder performs the following steps: modeling the global dependency of the interferometric phase map using the Transformer module to obtain a first feature map; extracting multi-scale features from the first feature map using the hollow spatial pyramid pooling module to obtain a second feature map; concatenating the first feature map and the second feature map along the channel dimension and performing convolutional compression to obtain a fused feature map; and applying channel attention weighting to the fused feature map using the efficient channel attention module to obtain a third feature map, which is then output to the decoder.

[0064] For example, Figure 2 A schematic diagram of a target phase unwrapping model is shown, as follows: Figure 2 As shown, the target phase unwrapping model adopts the encoder-decoder architecture of the Segformer-B0 model. The encoder includes an Overlap Patch Embedding module, followed by a feature extraction module. The feature extraction module comprises a four-stage Transformer module. Figure 2 The Transformer module consists of Transformer 1, Transformer 2, Transformer 3, and Transformer 4. Each Transformer module is followed by an Atrous Spatial Pyramid Pooling (ASPP) module and an Efficient Channel Attention (ECA) module. The output of the Efficient Channel Attention module is connected to the decoder.

[0065] In practical applications, after the control device acquires the interferometric phase image of the mining area, it inputs the image into the target phase unwrapping model. Upon receiving the interferometric phase image, the target phase unwrapping model performs downsampling and channel mapping processing on the image through an overlap patch embedding module. This converts the two-dimensional interferometric phase image into feature maps with different channel numbers. These feature maps with different channel numbers are then reassembled into a feature sequence adapted to the Transformer structure and output to the first of the four Transformer modules. The number of channels can be 64, 128, 320, and 512, corresponding to resolutions reduced to 1 / 4, 1 / 8, 1 / 16, and 1 / 32 of the input, respectively. The four Transformer modules sequentially model the global dependencies of the input feature sequences to obtain their respective first feature maps. Each Transformer module then outputs its first feature map to its corresponding dilated spatial pyramid pooling module. The dilated spatial pyramid pooling module performs multi-scale feature extraction on the first feature map, capturing feature information from different receptive fields using dilated convolution kernels with different dilation rates to obtain the second feature map. After obtaining the second feature map, the encoder can use a splicing layer set between the hollow space pyramid pooling module and the efficient channel attention module. Figure 2 (not shown in the image) and convolutional layers ( Figure 2 (Not shown in the image) The first and second feature maps of the same stage are concatenated and fused along the channel dimension to obtain a fused feature map, which is then output to the efficient channel attention module. After receiving the fused feature map, the efficient channel attention module applies channel attention weights to the fused feature map to obtain a third feature map, which is then output to the decoder.

[0066] In some embodiments, the Transformer module includes a layer normalization layer, an efficient self-attention mechanism layer, a hybrid feedforward network, and a residual connection layer. The overlapping patch embedding module outputs a feature sequence to the Transformer module. Upon receiving the feature sequence, the Transformer module performs layer normalization on the feature sequence using the layer normalization layer to obtain a normalized sequence, which is then output to the efficient self-attention mechanism layer. This efficient self-attention mechanism layer downsamples the key-value vectors, achieving global dependency modeling with lower computational complexity, thereby generating attention weights and performing weighted aggregation on the normalized sequence to output a self-attention feature sequence. Subsequently, the hybrid feedforward network performs a nonlinear transformation on the self-attention feature sequence to obtain the transformed feature sequence. The hybrid feedforward network, based on the standard feedforward network structure, introduces a 3×3 depthwise separable convolution module, which enhances the perception of local spatial features while performing nonlinear transformation. Finally, the residual connection layer adds the feature sequence to the transformed feature sequence obtained through the above processing to obtain the output feature sequence. Then, the output feature sequence is reshaped into a two-dimensional spatial format to obtain the first feature map and output to the corresponding hollow spatial pyramid pooling module.

