A mining subsidence basin information extraction method and system based on InSAR

Abstract

The present invention proposes a mining subsidence basin information extraction method and system based on InSAR, which relates to the field of mining area monitoring. The specific scheme includes: obtaining an InSAR interference map of the mining area to be monitored; inputting the InSAR interference map into a trained deep learning network model to extract mining subsidence basin information; wherein, the training data set of the deep learning network model includes a real training data set and a simulated training data set, wherein the real training data set is constructed by combining differential interferometry with a static probability integral method to obtain real mining subsidence basin information, and the simulated training data set is constructed by simulating the mining subsidence phase using a dynamic probability integral method and then combining Perlin noise and Gaussian noise models; the present invention fully trains the deep learning network model based on the real training data set and the simulated training data set to improve the accuracy and efficiency of mining subsidence basin information extraction.

Description

Technical Field

[0001] The invention belongs to the field of mining area monitoring, and in particular relates to a mining subsidence basin information extraction method and system based on InSAR. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] The geological disasters caused by coal mining are receiving more and more attention. In order to reduce the geological environmental disasters caused by coal mining, it is necessary to monitor mining subsidence in mining areas through scientific methods, analyze the temporal and spatial evolution of surface subsidence, and study the changing process and impact range of surface subsidence.

[0004] Traditional surface deformation measurement methods include leveling, total station measurement, and global satellite navigation system (GNSS) measurement. These three measurement methods generally only monitor the surface observation points in the mining area. This point-to-surface method has great defects and cannot fully reflect the complete deformation field of the surface and the dynamic process of deformation. Synthetic aperture radar interferometry (InSAR) technology is an important non-contact measurement method with advantages such as all-day, all-weather, high spatial resolution, and a wide range. In recent years, researchers have integrated InSAR with deep learning technology to automatically identify the location distribution of mining subsidence areas. However, these methods rely on manually selected training samples, and the amount of training data sets is limited. At the same time, it is often impossible to fully understand the underground mining situation only through location distribution information. In addition, the InSAR interferogram of the mining area is characterized by high noise and dense fringes, which makes it difficult to efficiently and accurately extract settlement information from the interferogram. Therefore, the existing schemes are highly complex and weakly robust, and cannot achieve efficient and accurate mining subsidence monitoring. Summary of the invention

[0005] In order to overcome the shortcomings of the above-mentioned prior art, the present invention provides a method and system for extracting mining subsidence basin information based on InSAR. Based on real training data sets and simulated training data sets, the deep learning network model is fully trained to improve the accuracy and efficiency of mining subsidence basin information extraction.

[0006] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions:

[0007] A first aspect of the present invention provides a method for extracting mining subsidence basin information based on InSAR.

[0008] A mining subsidence basin information extraction method based on InSAR, comprising:

[0009] Obtain the InSAR interferogram of the mining area to be monitored;

[0010] The InSAR interferogram is input into the trained deep learning network model to extract mining subsidence basin information;

[0011] Among them, the training data set of the deep learning network model includes a real training data set and a simulated training data set. The real training data set is constructed by combining differential interferometry with the static probability integration method to obtain real mining subsidence basin information. The simulated training data set is constructed by simulating the mining subsidence phase using the dynamic probability integration method and combining the Perlin noise and Gaussian noise models.

[0012] Furthermore, the method for acquiring the InSAR interference map is: using synthetic aperture radar SAR to perform high-resolution microwave imaging on the mining area to be monitored, and processing the microwave imaging by a differential radar interferometry method to obtain the InSAR interference map.

[0013] Furthermore, the deep learning network model includes a feature extraction module, a dilated convolution module, a residual network module and a feature output module;

[0014] The feature extraction module extracts multiple feature maps from the InSAR interferogram;

[0015] The dilated convolution module uses convolutional layers with different dilation rates to perform multi-scale information fusion on the extracted multiple feature maps;

[0016] The residual network module connects the input and output of the residual network module based on a plurality of convolutional layers of residual structures;

[0017] The feature output module extracts mining subsidence basin information from the feature map output by the residual network module;

[0018] The mining subsidence basin information is the surface subsidence value.

