Earth crust deformation monitoring method, electronic equipment and storage medium

Abstract

The invention provides an earth crust deformation monitoring method, electronic equipment and a storage medium, and relates to the technical field of geological investigation, an earth crust observation data set of a target area is obtained, and the earth crust observation data set comprises at least two different types of earth crust observation data; and inputting the earth crust observation data set into a pre-constructed long-short-term memory neural network model based on wavelet transformation to obtain a data fusion result, carrying out statistical analysis on the data fusion result, and obtaining earth crust abnormal information of the target area so as to monitor earth crust deformation change of the target area. Due to the fact that different types of crustal observation data are effectively fused, crustal deformation is monitored through the fused data fusion result, the crustal deformation monitoring precision is improved, the three-dimensional monitoring capacity of crustal deformation of the target area is expanded, crustal deformation changes of small areas can be detected more accurately, and the crustal deformation monitoring accuracy is improved. And the crustal deformation monitoring precision is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of geological exploration, and in particular, to a crustal deformation monitoring method, an electronic device, and a storage medium. Background Art

[0002] At present, the global navigation satellite system (GNSS) reference station network established globally provides a direct and effective means of observation for crustal deformation monitoring. The crustal movement monitoring network can be divided into three scales of monitoring networks: global, regional, and local, according to the size of its coverage. Since these GNSS reference stations were established, they have accumulated decades of observation data. The coordinate time series obtained by solving with high-precision GNSS data software provides important basic data for research in fields such as regional crustal deformation monitoring, global plate tectonic movement research, fault slip monitoring, and earthquake deformation monitoring.

[0003] The GNSS reference station coordinate time series contains long-term linear changes, seasonal periodic changes, earthquake displacement changes, and changes affected by other factors in the crust. Completing the analysis of the GNSS reference station coordinate time series and finding out the abnormal information therein is of great significance for studying the laws of regional crustal deformation. However, the analysis of the characteristics of the GNSS reference station coordinate time series mainly focuses on studying the noise model and seasonal changes of the time series. When analyzing the impact of some natural disasters on the time series, further research and analysis of non-seasonal periodic signals are still required. However, the GNSS reference stations are sparsely distributed and it is difficult to form a complete coverage of the region, thus affecting the monitoring accuracy of crustal deformation.

[0004] In summary, there is a certain need to improve the monitoring accuracy of crustal deformation at present. Summary of the Invention

[0005] The problem solved by the present invention is how to solve the problem of low monitoring accuracy caused by single data.

[0006] To solve the above problems, the present invention provides a crustal deformation monitoring method, an electronic device, and a storage medium.

[0007] In a first aspect, the present invention provides a crustal deformation monitoring method, including: Obtaining a crustal observation data set of a target area, where the crustal observation data set includes at least two different types of crustal observation data, and at least one type of crustal observation data is global navigation satellite system data; Input the crust observation data set into a pre-constructed long short-term memory neural network model based on wavelet transform to obtain a data fusion result output by the long short-term memory neural network model based on wavelet transform. The data fusion result is the result of fusing at least two different types of crust observation data in the crust observation data set. Perform statistical analysis on the data fusion result to obtain crust anomaly information of the target area, where the crust anomaly information is used to characterize crustal deformation changes.

[0008] Optionally, inputting the crust observation data set into a pre-constructed long short-term memory neural network model based on wavelet transform to obtain the data fusion result output by the long short-term memory neural network model based on wavelet transform includes: Perform wavelet transform processing on at least two different types of crust observation data in the crust observation data set to obtain weight information of the crust observation data set; Use a convolutional neural network to extract features from at least two different types of crust observation data in the crust observation data set to obtain feature information of the crust observation data set; Input the weight information of the crust observation data set and the feature information of the crust observation data set into a long short-term memory neural network model to obtain a data fusion result output by the long short-term memory neural network model.

[0009] In a second aspect, the present invention provides an electronic device, including a memory and a processor; The memory is used to store a computer program; The processor is used to implement the crustal deformation monitoring method as described in the first aspect when executing the computer program.

[0010] In a third aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the crustal deformation monitoring method as described in the first aspect is implemented.

[0011] The beneficial effects of the crustal deformation monitoring method, electronic device and storage medium of the present invention are as follows: obtaining a crustal observation data set of a target area, where the crustal observation data set includes at least two different types of crustal observation data, and at least one type of crustal observation data is global navigation satellite system data. By obtaining multiple types of observation data of different types and inputting the crustal observation data set into a pre-constructed long short-term memory neural network model based on wavelet transform to obtain a data fusion result output by the long short-term memory neural network model based on wavelet transform. Through the long short-term memory neural network model based on wavelet transform to perform fusion processing on the obtained multi-source observation data, and perform statistical analysis on the data fusion result to obtain crustal anomaly information of the target area; due to the effective fusion of different types of crustal observation data, monitoring the crustal deformation through the data fusion result after fusion improves the accuracy of crustal deformation monitoring, and expands the three-dimensional monitoring ability of crustal deformation in the target area, and can more accurately detect the crustal deformation changes in small areas, realizing the improvement of the accuracy of crustal deformation monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 It is a flowchart of a crustal deformation monitoring method according to an embodiment of the present invention; Figure 2 It is a flowchart for preprocessing and standardizing crustal observation data; Figure 3 It is a processing flowchart for inputting the crustal observation data set into a pre-constructed long short-term memory neural network model based on wavelet transform; Figure 4 It is a schematic diagram of wavelet transform processing and convolutional neural network processing for GNSS site displacement data, InSAR displacement data and remote sensing image data respectively; Figure 5 It is a flowchart for wavelet transform processing of at least two different types of crustal observation data in the crustal observation data set; Figure 6 It is a schematic diagram of the structure of the gated convolutional layer; Figure 7 It is a schematic diagram of the structure of the convolutional neural network; Figure 8 It is a flowchart for feature extraction of at least two different types of crustal observation data in the crustal observation data set by using a convolutional neural network; Figure 9 It is a processing flowchart for inputting the weight information and feature information of the crustal observation data set into the long short-term memory neural network model; Figure 10 It is a schematic diagram of the structure of the cell structure in the long short-term memory neural network model; Figure 11A flowchart for statistical analysis of data fusion results; Figure 12 A schematic structural diagram of a crustal deformation monitoring device according to an embodiment of the present invention; Figure 13 A schematic structural diagram of an electronic device according to an embodiment of the present invention. Detailed implementation manners

[0013] To make the above objects, features, and advantages of the present invention more apparent and understandable, the following describes in detail the specific embodiments of the present invention with reference to the accompanying drawings. Although some embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be construed as limited to the embodiments described herein. On the contrary, these embodiments are provided to more thoroughly and completely understand the present invention. It should be understood that the drawings and embodiments of the present invention are only for exemplary purposes and are not used to limit the protection scope of the present invention.

