A hydroelectric slope deformation monitoring method, system and electronic device

Abstract

The embodiment of the specification discloses a kind of hydropower slope deformation monitoring method, system and electronic equipment, and the deformation monitoring method therein includes: obtaining optical image stream, radar image sequence, weather data set;Feature extraction is carried out to optical image, and target displacement sequence is obtained;Lens tilt distortion correction is carried out based on target displacement sequence to obtain optical correction data;Permanent scatterer deformation sequence is obtained using SBAS-InSAR technology based on radar image sequence, and space interpolation algorithm is used to generate global deformation map sequence, target displacement correction sequence and weather data set are spatiotemporally aligned with global deformation map sequence, atmospheric delay correction is carried out based on the time distribution matrix obtained, and radar data is calibrated based on optical data, to obtain corrected radar deformation data;Hydropower slope deformation monitoring information is output by fusing optical displacement data and radar deformation data.The precision, timeliness and environmental adaptability of deformation monitoring are enhanced.

Description

Technical Field

[0001] Several embodiments in this specification relate to the field of deformation monitoring technology for hydropower projects, specifically to the optimization of the accuracy of hydropower slope deformation monitoring. Background Technology

[0002] Hydropower projects, as vital national infrastructure, directly impact the safe operation of core structures such as dams and power stations due to the stability of their slopes. Because hydropower projects are often built in mountainous areas with complex geological conditions, slopes are prone to deformation and even instability under the influence of long-term natural forces (such as rainfall and earthquakes) and engineering activities, potentially triggering major disasters such as landslides and collapses. Therefore, continuous and precise deformation monitoring of hydropower slopes is crucial for timely detection of safety hazards, assessment of project risks, and implementation of preventative measures, and is of great significance for ensuring project safety and the safety of people's lives and property.

[0003] Currently, hydropower slope deformation monitoring mainly relies on remote sensing technology, with Synthetic Aperture Radar Interferometry (InSAR) and optical imaging technology being the two most widely used methods. InSAR technology, by processing time-series radar satellite imagery, can monitor surface deformation over a wide area with high precision, and is particularly suitable for extracting deformation rate information across the entire slope area, offering advantages such as all-weather operation and independence from lighting conditions. On the other hand, optical imaging technology, relying on sequence images captured by high-resolution cameras, can achieve fine deformation measurement of local areas through feature point matching and displacement extraction, demonstrating good applicability in deformation identification of key slope locations (such as artificially placed target areas).

[0004] However, the aforementioned technologies still have significant limitations in practical applications. While InSAR technology can provide wide-area deformation information, its results are easily affected by environmental factors such as atmospheric delay and ionospheric disturbances, leading to systematic errors in the deformation data, especially under complex weather conditions in mountainous areas. Although optical imaging technology offers high local measurement accuracy, its effectiveness is heavily dependent on lighting conditions and image quality. It is difficult to obtain effective data in cloudy, foggy, rainy, snowy, or shadowy conditions, and it is prone to feature point loss or mismatch problems. Summary of the Invention

[0005] This specification provides a method, system, and electronic device for monitoring the deformation of hydropower slopes. It solves the problems of feature point drift in optical images and inaccurate deformation data correction in InSAR technology. It also integrates deformation data from the two monitoring methods to achieve deformation monitoring that takes into account both global and local spatiotemporal information, thereby improving the accuracy and reliability of hydropower slope deformation identification.

[0006] The technical solution is as follows:

[0007] Firstly, this specification provides a method for monitoring the deformation of hydropower slopes, comprising the following steps:

[0008] Acquire optical image streams including multiple frames of continuous optical images, radar image sequences including multiple frames of continuous radar synthetic aperture images, and meteorological datasets including various meteorological elements and their spatiotemporal distribution information.

[0009] Feature extraction is performed on each frame of optical image based on the trained convolutional neural network. The temporal attention mechanism is used to obtain the target feature point coordinates corresponding to each frame of optical image, and the target displacement sequence is obtained based on the target feature point coordinates corresponding to each frame of optical image.

[0010] The target displacement correction sequence is obtained by correcting lens tilt distortion based on the target displacement sequence;

[0011] Based on radar image sequences, SBAS-InSAR technology is used to identify permanent scatterers and obtain permanent scatterer deformation sequences.

[0012] Based on the permanent scatterer deformation sequence, a spatial interpolation algorithm is used to generate a global deformation map sequence. The target displacement correction sequence and meteorological dataset are then spatiotemporally aligned with the global deformation map sequence to obtain a time distribution matrix.

[0013] Based on the time distribution matrix, atmospheric delay correction is performed on the target displacement correction sequence and the permanent scatterer deformation sequence using meteorological datasets to obtain the atmospheric delay-corrected target displacement sequence and the atmospheric delay-corrected permanent scatterer deformation sequence. The atmospheric delay-corrected permanent scatterer deformation sequence is then calibrated based on the atmospheric delay-corrected target displacement sequence to obtain the permanent scatterer deformation correction sequence.

[0014] Deformation monitoring information of hydropower slopes is obtained based on target displacement correction sequences and permanent scatterer deformation correction sequences.

[0015] As a preferred embodiment, the step of extracting features from each frame of optical image based on a trained convolutional neural network and obtaining the target feature point coordinates corresponding to each frame of optical image using a temporal attention mechanism includes:

[0016] Use any frame of optical imagery as the current frame of optical imagery:

[0017] In both the previous and current optical images, multi-scale image blocks are extracted centered on the target feature point coordinates in the previous optical image.

[0018] Based on the trained convolutional neural network, feature extraction is performed on the multi-scale image patches corresponding to the previous frame optical image and the current frame optical image respectively to obtain the depth features corresponding to the previous frame optical image and the current frame optical image respectively.

[0019] A temporal attention map is obtained by using a temporal attention mechanism based on the depth features corresponding to the previous and current optical images.

[0020] The coordinates of target feature points corresponding to the current frame of optical image are obtained based on the temporal attention map.

[0021] Repeat the above steps until the coordinates of the target feature points corresponding to each frame of optical image are obtained.

[0022] As a preferred embodiment, obtaining the target feature point coordinates corresponding to the current frame optical image based on the temporal attention map includes:

[0023] Initial sliding coordinates are obtained based on the time attention graph;

[0024] A sliding window of fixed size starts from the initial sliding coordinates in the current frame of optical image and adjusts the position of the sliding window based on the image gradient information of the sliding window to determine the candidate search area;

[0025] The sliding window traverses the candidate search area and matches the target image patch with the target feature point coordinates in the previous frame optical image with the image of the sliding window to obtain multiple high-confidence sliding windows.

[0026] Curve fitting and deformation compensation are performed on the image contours corresponding to each of the multiple high-confidence sliding windows, and the coordinates of the target feature points corresponding to the current frame optical image are determined based on the maximum response principle.

[0027] As a preferred embodiment, obtaining the target displacement sequence based on the target feature point coordinates corresponding to each frame of optical image includes:

[0028] Based on the target feature point coordinates corresponding to each frame of optical image, the horizontal and vertical displacements of coordinates between any two adjacent frames of optical image in the optical image stream are obtained sequentially.

[0029] Arrange the horizontal and vertical coordinate displacements between all adjacent optical images in chronological order to obtain the original displacement sequence that includes the corresponding time information for each frame.

[0030] The original displacement sequence is time-series smoothed to obtain the target displacement sequence.