[0019] Furthermore, the method of combining differential interferometry with the static probability integration method is specifically as follows:

[0020] For small-scale mining subsidence areas, the subsidence monitoring results of D-InSAR are used as sample label data;

[0021] For large-scale mining subsidence areas, based on sample label data, the corresponding sample label data are obtained by combining InSAR with the static probability integral method.

[0022] Furthermore, the dynamic probability integral method is used to simulate the mining subsidence phase, specifically:

[0023] The probability integral method is used to dynamically predict mining subsidence in the mining area. By adjusting the mining parameters and using the Knothe time function, the surface settlement values ​​of different intensities and forms within the predicted time of different working faces are obtained, and then a subsidence prediction model is constructed.

[0024] Furthermore, the subsidence prediction model is specifically:

[0025] ;

[0026] in, For the The calculation formula for the maximum impact value of surface subsidence after mining of a dynamic unit is: For the The time coefficient corresponding to the dynamic unit at time t, The dynamic unit is obtained by dividing the mining area to be monitored into units. units.

[0027] Furthermore, the combination of Perlin noise and Gaussian noise model is specifically:

[0028] By superimposing Perlin noises of different frequencies and amplitudes, fractal Perlin noise is obtained as the atmospheric delay phase in the simulated InSAR interferogram.

[0029] A Gaussian noise model is introduced, and the noise intensity is randomly set during the simulation process to generate independent noise matrices with the same magnitude as the real and imaginary parts. The values ​​are wrapped between (-π,π] to obtain the wrapped phase of the InSAR interferogram.

[0030] A second aspect of the present invention provides an InSAR-based mining subsidence basin information extraction system.

[0031] An InSAR-based mining subsidence basin information extraction system includes an acquisition module and an extraction module:

[0032] The acquisition module is configured to: acquire an InSAR interferogram of the mining area to be monitored;

[0033] The extraction module is configured to: input the InSAR interferogram into the trained deep learning network model to extract mining subsidence basin information;

[0034] Among them, the training data set of the deep learning network model includes a real training data set and a simulated training data set. The real training data set is constructed by combining differential interferometry with the static probability integration method to obtain real mining subsidence basin information. The simulated training data set is constructed by simulating the mining subsidence phase using the dynamic probability integration method and combining the Perlin noise and Gaussian noise models.

[0035] The third aspect of the present invention provides a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the steps of the method for extracting mining subsidence basin information based on InSAR as described in the first aspect of the present invention.

[0036] The fourth aspect of the present invention provides an electronic device, including a memory, a processor, and a program stored in the memory and executable on the processor. When the processor executes the program, the steps in the method for extracting mining subsidence basin information based on InSAR as described in the first aspect of the present invention are implemented.

[0037] One or more of the above technical solutions have the following beneficial effects:

[0038] The present invention adopts a combination of differential interferometry and static probability integration method to obtain real mining subsidence basin information to construct a real training data set, uses dynamic probability integration method to simulate mining subsidence phase, and then combines Perlin noise and Gaussian noise models to construct a simulated training data set, thereby obtaining a diverse and comprehensive data set, thereby improving the training effect of the deep learning network model and providing the extraction accuracy of the model.

[0039] The present invention integrates feature extraction, dilated convolution, residual network and feature output sub-networks to construct a deep learning network model for extracting mining subsidence basin information, and trains the network using simulated training data sets and real training data sets, giving full play to the advantages of deep learning networks with strong feature extraction capabilities and good noise robustness, overcoming the shortcomings of the prior art that it is difficult to extract dense fringe information from high-noise interference patterns, and can better extract medium-scale subsidence basin information.