[0014] It should be understood that the steps recorded in the method embodiments of the present invention can be executed in different orders and / or in parallel. In addition, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present invention is not limited in this regard.

[0015] The following explains the terms involved in the present application.

[0016] GNSS (Global Navigation Satellite System) is a satellite system for global autonomous geospatial positioning, allowing small electronic receivers (stations) to determine their locations (longitude, latitude, and altitude) and accurately determining the time signals transmitted along the line of sight via satellite broadcasts within a range of 10 meters.

[0017] INSAR (Interferometric Synthetic Aperture Radar) is a remote sensing technology for monitoring surface deformation and topographic mapping using synthetic aperture radar images. Two or more synthetic aperture radar imagery maps are used to generate a digital elevation model or a surface deformation map based on the phase difference of the echoes received by satellites or aircraft.

[0018] Remote sensing images are images of the Earth's surface obtained through remote sensing platforms such as satellites, unmanned aerial vehicles, and aircraft.

[0019] In the related art, when processing the observation data for monitoring crustal deformation, it is often limited by the singularity of the observation data and the complexity of the model, making it difficult to accurately capture the subtle changes in crustal deformation. Multiple different types of observation data can provide more comprehensive and rich information. For example, the fusion of multiple different types of data such as GNSS and InSAR can reflect the crustal deformation situation from different angles and levels, effectively improving the reliability and integrity of the data. When using multiple different types of observation data, it is necessary to fuse the multiple data. Using traditional fusion methods to fuse the data will be limited by the different resolutions between the data, resulting in a significant reduction in the fusion performance and accuracy of the data.

[0020] In view of the problems existing in the above related art, this embodiment provides a crustal deformation monitoring method, an electronic device, and a storage medium.

[0021] As Figure 1 shown, a crustal deformation monitoring method provided by an embodiment of the present invention includes: S100, obtaining a crustal observation data set of a target area, where the crustal observation data set includes at least two different types of crustal observation data, and at least one type of crustal observation data is global navigation satellite system data.

[0022] Specifically, different types of crustal observation data are observation data obtained by different acquisition source methods, that is, multi-source observation data. For example, it can be global navigation satellite system data, interferometric synthetic aperture radar data, and remote sensing image data. It should be noted that the crustal observation data set of the embodiment of the present invention needs to include at least global navigation satellite system data. Other observation data is not limited to the three types of data exemplified above. If some of the data is missing in a certain area, the method of the embodiment of the present invention can also be used for reconstruction calculation. However, in the case of less data volume, the accuracy will be reduced. In addition, other observation data can be used to supplement the data volume, thereby improving the reconstruction accuracy. For example, the Gravity Recovery and Climate Experiment gravity satellite data and the Global Land Data Assimilation System data can be added. By jointly fusing these supplementary data with the global navigation satellite system data, the monitoring accuracy and data integrity can be effectively improved, and it has strong flexibility and scalability. The problem of data loss can be compensated by adding other data types, so as to achieve efficient and accurate crustal deformation monitoring in a wider range of application scenarios.

[0023] Specifically, the target area can be selected according to specific needs, and usually includes earthquake-active areas, volcanic areas, geological disaster-prone areas, mining areas and oil and gas extraction areas, urban and infrastructure areas, plate boundaries and active fault areas, and other special areas (such as high mountain and canyon areas, deserts, swamps, permafrost, etc.).

[0024] S200, Input the crust observation dataset into a pre-constructed long short-term memory neural network model based on wavelet transform to obtain the data fusion result output by the long short-term memory neural network model based on wavelet transform; wherein, the data fusion result is the result of fusing at least two different types of crust observation data in the crust observation dataset.

[0025] Specifically, the long short-term memory neural network model based on wavelet transform can assign different weights to data according to its different frequency domain dimensions, which helps to improve the accuracy of the fused data, thereby enriching the observation data of crustal deformation in the target area.

[0026] S300, Conduct statistical analysis on the data fusion result to obtain the crust anomaly information of the target area, and the crust anomaly information is used to characterize the change of crustal deformation.

[0027] Specifically, conduct statistical analysis on the data fusion result, and comprehensively statistically analyze the crust anomaly information in the target area to describe the deformation characteristics of the crust.

[0028] In this embodiment, obtain the crust observation dataset of the target area. The crust observation dataset includes at least two different types of crust observation data, and at least one type of crust observation data is global navigation satellite system data. By obtaining multi-source observation data, it is better to completely cover the target area. Input the crust observation dataset into a pre-constructed long short-term memory neural network model based on wavelet transform to obtain the data fusion result output by the long short-term memory neural network model based on wavelet transform, and perform fusion processing on the obtained multi-source observation data through the long short-term memory neural network model based on wavelet transform. Conduct statistical analysis on the data fusion result to obtain the crust anomaly information of the target area to monitor the change of crustal deformation in the target area. Due to the effective fusion of multi-source observation data, monitor the crustal deformation through the fused data fusion result, which improves the accuracy of crustal deformation monitoring, and expands the three-dimensional monitoring ability of the deformation in the target area, and can more accurately detect the change of crustal deformation in a small area, realizing the improvement of the accuracy of crustal deformation monitoring.

[0029] Optionally, before inputting the crust observation dataset into a pre-constructed long short-term memory neural network model based on wavelet transform, it further includes: Perform preprocessing and standardization processing on at least two different types of crust observation data in the crust observation dataset respectively, so that at least two different types of crust observation data are consistent in the spatial coordinate system.

[0030] In some embodiments, the crust observation dataset includes: global navigation satellite system (hereinafter referred to as GNSS) data, interferometric synthetic aperture radar data (hereinafter referred to as InSAR image data), and remote sensing image data, such asFigure 2 As shown, preprocessing and standardizing at least two different types of crustal observation data in the crustal observation dataset include: S210: Calculate the GNSS data to obtain displacement data of corresponding GNSS stations in the target area.

[0031] Specifically, the displacement data is displacement data of a preset length time series, and the displacement data includes an east component, a north component, and a sky component. The preset length time series may be a natural day or other time lengths.

[0032] Specifically, the GNSS data site corresponds to the GNSS data. For example, for the GNSS data A1 received by the GNSS site A, the displacement data of the GNSS site A can be obtained by solving the GNSS data A1.