[0031] As a preferred embodiment, the step of obtaining a target displacement correction sequence by performing lens tilt distortion correction based on the target displacement sequence includes:

[0032] The target feature point coordinates based on a single frame of optical image are matched with the ground using a spatial registration method to obtain the image-ground mapping relationship;

[0033] Geometric offset features that characterize the spatial offset trend of the target region are obtained based on the target displacement sequence corresponding to each of the consecutive frames of optical images, and the image-ground mapping relationship is corrected based on the geometric offset features to obtain the corrected mapping relationship.

[0034] The target displacement correction sequence is obtained based on the target displacement sequence and the correction mapping relationship.

[0035] As a preferred embodiment, the radar image sequence also includes the acquisition time of each radar synthetic aperture image and the corresponding satellite orbit parameters at the time of acquisition; the step of identifying permanent scatterers and obtaining permanent scatterer deformation sequences using SBAS-InSAR technology based on the radar image sequence includes:

[0036] Based on the acquisition time and corresponding satellite orbit parameters of each radar synthetic aperture image in the radar image sequence, all interferometric image pairs that meet the constraints are obtained from the radar image sequence, and the interferogram corresponding to each interferometric image pair is calculated.

[0037] Arrange all interferograms in chronological order to obtain the interferometric phase sequence corresponding to each pixel, and identify permanent scatterers with long-term stable coherence in the radar image sequence;

[0038] The deformation sequence of the permanent scatterer is obtained by least-squares inversion based on the interference phase sequence corresponding to the permanent scatterer.

[0039] As a preferred embodiment, the calibration of the atmospheric delay-corrected permanent scatterer deformation sequence based on the atmospheric delay-corrected target displacement sequence to obtain the permanent scatterer deformation correction sequence includes:

[0040] Registration error was calculated based on atmospheric delay-corrected target displacement sequence and atmospheric delay-corrected permanent scatterer deformation sequence.

[0041] Based on the registration error, the least squares fitting algorithm is used to reverse-calibrate the atmospheric delay-corrected permanent scatterer deformation sequence, and then the permanent scatterer deformation correction sequence is obtained by weighting and fusing it with the atmospheric delay-corrected target displacement sequence according to the data confidence.

[0042] As a preferred embodiment, the acquisition of hydropower slope deformation monitoring information based on the target displacement correction sequence and the permanent scatterer deformation correction sequence includes:

[0043] The target displacement correction sequence and the permanent scatterer deformation correction sequence are spatially aligned and rasterized to obtain gridded data including target displacement and permanent scatterer deformation.

[0044] The fusion weights are obtained based on the spatial coherence and temporal stability of gridded data. The spatial coherence characterizes the consistency of the changing trends of the target displacement and the permanent scattering volume shape at the same moment, and the temporal stability characterizes the degree of fluctuation of the target displacement and the permanent scattering volume shape in the time series.

[0045] The fused deformation value of each grid is obtained based on the gridded data of each grid and its corresponding fusion weight, and the fused deformation value of all grids is used as the deformation monitoring information of the hydropower slope.

[0046] Secondly, this specification provides a hydropower slope deformation monitoring system, including a data acquisition module, an optical measurement module, an optical correction module, a radar measurement module, a spatiotemporal alignment module, a radar correction module, and a data fusion module.

[0047] The data acquisition module acquires an optical image stream including multiple frames of continuous optical images, a radar image sequence including multiple frames of continuous radar synthetic aperture images, and a meteorological dataset including various meteorological elements and their spatiotemporal distribution information.

[0048] The optical measurement module extracts features from each frame of optical image based on a trained convolutional neural network, uses a temporal attention mechanism to obtain the target feature point coordinates corresponding to each frame of optical image, and obtains the target displacement sequence based on the target feature point coordinates corresponding to each frame of optical image.

[0049] The optical correction module obtains a target displacement correction sequence by performing lens tilt distortion correction based on the target displacement sequence.

[0050] The radar measurement module uses SBAS-InSAR technology to identify permanent scatterers and obtain permanent scatterer deformation sequences based on radar image sequences.

[0051] The spatiotemporal alignment module generates a global deformation map sequence based on the permanent scatterer deformation sequence using a spatial interpolation algorithm, and spatiotemporally aligns the target displacement correction sequence and meteorological dataset with the global deformation map sequence to obtain a time distribution matrix;

[0052] The radar correction module, based on the time distribution matrix, uses meteorological datasets to perform atmospheric delay correction on the target displacement correction sequence and the permanent scatterer deformation sequence, respectively, to obtain the atmospheric delay-corrected target displacement sequence and the atmospheric delay-corrected permanent scatterer deformation sequence. Based on the atmospheric delay-corrected target displacement sequence, the atmospheric delay-corrected permanent scatterer deformation sequence is calibrated to obtain the permanent scatterer deformation correction sequence.

[0053] The data fusion module acquires hydropower slope deformation monitoring information based on the target displacement correction sequence and the permanent scatterer deformation correction sequence.

[0054] Thirdly, embodiments of this specification provide an electronic device, including a processor and a memory; the processor is connected to the memory; the memory is used to store executable program code; the processor reads the executable program code stored in the memory to run a program corresponding to the executable program code, so as to perform the steps described in the first aspect of the above embodiments.

[0055] Fourthly, embodiments of this specification provide a computer storage medium storing a plurality of instructions adapted for loading by a processor and executing the steps described in the first aspect of the above embodiments.

[0056] The beneficial effects of the technical solutions provided in some embodiments of this specification include at least the following: by constructing a target extraction method based on dynamic local windows and attention mechanisms, high-precision identification of subtle deformations in hydropower slopes is achieved, improving the accuracy and robustness of image tracking; simultaneously, atmospheric delay correction of permanent scatterers is performed by combining a time distribution matrix, and radar deformation values ​​are calibrated in reverse, effectively achieving physical consistency alignment of multi-source data. High-reliability hydropower slope deformation monitoring information is output jointly from optical image streams and radar image sequences, significantly enhancing the accuracy, timeliness, and environmental adaptability of slope deformation monitoring. Attached Figure Description

[0057] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0058] Figure 1 This is a flowchart illustrating a method for monitoring the deformation of hydroelectric slopes provided in the embodiments of this specification.

[0059] Figure 2 This is a structural schematic diagram of a hydropower slope deformation monitoring system provided in the embodiments of this specification.

[0060] Figure 3 This is a schematic diagram of the structure of an electronic device provided in the embodiments of this specification. Detailed Implementation

[0061] The technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings.

[0062] The terms "first," "second," "third," etc., in the description, claims, and accompanying drawings are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to such processes, methods, products, or apparatus.

[0063] The following description provides examples and does not limit the scope, applicability, or examples set forth in the claims. Changes may be made to the function and arrangement of the described elements without departing from the scope of this specification. Various processes or components may be appropriately omitted, substituted, or added to the examples. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Furthermore, features described with respect to some examples may be combined into other examples.

[0064] Reference Figure 1 As shown, Figure 1 A flowchart illustrating a method for monitoring the deformation of hydropower slopes, provided as an embodiment of this specification, may include at least the following steps:

[0065] Step 102: Obtain an optical image stream including multiple frames of continuous optical images, a radar image sequence including multiple frames of continuous radar synthetic aperture images, and a meteorological dataset including various meteorological elements and their spatiotemporal distribution information.

