CN-121995381-A - InSAR time sequence image-based intelligent interpretation method and system for slope deformation

CN121995381ACN 121995381 ACN121995381 ACN 121995381ACN-121995381-A

Abstract

The invention discloses a slope deformation intelligent interpretation method and system based on InSAR time sequence images, belonging to the technical field of geological disaster monitoring, wherein the method comprises the following steps: acquiring multi-time-phase SAR images and generating a deformation rate field and a time sequence; the deformation signal quality of the low-coherence region is improved by adopting a space-time double-domain self-adaptive enhancement algorithm; extracting space-time characteristics of deformation based on a multi-scale space-time diagram attention-transducer coupling network; fusing deformation characteristics, geological topography factors and environment triggering factors to perform multi-modal risk intelligent assessment; according to the system, the parameters of the front-end modules are reversely optimized according to the evaluation confidence coefficient, closed-loop collaborative optimization is realized, and the collaborative improvement of deformation signal quality, characteristic extraction effect and risk evaluation accuracy is realized through the deep coupling and closed-loop feedback among the modules, so that a high-precision and high-reliability technical means is provided for large-scale slope monitoring and geological disaster prevention.

Inventors

ZHAO JING
FAN JIANGTAO
LU YONGHUI
QU JINGYU
LIU HU
ZHANG CHEN

Assignees

西安航空学院

Dates

Publication Date: 20260508
Application Date: 20260408

Claims (10)