[0033] Specifically, GNSS data processing software (such as GAMIT) can be used to resolve GNSS data. In one embodiment, TEQC software is used to perform quality checks on GNSS data, eliminating substandard station data. GAMIT, an open-source data processing tool, is then used to resolve the GNSS data based on precise satellite ephemeris files and other parameter files requiring corrections. This results in long-term daily displacement data for the station in the east, north, and sky directions. Because some stations lack complete time series data due to weather or instrument issues, GNSS missing data interpolation software is used to interpolate the missing data, resulting in complete time series displacement data for the corresponding GNSS station of a preset length.

[0034] S220: Perform interferometric processing on at least two InSAR images in the InSAR image data to obtain InSAR displacement data of the target area.

[0035] Specifically, radiometric correction is performed on InSAR images to eliminate radiometric errors caused by factors such as sensor characteristics and atmospheric conditions, so that the radiometric values of the images can more accurately reflect the scattering characteristics of ground objects; using reference DEM data (digital terrain model for representing the altitude of the Earth's surface), terrain correction is performed on InSAR images to transform the InSAR images from the radar coordinate system to the geographic coordinate system, eliminating the influence of terrain undulation on the geometric shape of InSAR images; through multi-looking processing of InSAR images, the spatial resolution of the images is reduced, while the signal-to-noise ratio of scatterers is increased, improving the quality of the interferogram; for multi-scene InSAR images in different time periods of the coverage area, the temporal and spatial baselines are calculated, and usually the interferometric pairs with shorter spatio-temporal baselines are selected to extract deformation information; two or more preprocessed InSAR images are interferometrically processed to generate an interferogram, which is used to display the phase difference between different InSAR images; coherence calculation is performed on the generated interferogram to obtain a coherence map, which reflects the correlation between different InSAR images, and coherence reflects the interference quality of each pixel in the interferogram. The interference phase in the high-coherence region is more stable and reliable; filtering processing is performed on the coherence map to further enhance the signal in the high-coherence region, suppress the noise in the low-coherence region, and improve the overall quality of the interferogram; the interferogram usually contains a 2π periodic phase, and the least squares method is used for phase unwrapping to obtain the true displacement information of the ground surface; according to the situation of forming subsets by freely combining interferograms, a phase equation is formed for all interferograms, and the least squares solution method is used to solve the phase equation to estimate the InSAR displacement data of the final target area.

[0036] S230, perform cloud removal processing and band fusion processing on remote sensing image data.

[0037] Specifically, use ENVI software to load remote sensing image data, select the Cloud Masking tool, use the exponential masking algorithm, set the corresponding threshold parameters, and perform cloud removal processing on the remote sensing image data. In ENVI software, select the Color Normalized Sharpening tool, select the band names to be fused, set the resampling method, and the fused remote sensing image data can be obtained.

[0038] S240, fuse the displacement data of GNSS stations and remote sensing image data to obtain the displacement data of GNSS stations with unified spatial coordinates with the remote sensing images.

[0039] Specifically, according to the longitude and latitude information of the station coordinates, add the displacement information of its station to the corresponding position of the remote sensing image and perform familiar superposition in the form of multi-bands; then perform standardization processing on the bands of the remote sensing image and set its pixel values to data that conforms to the normal distribution.

[0040] S250: Interpolate the GNSS site displacement data to obtain the displacement distribution within the region.

[0041] In some embodiments, the inverse distance weighting method is used to interpolate the site displacement data. Specifically, according to the coordinate positions of the existing GNSS sites, each GNSS site is processed as follows: for each position to be interpolated, calculate its distance from all known GNSS sites: , where and respectively represent the longitude and latitude to be interpolated; and respectively represent the longitude and latitude positions of the known sample points, and then convert the distance into a weight: , p is an exponential parameter. To avoid division by zero, a minimum value is set for the distance, and interpolation calculations are performed for each grid using the weight information: , where N table represents the number of known GNSS sites within the search radius, represents the i th displacement value of the GNSS site, represents element-wise multiplication. Interpolation is performed for each grid within the target area to obtain the displacement data within the area.

[0042] S260: Perform fusion processing on the InSAR displacement data and remote sensing image data of the target area to obtain InSAR displacement data with the same spatial coordinates as the remote sensing image.

[0043] Specifically, similar to the method of fusing the GNSS site displacement data and remote sensing image data, the InSAR displacement data is added to the corresponding position of the remote sensing image to obtain InSAR displacement data with the same spatial coordinates as the remote sensing image.

[0044] In this alternative embodiment, preprocessing and standardization processing are performed on different types of observation data to ensure the spatial unity of different types of observation data, so as to facilitate subsequent input into the neural network model for fusion processing.

[0045] Optionally, as Figure 3 shown, input the crustal observation data set into a pre-constructed long short-term memory neural network model based on wavelet transform to obtain the data fusion result output by the long short-term memory neural network model based on wavelet transform, including: S310: Perform wavelet transform processing on at least two different types of crustal observation data in the crustal observation data set to obtain the weight information of the crustal observation data set.

[0046] S320, Use a convolutional neural network to extract features from at least two different types of crustal observation data in the crustal observation dataset to obtain the feature information of the crustal observation dataset.

[0047] S330, Input the weight information and feature information of the crustal observation dataset into a long short-term memory neural network model to obtain the data fusion result output by the long short-term memory neural network model.

[0048] In some embodiments, the crustal observation dataset includes GNSS data, InSAR data, and remote sensing image data. After preprocessing and standardization, GNSS station displacement data, InSAR displacement data, and remote sensing image data are obtained. As Figure 4 shown, perform wavelet transform processing and convolutional neural network on the GNSS station displacement data, InSAR displacement data, and remote sensing image data respectively to obtain weight information and feature information, and then input the weight information and feature information into a long short-term memory neural network model (LSTM neural network model) for data fusion to obtain the data fusion result.

[0049] In this alternative embodiment, compared with traditional fusion methods, the long short-term memory neural network model based on wavelet transform can assign different weights to data according to its different frequency domain dimensions, which helps to improve the accuracy of the data fusion result, thereby enriching the observation data of crustal deformation in the target area.

[0050] Optionally, as Figure 5 shown, performing wavelet transform processing on at least two different types of crustal observation data in the crustal observation dataset, the weight information of the crustal observation dataset obtained includes: S510, Perform wavelet analysis on at least two different types of crustal observation data in the crustal observation dataset to obtain high-frequency feature data and low-frequency feature data. The high-frequency feature data is used to characterize the short-term changes and noise of the crustal observation dataset, and the low-frequency feature data is used to characterize the long-term trend and low-frequency information of the crustal observation dataset.