[0066] It should be noted that the optical image stream comprises multiple frames of continuously captured optical images, recording changes in the visual characteristics of the monitored area. The radar image sequence includes radar synthetic aperture images acquired at multiple time points, used to extract surface deformation information. The meteorological dataset includes spatiotemporal distribution information of key meteorological elements such as temperature, humidity, air pressure, and wind speed during the monitoring period.

[0067] Ensuring the comprehensiveness and diversity of data sources lays the foundation for multi-source collaborative monitoring. Optical data provides high-resolution local information, radar data covers a wide area, and meteorological data assists in error correction, thus enhancing the overall spatiotemporal coverage and environmental adaptability of monitoring.

[0068] Step 104: Based on the trained convolutional neural network, feature extraction is performed on each frame of optical image. The temporal attention mechanism is used to obtain the target feature point coordinates corresponding to each frame of optical image, and the target displacement sequence is obtained based on the target feature point coordinates corresponding to each frame of optical image.

[0069] Interpretive deep feature extraction is performed on each frame of optical imagery using a pre-trained convolutional neural network (CNN) to capture the texture and structural information of the target. A temporal attention mechanism is employed to focus on stable features of the target across consecutive frames, reducing interference from illumination changes and occlusion. The coordinates of target feature points are extracted from the optical image stream, and the displacement sequence is calculated.

[0070] Achieving high-precision identification of local slope deformation, with displacement sequence accuracy down to the sub-pixel level. The temporal attention mechanism enhances robustness to environmental disturbances (such as clouds and shadows), ensuring the continuity and accuracy of feature point tracking and providing reliable input for subsequent correction.

[0071] Illustratively, using manually labeled target regions as supervisory signals, multi-scale feature extraction training was performed using the ResNet-18 architecture. The cross-entropy loss function was used to evaluate the accuracy of feature recognition, and a stochastic gradient descent optimizer (example values: initial learning rate 0.001, momentum 0.9) was used for parameter updates. During training, a dynamic learning rate decay strategy was adopted (example values: decay factor 0.1, patience parameter 5 epochs). Finally, on the validation set, the target localization error was ≤1.5 pixels, thus completing the training of the convolutional neural network.

[0072] In one embodiment of this specification, feature extraction is performed on each frame of optical image based on a trained convolutional neural network, and a temporal attention mechanism is used to obtain the target feature point coordinates corresponding to each frame of optical image, including:

[0073] Use any frame of optical imagery as the current frame of optical imagery:

[0074] In both the previous and current optical images, multi-scale image blocks are extracted centered on the target feature point coordinates in the previous optical image.

[0075] Based on the trained convolutional neural network, feature extraction is performed on the multi-scale image patches corresponding to the previous frame optical image and the current frame optical image respectively to obtain the depth features corresponding to the previous frame optical image and the current frame optical image respectively.

[0076] A temporal attention map is obtained by using a temporal attention mechanism based on the depth features corresponding to the previous and current optical images.

[0077] The coordinates of target feature points corresponding to the current frame of optical image are obtained based on the temporal attention map.

[0078] Repeat the above steps until the coordinates of the target feature points corresponding to each frame of optical image are obtained.

[0079] Interpretively, multiple image patches are extracted at different spatial scales centered on the target feature point coordinates. Each patch covers a different observation range, for example, windows with scales of 11×11, 21×21, and 31×31 are used. Each patch is input into a convolutional neural network, and spatial texture information and structural features are extracted from the multi-scale patches through successive convolutional and pooling layers, forming corresponding depth features. When aligning the extracted depth features with the depth features of the corresponding image patch in the current frame of the optical image stream, the depth features extracted from the multi-scale image patches in the previous frame are arranged in a one-to-one correspondence with the depth features of the image patches at the same location in the current frame. By performing channel-level combination of feature vectors at the same spatial location, a cross-frame feature pair set is formed. This cross-frame feature pair set is then input into a cross-frame attention mechanism to extract the attention weights of each location in the temporal dimension, generating a temporal attention map.

[0080] For illustrative purposes, the first frame is based on manually labeled target coordinates. By extracting the contour and identifying feature points of the manually labeled target region in the known reference frame, and combining the pixel coordinates of the target center point or significant structural points with the image coordinate system, the target feature point coordinates of the previous frame of the optical image in the optical image stream are obtained.

[0081] In one embodiment of this specification, obtaining the target feature point coordinates corresponding to the current frame optical image based on a temporal attention map includes:

[0082] Initial sliding coordinates are obtained based on the time attention graph;

[0083] A sliding window of fixed size starts from the initial sliding coordinates in the current frame of optical image and adjusts the position of the sliding window based on the image gradient information of the sliding window to determine the candidate search area;

[0084] The sliding window traverses the candidate search area and matches the target image patch with the target feature point coordinates in the previous frame optical image with the image of the sliding window to obtain multiple high-confidence sliding windows.

[0085] Curve fitting and deformation compensation are performed on the image contours corresponding to each of the multiple high-confidence sliding windows, and the coordinates of the target feature points corresponding to the current frame optical image are determined based on the maximum response principle.

[0086] Interpretively, regions with high response values ​​in the temporal attention map are selected as initial sliding region reference positions. Within the current frame of the optical image stream, a fixed window size, e.g., 15×15 pixels, is used to progressively slide around this reference position horizontally and vertically. During each sliding operation, image gradient information within the current window region is extracted, and the degree of local variation in image grayscale values ​​is analyzed. The center position of the sliding window is automatically fine-tuned based on the direction and magnitude of the gradient change. This adjustment enhances the overlap between the sliding window and the high-response regions of the temporal attention map, dynamically adjusting the sliding window path. After initial sliding and fine-tuning, all sliding window regions with significant response intensity, significant image gradient, and continuity are summarized as candidate search regions in the current frame's optical image stream.

[0087] Furthermore, fine-grained sliding is performed within each candidate search region with a small step size, such as 1 pixel. Image patches covered by the sliding window are extracted sequentially and matched with the target image patches located in the previous frame's optical image stream. The matching process uses a lightweight matching algorithm. By calculating the similarity between the sliding window image patch and the reference image patch in dimensions such as texture, grayscale distribution, or feature vector, a confidence score is generated for each sliding window. The confidence scores of all sliding windows are compared with a confidence threshold, and only sliding windows with confidence scores higher than the confidence threshold are retained to form a high-confidence sliding window set.

[0088] For each high-confidence sliding window, edge pixels are extracted and contour extraction is performed. Contour boundary point sets are obtained through polygon approximation or image gradient direction constraints. Spline curves or polynomial functions are applied to the contour boundary point sets for smooth curve fitting, making the contour representation more continuous and accurate. After curve fitting, sub-pixel-level offset estimation is performed by combining pixel-level grayscale changes or matching score distribution. The positions of the curve control points are then finely adjusted at the sub-pixel level to perform sub-pixel-level deformation compensation. Among all fitted and compensated high-confidence sliding windows, based on the maximum response principle—comparing the response intensity, confidence, or texture matching quality within each window region—the spatial location corresponding to the maximum value is selected, and the target feature point coordinates are extracted and output.

[0089] Illustratively, by statistically analyzing the confidence distribution of lightweight matching algorithms under correct and incorrect matching conditions in existing sample data, a threshold value outside the overlapping area of ​​these distributions is selected as the benchmark. This threshold is then empirically adjusted based on the requirements of matching accuracy and computational efficiency for different monitoring scenarios, thus setting a confidence threshold for selecting high-confidence sliding windows. The maximum response principle refers to selecting the position with the highest response intensity from multiple candidate regions as the basis for the final feature point output. Response intensity is obtained by analyzing indicators such as matching score, image gradient magnitude, or feature similarity within the sliding window region.