[0067] In some embodiments, the dilated spatial pyramid pooling module includes a convolutional branch, multiple dilated convolutional branches, a global average pooling branch, a channel concatenation layer, and a convolutional compression layer; wherein the multiple dilated convolutional branches correspond to different dilation rates. Specifically, the dilated spatial pyramid pooling module performs the following steps: receiving a first feature map output by the Transformer module; extracting local features from the first feature map and adjusting the channel dimensions through the convolutional branch to obtain a local feature map; performing multi-scale dilated convolution operations on the first feature map through multiple dilated convolutional branches to obtain a multi-scale contextual feature map; extracting global features from the first feature map through the global average pooling branch to obtain a global feature map; and concatenating and compressing the local feature map, multi-scale contextual feature map, and global feature map through the channel concatenation layer and the convolutional compression layer to obtain a second feature map.

[0068] For example, Figure 3 A schematic diagram of a hollow space pyramid pooling module is shown, such as Figure 3As shown, the dilated spatial pyramid pooling module comprises five parallel processing branches, followed by a channel stitching layer and a convolutional compression layer. The five parallel processing branches are: one convolutional branch (with a 1×1 kernel), three dilated convolutional branches, and one global average pooling branch. The three dilated convolutional branches are designated as the first, second, and third dilated convolutional branches. Branches with smaller dilation rates are used to capture mesoscale interference fringe patterns and local structural features; branches with medium dilation rates are used to perceive a wider range of phase correlations and adapt to regions of varying fringe density; branches with larger dilation rates are used to model long-range contextual dependencies to identify large-scale terrain distortions or complex noise distribution patterns.

[0069] In some application scenarios, the dilation rate of the first dilated convolution branch can be 3. Performing dilated convolution on the first feature map yields a first-scale context feature map. The dilation rate of the second dilated convolution branch can be 5. Performing dilated convolution on the first feature map yields a second-scale context feature map. The dilation rate of the third dilated convolution branch can be 7. Performing dilated convolution on the first feature map yields a third-scale context feature map. That is, the multi-scale context feature map includes a first-scale context feature map, a second-scale context feature map, and a third-scale context feature map. After the first feature map is input into the dilated spatial pyramid pooling module, the five parallel branches in the dilated spatial pyramid pooling module process the first feature map to obtain local feature maps, first-scale context feature maps, second-scale context feature maps, third-scale context feature maps, and global feature maps. Subsequently, the channel stitching layer performs channel stitching on the aforementioned local feature map, first-scale context feature map, second-scale context feature map, third-scale context feature map, and global feature map to obtain a stitched feature map; the convolutional compression layer performs convolutional compression on the stitched feature map to obtain a second feature map.

[0070] Understandably, introducing a void space pyramid pooling module after the Transformer module at each stage can effectively expand the comprehensive receptive field of the target phase unwrapping model, enhance the target phase unwrapping model's ability to identify phase edges, deformation regions and noise patterns and its robustness, and improve the accuracy of mine surface subsidence detection.

[0071] Since the channel information of the fused feature map obtained by fusing the Transformer backbone features (i.e., the first feature map mentioned above) with the multi-scale context features (i.e., the second feature map) obtained by the hollow spatial pyramid pooling module may be redundant or uneven in importance, the fused feature map can be input into the efficient attention channel module for processing. In some embodiments, the efficient channel attention module is specifically used to perform the following steps: receiving the fused feature map, performing global average pooling on the fused feature map to obtain a global feature vector; performing cross-channel information interaction on the global feature vector to generate channel attention weights; multiplying the channel attention weights with the fused feature map channel by channel to obtain a third feature map, and outputting the third feature map to the decoder.

[0072] For example, Figure 4 A schematic diagram of a high-efficiency channel attention module is shown, as follows: Figure 4 As shown, the efficient channel attention module includes a global average pooling layer, a first convolutional layer, and an activation function layer. After receiving the fused feature map, the efficient channel attention module performs global average pooling on the fused feature map through the global average pooling layer to compress spatial information and generate a global feature vector. Then, the first convolutional layer (with a 1×1 kernel) performs cross-channel information interaction and modeling on the global feature vector, thereby learning the correlation between channels. The vector is then normalized by the sigmoid activation function in the activation function layer to generate channel attention weights. Finally, the channel attention weights are multiplied and weighted with the fused feature map channel by channel to highlight key feature channels and suppress secondary or interfering channels, resulting in a third feature map.

[0073] It is understood that by introducing an efficient channel attention module in the embodiments of this application, the key phase feature channels can be enhanced, while invalid or redundant features are suppressed, thereby improving the expression quality and untangling robustness of the fused features.