[0051] Specifically, first decompose multiple observation data using wavelet analysis. Wavelet analysis usually decomposes data into three layers. The core of the decomposition is to decompose the data into approximate coefficients (low-frequency feature data) and detail coefficients (high-frequency feature data) at different scales. The expression for data decomposition using wavelet analysis is as follows: ; where represents the decomposition coefficient of the j th layer, h and g represent the coefficients of the filter, j and nrespectively represent the decomposition level and the node number, l represents a discrete index related to the time axis, k represents the position index in signal processing.

[0052] In this embodiment, various observation data of different types are decomposed into 3 layers, and the approximate coefficients (low-frequency feature data) of each layer mainly include the long-term trend and low-frequency information of the information, and the detail coefficients (high-frequency feature data) mainly include the short-term changes and noise of the information.

[0053] S520, input the high-frequency feature data and the low-frequency feature data into the first double convolutional layer to obtain the first high-frequency data feature map and the first low-frequency data feature map output by the first double convolutional layer.

[0054] Specifically, input the high-frequency feature data and the low-frequency feature data into the first double convolutional layer. In one embodiment, the first double convolutional layer is composed of two convolutional kernels with a size of 3, the stride is set to 1, the padding is set to 1, and the output channels are set to 32. By passing through the first double convolutional layer, the size of the data can be kept unchanged, and more feature data can be extracted to obtain the first high-frequency data feature map and the first low-frequency data feature map.

[0055] S530, input the first high-frequency data feature map and the first low-frequency data feature map into the max pooling layer to obtain the second high-frequency data feature map and the second low-frequency data feature map output by the max pooling layer.

[0056] Specifically, after the data is subjected to feature extraction through S520, its size needs to be changed to half of the original. In one embodiment, the convolutional kernel size of the max pooling layer is set to 2, the stride is set to 2, and the padding is set to 0, so that the first high-frequency data feature map and the first low-frequency data feature map will have their sizes reduced by half when passing through the max pooling layer, so that the next convolutional operation can extract global relevant features.

[0057] S540, input the second high-frequency data feature map and the second low-frequency data feature map into the first gated convolutional layer to obtain the third high-frequency data feature map and the third low-frequency data feature map output by the first gated convolutional layer.

[0058] Specifically, set the convolution kernel size of the first gated convolutional layer to 3, the stride to 1, the padding to 1, and the number of output channels to 64. When performing gated convolution, similar to standard convolution, first set the number of output channels to 128. After the first convolution of S520, the size of the data remains unchanged, but the number of output channels becomes four times that of the input channels. Then, through S530, the convolved data is evenly divided into two parts in the channel dimension. One part is input into the sigmoid activation function and then multiplied by the remaining half of the data to obtain the output result of the first gated convolutional layer, that is, the third high-frequency data feature map and the third low-frequency data feature map are obtained.

[0059] Specifically, input the high-frequency feature data and low-frequency feature data after S510 wavelet analysis into the encoding layer for feature extraction. Among them, the first gated convolutional layer is used to further extract key features. The gated convolutional layer is as Figure 6 shown. The gated convolution structure mainly includes a convolutional layer 610 and a gating layer 620. The convolution expression of the convolutional layer 610 is: A = E * W + b, B = E * V + c, which is the expression for double convolution of the input data. Among them, A and B are the results after convolution, E represents the data input into the convolutional layer 610, W and V represent the convolution weights, b and c represent the convolution biases. The convolutional layer 610 applies filters to tensors to generate output tensors, while gated convolution modifies the ordinary convolution process by introducing a gating mechanism, so that the output is calculated as a combination of the convolution output and the gating signal. The processing expression of the gating layer 620 is: , where represents the result after gated convolution, represents element-wise multiplication, represents the sigmoid activation function.

[0060] The convolutional layer 610 convolves the input high-frequency feature data and low-frequency feature data to obtain feature data twice the number of output channels, and evenly divides the feature data into two parts. One part is input into the sigmoid activation function, and the other part is multiplied by the result of the input activation function to obtain the final result. The gating layer 620 is an advanced convolution operation that introduces a gating mechanism to enhance the model's ability to capture complex dependencies in the data. Compared with ordinary convolutional layers, gated convolution is particularly effective in capturing long-range dependencies in sequences and provides flexibility in processing information, enabling the model to adaptively learn which features are important for the current task and improving the model's robustness. Using an encoding layer with gated convolution can effectively extract important feature data.

[0061] S550, upsample the third high-frequency data feature map and the third low-frequency data feature map and input them into the second double convolutional layer to obtain the fourth high-frequency data feature map and the fourth low-frequency data feature map output by the second double convolutional layer.

[0062] Specifically, the third high-frequency data feature map and the third low-frequency data feature map are input into the upsampling function, the size is set to twice the current size, and the sampling method is set to the nearest neighbor interpolation method to restore its size to the original size; subsequently, the feature map with the restored size is input into the second double convolutional layer. The convolutional kernel size of the second double convolutional layer is set to 3, the stride is set to 1, the padding is set to 1, and the number of output channels is set to 64. The fourth high-frequency data feature map and the fourth low-frequency data feature map are obtained through the second double convolutional layer.

[0063] S560, the fourth high-frequency data feature map is concatenated with the first high-frequency data feature map to obtain the first high-frequency concatenated feature map; the fourth low-frequency data feature map is concatenated with the first low-frequency data feature map to obtain the first low-frequency concatenated feature map; the first high-frequency concatenated feature map and the first low-frequency concatenated feature map are input into the third double convolutional layer to obtain the fifth high-frequency data feature map and the fifth low-frequency data feature map output by the third double convolutional layer.

[0064] Specifically, since the fourth high-frequency data feature map and the first high-frequency data feature map, and the fourth low-frequency data feature map and the first low-frequency data feature map are inconsistent in the channel dimension, in this embodiment, the two feature maps with inconsistent channel dimensions are merged in the dimension scale through concatenation to obtain a feature map with consistent channel dimensions, and then the concatenated feature map is input into the third double convolutional layer. In one embodiment, if the number of channels of the concatenated feature map is 96, the convolutional kernel of the third double convolutional layer is set to 3, the stride is 1, the padding is 1, and the output channel is 48, and finally the fifth high-frequency data feature map and the fifth low-frequency data feature map are obtained.

[0065] S570, the fifth high-frequency data feature map and the fifth low-frequency data feature map are input into the second gated convolutional layer to obtain the sixth high-frequency data feature map and the sixth low-frequency data feature map output by the second gated convolutional layer; wherein, the sixth high-frequency data feature map and the sixth low-frequency data feature map are used to characterize the weight information of the crust observation data set.

[0066] Specifically, the convolutional kernel size of the second gated convolutional layer is set to 3, the stride is 1, the padding is 1, and the output channel is set to 1. The sixth high-frequency data feature map and the sixth low-frequency data feature map are obtained through the second gated convolutional layer.