[0090] In one embodiment of this specification, obtaining the target displacement sequence based on the target feature point coordinates corresponding to each frame of optical image includes:

[0091] Based on the target feature point coordinates corresponding to each frame of optical image, the horizontal and vertical displacements of coordinates between any two adjacent frames of optical image in the optical image stream are obtained sequentially.

[0092] Arrange the horizontal and vertical coordinate displacements between all adjacent optical images in chronological order to obtain the original displacement sequence that includes the corresponding time information for each frame.

[0093] The original displacement sequence is time-series smoothed to obtain the target displacement sequence.

[0094] Interpretatively, the image coordinates of corresponding target feature points in two adjacent frames of the optical image stream are recorded to obtain the pixel positions of the target feature point coordinates in the previous frame and the target feature point coordinates in the current frame. The horizontal coordinates of the target feature point in the current frame are compared with the corresponding horizontal coordinates of the feature point in the previous frame to determine the magnitude of their positional change in the horizontal direction. Similarly, the vertical coordinates of the target feature point in the current frame are compared with the corresponding vertical coordinates of the feature point in the previous frame to determine the magnitude of their positional change in the vertical direction, thereby obtaining the pixel displacement components in the horizontal and vertical directions.

[0095] To illustrate, based on the acquisition time of each frame of optical image in the optical image stream, the horizontal and vertical displacements of the corresponding frames are arranged in chronological order. The displacement components in the two directions are sequentially combined to form a displacement sequence containing temporal information. To ensure temporal consistency, a corresponding timestamp is assigned to each position in the displacement sequence according to the specific acquisition time of each frame, marking the time point corresponding to each frame. Finally, an original displacement sequence containing horizontal displacement, vertical displacement, and time information is generated.

[0096] Furthermore, the displacement sequence is traversed chronologically. For the displacement value of each frame, combined with the displacement information of adjacent frames, a smoothed displacement value is calculated to remove occasional abnormal fluctuations and enhance the continuity and stability of the sequence. After smoothing, the target displacement sequence containing the temporally smoothed horizontal and vertical displacements is output.

[0097] Step 106: Correct the lens tilt distortion based on the target displacement sequence to obtain the target displacement correction sequence.

[0098] Interpretive lens tilt distortion correction is applied to the target displacement sequence to eliminate imaging geometric errors caused by camera tilt and lens distortion. This ensures that the displacement data is consistent with the actual geographic coordinates. This step improves the geometric accuracy of the optical data and provides a precise benchmark for multi-source data fusion.

[0099] In one embodiment of this specification, lens tilt distortion correction is performed based on the target displacement sequence to obtain a target displacement correction sequence, including:

[0100] The target feature point coordinates based on a single frame of optical image are matched with the ground using a spatial registration method to obtain the image-ground mapping relationship;

[0101] Geometric offset features that characterize the spatial offset trend of the target region are obtained based on the target displacement sequence corresponding to each of the consecutive frames of optical images, and the image-ground mapping relationship is corrected based on the geometric offset features to obtain the corrected mapping relationship.

[0102] The target displacement correction sequence is obtained based on the target displacement sequence and the correction mapping relationship.

[0103] Interpretive, based on the pixel coordinates of target feature points in the optical image stream, combined with the coordinates of ground reference points provided by field measurements or existing geographic information systems, a spatial registration method is used to match the target feature points in the optical image stream with their corresponding ground reference locations. During the matching process, by comparing the spatial distribution of target feature points with the geographical distribution of ground reference points, the correspondence rules of the mapping relationship are determined, and the transformation relationship between the two-dimensional pixel coordinate system of the optical image stream and the three-dimensional spatial coordinate system of the ground is established. This achieves a one-to-one correspondence mapping between the target feature points in the optical image stream and the actual ground locations, thus constructing a complete image-ground mapping relationship.

[0104] Furthermore, based on the established image-ground mapping relationship, the corresponding target region contours in the continuous frame optical image stream are sequentially located. Image registration technology is used to achieve continuous contour tracking, ensuring the spatial consistency of the target region position in each frame. During contour tracking, the geometric position changes of contour points between consecutive frames are calculated, and gradient information of contour deformation is extracted to reflect the local and overall spatial offset trends of the target region. Using the extracted geometric offset features, the image-ground mapping relationship is adjusted and optimized. By introducing a deformation gradient compensation term, a more accurate corrected mapping relationship is constructed. Based on the constructed corrected mapping relationship, for the displacement data of each frame in the original target displacement sequence, the spatial transformation rules in the mapping relationship are used to adjust the displacement values ​​in terms of position and scale, eliminating deviations caused by image-ground mapping errors and deformation. By applying the correction operation frame by frame, the corrected horizontal and vertical displacement data for each frame are obtained. Following the chronological order of acquisition time, all corrected displacement data are sequentially combined to form a complete and continuous target displacement correction sequence.

[0105] It's important to note that when using a fixed optical camera to acquire an optical image stream, the error primarily stems from the geometric distortion of the lens itself. In this case, once the image-ground mapping is established, the model itself is relatively stable because the camera remains stationary, and the external parameters (rotation, translation) in the mapping are constants. However, when using a moving optical camera to acquire an optical image stream, the error arises from the combined effect of lens distortion and changes in the camera's external parameters (position, attitude). The camera's own movement generates significant, inter-frame varying perspective distortion. Assume that the initial mapping compresses the coordinates of the image edges due to lens distortion. By analyzing the trajectory deviation of the target moving from the image center to the edge across multiple frames, correcting the mapping can quantify and compensate for this compression effect. Therefore, correcting the displacement sequence with the optimized mapping can eliminate systematic geometric errors and preserve the true deformation signal.

[0106] This ensures that optical displacement data and radar deformation data can be fused under a unified geographic reference system, avoiding registration errors caused by inconsistencies in coordinate systems.

[0107] Step 108: Based on the radar image sequence, use SBAS-InSAR technology to identify permanent scatterers and obtain the deformation sequence of permanent scatterers.

[0108] Interpretive analysis utilizes SBAS-InSAR (Small Baseline Subset InSAR) technology to interferometrically process temporal radar imagery, identifying highly coherent permanent scatterers and extracting the deformation time series of each permanent scatterer to reflect the cumulative deformation of the slope surface. This provides global deformation information, with permanent scatterers serving as stable benchmarks to effectively suppress decorrelation noise. SBAS-InSAR technology ensures high spatial coverage and high accuracy in deformation monitoring, compensating for the limitations of optical data.

[0109] For illustrative purposes, a permanent scatterer is a single pixel of a permanent scatterer in a time-series radar image sequence that exhibits stable reflected signals and maintains a high coherence value over a long period. These pixels typically correspond to stable features on the slope surface (such as exposed rock, corners of artificial structures, or fixed targets).

[0110] In one embodiment of this specification, the radar image sequence further includes the acquisition time of each radar synthetic aperture image and the corresponding satellite orbit parameters at the time of acquisition; based on the radar image sequence, SBAS-InSAR technology is used to identify permanent scatterers and obtain permanent scatterer deformation sequences, including:

[0111] Based on the acquisition time and corresponding satellite orbit parameters of each radar synthetic aperture image in the radar image sequence, all interferometric image pairs that meet the constraints are obtained from the radar image sequence, and the interferogram corresponding to each interferometric image pair is calculated.