[0074] In some embodiments, the decoder is configured to perform the following operations: receive the third feature maps output by four efficient channel attention modules, and project the four third feature maps onto the same embedding dimension to obtain four projected feature maps; upsample the four projected feature maps to the same spatial resolution to obtain four upsampled feature maps; concatenate the four upsampled feature maps along the channel dimension to obtain a concatenated feature map, and perform fusion and regression on the concatenated feature map to obtain an absolute phase map.

[0075] For example, such as Figure 2As shown, the decoder includes a multi-layer perceptron layer (MLP layer), an upsampling layer, a concatenation layer, a third convolutional layer, and a fourth convolutional layer. After receiving the third feature maps from the encoder's four stages (i.e., the third feature maps from the four efficient channel attention modules), the decoder first projects these four third feature maps onto the same embedding dimension using the MLP layer to standardize the feature dimensions, resulting in four projected feature maps. Next, the upsampling layer uses an upsampling operation to increase the spatial resolution of the four projected feature maps to the highest level, aligning them with the size of the shallowest feature map (the first feature map obtained in the first stage), ensuring that all features correspond pixel-by-pixel in spatial location, resulting in four upsampled feature maps. Then, the concatenation layer concatenates the four upsampled feature maps along the channel dimension, resulting in a concatenated feature map. Finally, the second convolutional layer performs a convolution operation on this concatenated feature map to fuse multi-scale contextual semantics, obtaining fused features; the third convolutional layer then acts as a regression head to map the fused features into a single-channel, continuous absolute phase map. The kernels of the second and third convolutional layers are both 1×1.

[0076] Understandably, the decoder structure, through linear projection, upsampling alignment, and lightweight convolution operations, can achieve the fusion of multi-scale features, ensuring the accuracy and efficiency of phase unwrapping.

[0077] S103. Extract deformation information from the absolute phase map to obtain the detection results of surface subsidence in the mining area.

[0078] In some embodiments, feature parameters (such as deformation displacement, deformation rate, etc.) related to surface deformation can be extracted from the absolute phase map using image processing algorithms and geological deformation analysis models. Then, the subsidence of the surface in the mining area can be determined based on the aforementioned feature parameters, thereby obtaining the detection results of surface subsidence in the mining area.

[0079] It should be noted that the method of extracting feature parameters related to surface deformation from the absolute phase map through image processing algorithms and geological deformation analysis models, and determining the subsidence of the surface in the mining area based on the above feature parameters, is an existing technology and will not be elaborated here.

[0080] It is understood that the embodiments of this application employ a multi-layer Transformer structure (i.e., a four-stage Transformer module) to model the global dependencies of the image, introduce a void space pyramid pooling module to extract multi-scale features, enhance the local structural expression capability, and introduce an efficient channel attention mechanism (ECA) in the feature fusion stage to guide the target phase unwrapping model to focus on key features and suppress redundant information, thereby improving the accuracy of surface subsidence detection in mining areas.

[0081] In some embodiments, the underground mine surface subsidence detection method provided in this application further includes: generating a training sample set by simulating the deformation field and interference phase noise field of the mining area; wherein the training sample set includes a wrapped phase map and a true absolute phase map corresponding to the wrapped phase map. Using the wrapped phase map as input and the true absolute phase map as supervision label, the initial phase unwrapping model is iteratively trained until the initial phase unwrapping model converges to obtain the target phase unwrapping model.

[0082] In some embodiments, generating a training sample set by simulating the deformation field and interferometric phase noise of a mining area includes: obtaining a simulated settlement funnel by superimposing multiple two-dimensional Gaussian functions based on the preset geometric features of the mining area settlement funnel; randomly superimposing the simulated settlement funnel onto a real digital elevation model to obtain a simulated settlement model; generating a mining area deformation field and an interferometric phase noise field based on preset radar system parameters and the simulated settlement model; superimposing the mining area deformation field and the interferometric phase noise field to obtain a continuous interferometric phase field; performing phase wrapping processing on the continuous interferometric phase field to obtain a wrapped phase map; and using the mining area deformation field as the true absolute phase map. The geometric features include at least the width, fringe density, shape, and distribution pattern of the mining area settlement funnel; the interferometric phase noise field includes geometrically decoherent noise, temporally decoherent noise, and thermally decoherent noise. In some application scenarios, the interferometric phase noise field can be represented as: ;in, For interference phase noise, For geometric incoherence, For thermal noise coherence, This is thermal noise and phase-decoherent noise.