[0067] In this optional embodiment, after the above convolution, the weight information of various types of data under wavelet transform is obtained.

[0068] Optionally, as Figure 7As shown, the convolutional neural network consists of an encoding module, a standard convolutional layer, a decoding module, and skip connections. The encoding module is jointly composed of a wavelet downsampling layer and a double convolutional layer (including gated convolution), and the decoding module is composed of a transposed convolutional layer and a double convolutional layer.

[0069] Optionally, as Figure 8 shown, the convolutional neural network is used to extract features from at least two different types of crustal observation data in the crustal observation dataset, and the obtained feature information of the crustal observation dataset includes: S810, input at least two different types of crustal observation data in the crustal observation dataset into the fourth double convolutional layer to obtain the first fusion feature map output by the fourth double convolutional layer.

[0070] In one embodiment, the convolutional kernel size of the fourth double convolutional layer is set to 3, the stride is 1, the padding is 1, and the number of output channels is 32. The first fusion feature map can be obtained through the fourth double convolutional layer.

[0071] S820, input the first fusion feature map into the first wavelet downsampling layer, and input the output result of the first wavelet downsampling layer into the fifth double convolutional layer to obtain the second fusion feature map output by the fifth double convolutional layer.

[0072] In one embodiment, the number of channels of the first wavelet downsampling layer is set to 32, the convolutional kernel size of the fifth double convolutional layer is set to 3, the stride is 1, the padding is 1, and the number of output channels is 64.

[0073] S830, input the second fusion feature map into the second wavelet downsampling layer, and input the output result of the second wavelet downsampling layer into the sixth double convolutional layer to obtain the third fusion feature map output by the sixth double convolutional layer.

[0074] In one embodiment, the number of channels of the second wavelet downsampling layer is set to 64, the convolutional kernel size of the sixth double convolutional layer is set to 3, the stride is 1, the padding is 1, and the number of output channels is 128.

[0075] S840, input the third fusion feature map into the third wavelet downsampling layer, and input the output result of the third wavelet downsampling layer into the seventh double convolutional layer to obtain the fourth fusion feature map output by the seventh double convolutional layer.

[0076] In one embodiment, the convolutional kernel size of the seventh double convolutional layer is set to 3, the stride is 1, the padding is 1, and the number of output channels is 256.

[0077] S850, input the fourth fusion feature map into the eighth double convolutional layer to obtain the fifth fusion feature map output by the eighth double convolutional layer.

[0078] In one embodiment, the convolutional kernel size of the eighth double convolutional layer is set to 3, the stride is 1, the padding is 1, and the number of output channels is 512.

[0079] S860, input the fifth fused feature map into the first transposed convolutional layer, concatenate the output result of the first transposed convolutional layer with the third fused feature map in dimension, and input the dimension concatenation result into the ninth double convolutional layer to obtain the sixth fused feature map output by the ninth double convolutional layer.

[0080] In one embodiment, the convolutional kernel size of the first transposed convolutional layer is 3, the stride is 1, and the padding is 2.

[0081] S870, input the sixth fused feature map into the second transposed convolutional layer, concatenate the output result of the second transposed convolutional layer with the second fused feature map in dimension, and input the dimension concatenation result into the tenth double convolutional layer to obtain the seventh fused feature map output by the tenth double convolutional layer.

[0082] In one embodiment, the convolutional kernel size of the second transposed convolutional layer is 3, the stride is 1, and the padding is 2. The number of output channels of the tenth double convolutional layer is 32.

[0083] S880, input the seventh fused feature map into the third transposed convolutional layer, concatenate the output result of the third transposed convolutional layer with the first fused feature map in dimension, and input the dimension concatenation result into the eleventh double convolutional layer to obtain the eighth fused feature map output by the eleventh double convolutional layer. The eighth fused feature map is used to represent the feature information of the crust observation data set.

[0084] In one embodiment, the convolutional kernel size of the third transposed convolutional layer is 3, the stride is 1, and the padding is 2. The number of output channels of the eleventh double convolutional layer is 32.

[0085] In this alternative embodiment, using the wavelet downsampling layer can reduce the dimension and compress the data through discrete wavelet transform, retain the key information while reducing the data volume, thereby improving the storage efficiency and analysis speed. In addition, wavelet transform can effectively separate noise and useful signals, and realize noise filtering and improve data quality by selectively retaining low-frequency approximation coefficients or performing threshold processing on high-frequency detail coefficients. At the same time, the multi-resolution analysis characteristic of wavelets allows observing time series data at different time scales, which can not only capture long-term trends but also analyze short-term fluctuations, providing a more comprehensive perspective for complex data. The extracted global and local features can be used for classification and pattern recognition, enhancing the discrimination ability of the model, and can also be used for anomaly detection, which not only improves the analysis efficiency but also enhances the interpretability and accuracy of time series data.

[0086] Optionally, as Figure 9As shown in the figure, the weight information of the crust observation data set and the feature information of the crust observation data set are input into the long short-term memory neural network model to obtain the data fusion result output by the long short-term memory neural network model, including: S910, merge the weight information of the crust observation data set and the feature information of the crust observation data set to obtain merged data.

[0087] Specifically, the sixth high-frequency data feature map and the sixth low-frequency data feature map are respectively multiplied pixel by pixel with the eighth fusion feature map, and then the results after pixel-by-pixel multiplication are concatenated at the channel scale to obtain a weighted fusion feature map.

[0088] S920, input the merged data into the third gated convolutional layer to obtain the merged feature data output by the third gated convolutional layer.

[0089] Specifically, input the weighted fusion feature map into the third gated convolutional layer. In one embodiment, the convolutional kernel size of the third gated convolutional layer is 3, the stride is 1, the padding is 1, and the output channels are 3; through the third gated convolutional layer, the feature data of the weighted fusion feature map is extracted to obtain the merged feature data input into the long short-term memory neural network model, and the merged feature data is long-time series data.

[0090] S930, input the merged feature data into the long short-term memory neural network model (LSTM) to obtain the data fusion result output by the long short-term memory neural network model.