[0040] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0042] Figure 1 It is a schematic diagram of the overall process of the first embodiment.

[0043] Figure 2 Schematic diagram of the construction of the simulated training data set in the first embodiment, wherein (a) is the InSAR phase image, (b) is the InSAR phase image disturbed by the atmosphere, and (c) is the InSAR interferogram.

[0044] Figure 3This is a structural diagram of the deep learning network model of the first embodiment, where (a) is feature extraction, (b) is dilated convolution, (c) is the residual network, and (d) is feature output.

[0045] Figure 4 A comparison chart of extraction results in a small-scale mining subsidence area.

[0046] Figure 5 A comparison chart of extraction results in a medium-scale mining subsidence area. DETAILED DESCRIPTION

[0047] It should be noted that the following detailed descriptions are illustrative and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used in the present invention have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.

[0048] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0049] First, let's briefly explain SAR, InSAR, and D-InSAR:

[0050] Synthetic Aperture Radar (SAR) is a radar system that can produce high-resolution images. SAR is derived from radar, but it is different from radar. The radar that everyone is familiar with generally refers to ground-based radar, which can use antennas to send and receive pulses to targets, and then calculate the target distance based on the pulse time interval. This type of radar can accurately measure distance, but cannot produce fine images. Synthetic aperture radar uses a "virtual" antenna length. During the flight of the satellite, it can detect the target from a long distance, and the target will completely leave the observation area after flying for a period of time. During this observation time, the satellite transmits a series of pulse signals from different positions. Combining this series of pulse signals is equivalent to transmitting these signals from a very long aperture. This process of synthesizing an equivalent "large antenna" through a small antenna is called "synthetic aperture", and synthetic aperture radar is to combine the echo signals received by the small antenna from different positions, and synthesize the real antenna length of several meters into a virtual antenna length of several kilometers, so as to obtain a high-resolution imaging radar.

[0051] Interferometric Synthetic Aperture Radar (InSAR) is one of the most promising space earth observation technologies and an extension of SAR technology. Its working principle is to use SAR to take two images of the same area and obtain the interference map of the area through interference. The interference map contains the terrain information of the area.

[0052] Differential InSAR (D-InSAR) is an extension of InSAR technology. It uses two SAR images combined with external terrain data to obtain tiny deformations of the surface. The differential interferogram is composed of two SAR images acquired at different times. If the surface is deformed during this period, the differential interferogram will record the relevant information. D-InSAR technology is extremely sensitive to deformation. For the Land Exploration Satellite No. 1, a stripe change is close to a deformation change of about 12 cm. It can be seen that although they are both stripes, the stripes used to measure terrain by InSAR and the stripes used to measure deformation by DInSAR have an order of magnitude difference in the degree of depiction of surface information.

[0053] Embodiment 1

[0054] In one embodiment of the present disclosure, a method for extracting mining subsidence basin information based on InSAR is provided, comprising the following steps:

[0055] Step S1: Obtaining the InSAR interferogram of the mining area to be monitored;

[0056] Step S2: Input the InSAR interferogram into the trained deep learning network model to extract mining subsidence basin information;

[0057] Among them, the training data set of the deep learning network model includes a real training data set and a simulated training data set. The real training data set is constructed by combining differential interferometry with the static probability integration method to obtain real mining subsidence basin information. The simulated training data set is constructed by simulating the mining subsidence phase using the dynamic probability integration method and combining the Perlin noise and Gaussian noise models.

[0058] In the above scheme, the construction and training of the deep learning network model is the key to extracting mining subsidence information from the InSAR interferogram in one step. The construction and training of the model are described in detail below. Figure 1 As shown, the specific steps include:

[0059] Step 1: Use D-InSAR combined with the probability integral method to obtain the mining surface settlement values ​​of real mining areas with different time baselines and create a real training data set;

[0060] Specifically, the construction of the training data set is based on the real data collected from the subsidence area, namely the InSAR interferogram, which can be divided into two cases:

[0061] In areas with small-scale subsidence, D-InSAR is used to obtain surface subsidence values. Specifically, D-InSAR compares the phase information of two (or more) SAR images at the same location in the InSAR interferogram, calculates the displacement change of the surface between the two SAR images, that is, the surface subsidence value, and uses it as the subsidence monitoring result in small-scale subsidence areas as sample label data.