[0112] Arrange all interferograms in chronological order to obtain the interferometric phase sequence corresponding to each pixel, and identify permanent scatterers with long-term stable coherence in the radar image sequence;

[0113] The deformation sequence of the permanent scatterer is obtained by least-squares inversion based on the interference phase sequence corresponding to the permanent scatterer.

[0114] Interpretive methods are employed, extracting the acquisition time and satellite orbit parameters of each image from the radar image sequence, and calculating the temporal and spatial baselines between any two images. By pairwise combining any two images from the radar image sequence, the temporal and spatial baseline values ​​for each pair are calculated. The expression for the temporal baseline value is as follows:

[0115]

[0116] In the formula, It is the temporal baseline value between the i-th reference image and the j-th paired image. It is the acquisition time of the i-th reference image. is the acquisition time between the j-th paired images, where i is the index of the reference image and j is the index of the paired image.

[0117] The spatial value expression is:

[0118]

[0119] In the formula, It is the first The spatial baseline value between the j-th reference image and the j-th paired image. These are the three-dimensional coordinates of the satellite's orbital position in the i-th reference image. is the three-dimensional coordinate of the satellite orbit position of the j-th paired image.

[0120] For illustrative purposes, the constraints require that the time interval between the two images cannot be too long, and the shooting angles (satellite positions) of the two images cannot differ too much. The temporal and spatial baseline values ​​corresponding to all combinations are compared with the temporal and spatial baseline thresholds, respectively. Only image pairs that simultaneously satisfy the condition of having a temporal baseline not exceeding the temporal baseline threshold and a spatial baseline not exceeding the spatial baseline threshold are retained as qualified interferometric image pairs.

[0121] From a massive amount of temporal radar imagery, image combinations with the highest temporal and spatial matching are selected—these are called interferometric image pairs. Further, these selected image pairs are matched, and a registration method is used to precisely align the two radar images spatially, achieving pixel-level correspondence and generating interferograms for subsequent interferometric processing. An interferogram is a set of phase difference images; each interferogram corresponds to an interferometric image pair and encodes information such as surface deformation, topography, and atmospheric delay. The interferogram is the foundation for subsequent deformation analysis; it reveals minute surface deformations by comparing the phase differences between two images. Improper pairing, like comparing blurry or misaligned photographs, can lead to errors in deformation measurement.

[0122] It should be noted that, based on the relationship between the radar image acquisition time interval and the rate of surface deformation, the aim is to ensure that the selected image pairs have appropriate time intervals to avoid distortion of deformation information due to excessive time spans. The time baseline threshold is typically determined by analyzing the deformation periodic characteristics of the monitoring area and the stability of the time-series data. The spatial baseline threshold is set based on satellite orbital parameters and radar imaging geometry to limit the spectral differences between image pairs, ensuring that the interferometric images have high coherence and low geometric distortion.

[0123] Furthermore, radar images acquired at different times are precisely registered to eliminate geometric distortions between images. By determining the pixel-level correspondence of each image pair, the spatial alignment of all interferometric images is ensured. Common registration algorithms, such as those based on feature point matching or gray-level similarity matching, are used during the registration process. After image registration, interference fringes between the two images are calculated using interferometric processing techniques. These fringes reflect the surface deformation between the images. The interference phase sequence is extracted from the interference fringes, representing the surface deformation at different time points between the two images. By analyzing the spatial regions of the images, high-coherence areas are extracted. These high-coherence areas show smaller phase sequence changes, reflecting the deformation of a stable surface. Using the extracted interference phase sequence, the coherence of multi-temporal images is statistically analyzed to calculate the multi-temporal coherence coefficient between each image pair. Using the multi-temporal coherence coefficient, the coherence of each pixel in the interferogram is analyzed, and regions with long-term stable coherence and coherence coefficients higher than the coherence coefficient threshold are selected to determine the location of permanent scatterers on the hydropower slope.

[0124] For illustration, assume there are N radar images, generating M interferograms (M being the number of image pairs satisfying baseline constraints). For a specific pixel, its interferometric phase sequence is a vector of length M. Each element in the vector represents the phase value of that pixel in different interferograms over time. The multi-temporal coherence coefficient is calculated by statistically analyzing the phase consistency of the target pixels across multiple interferograms, with higher values ​​indicating better stability. Based on historical monitoring experience and statistical analysis, and considering the coherence coefficient distribution characteristics and noise levels in the actual monitoring environment, a critical value that effectively distinguishes between permanent and non-permanent scatterers is selected to obtain the coherence coefficient threshold.

[0125] The transformation from interferograms to interferometric phase sequences involves a reorganization of data dimensions, changing from "all pixels in each image" to "all images per pixel." The multi-temporal coherence coefficient is a statistic that assesses pixel stability, obtained by calculating the normalized cross-coherence among all image pairs.

[0126] Furthermore, for permanent scatterers, a linear deformation inversion equation is established based on the interferometric phase sequence. The equation is then solved using the least squares method to calculate the linear deformation rate of the permanent scatterer. The expression for the linear deformation rate is as follows:

[0127]

[0128] In the formula, It is a linear deformation rate value. It is the transpose of the coefficient matrix of the inversion equation. It is the coefficient matrix of the inversion equation. It is the phase observation vector of the selected permanent scatterer in the interferometric phase sequence.

[0129] The cumulative linear deformation rate of the permanent scatterer at each time point is the deformation, thus obtaining the deformation sequence of the permanent scatterer.

[0130] Step 110: Generate a global deformation map sequence based on the permanent scatterer deformation sequence using a spatial interpolation algorithm, and align the target displacement correction sequence and meteorological dataset with the global deformation map sequence in time and space to obtain a time distribution matrix.

[0131] Interpretive methods first construct a set of control points based on the spatial coordinates of permanent scatterers and linear deformation rate values. Then, a spatial interpolation algorithm is used to estimate the deformation rate between control points within the monitoring area, generating a continuous deformation rate distribution. The spatial interpolation algorithm calculates the spatial distance weights between each control point and combines them with the deformation rate values ​​to predict the deformation rate of unobserved areas, thereby obtaining a global deformation rate map covering the entire hydropower slope area, and finally obtaining a global deformation map sequence.

[0132] Furthermore, the target displacement correction sequence and the global deformation map sequence are projected onto the same spatial reference frame to ensure consistency in their spatial coordinates. Based on the grid resolution of the global deformation map, the target displacement correction sequence is spatially rasterized, mapping the target displacement point data to the corresponding grid cells. During rasterization, the grid cell to which the target displacement point data belongs is determined based on its spatial coordinates, and a unique grid index is generated for each grid cell. An attribute table is constructed to record all corresponding target displacement values ​​and related information within the grid cell. Combined with the acquisition time information, the displacement data at different time points are summarized and integrated according to the timestamps to form a preliminary spatiotemporal record containing spatial grid indexes and time information. Then, key meteorological elements during the monitoring period are extracted from the meteorological dataset, including temperature, humidity, air pressure, and wind speed. The data for each element is matched according to the corresponding time point in the preliminary spatiotemporal record. For each grid cell and timestamp in the spatiotemporal record, spatial interpolation and temporal interpolation methods are used to supplement the meteorological element values ​​to the corresponding time, achieving a continuous spatiotemporal distribution of meteorological elements. After interpolation is completed, the spatiotemporal records of meteorological elements are checked for completeness according to the chronological order. When missing values ​​are found, interpolation is used to fill them in to ensure the continuity and completeness of the records. Based on the time index and raster number, the interpolated and checked spatiotemporal records of meteorological elements are reorganized into a row-column matrix to form a structured time distribution matrix with time as the row index and spatial raster as the column index.