[0091] Specifically, the structure of the long short-term memory neural network model is composed of a cell sequence, and the cell structure is as Figure 10 shown. This structure mainly includes three parts: the forget gate, the input gate, and the output gate. In this cell structure, is the input of the time series, represents the output of the time series, t represents the time; the forget gate is used to control the discarding (forgetting) or retention of some information in the memory unit, and its calculation formula is: ; the input gate determines the information to update the memory unit, including two parts: Sigmoid and Tanh. Among them, the formula of Sigmoid is: , and the formula of Tanh is: , where is the Sigmoid function, x represents the input data, Tanh( x ) is the Tanh function; both the Sigmoid and Tanh parts contain the input at the current moment and the output at the previous moment, which is composed of: , These two formulas determine; the function of the output gate is to read the just-updated neural network state and output to the memory cell, and which specific information can be output is controlled by the output gate. Its main calculation formula is: , , ; in the above formula, , , , are all internal functions of the long short-term memory cell, , and are memory vectors, , , , are weight parameters, , , , are convolutional biases. It should be noted that the weight parameters are obtained through the training of the neural network.

[0092] In this optional embodiment, the long short-term memory neural network model, as a kind of recurrent neural network architecture, is widely used in time series modeling and can also be applied to deformation monitoring based on monitoring data. By constructing a multi-layer stacked long short-term memory neural network model and establishing displacement transformation monitoring data through data training, data fusion and deformation reconstruction are realized. Moreover, the long short-term memory neural network model has the characteristic of long-term memory of information, can extract information not only from a single data point but also from the entire series of data, and allows information to persist and circulate.

[0093] Optionally, as Figure 11 shown, statistical analysis is performed on the data fusion result to obtain crustal anomaly information of the target area, including: S1110, perform statistical analysis on the data fusion result to obtain the mean and standard deviation of the data fusion result.

[0094] S1120, perform a normal distribution test on the data fusion result and perform a normality transformation on the data in the data fusion result that does not conform to the normal distribution to obtain a data fusion result that conforms to the normal distribution.

[0095] Specifically, perform a normal distribution test on the data fusion result. If there is data in the data fusion result that does not conform to the normal distribution, the Q statistic method is used to transform the data that does not conform to the normal distribution. The Q statistic is usually used in the process of normality transformation and can effectively process variables that do not conform to the normal distribution to make them more in line with the normal distribution hypothesis.

[0096] S1130, calculate the mean and standard deviation of the data fusion result using the cumulative sum control chart formula to obtain the test data, and convert the test data into a time-type sequence to obtain the time series test data.

[0097] Specifically, input the mean and standard deviation of the data fusion result statistically obtained in S1110 into the cumulative sum control chart (CUSUM) formula for calculation to obtain the upper offset statistic and lower offset statistic in CUSUM. Take the upper offset statistic and lower offset statistic in the above CUSUM as the test data and convert them into a time-type sequence.

[0098] S1140, perform a linear trend fitting on the time series test data using the least squares method to obtain the linear change trend of the target area.

[0099] S1150, remove the linear change trend from the time series test data and fit the periodic signal in the data after removal to obtain the periodic motion trend of the target area in one or more predetermined time periods.

[0100] Specifically, the predetermined time period can be an annual period and / or a semi-annual period.

[0101] S1160, obtain the crustal anomaly information of the target area based on the crustal linear change trend and periodic crustal movement trend of the target area.

[0102] Specifically, detect the abnormal points of the crustal linear change trend. For each grid point, calculate the difference between its actual displacement and the fitted linear change trend. If the difference exceeds a certain set threshold (such as 3 times the standard deviation), it is considered that there is an abnormality in the linear trend of this point; detect the abnormal points of the periodic crustal movement trend by comparing the actual periodic crustal movement trend of each grid point with the fitted periodic crustal movement trend. If the period amplitude and frequency exceed the preset range, it may indicate that the periodic change in this area is abnormal. Use standardized residual detection. For each point, calculate the standardized residual between the actual value and the fitted value. If the residual is too large, it may mean that there is an abnormality at this point. For the obtained abnormal points, use cluster analysis to identify the possible abnormal areas within the target area. These abnormal areas may have abnormal movement trends or periodic fluctuations. Statistics on the distribution of abnormal points, such as the number of abnormal grid points and the spatial distribution characteristics of abnormal points, helps to understand whether the abnormal phenomena are concentrated in certain areas.

[0103] In this alternative embodiment, statistical analysis is performed on the data fusion result after fusion. The CUBFASET method can effectively detect abnormal information in the data. Then, the least squares method is used to fit the fused data to obtain the linear velocity and periodic change data of crustal deformation. The linear change trend and periodic movement trend in the target area are comprehensively statistically analyzed to describe the deformation characteristics of the crust.

[0104] As Figure 12 shown, a crustal deformation monitoring device 1200 provided by an embodiment of the present invention includes: An acquisition module 1210, configured to acquire a crustal observation data set of a target area, where the crustal observation data set includes at least two different types of crustal observation data, and at least one type of crustal observation data is global navigation satellite system data.

[0105] A fusion module 1220, configured to input the crustal observation data set into a pre-constructed long short-term memory neural network model based on wavelet transform to obtain a data fusion result output by the long short-term memory neural network model based on wavelet transform, where the data fusion result is the result of fusing at least two different types of crustal observation data in the crustal observation data set.

[0106] A statistical analysis module 1230, configured to perform statistical analysis on the data fusion result to obtain crustal abnormal information of the target area, where the crustal abnormal information is used to characterize crustal deformation changes.

[0107] Optionally, inputting the crustal observation data set into a pre-constructed long short-term memory neural network model based on wavelet transform to obtain a data fusion result output by the long short-term memory neural network model based on wavelet transform includes: Performing wavelet transform processing on at least two different types of crustal observation data in the crustal observation data set to obtain weight information of the crustal observation data set; Using a convolutional neural network to extract features from at least two different types of crustal observation data in the crustal observation data set to obtain feature information of the crustal observation data set; Inputting the weight information of the crustal observation data set and the feature information of the crustal observation data set into a long short-term memory neural network model to obtain a data fusion result output by the long short-term memory neural network model.