[0062] InSAR cannot accurately obtain a large number of subsidence areas, and the probability integral method is used to obtain the surface settlement value. The probability integral method here is a static probability integral method and does not involve time, because the settlement value it inverts is the settlement value after the working face reaches full stability. Therefore, this method first inverts the optimal 8 probability integral parameters based on the ant colony optimization algorithm, which are the subsidence coefficient q , Mainly affecting the tangent angle tanβ , horizontal shift coefficient b , mining impact propagation angle θ ０ , Left turning point offset S L , right turning point offset SR , Upper inflection point offset SU , lower inflection point offset SD , and then substitute the solved parameters into the mining influence function to calculate the surface shape variables in the mining subsidence area, that is, the surface settlement value, so as to obtain the corresponding settlement monitoring results; combined with the actual mining subsidence monitoring results, the corresponding sample label data is produced.

[0063] The mining impact function is expressed as follows:

[0064] (1)

[0065] In the formula, is the surface settlement value to be calculated, x and y are the location information of the working surface, and j represents the jth working surface; Indicates the maximum sinking value, , is the coal seam inclination; represents the main influence radius, ; H is the mining depth; q is the sinking coefficient; b is the horizontal movement coefficient; m is the mining thickness; is the main influencing angle tangent; D is the mining area of ​​the working face, that is, considering the inflection point offset, using the left inflection point offset SL , right turning point offset SR , Upper inflection point offset S U , lower inflection point offset SD Constructed mining area.

[0066] Here, the 8 optimal probability integral parameters are inverted according to the ant colony optimization algorithm, specifically:

[0067] The displacement back analysis calculation model of mining subsidence calculation parameters is:

[0068] (2)

[0069] in, [K] is the stiffness matrix related to the mining subsidence calculation parameters; {A} An array of sinking values ​​or horizontal movement values ​​of measuring points; {F} An array of equivalent sinking values ​​or horizontal movement values ​​of measuring points; T (P) is the objective function, P Calculate parameter vectors for mining subsidence; M is the total number of measuring points; W ｄ (P) For the d The measurement points are based on P Calculated sinking value or horizontal shift value of; W′ ｄ For the d The measured subsidence value or horizontal movement value of each measuring point.

[0070] set up P=(q,tanβ,b,θ ０ , SL, SR, SU, SD) , B for P The search space of mining subsidence calculation parameters displacement back analysis calculation model can be described as in 8-dimensional space B Find a vector in P ０ , so that the objective function T (P ０ ） The value of is the smallest.

[0071] Ant colony algorithm is suitable for solving combinatorial optimization problems. Therefore, it is necessary to transform the continuous variable optimization problem of the displacement back analysis calculation model of mining subsidence calculation parameters into a combinatorial optimization problem. Discrete search space B Method: The nth (n=1, 2, ..., 8) calculation parameter P in P ｎ The value range is divided into M ｎ The search space B is discretized into a discrete space containing a finite number of vectors; the parameterＰｎ The higher the discreteness, the higher the calculation accuracy.

[0072] Based on the above settings, the ant colony algorithm is used to solve the problem. The specific steps are as follows:

[0073] (1) Initialization: Set up multiple ants, and each ant searches the solution space in parallel. Initially, the amount of pheromone on all paths is equal.

[0074] (2) Ant search: After each ant completes a tour (i.e. traversing all cities), it releases pheromones on the way, and the amount of pheromones is proportional to the quality of the solution. The ant path is selected based on the current pheromone intensity, while taking into account the distance between two points, and a random local search strategy is adopted.