[0083] For example, since surface subsidence in mining areas is spatially concentrated and often manifests as approximately circular or elliptical subsidence funnels, whose extent is typically much larger than the underground goaf, a two-dimensional Gaussian function with rotational symmetry can be used as the basic unit. The parameters of the two-dimensional Gaussian function (such as center position, amplitude, and covariance) are determined based on the pre-defined geometric characteristics of the mining area subsidence funnels. Then, by superimposing multiple two-dimensional Gaussian functions with different parameters, multiple simulated subsidence funnels (i.e., vertical displacements) are generated. These simulated subsidence funnels are then randomly superimposed on a real Digital Elevation Model (DEM) to simulate subsidence areas of different locations and scales, thereby constructing a simulated subsidence model. Then, based on pre-defined radar system parameters (such as wavelength, incident angle, and baseline), the elevation changes in the simulated subsidence model are converted into absolute phase changes in the radar line-of-sight direction, thus obtaining the mining area deformation field. This mining area deformation field is the true absolute phase map from the sample dataset.

[0084] In some embodiments, geometrical incoherence can be calculated using the following formula 1 based on the surface slope represented by the aforementioned real digital elevation model and preset radar system parameters (such as vertical baseline, incident angle, etc.), thereby generating geometrical incoherence noise:

[0085] (Formula 1);

[0086] in, For geometric incoherence, At the speed of light, For vertical baseline, The preset incident angle of the radar system, The surface slope is represented by a true digital elevation model. It is the preset wavelength of the radar system. It is the slant distance. This refers to the bandwidth frequency.

[0087] Thermal noise coherence can be determined based on the signal-to-noise ratio of a radar system (such as Sentinel-1A) using the following formula 2, and then thermal noise coherence noise can be generated:

[0088] (Formula 2);

[0089] in, For thermal noise coherence, This represents the absolute value of the thermal noise coherence. This represents the signal-to-noise ratio of the preset radar system in Interferometric Wide swath (IW) imaging mode. It is the reciprocal of the signal-to-noise ratio.

[0090] Based on real land cover classification data and a preset imaging time interval, temporal decoherence can be determined using the following formula 3, thereby simulating temporal decoherence noise:

[0091] (Formula 3);

[0092] in, This is time-decoherent noise. For time intervals, Static coherence of land cover under different land cover classifications. The time of temporal incoherence characteristic represents the difference in the impact of land cover classification on temporal coherence.

[0093] It is understood that the embodiments of this application combine the simulation of deformation phase in the mining area with the simulation of multi-source interference phase noise to generate a wrap-around phase map and the corresponding absolute phase map that are close to the real scene in the mining area. This can solve the problems of difficulty in obtaining real mining area data, high annotation cost and insufficient data diversity.