[0133] The optically corrected displacement sequence, meteorological data, and radar-generated global deformation map sequence are spatiotemporally aligned to achieve multi-source data fusion. The temporal distribution matrix, as a structured data cube, integrates deformation, displacement, and meteorological information, ensuring consistency of multi-source data across the spatiotemporal dimensions. This provides a unified framework for subsequent atmospheric correction and calibration, enhancing data collaboration efficiency.

[0134] Step 112: Based on the time distribution matrix, atmospheric delay correction is performed on the target displacement correction sequence and the permanent scatterer deformation sequence using the meteorological dataset to obtain the atmospheric delay-corrected target displacement sequence and the atmospheric delay-corrected permanent scatterer deformation sequence. The atmospheric delay-corrected permanent scatterer deformation sequence is then calibrated based on the atmospheric delay-corrected target displacement sequence to obtain the permanent scatterer deformation correction sequence.

[0135] Interpretively, based on meteorological element data in the time distribution matrix, for the permanent scatterer deformation sequence, meteorological factors such as temperature, humidity, air pressure, and wind speed are mapped hourly to the corresponding deformation observation time. Utilizing the relationship between meteorological factors and electromagnetic wave propagation delay, combined with a physical atmospheric delay model, the impact of atmospheric delay error on the permanent scatterer deformation observation at each time step is estimated. Based on the estimation results, error correction is applied to the observed values ​​in the permanent scatterer deformation sequence. The corrected deformation sequence undergoes phase unwrapping to eliminate phase ambiguity and jumps, achieving the recovery of a continuous phase sequence. Based on the unwrapped phase sequence, time series analysis is conducted to extract the cumulative deformation corresponding to phase changes, ultimately forming an atmospheric delay-corrected permanent scatterer deformation sequence. This eliminates interference from atmospheric turbulence and humidity changes on radar signals, improves the physical accuracy of deformation data, and provides clean input for subsequent calibration. Furthermore, the radar phase data is converted into quantifiable deformation variables, providing a benchmark for registration with optical data.

[0136] Similarly, based on the corresponding time-space meteorological records in the time distribution matrix, the influence of meteorological factors on the displacement value of each point in the target displacement correction sequence is estimated point by point. The displacement data is then corrected in time series in combination with the changes in meteorological elements. The amount of influence of the optical target displacement on the meteorological environment (such as thermal expansion and contraction caused by temperature) is calculated point by point to eliminate the interference of atmospheric delay on displacement observation.

[0137] In one embodiment of this specification, the atmospheric delay-corrected permanent scatterer deformation sequence is calibrated based on the atmospheric delay-corrected target displacement sequence to obtain the permanent scatterer deformation correction sequence, including:

[0138] Registration error was calculated based on atmospheric delay-corrected target displacement sequence and atmospheric delay-corrected permanent scatterer deformation sequence.

[0139] Based on the registration error, the least squares fitting algorithm is used to reverse-calibrate the atmospheric delay-corrected permanent scatterer deformation sequence, and then the permanent scatterer deformation correction sequence is obtained by weighting and fusing it with the atmospheric delay-corrected target displacement sequence according to the data confidence.

[0140] Interpretive, the atmospheric delay-corrected target displacement sequence and the atmospheric delay-corrected permanent scatterer deformation sequence are projected onto a unified spatiotemporal coordinate system to ensure the correspondence between the two in spatial location and time point. Based on the spatial registration method, the deformation observations in the atmospheric delay-corrected target displacement sequence and the atmospheric delay-corrected permanent scatterer deformation sequence are matched point by point, and the differences between the two sets of deformation data at the same location and time point are analyzed to obtain the registration error.

[0141] A dataset quantifying the differences between radar and optical data based on registration errors is used to assess data consistency. Residual errors among multi-source data are revealed, providing a basis for reverse calibration and ensuring fusion accuracy. Based on the registration errors, a least-squares fitting algorithm is used to reverse-calibrate the atmospheric delay-corrected permanent scatterer deformation sequence, adjusting the deviations in the atmospheric delay-corrected permanent scatterer deformation values ​​to generate a corrected atmospheric delay-corrected permanent scatterer deformation sequence. The atmospheric delay-corrected target displacement sequence and the corrected atmospheric delay-corrected permanent scatterer deformation sequence are then fused point-by-point in the same spatial coordinates according to confidence weights. During the fusion process, weights are allocated based on the confidence level of each data point to improve the accuracy and stability of the fusion result. Finally, the fused permanent scatterer deformation correction sequence is output, solving the problem of inaccurate radar deformation data correction.

[0142] Step 114: Obtain hydropower slope deformation monitoring information based on target displacement correction sequence and permanent scatterer deformation correction sequence.

[0143] Interpretive, based on corrected optical and radar sequences, it generates hydropower slope deformation monitoring information. It provides comprehensive and accurate slope deformation monitoring results, including localized minor deformations and overall trends.

[0144] Furthermore, based on the corrected optical and radar sequences, a continuous three-dimensional deformation field can be reconstructed, and combined with topographic elevation information, a three-dimensional terrain model covering the monitoring area can be constructed. Based on the deformation values ​​in the three-dimensional terrain model, the deformation degree of each grid cell is color-mapped according to a linear gradient color band, mapping smaller deformation values ​​to cooler colors and larger deformation values ​​to warmer colors. This color gradient visually reflects the deformation distribution characteristics. Three-dimensional rendering technology is used to visualize the color-mapped three-dimensional terrain model, generating a three-dimensional deformation heat map of the hydropower slope with spatial deformation information. The three-dimensional heat map supports rapid risk identification (such as landslide early warning), providing a scientific basis for engineering safety assessment and decision-making, and significantly improving the timeliness and environmental adaptability of monitoring.

[0145] In one embodiment of this specification, hydropower slope deformation monitoring information is obtained based on a target displacement correction sequence and a permanent scatterer deformation correction sequence, including:

[0146] The target displacement correction sequence and the permanent scatterer deformation correction sequence are spatially aligned and rasterized to obtain gridded data including target displacement and permanent scatterer deformation.

[0147] The fusion weights are obtained based on the spatial coherence and temporal stability of gridded data. Spatial coherence characterizes the consistency of the changing trends of target displacement and permanent scattering volume shape at the same moment, while temporal stability characterizes the degree of fluctuation of target displacement and permanent scattering volume shape in the time series.

[0148] The fused deformation value of each grid is obtained based on the gridded data of each grid and its corresponding fusion weight, and the fused deformation value of all grids is used as the deformation monitoring information of the hydropower slope.

[0149] Interpretively, the target displacement correction sequence and the permanent scatterer deformation correction sequence are projected onto the same spatial reference frame to ensure their spatial consistency. Based on the grid resolution, the projected target displacement correction sequence and permanent scatterer deformation correction sequence are rasterized, dividing the continuous spatial information into regular grid cells. Each grid cell contains the target displacement and permanent scatterer deformation within the corresponding spatial range, forming structured gridded data.