[0108] Optionally, performing wavelet transform processing on at least two different types of crustal observation data in the crustal observation data set to obtain weight information of the crustal observation data set includes: Perform wavelet analysis on at least two different types of crustal observation data in the crustal observation dataset to obtain high-frequency feature data and low-frequency feature data. The high-frequency feature data is used to characterize the short-term changes and noise of the crustal observation dataset, and the low-frequency feature data is used to characterize the long-term trend and low-frequency information of the crustal observation dataset; Input the high-frequency feature data and the low-frequency feature data into the first double convolutional layer to obtain a first high-frequency data feature map and a first low-frequency data feature map output by the first double convolutional layer; Input the first high-frequency data feature map and the first low-frequency data feature map into the max pooling layer to obtain a second high-frequency data feature map and a second low-frequency data feature map output by the max pooling layer; Input the second high-frequency data feature map and the second low-frequency data feature map into the first gated convolutional layer to obtain a third high-frequency data feature map and a third low-frequency data feature map output by the first gated convolutional layer; Upsample the third high-frequency data feature map and the third low-frequency data feature map and then input them into the second double convolutional layer to obtain a fourth high-frequency data feature map and a fourth low-frequency data feature map output by the second double convolutional layer; Concatenate the fourth high-frequency data feature map with the first high-frequency data feature map to obtain a first high-frequency concatenated feature map; concatenate the fourth low-frequency data feature map with the first low-frequency data feature map to obtain a first low-frequency concatenated feature map; input the first high-frequency concatenated feature map and the first low-frequency concatenated feature map into the third double convolutional layer to obtain a fifth high-frequency data feature map and a fifth low-frequency data feature map output by the third double convolutional layer; Input the fifth high-frequency data feature map and the fifth low-frequency data feature map into the second gated convolutional layer to obtain a sixth high-frequency data feature map and a sixth low-frequency data feature map output by the second gated convolutional layer; wherein, the sixth high-frequency data feature map and the sixth low-frequency data feature map are used to characterize the weight information of the crustal observation dataset.

[0109] Optionally, the using a convolutional neural network to extract features from at least two different types of crustal observation data in the crustal observation dataset to obtain the feature information of the crustal observation dataset includes: Input at least two different types of crustal observation data in the crustal observation dataset into the fourth double convolutional layer to obtain a first fusion feature map output by the fourth double convolutional layer; Input the first fusion feature map into the first wavelet downsampling layer, and input the output result of the first wavelet downsampling layer into the fifth double convolutional layer to obtain a second fusion feature map output by the fifth double convolutional layer; Input the second fused feature map into the second wavelet downsampling layer, and input the output result of the second wavelet downsampling layer into the sixth double convolutional layer to obtain the third fused feature map output by the sixth double convolutional layer; Input the third fused feature map into the third wavelet downsampling layer, and input the output result of the third wavelet downsampling layer into the seventh double convolutional layer to obtain the fourth fused feature map output by the seventh double convolutional layer; Input the fourth fused feature map into the eighth double convolutional layer to obtain the fifth fused feature map output by the eighth double convolutional layer; Input the fifth fused feature map into the first transposed convolutional layer, concatenate the output result of the first transposed convolutional layer with the third fused feature map in dimension, and input the dimension concatenation result into the ninth double convolutional layer to obtain the sixth fused feature map output by the ninth double convolutional layer; Input the sixth fused feature map into the second transposed convolutional layer, concatenate the output result of the second transposed convolutional layer with the second fused feature map in dimension, and input the dimension concatenation result into the tenth double convolutional layer to obtain the seventh fused feature map output by the tenth double convolutional layer; Input the seventh fused feature map into the third transposed convolutional layer, concatenate the output result of the third transposed convolutional layer with the first fused feature map in dimension, and input the dimension concatenation result into the eleventh double convolutional layer to obtain the eighth fused feature map output by the eleventh double convolutional layer, and the eighth fused feature map is used to represent the feature information of the crust observation data set.

[0110] Optionally, input the weight information of the crust observation data set and the feature information of the crust observation data set into the long short-term memory neural network model to obtain the data fusion result output by the long short-term memory neural network model, including: Merge the weight information of the crust observation data set and the feature information of the crust observation data set to obtain merged data; Input the merged data into the third gated convolutional layer to obtain the merged feature data output by the third gated convolutional layer; Input the merged feature data into the long short-term memory neural network model to obtain the data fusion result output by the long short-term memory neural network model.

[0111] Optionally, perform statistical analysis on the data fusion result to obtain the crust anomaly information of the target area, including: Perform statistical analysis on the data fusion result to obtain the mean and standard deviation of the data fusion result; Perform a normal distribution test on the data fusion result, and perform a normality transformation on the data in the data fusion result that does not conform to the normal distribution to obtain a data fusion result that conforms to the normal distribution; Calculate the mean and standard deviation of the data fusion result using the cumulative sum control chart formula to obtain test data, and convert the test data into a time-type sequence to obtain time-series test data; Perform linear trend fitting on the time-series test data using the least squares method to obtain the linear change trend of the target area; Remove the linear change trend from the time-series test data, and fit the periodic signals in the remaining data to obtain the periodic motion trend of the target area in one or more predetermined time periods; Obtain the crustal anomaly information of the target area based on the linear change trend and periodic motion trend of the target area.

[0112] Optionally, the crustal observation data set at least includes: global navigation satellite system data, interferometric synthetic aperture radar data, and remote sensing image data.

[0113] Optionally, it further includes: a normalization processing module for respectively performing preprocessing and normalization processing on at least two different types of crustal observation data in the crustal observation data set, so that the at least two different types of crustal observation data are consistent in the spatial coordinate system.

[0114] As Figure 13 shown, an electronic device 1300 provided by an embodiment of the present invention includes a memory 1310 and a processor 1320; the memory 1310 is used to store a computer program; the processor 1320 is used to implement the crustal deformation monitoring method as described above when executing the computer program.

[0115] A computer-readable storage medium provided by an embodiment of the present invention has a computer program stored thereon, and when the computer program is executed by a processor, the crustal deformation monitoring method as described above is implemented.

[0116] Now, an electronic device 1300 that can be used as a server or a client of the present invention will be described. It is an example of a hardware device that can be applied to various aspects of the present invention. The electronic device 1300 is intended to represent various forms of digital electronic computer devices, such as, laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device 1300 can also represent various forms of mobile devices, such as, personal digital processors, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described herein and / or claimed.

[0117] Although the present invention is disclosed as above, the scope of protection of the present invention is not limited thereto. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the present invention, and these changes and modifications will all fall within the scope of protection of the present invention.

Claims

1. A crustal deformation monitoring method, characterized in that, Including: Obtaining a crust observation data set of a target area, the crust observation data set including at least two different types of crust observation data, wherein at least one type of crust observation data is global navigation satellite system data; Inputting the crust observation data set into a pre-constructed long short-term memory neural network model based on wavelet transform to obtain a data fusion result output by the long short-term memory neural network model based on wavelet transform, the data fusion result being the result of fusing at least two different types of crust observation data in the crust observation data set; Performing statistical analysis on the data fusion result to obtain crust anomaly information of the target area, the crust anomaly information being used to characterize crustal deformation changes.

2. The crustal deformation monitoring method according to claim 1, characterized in that Inputting the crust observation data set into a pre-constructed long short-term memory neural network model based on wavelet transform to obtain a data fusion result output by the long short-term memory neural network model based on wavelet transform includes: Performing wavelet transform processing on at least two different types of crust observation data in the crust observation data set to obtain weight information of the crust observation data set; Using a convolutional neural network to extract features from at least two different types of crust observation data in the crust observation data set to obtain feature information of the crust observation data set; Inputting the weight information of the crust observation data set and the feature information of the crust observation data set into a long short-term memory neural network model to obtain a data fusion result output by the long short-term memory neural network model.