[0094] In some embodiments, the target phase unwrapping model (Lightweight Transformer-CNNHybrid network, LTCHnet) can be compared with Quantized Deep Long Short-Term Memory (DLSTM), Phase Unwrapping Network (PUnet), U-Net Convolutional Neural Network (U-Net), Statistical-cost, Network-flow Algorithm for Phase Unwrapping (SNAPHU), and Cascaded Deep Convolutional Neural Network (CDCNN). Figure 5 A schematic diagram of comparative experimental results is shown, such as... Figure 5 As shown, Figure 5 The first column from left to right is the interferometric phase diagram of the AF ore area; the second column is the phase unwrapping result (i.e., absolute phase diagram) obtained by phase unwrapping the AF ore area using DLSTM; the third column is the phase unwrapping result obtained by phase unwrapping the AF ore area using PUnet; the fourth column is the phase unwrapping result obtained by phase unwrapping the AF ore area using U-Net; the fifth column is the phase unwrapping result obtained by phase unwrapping the AF ore area using SNAPHU; the sixth column is the phase unwrapping result obtained by phase unwrapping the AF ore area using CDCNN; and the seventh column is the phase unwrapping result obtained by phase unwrapping the AF ore area using LTCHnet. From top to bottom, the first row 'a' is the phase unwrapping result for ore area A (i.e., absolute phase diagram); the second row 'b' is the phase unwrapping result for ore area B; the third row 'c' is the phase unwrapping result for ore area C; the fourth row 'd' is the phase unwrapping result for ore area D; the fifth row 'e' is the phase unwrapping result for ore area E; and the sixth row 'f' is the phase unwrapping result for ore area F. Figure 5The comparative experimental results show that DLSTM tends to over-smooth, resulting in blurred deformation edges and underestimated amplitude; PUnet, while suppressing noise, loses high-frequency details and stretches the deformation range; U-Net, while maintaining overall continuity, easily introduces checkerboard artifacts and weakens deformation gradients; SNAPHU can generate distinct stripes, but generally produces voids and fractures that disrupt terrain continuity in low coherence areas; CDCNN is prone to overfitting noise, producing block artifacts, step errors, and even erroneous stripe structures; while the target phase unwrapping model provided in this application can capture deformation features more accurately, maintaining high-accuracy phase unwrapping effects in continuous regions and effectively suppressing the generation of voids and fractures in low coherence areas, making it more suitable for settlement monitoring and analysis under complex geological conditions.

[0095] Based on the same inventive concept, this application also provides an underground mine surface settlement detection system for implementing a method for detecting surface settlement in underground mines. The solution provided by this system is similar to the implementation scheme described in the above-mentioned method; therefore, please refer to the limitations of the underground mine surface settlement detection method described above, and will not be repeated here. Specifically, Figure 6 This is a schematic diagram of the structure of an underground mine surface subsidence detection system according to an embodiment of this application. Figure 6 As shown, the underground mine surface settlement detection system includes:

[0096] The acquisition module 610 is used to acquire an interference phase map of the mining area, wherein the interference phase map contains phase entanglement information caused by mining disturbance;

[0097] The phase unwrapping module 620 is used to perform feature extraction and phase recovery processing on the interference phase map through a preset target phase unwrapping model to obtain the unwrapped absolute phase map; wherein, the target phase unwrapping model is based on the encoder-decoder architecture of the Segformer-B0 model, and a hole space pyramid pooling module and an efficient channel attention module are added after the four Transformer modules included in the encoder.

[0098] The settlement detection module 630 is used to extract deformation information from the absolute phase map to obtain the detection results of surface settlement in the mining area.

[0099] In some embodiments, the phase unwrapping module 620 is further configured to perform global dependency modeling on the interferometric phase map through the Transformer module to obtain a first feature map; perform multi-scale feature extraction on the first feature map through the hollow spatial pyramid pooling module to obtain a second feature map; concatenate the first feature map and the second feature map in the channel dimension and perform a convolution operation to obtain a fused feature map; and after the efficient channel attention module performs channel attention weighting on the fused feature map, a third feature map is obtained and output to the decoder.

[0100] In some embodiments, the dilated spatial pyramid pooling module includes a convolutional branch, multiple dilated convolutional branches, a global average pooling branch, a channel concatenation layer, and a convolutional compression layer; wherein the multiple dilated convolutional branches correspond to different dilation rates; the phase unwrapping module 620 is further configured to receive the first feature map, perform local feature extraction on the first feature map through the convolutional branches and adjust the channel dimensions to obtain a local feature map; perform multi-scale dilated convolution operations on the first feature map through the multiple dilated convolutional branches to obtain a multi-scale context feature map; perform global feature extraction on the first feature map through the global average pooling branch to obtain a global feature map; and perform channel concatenation and convolutional compression on the local feature map, the multi-scale context feature map, and the global feature map through the channel concatenation layer and the convolutional compression layer to obtain the second feature map.

[0101] In some embodiments, the efficient channel attention module includes a global average pooling layer, a first convolutional layer, and an activation function layer; the phase unwrapping module 620 is further configured to receive the fused feature map, perform global average pooling on the fused feature map through the global average pooling layer to obtain a global feature vector; perform cross-channel information interaction on the global feature vector through the first convolutional layer and the activation function layer to generate channel attention weights; multiply the channel attention weights with the fused feature map channel by channel to obtain a third feature map, and output the third feature map to the decoder.