[0150] In illustrative terms, grid resolution is typically chosen to balance data accuracy and computational efficiency, such as by setting a reasonable grid cell size based on the size of the monitored target features or the sensor resolution.

[0151] Furthermore, by evaluating the consistency of the changing trends of target displacement and permanent scattering body deformation at the same moment, if the directions and amplitudes of their changes are consistent, they are considered to have high coherence, and a coherence index is obtained. The fluctuation of the deformation values ​​of target displacement and permanent scattering body deformation in the time series is also evaluated. If the deformation changes of target displacement and permanent scattering body deformation are stable and without abrupt changes or abnormal jumps over a continuous time period, it indicates high temporal stability, and a temporal stability index is obtained. The fusion weight of the grid cells is obtained by combining the coherence index and the temporal stability index. The fusion weight reflects the reliability of each data source. Based on the fusion weight, a weighted average of the target displacement and permanent scattering body deformation is performed to obtain the fused deformation value of each grid cell. This results in the final hydropower slope deformation monitoring information.

[0152] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0153] Please refer to the following. Figure 2 , Figure 2 A schematic diagram of the structure of a hydropower slope deformation monitoring system provided in an embodiment of this specification is shown.

[0154] The deformation monitoring system 200 includes a data acquisition module 201, an optical measurement module 202, an optical correction module 203, a radar measurement module 204, a spatiotemporal alignment module 205, a radar correction module 206, and a data fusion module 207.

[0155] The data acquisition module 201 acquires an optical image stream including multiple frames of continuous optical images, a radar image sequence including multiple frames of continuous radar synthetic aperture images, and a meteorological dataset including various meteorological elements and their spatiotemporal distribution information.

[0156] The optical measurement module 202 extracts features from each frame of optical image based on a trained convolutional neural network, uses a temporal attention mechanism to obtain the target feature point coordinates corresponding to each frame of optical image, and obtains the target displacement sequence based on the target feature point coordinates corresponding to each frame of optical image.

[0157] Optical correction modules 20 and 3 perform lens tilt distortion correction based on target displacement sequence to obtain target displacement correction sequence;

[0158] Radar measurement module 204 uses SBAS-InSAR technology to identify permanent scatterers and obtain permanent scatterer deformation sequences based on radar image sequences.

[0159] The spatiotemporal alignment module 205 generates a global deformation map sequence based on the permanent scatterer deformation sequence using a spatial interpolation algorithm, and spatiotemporally aligns the target displacement correction sequence and meteorological dataset with the global deformation map sequence to obtain a time distribution matrix;

[0160] The radar correction module 206, based on the time distribution matrix, uses meteorological datasets to perform atmospheric delay correction on the target displacement correction sequence and the permanent scatterer deformation sequence, respectively, to obtain the atmospheric delay-corrected target displacement sequence and the atmospheric delay-corrected permanent scatterer deformation sequence. Based on the atmospheric delay-corrected target displacement sequence, the atmospheric delay-corrected permanent scatterer deformation sequence is calibrated to obtain the permanent scatterer deformation correction sequence.

[0161] The data fusion module 207 acquires hydropower slope deformation monitoring information based on the target displacement correction sequence and the permanent scatterer deformation correction sequence.

[0162] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the deformation monitoring system embodiments are basically similar to the deformation monitoring method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the deformation monitoring method embodiments.

[0163] Please see Figure 3The diagram shown is a structural schematic of an electronic device provided in an embodiment of this specification.

[0164] like Figure 3 As shown, the electronic device 300 may include at least one processor 301, at least one network interface 304, a user interface 303, a memory 305, and at least one communication bus 302.

[0165] The communication bus 302 can be used to realize the connection and communication of the above components.

[0166] The user interface 303 may include buttons, and the optional user interface may also include a standard wired interface or a wireless interface.

[0167] The network interface 304 may include, but is not limited to, Bluetooth modules, NFC modules, Wi-Fi modules, etc.

[0168] The processor 301 may include one or more processing cores. The processor 301 connects to various parts within the electronic device 300 using various interfaces and lines. It executes various functions and processes data by running or executing instructions, programs, code sets, or instruction sets stored in the memory 305, and by calling data stored in the memory 305. Optionally, the processor 301 may be implemented using at least one hardware form selected from DSP, FPGA, and PLC. The processor 301 may integrate one or more of the following: CPU, GPU, and modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the content required for display; and the modem handles wireless communication. It is understood that the modem may also be implemented as a separate chip, without being integrated into the processor 301.

[0169] The memory 305 may include RAM or ROM. Optionally, the memory 305 may include a non-transitory computer-readable medium. The memory 305 may be used to store instructions, programs, code, code sets, or instruction sets. The memory 305 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the above-described method embodiments, etc.; the data storage area may store data involved in the above-described method embodiments, etc. Optionally, the memory 305 may also be at least one storage device located remotely from the aforementioned processor 301. As a computer storage medium, the memory 305 may include an operating system, a network communication module, a user interface module, and a deformation monitoring application. The processor 301 may be used to call the deformation monitoring application stored in the memory 305 and execute the steps of the deformation monitoring method mentioned in the foregoing embodiments.

[0170] This specification also provides a computer-readable storage medium storing instructions that, when executed on a computer or processor, cause the computer or processor to perform one or more steps in the above-described deformation monitoring method embodiments. If the constituent modules of the above-described electronic device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium.

[0171] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to the embodiments of this specification is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in or transmitted through a computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, Digital Subscriber Line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., Digital Versatile Discs (DVDs)), or semiconductor media (e.g., Solid State Disks (SSDs)).

[0172] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. The aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks. Unless otherwise specified, the technical features of this embodiment and its implementation can be combined arbitrarily.

[0173] The above embodiments are merely preferred embodiments described in this specification and are not intended to limit the scope of this specification. Any modifications and improvements made by those skilled in the art to the technical solutions of this specification without departing from the spirit of this specification should fall within the protection scope defined by the claims of this specification.

Claims

1. A method for monitoring the deformation of hydropower slopes, characterized in that, Includes the following steps: Acquire optical image streams including multiple frames of continuous optical images, radar image sequences including multiple frames of continuous radar synthetic aperture images, and meteorological datasets including various meteorological elements and their spatiotemporal distribution information. Feature extraction is performed on each frame of optical image based on the trained convolutional neural network. The temporal attention mechanism is used to obtain the target feature point coordinates corresponding to each frame of optical image, and the target displacement sequence is obtained based on the target feature point coordinates corresponding to each frame of optical image. The target displacement correction sequence is obtained by correcting lens tilt distortion based on the target displacement sequence; Based on radar image sequences, SBAS-InSAR technology is used to identify permanent scatterers and obtain permanent scatterer deformation sequences. Based on the permanent scatterer deformation sequence, a spatial interpolation algorithm is used to generate a global deformation map sequence. The target displacement correction sequence and meteorological dataset are then spatiotemporally aligned with the global deformation map sequence to obtain a time distribution matrix. Based on the time distribution matrix, atmospheric delay correction is performed on the target displacement correction sequence and the permanent scatterer deformation sequence using meteorological datasets to obtain the atmospheric delay-corrected target displacement sequence and the atmospheric delay-corrected permanent scatterer deformation sequence. The atmospheric delay-corrected permanent scatterer deformation sequence is then calibrated based on the atmospheric delay-corrected target displacement sequence to obtain the permanent scatterer deformation correction sequence. Deformation monitoring information of hydropower slopes is obtained based on target displacement correction sequences and permanent scatterer deformation correction sequences.