[0075] (3) Pheromone update: After all ants have completed their search, an iteration is performed; each iteration updates all edges with pheromones, including the evaporation of the original pheromones and the increase of pheromones on the passed paths; the update of pheromones is a positive feedback mechanism in the ant colony algorithm, and the better path will become more attractive to ants due to the accumulation of pheromones.

[0076] (4) Taboo table control: Ants can only take legal routes (for example, passing through each city only once), and a taboo table is set to control the movement of ants.

[0077] (5) Algorithm termination condition: The algorithm ends when the predetermined number of iterations is reached or stagnation occurs (i.e., the optimal solution does not improve significantly after multiple consecutive iterations).

[0078] Step 2: Use the probability integral method to simulate the mining subsidence basin information and convert it into InSAR phase. Then combine the Perlin noise and Gaussian noise models to construct a simulated training data set.

[0079] The simulation data is simulated based on the theoretical knowledge of real data and some of its characteristics. It can be used as a supplement to the network model data set. The dynamic prediction method of probability integral method is adopted to predict the mining subsidence in the mining area, that is, the dynamic probability integral method, introduces the Knothe time function, adjusts the mining parameters, and inverts the surface settlement values ​​of different intensities and forms at any predicted time within the total time of surface movement. The Knothe time function is as follows:

[0080] (3)

[0081] Where t is the estimated time; represents the time coefficient of the expected time; τ is the time when the maximum sinking velocity occurs; c is the time coefficient related to the mechanical properties of the overburden rock; T is the total time of surface movement;

[0082] Based on formula (1), the subsidence prediction model of any point (x, y) on the surface can be expressed as:

[0083] (4)

[0084] in, is the vertical settlement value of any point (x, y) on the ground surface, For the The calculation formula for the maximum impact value of surface subsidence after mining of a dynamic unit is: For the The time coefficient corresponding to the dynamic unit at time t, The dynamic unit is obtained by dividing the mining area to be monitored into units. units.

[0085] Since SAR satellite imaging is side-view imaging, the vertical settlement value of any point on the surface needs to be Converted to satellite LOS deformation, assuming that the mining area deformation is only vertical settlement, it can be converted according to the following formula:

[0086] (5)

[0087] in, represents the deformation of the satellite line of sight, Represents the incident angle of the corresponding pixel.

[0088] The satellite LOS deformation is converted into Figure 2 The InSAR phase map shown in (a). Perlin noise is a natural noise algorithm that randomly selects some points on the array, assigns a certain range of random values ​​and gradient vectors to these points, and then calculates the value of the unassigned position based on the value and gradient of the random point. In order to add atmospheric disturbances to the phase map, Perlin noises of different frequencies and amplitudes are superimposed to obtain fractal Perlin noise as the atmospheric delay phase in the simulated InSAR interferogram. Figure 2 Adding the atmospheric delay phase to (a), we get Figure 2 (b) shows the InSAR phase image affected by atmospheric disturbances.

[0089] In order to simulate the decorrelation effect of SAR images, a Gaussian noise model is introduced. The noise intensity is randomly set during the simulation process to generate independent noise matrices with the same magnitude as the real and imaginary parts, and the values ​​are wrapped between (-π, π] to obtain Figure 2 InSAR interferogram shown in (c).

[0090] Finally, the generated InSAR interferogram is used as the model input, and the surface settlement value is used as the For labels, construct a simulated training dataset.

[0091] Step 3: Combining feature extraction, dilated convolution, residual network and feature output, a convolutional neural network for extracting mining subsidence information from InSAR interferograms is constructed as a deep learning network model;

[0092] like Figure 3 As shown in the figure, the network structure mainly consists of four parts: feature extraction (a), dilated convolution (b), residual network (c) and feature output (d).

[0093] First, in the feature extraction part, a convolution layer Conv with a smaller convolution kernel is used to extract multiple feature maps from the input InSAR interferogram, and then the activation function Relu is used to nonlinearize it.