[0102] In some embodiments, the phase unwrapping module 620 is further configured to receive the third feature maps output by the four efficient channel attention modules, project the four third feature maps onto the same embedding dimension to obtain four projected feature maps; upsample the four projected feature maps to the same spatial resolution to obtain four upsampled feature maps; concatenate the four upsampled feature maps in the channel dimension to obtain a concatenated feature map, and perform fusion and regression on the concatenated feature map to obtain the absolute phase map.

[0103] In some embodiments, the system further includes a simulation module for generating a training sample set by simulating the deformation field and interference phase noise field of the mining area; wherein the training sample set includes a wrapped phase map and a true absolute phase map corresponding to the wrapped phase map;

[0104] The training module is used to iteratively train the initial phase unwrapping model with the wrapped phase map as input and the true absolute phase map as supervision label until the initial phase unwrapping model converges, thereby obtaining the target phase unwrapping model.

[0105] In some embodiments, the simulation module is further configured to obtain a simulated settlement funnel by superimposing multiple two-dimensional Gaussian functions based on the preset geometric features of the mining area settlement funnel, and randomly superimpose the simulated settlement funnel onto a real digital elevation model to obtain a simulated settlement model; wherein the geometric features include at least the width, stripe density, shape, and distribution pattern of the mining area settlement funnel; based on preset radar system parameters and the simulated settlement model, generate the mining area deformation field and the interference phase noise field; wherein the interference phase noise field includes geometric decoherence noise, temporal decoherence noise, and thermal noise decoherence noise; superimpose the mining area deformation field and the interference phase noise field to obtain a continuous interference phase field, and perform phase wrapping processing on the continuous interference phase field to obtain the wrapped phase map, and use the mining area deformation field as the real absolute phase map.

[0106] It should be understood that the phrase "one embodiment" or "an embodiment" throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of the invention. Therefore, "in one embodiment" or "in an embodiment" appearing in every place throughout the specification does not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. It should be understood that in each embodiment of the invention, the sequence number of each process described above does not imply a sequential order of execution; the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the invention. The sequence numbers of the embodiments of the invention described above are merely descriptive and do not represent the superiority or inferiority of the embodiments.

[0107] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0108] In the several embodiments provided by this invention, it should be understood that the disclosed methods can be implemented in other ways. The methods disclosed in the several method embodiments provided by this invention can be arbitrarily combined without conflict to obtain new method embodiments. The features disclosed in the several method embodiments provided by this invention can be arbitrarily combined without conflict to obtain new method embodiments.

[0109] The above description is merely an embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for detecting surface subsidence in underground mines, characterized in that, include: Obtain the interferometric phase map of the mining area; The interference phase map is subjected to feature extraction and phase recovery processing by a preset target phase unwrapping model to obtain the unwrapped absolute phase map. Deformation information is extracted from the absolute phase map to obtain the detection results of surface subsidence in the mining area; The target phase unwrapping model is based on the encoder-decoder architecture of the Segformer-B0 model. It adds a dilated spatial pyramid pooling module and an efficient channel attention module after the four Transformer modules in the encoder. The dilated spatial pyramid pooling module is used to extract features from the input feature map at multiple scales. The efficient channel attention module is used to apply channel attention weights to the input feature map and output the weighted feature map to the decoder. The encoder performs the following steps: modeling the global dependency of the interference phase map using the Transformer module to obtain a first feature map; extracting features from the first feature map at multiple scales using the dilated spatial pyramid pooling module to obtain a second feature map; concatenating the first feature map and the second feature map along the channel dimension and performing a convolution operation to obtain a fused feature map. The efficient channel attention module performs channel attention weighting on the fused feature map to obtain a third feature map, which is then output to the decoder.