2. The method for monitoring the deformation of hydropower slopes according to claim 1, characterized in that, The process involves extracting features from each frame of optical image based on a pre-trained convolutional neural network, and employing a temporal attention mechanism to obtain the coordinates of the target feature points corresponding to each frame of optical image, including: Use any frame of optical imagery as the current frame of optical imagery: In both the previous and current optical images, multi-scale image blocks are extracted centered on the target feature point coordinates in the previous optical image. Based on the trained convolutional neural network, feature extraction is performed on the multi-scale image patches corresponding to the previous frame optical image and the current frame optical image respectively to obtain the depth features corresponding to the previous frame optical image and the current frame optical image respectively. A temporal attention map is obtained by using a temporal attention mechanism based on the depth features corresponding to the previous and current optical images. The coordinates of target feature points corresponding to the current frame of optical image are obtained based on the temporal attention map. Repeat the above steps until the coordinates of the target feature points corresponding to each frame of optical image are obtained.

3. The method for monitoring the deformation of hydropower slopes according to claim 2, characterized in that, The step of obtaining the target feature point coordinates corresponding to the current frame optical image based on the time attention map includes: Initial sliding coordinates are obtained based on the time attention graph; A sliding window of fixed size starts from the initial sliding coordinates in the current frame of optical image and adjusts the position of the sliding window based on the image gradient information of the sliding window to determine the candidate search area; The sliding window traverses the candidate search area and matches the target image patch with the target feature point coordinates in the previous frame optical image with the image of the sliding window to obtain multiple high-confidence sliding windows. Curve fitting and deformation compensation are performed on the image contours corresponding to each of the multiple high-confidence sliding windows, and the coordinates of the target feature points corresponding to the current frame optical image are determined based on the maximum response principle.

4. The method for monitoring the deformation of hydropower slopes according to claim 3, characterized in that, The process of obtaining the target displacement sequence based on the target feature point coordinates corresponding to each frame of optical image includes: Based on the target feature point coordinates corresponding to each frame of optical image, the horizontal and vertical displacements of coordinates between any two adjacent frames of optical image in the optical image stream are obtained sequentially. Arrange the horizontal and vertical coordinate displacements between all adjacent optical images in chronological order to obtain the original displacement sequence that includes the corresponding time information for each frame. The original displacement sequence is time-series smoothed to obtain the target displacement sequence.

5. The method for monitoring the deformation of hydropower slopes according to claim 4, characterized in that, The target displacement correction sequence obtained by performing lens tilt distortion correction based on the target displacement sequence includes: The target feature point coordinates based on a single frame of optical image are matched with the ground using a spatial registration method to obtain the image-ground mapping relationship; Geometric offset features that characterize the spatial offset trend of the target region are obtained based on the target displacement sequence corresponding to each of the consecutive frames of optical images, and the image-ground mapping relationship is corrected based on the geometric offset features to obtain the corrected mapping relationship. The target displacement correction sequence is obtained based on the target displacement sequence and the correction mapping relationship.

6. The method for monitoring the deformation of hydropower slopes according to claim 1, characterized in that, The radar image sequence also includes the acquisition time of each radar synthetic aperture image and the corresponding satellite orbit parameters at the time of acquisition; the identification of permanent scatterers and acquisition of permanent scatterer deformation sequences based on the radar image sequence using SBAS-InSAR technology includes: Based on the acquisition time and corresponding satellite orbit parameters of each radar synthetic aperture image in the radar image sequence, all interferometric image pairs that meet the constraints are obtained from the radar image sequence, and the interferogram corresponding to each interferometric image pair is calculated. Arrange all interferograms in chronological order to obtain the interferometric phase sequence corresponding to each pixel, and identify permanent scatterers with long-term stable coherence in the radar image sequence; The deformation sequence of the permanent scatterer is obtained by least-squares inversion based on the interference phase sequence corresponding to the permanent scatterer.

7. The method for monitoring the deformation of hydropower slopes according to claim 6, characterized in that, The calibration of the atmospheric delay-corrected permanent scatterer deformation sequence based on the atmospheric delay-corrected target displacement sequence yields the permanent scatterer deformation correction sequence, including: Registration error was calculated based on atmospheric delay-corrected target displacement sequence and atmospheric delay-corrected permanent scatterer deformation sequence. Based on the registration error, the least squares fitting algorithm is used to reverse-calibrate the atmospheric delay-corrected permanent scatterer deformation sequence, and then the permanent scatterer deformation correction sequence is obtained by weighting and fusing it with the atmospheric delay-corrected target displacement sequence according to the data confidence.

8. The method for monitoring the deformation of hydropower slopes according to claim 1, characterized in that, The acquisition of hydropower slope deformation monitoring information based on target displacement correction sequence and permanent scatterer deformation correction sequence includes: The target displacement correction sequence and the permanent scatterer deformation correction sequence are spatially aligned and rasterized to obtain gridded data including target displacement and permanent scatterer deformation. The fusion weights are obtained based on the spatial coherence and temporal stability of gridded data. The spatial coherence characterizes the consistency of the changing trends of the target displacement and the permanent scattering volume shape at the same moment, and the temporal stability characterizes the degree of fluctuation of the target displacement and the permanent scattering volume shape in the time series. The fused deformation value of each grid is obtained based on the gridded data of each grid and its corresponding fusion weight, and the fused deformation value of all grids is used as the deformation monitoring information of the hydropower slope.

9. A hydropower slope deformation monitoring system, characterized in that, It includes a data acquisition module, an optical measurement module, an optical correction module, a radar measurement module, a spatiotemporal alignment module, a radar correction module, and a data fusion module. The data acquisition module acquires an optical image stream including multiple frames of continuous optical images, a radar image sequence including multiple frames of continuous radar synthetic aperture images, and a meteorological dataset including various meteorological elements and their spatiotemporal distribution information. The optical measurement module extracts features from each frame of optical image based on a trained convolutional neural network, uses a temporal attention mechanism to obtain the target feature point coordinates corresponding to each frame of optical image, and obtains the target displacement sequence based on the target feature point coordinates corresponding to each frame of optical image. The optical correction module obtains a target displacement correction sequence by performing lens tilt distortion correction based on the target displacement sequence. The radar measurement module uses SBAS-InSAR technology to identify permanent scatterers and obtain permanent scatterer deformation sequences based on radar image sequences. The spatiotemporal alignment module generates a global deformation map sequence based on the permanent scatterer deformation sequence using a spatial interpolation algorithm, and spatiotemporally aligns the target displacement correction sequence and meteorological dataset with the global deformation map sequence to obtain a time distribution matrix; The radar correction module, based on the time distribution matrix, uses meteorological datasets to perform atmospheric delay correction on the target displacement correction sequence and the permanent scatterer deformation sequence, respectively, to obtain the atmospheric delay-corrected target displacement sequence and the atmospheric delay-corrected permanent scatterer deformation sequence. Based on the atmospheric delay-corrected target displacement sequence, the atmospheric delay-corrected permanent scatterer deformation sequence is calibrated to obtain the permanent scatterer deformation correction sequence. The data fusion module acquires hydropower slope deformation monitoring information based on the target displacement correction sequence and the permanent scatterer deformation correction sequence.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1-8.

Images