[0094] Then, the dilated convolution part uses three convolutional layers Conv with different dilation rates to process the results of feature extraction, concatenates the feature maps output by the three dilated convolutional layers with different sampling rates, and passes through a convolutional layer Conv to expand the receptive field of the convolutional neural network to fuse multi-scale information.

[0095] Secondly, the residual network contains multiple convolutional layers Conv with residual structures, that is, a shortcut connection is set between the input and output of the residual network to speed up network training and improve network performance.

[0096] Finally, in the feature output part, a convolutional layer Conv is used to output the mining subsidence basin information.

[0097] Step 4: Use the training data set generated in step 1 and step 2 as the input of the convolutional neural network in step 3 for training, where the training data set contains the InSAR interferogram of the mining subsidence area and the actual mining subsidence information on the surface;

[0098] The trained network model is used to quickly and accurately extract the mining subsidence information (i.e., surface subsidence value) contained in the InSAR interferogram.

[0099] In the above implementation, at the two levels of small-scale mining subsidence and medium-scale mining subsidence, the method of this embodiment and the D-InSAR method respectively solve and compare a group of InSAR interferograms collected in time sequence. The solution results for small-scale mining subsidence are as follows: Figure 4 As shown, the middleweight results are as follows Figure 5As shown in the figure, DM-Net is the method of this embodiment, mm represents the amount of settlement in millimeters; it can be intuitively found from the figure that with the increase of settlement, the solution result of the traditional D-InSAR method gradually becomes worse, while the deep learning network can still extract information about the subsidence basin when there are obvious interference fringes in the interference pattern, which is significantly improved in the accuracy of extracting medium-scale settlement information compared with the D-InSAR method.

[0100] Embodiment 2

[0101] In one embodiment of the present disclosure, a mining subsidence basin information extraction system based on InSAR is provided, including an acquisition module and an extraction module:

[0102] The acquisition module is configured to: acquire an InSAR interferogram of the mining area to be monitored;

[0103] The extraction module is configured to: input the InSAR interferogram into the trained deep learning network model to extract mining subsidence basin information;

[0104] Among them, the training data set of the deep learning network model includes a real training data set and a simulated training data set. The real training data set is constructed by combining differential interferometry with the static probability integration method to obtain real mining subsidence basin information. The simulated training data set is constructed by simulating the mining subsidence phase using the dynamic probability integration method and combining the Perlin noise and Gaussian noise models.

[0105] Embodiment 3

[0106] The purpose of this embodiment is to provide a computer-readable storage medium.

[0107] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in a method for extracting mining subsidence basin information based on InSAR as described in the first embodiment of the present disclosure.

[0108] Embodiment 4

[0109] The purpose of this embodiment is to provide an electronic device.

[0110] An electronic device includes a memory, a processor, and a program stored in the memory and executable on the processor. When the processor executes the program, the steps in the method for extracting mining subsidence basin information based on InSAR as described in the first embodiment of the present disclosure are implemented.

[0111] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A mining subsidence basin information extraction method based on InSAR, characterized in that: include: Obtain the InSAR interferogram of the mining area to be monitored; Inputting the InSAR interferogram into the trained deep learning network model to extract mining subsidence basin information, wherein the mining subsidence basin information is the surface subsidence value; The training data set of the deep learning network model includes a real training data set and a simulated training data set. The real training data set is constructed by combining differential interferometry with a static probability integral method to obtain real mining subsidence basin information. The simulated training data set is constructed by simulating the mining subsidence phase using a dynamic probability integral method and combining Perlin noise and Gaussian noise models. The method of combining differential interferometry with static probability integration method is specifically as follows: For small-scale mining subsidence areas, the subsidence monitoring results of D-InSAR are used as sample label data; For large-scale mining subsidence areas, based on sample label data, the corresponding sample label data is obtained by combining InSAR with the static probability integral method; The dynamic probability integral method is used to simulate the mining subsidence phase, specifically: The probability integral method is used to dynamically predict mining subsidence in the mining area. By adjusting the mining parameters and using the Knothe time function, the surface settlement values ​​of different intensities and forms within the predicted time of different working faces are obtained, and then a subsidence prediction model is constructed.