3. The crustal deformation monitoring method according to claim 2, characterized in that, The performing wavelet transform processing on at least two different types of crust observation data in the crust observation data set to obtain weight information of the crust observation data set includes: Performing wavelet analysis on at least two different types of crust observation data in the crust observation data set to obtain high-frequency feature data and low-frequency feature data, the high-frequency feature data being used to characterize the short-term changes and noise of the crust observation data set, and the low-frequency feature data being used to characterize the long-term trend and low-frequency information of the crust observation data set; Inputting the high-frequency feature data and the low-frequency feature data into a first double convolutional layer to obtain a first high-frequency data feature map and a first low-frequency data feature map output by the first double convolutional layer; Inputting the first high-frequency data feature map and the first low-frequency data feature map into a max pooling layer to obtain a second high-frequency data feature map and a second low-frequency data feature map output by the max pooling layer; Inputting the second high-frequency data feature map and the second low-frequency data feature map into a first gated convolutional layer to obtain a third high-frequency data feature map and a third low-frequency data feature map output by the first gated convolutional layer; Performing upsampling on the third high-frequency data feature map and the third low-frequency data feature map and then inputting them into a second double convolutional layer to obtain a fourth high-frequency data feature map and a fourth low-frequency data feature map output by the second double convolutional layer; Concatenate the fourth high-frequency data feature map and the first high-frequency data feature map to obtain a first high-frequency concatenated feature map; concatenate the fourth low-frequency data feature map and the first low-frequency data feature map to obtain a first low-frequency concatenated feature map; input the first high-frequency concatenated feature map and the first low-frequency concatenated feature map into a third double convolutional layer to obtain a fifth high-frequency data feature map and a fifth low-frequency data feature map output by the third double convolutional layer; Input the fifth high-frequency data feature map and the fifth low-frequency data feature map into a second gated convolutional layer to obtain a sixth high-frequency data feature map and a sixth low-frequency data feature map output by the second gated convolutional layer; wherein, the sixth high-frequency data feature map and the sixth low-frequency data feature map are used to represent the weight information of the crust observation data set.

4. The crustal deformation monitoring method according to claim 2, wherein The feature extraction of at least two different types of crust observation data in the crust observation data set by using a convolutional neural network to obtain the feature information of the crust observation data set includes: Input at least two different types of crust observation data in the crust observation data set into a fourth double convolutional layer to obtain a first fusion feature map output by the fourth double convolutional layer; Input the first fusion feature map into a first wavelet downsampling layer, and input the output result of the first wavelet downsampling layer into a fifth double convolutional layer to obtain a second fusion feature map output by the fifth double convolutional layer; Input the second fusion feature map into a second wavelet downsampling layer, and input the output result of the second wavelet downsampling layer into a sixth double convolutional layer to obtain a third fusion feature map output by the sixth double convolutional layer; Input the third fusion feature map into a third wavelet downsampling layer, and input the output result of the third wavelet downsampling layer into a seventh double convolutional layer to obtain a fourth fusion feature map output by the seventh double convolutional layer; Input the fourth fusion feature map into an eighth double convolutional layer to obtain a fifth fusion feature map output by the eighth double convolutional layer; Input the fifth fusion feature map into a first deconvolutional layer, concatenate the output result of the first deconvolutional layer with the third fusion feature map in dimension, and input the dimension concatenation result into a ninth double convolutional layer to obtain a sixth fusion feature map output by the ninth double convolutional layer; Input the sixth fusion feature map into a second deconvolutional layer, concatenate the output result of the second deconvolutional layer with the second fusion feature map in dimension, and input the dimension concatenation result into a tenth double convolutional layer to obtain a seventh fusion feature map output by the tenth double convolutional layer; Input the seventh fusion feature map into a third deconvolutional layer, concatenate the output result of the third deconvolutional layer with the first fusion feature map in dimension, and input the dimension concatenation result into an eleventh double convolutional layer to obtain an eighth fusion feature map output by the eleventh double convolutional layer, and the eighth fusion feature map is used to represent the feature information of the crust observation data set.

5. The crustal deformation monitoring method according to claim 2, characterized in that, The input of the weight information of the crust observation data set and the feature information of the crust observation data set into a long short-term memory neural network model to obtain the data fusion result output by the long short-term memory neural network model includes: Merge the weight information and the feature information of the crust observation data set to obtain merged data; Input the merged data into the third gated convolutional layer to obtain the merged feature data output by the third gated convolutional layer; Input the merged feature data into the long short-term memory neural network model to obtain the data fusion result output by the long short-term memory neural network model.

6. The crustal deformation monitoring method according to claim 1, characterized in that, The statistical analysis of the data fusion result to obtain the crust anomaly information of the target area includes: Perform statistical analysis on the data fusion result to obtain the mean and standard deviation of the data fusion result; Perform a normal distribution test on the data fusion result, and perform a normality transformation on the data that does not conform to the normal distribution in the data fusion result to obtain a data fusion result with a normal distribution; Use the cumulative sum control chart formula to calculate the mean and standard deviation of the data fusion result to obtain test data, and convert the test data into a time-type sequence to obtain time series test data; Use the least squares method to perform a linear trend fitting on the time series test data to obtain the linear change trend of the target area; Remove the linear change trend from the time series test data, and fit the periodic signal in the data after removal to obtain the periodic motion trend of the target area in one or more predetermined time periods; Obtain the crust anomaly information of the target area according to the linear change trend and the periodic motion trend of the target area.

7. The crustal deformation monitoring method according to any one of claims 1 to 6, characterized in that, The crust observation data set at least includes: global navigation satellite system data, interferometric synthetic aperture radar data, and remote sensing image data.

8. The crustal deformation monitoring method according to any one of claims 1 to 6, characterized in that, Before inputting the crust observation data set into the pre-constructed long short-term memory neural network model based on wavelet transform, it further includes: Perform preprocessing and standardization processing on at least two different types of crust observation data in the crust observation data set so that the at least two different types of crust observation data are consistent in the spatial coordinate system.

9. An electronic device, characterized in that, It includes a memory and a processor; The memory is used to store a computer program; The processor is used to implement the crust deformation monitoring method according to any one of claims 1 to 8 when executing the computer program.

10. A computer-readable storage medium, characterized in that, A computer program is stored on the storage medium, and when the computer program is executed by the processor, the crust deformation monitoring method according to any one of claims 1 to 8 is implemented.

Images