2. The method according to claim 1, characterized in that, The hollow spatial pyramid pooling module includes a convolutional branch, multiple hollow convolutional branches, a global average pooling branch, a channel splicing layer, and a convolutional compression layer; wherein, the multiple hollow convolutional branches correspond to different dilation rates; The void space pyramid pooling module is specifically used to perform the following steps: Receive the first feature map, extract local features from the first feature map through the convolution branch and adjust the channel dimension to obtain a local feature map; A multi-scale context feature map is obtained by performing multi-scale dilated convolution operations on the first feature map through multiple dilated convolution branches. Global feature extraction is performed on the first feature map using the global average pooling branch to obtain a global feature map; The local feature map, the multi-scale context feature map, and the global feature map are concatenated and convolved by the channel concatenation layer and the convolutional compression layer to obtain the second feature map.

3. The method according to claim 1, characterized in that, The efficient channel attention module includes a global average pooling layer, a first convolutional layer, and an activation function layer; The high-efficiency channel attention module is specifically used to perform the following steps: The fused feature map is received, and global average pooling is performed on the fused feature map through the global average pooling layer to obtain a global feature vector; The global feature vector is subjected to cross-channel information interaction through the first convolutional layer and the activation function layer to generate channel attention weights. The channel attention weights are multiplied channel by channel by the fused feature map to obtain a third feature map, which is then output to the decoder.

4. The method according to claim 1 or 3, characterized in that, The decoder is used to perform the following operations: The third feature maps output by the four efficient channel attention modules are received, and the four third feature maps are projected onto the same embedding dimension to obtain four projected feature maps. The four projected feature maps are upsampled to the same spatial resolution to obtain four upsampled feature maps. The four upsampled feature maps are concatenated along the channel dimension to obtain a concatenated feature map. The concatenated feature map is then fused and regressed to obtain the absolute phase map.

5. The method according to claim 1, characterized in that, The method further includes: A training sample set is generated by simulating the deformation field and interference phase noise field of the mining area; wherein, the training sample set includes a wrapped phase map and the true absolute phase map corresponding to the wrapped phase map; Using the wrapped phase map as input and the true absolute phase map as supervision label, the initial phase unwrapping model is iteratively trained until the initial phase unwrapping model converges, thus obtaining the target phase unwrapping model.

6. The method according to claim 5, characterized in that, The process of generating a training sample set by simulating the deformation field and interference phase noise field of the mining area includes: Based on the preset geometric features of the mining area settling funnel, a simulated settling funnel is obtained by superimposing multiple two-dimensional Gaussian functions. The simulated settling funnel is then randomly superimposed on a real digital elevation model to obtain a simulated settling model. The geometric features include at least the width, stripe density, shape, and distribution pattern of the mining area settling funnel. Based on preset radar system parameters and the simulated subsidence model, the deformation field of the mining area and the interferometric phase noise field are generated; wherein, the interferometric phase noise field includes geometric decoherence noise, temporal decoherence noise and thermal decoherence noise; The deformation field of the mining area is superimposed with the interference phase noise field to obtain a continuous interference phase field. The continuous interference phase field is then subjected to phase wrapping to obtain the wrapped phase map. The deformation field of the mining area is used as the true absolute phase map.

7. A surface settlement detection system for underground mines, characterized in that, include: The acquisition module is used to acquire an interference phase map of the mining area, wherein the interference phase map contains phase entanglement information caused by mining disturbance; A phase unwrapping module is used to perform feature extraction and phase recovery processing on the interferometric phase map using a preset target phase unwrapping model to obtain an unwrapped absolute phase map. The target phase unwrapping model is based on the encoder-decoder architecture of the Segformer-B0 model, with a dilated spatial pyramid pooling module and an efficient channel attention module added after the four Transformer modules in the encoder. The dilated spatial pyramid pooling module is used to perform multi-scale feature extraction on the input feature map. The efficient channel attention module is used to perform channel attention weighting on the input feature map and output the weighted feature map to the decoder. The encoder performs the following steps: modeling global dependencies in the interferometric phase map using the Transformer module to obtain a first feature map; performing multi-scale feature extraction on the first feature map using the dilated spatial pyramid pooling module to obtain a second feature map; concatenating the first and second feature maps along the channel dimension and performing a convolution operation to obtain a fused feature map; and performing channel attention weighting on the fused feature map using the efficient channel attention module to obtain a third feature map and outputting it to the decoder. The settlement detection module is used to extract deformation information from the absolute phase map to obtain the detection results of surface settlement in the mining area.

Images