2. The method for extracting mining subsidence basin information based on InSAR as claimed in claim 1, characterized in that: The method for acquiring the InSAR interference map is as follows: using synthetic aperture radar (SAR) to perform high-resolution microwave imaging on the mining area to be monitored, and processing the microwave imaging by a differential radar interferometry method to obtain the InSAR interference map.

3. The method for extracting mining subsidence basin information based on InSAR as claimed in claim 1, characterized in that: The deep learning network model includes a feature extraction module, a dilated convolution module, a residual network module and a feature output module; The feature extraction module extracts multiple feature maps from the InSAR interferogram; The dilated convolution module uses convolutional layers with different dilation rates to perform multi-scale information fusion on the extracted multiple feature maps; The residual network module connects the input and output of the residual network module based on a plurality of convolutional layers of residual structures; The feature output module extracts mining subsidence basin information from the feature map output by the residual network module.

4. The method for extracting mining subsidence basin information based on InSAR as claimed in claim 1, characterized in that: The subsidence prediction model is specifically: ； in, For the The calculation formula for the maximum impact value of surface subsidence after mining of a dynamic unit is: For the The time coefficient corresponding to the dynamic unit at time t, The dynamic unit is obtained by dividing the mining area to be monitored into units. units.

5. The method for extracting mining subsidence basin information based on InSAR as claimed in claim 1, characterized in that: The combined Perlin noise and Gaussian noise model is specifically: By superimposing Perlin noises of different frequencies and amplitudes, fractal Perlin noise is obtained as the atmospheric delay phase in the simulated InSAR interferogram. A Gaussian noise model is introduced, and the noise intensity is randomly set during the simulation process to generate independent noise matrices with the same magnitude as the real and imaginary parts. The values ​​are wrapped between (-π,π] to obtain the wrapped phase of the InSAR interferogram.

6. A mining subsidence basin information extraction system based on InSAR, characterized in that: Including acquisition module and extraction module: The acquisition module is configured to: acquire an InSAR interferogram of the mining area to be monitored; The extraction module is configured to: input the InSAR interference pattern into the trained deep learning network model to extract mining subsidence basin information, wherein the mining subsidence basin information is a surface subsidence value; The training data set of the deep learning network model includes a real training data set and a simulated training data set. The real training data set is constructed by combining differential interferometry with a static probability integral method to obtain real mining subsidence basin information. The simulated training data set is constructed by simulating the mining subsidence phase using a dynamic probability integral method and combining Perlin noise and Gaussian noise models. The method of combining differential interferometry with static probability integration method is specifically as follows: For small-scale mining subsidence areas, the subsidence monitoring results of D-InSAR are used as sample label data; For large-scale mining subsidence areas, based on sample label data, the corresponding sample label data is obtained by combining InSAR with the static probability integral method; The dynamic probability integral method is used to simulate the mining subsidence phase, specifically: The probability integral method is used to dynamically predict mining subsidence in the mining area. By adjusting the mining parameters and using the Knothe time function, the surface settlement values ​​of different intensities and forms within the predicted time of different working faces are obtained, and then a subsidence prediction model is constructed.

7. An electronic device, comprising: a memory for non-transitory storage of computer readable instructions; a processor for executing the computer-readable instructions; Wherein, when the computer-readable instructions are executed by the processor, the method according to any one of claims 1 to 5 is executed.

8. A storage medium, characterized in that: The computer-readable instructions are non-transitorily stored, wherein when the computer-readable instructions are executed by a computer, the method according to any one of claims 1 to 5 is performed.

Images