1. The intelligent slope deformation interpretation method based on the InSAR time sequence image is characterized by comprising the following steps of: Acquiring multi-temporal SAR images covering a target monitoring area, and processing the multi-temporal SAR images through a time sequence interferometry technology to generate a deformation rate field and an accumulated deformation time sequence, wherein the deformation rate field comprises an average deformation rate value and space coordinate information of each pixel, and the deformation time sequence comprises deformation variables at each moment; The deformation signal self-adaptive enhancement step is that for InSAR deformation data in a low coherence mountain area, a space-time double-domain self-adaptive enhancement algorithm is adopted to enhance the deformation signal, a space domain self-adaptive enhancement weight is calculated in a space domain according to the coherence and deformation gradient of a neighborhood pixel, a time domain self-adaptive enhancement weight is calculated in a time domain according to the smoothness and the change trend of a time sequence deformation curve, the space domain enhancement weight and the time domain enhancement weight are fused to generate a comprehensive enhancement weight, the raw deformation data is subjected to weighted filtering by utilizing the comprehensive enhancement weight, and an enhanced deformation rate field and a deformation time sequence are generated; Constructing a side slope unit space adjacency graph based on the enhanced deformation data, adopting a multi-scale time-space graph attention-transducer coupling network to extract deformation characteristics, wherein a graph attention network layer captures the space dependency relationship between adjacent side slope units, a multi-head time sequence transducer layer captures the long-range time sequence dependency relationship of a deformation time sequence, and fusing deformation characteristics of different time-space scales through a multi-scale characteristic pyramid to output multi-scale time-space characteristic tensors; Establishing a multi-mode risk intelligent evaluation network, fusing the multi-scale space-time characteristic tensor, the geological topography factor and the environment trigger factor, recognizing the deformation mode type and predicting the deformation development trend through the deep semantic understanding network, calculating the risk level of the hidden danger of the side slope according to the deformation rate, the deformation acceleration, the space aggregation characteristic and the deformation mode type, and outputting the confidence index of the risk evaluation.
2. The intelligent interpretation method of slope deformation based on InSAR time sequence images as set forth in claim 1, wherein in the deformation signal adaptive enhancement step, a space-time dual-domain adaptive enhancement algorithm comprises: calculating the spatial domain enhancement weight of each pixel, wherein the weight is positively correlated with the average coherence coefficient of the neighborhood pixels and negatively correlated with the variance of the neighborhood deformation gradient; Calculating the time domain enhancement weight of each pixel, wherein the weight is positively correlated with the linear fitting goodness of a time sequence deformation curve and is negatively correlated with a deformation mutation detection index; Generating comprehensive enhancement weights from the spatial domain enhancement weights and the time domain enhancement weights through a self-adaptive fusion algorithm, dynamically adjusting fusion coefficients according to coherence levels of local areas, increasing the spatial domain weight duty ratio when the local average coherence coefficient is lower than a coherence coefficient threshold value, and increasing the time domain weight duty ratio when the local average coherence coefficient is higher than or equal to the coherence coefficient threshold value.
3. The intelligent interpretation method of slope deformation based on InSAR time sequence images as set forth in claim 1, wherein in the multi-scale spatio-temporal feature extraction step, the multi-scale spatio-temporal graph attention-transducer coupling network comprises: The graph attention network layer is used for constructing a dynamic attention graph according to the spatial adjacent relation and the deformation correlation among slope units, calculating the spatial dependency strength among the slope units in different neighborhood ranges through a multi-head attention mechanism and outputting a spatial feature matrix; The multi-head time sequence Transformer layer is used for encoding the deformation time sequence of each slope unit, capturing deformation modes of different time scales through a multi-head self-attention mechanism, including short-term fluctuation, medium-term trend and long-term evolution characteristics, and outputting a time sequence characteristic matrix; And the multi-scale feature pyramid module is used for respectively extracting deformation features on different time-space scales, realizing multi-scale feature fusion through bidirectional feature transfer from bottom to top and from top to bottom, and generating multi-scale time-space feature tensors containing fine-granularity local deformation and coarse-granularity integral trend.
4. The intelligent interpretation method of slope deformation based on InSAR time sequence images as set forth in claim 1, wherein in the step of multi-modal risk intelligent assessment, the multi-modal risk assessment network comprises: The deformation mode identification branch is used for classifying deformation modes based on a deep convolutional neural network and carrying out multi-scale space-time characteristic tensors, and identifying linear creep deformation, acceleration deformation, periodic deformation and mutation deformation; The geological topography factor fusion branch extracts the geological topography factors of gradient, slope direction, elevation, lithology and geological structure, encodes the geological topography factors into high-dimensional feature vectors through the feature embedding layer, and performs depth fusion with the deformation features; the environmental trigger factor fusion branch is used for acquiring environmental trigger factors of rainfall, reservoir water level change, earthquake activities and human engineering activities, analyzing time lag correlation between the environmental factors and deformation acceleration, and identifying a deformation trigger mechanism; And the risk level prediction module is used for integrating the deformation mode type, the geological terrain susceptibility and the environment triggering sensitivity, outputting the risk level by adopting an integrated learning algorithm, and calculating a confidence index based on the consistency of a prediction result and the historical verification accuracy.
5. The intelligent interpretation method for slope deformation based on InSAR time sequence images according to claim 1, further comprising a closed loop parameter optimization step, wherein a parameter optimization mechanism is triggered according to a risk assessment result and a confidence index, when the confidence is lower than a confidence lower limit threshold, the enhancement weight calculation parameter in the deformation signal self-adaptive enhancement step and the attention weight in the multi-scale space-time feature extraction step are reversely adjusted, and the confidence of risk assessment is improved to be higher than the confidence upper limit threshold through iterative optimization; A step of automatically extracting the hidden danger areas of the side slope, wherein the step of automatically extracting the space boundaries of the high-risk and medium-risk areas is based on the risk assessment result, and a continuous hidden danger area is formed by connecting discrete risk pixels through morphological operation; A deformation prediction step, for the identified hidden danger area, predicting deformation quantity of a future period by utilizing a time sequence transducer network, and giving a deformation development trend by combining predicted data of environmental factors; And dynamically adjusting the early warning threshold value according to historical monitoring data and prediction accuracy statistics, and dynamically adjusting the early warning threshold value of each risk level to realize self-adaptive early warning.
6. The intelligent interpretation method of slope deformation based on InSAR time sequence images as set forth in claim 1, wherein in the step of InSAR image data acquisition and preprocessing: Adopting a permanent scatterer interferometry technology or a distributed scatterer interferometry technology to perform joint processing on long-time sequence SAR images to obtain high-density deformation monitoring points; performing atmospheric delay correction and orbit error correction on the interference phase, and separating the atmospheric phase and the deformation phase by adopting a space-time filtering method; The multi-temporal SAR image comprises a track lifting image and a track descending image, sight line deformation of the track lifting image and the track descending image is decomposed into vertical deformation and horizontal deformation, and three-dimensional accuracy of deformation monitoring is improved by utilizing a track lifting fusion technology.
7. The intelligent interpretation method of slope deformation based on InSAR time sequence images as set forth in claim 3, wherein the construction of the slope unit space adjacency graph comprises the following steps: Dividing a target monitoring area into a plurality of slope units based on a terrain segmentation algorithm, wherein each slope unit has relatively homogeneous terrain and geological conditions; Establishing connecting edges between nodes according to the spatial adjacent relation of the side slope units, wherein the weights of the edges are jointly determined by Euclidean distances and deformation correlation coefficients between the side slope units; And constructing a dynamic graph structure, dynamically updating the edge weights according to the deformation evolution process, and strengthening the connection strength between slope units with high deformation correlation.
8. The intelligent interpretation method of slope deformation based on InSAR time sequence images as claimed in claim 4, wherein the risk level prediction module adopts the following judgment criteria: when the deformation rate is greater than a first rate threshold value, the first rate threshold value is statistically determined according to the slope geological condition and the historical deformation data, the deformation acceleration is greater than an acceleration threshold value, the acceleration threshold value is statistically determined according to the change rate of the deformation time sequence, and the deformation mode is identified as an acceleration deformation type, the deformation mode is judged to be a high risk level; when the deformation rate is between the first rate threshold and the second rate threshold, the deformation presents linear creep or periodical deformation characteristics, and an environment trigger factor acts, the risk level is judged; And when the deformation rate is smaller than the second rate threshold, the deformation curve is stable, and no obvious environment triggering factor exists, the risk level is judged to be low.
9. The intelligent interpretation method of slope deformation based on InSAR timing images as set forth in claim 5, wherein the closed loop parameter optimization step includes: The confidence coefficient monitoring mechanism monitors confidence coefficient indexes of the risk assessment in real time, and when the confidence coefficient is lower than a confidence coefficient lower limit threshold value, a parameter optimization flow is triggered; gradient back propagation optimization, calculating parameter gradients according to a confidence coefficient loss function, and reversely adjusting a spatial domain weight coefficient and a time domain weight coefficient in a deformation signal self-adaptive enhancement step, and the number of attention heads and the number of feature pyramid layers in a multi-scale space-time feature extraction step; And (3) carrying out iteration convergence judgment, re-executing deformation signal enhancement, feature extraction and risk assessment processes after parameter adjustment, terminating optimization if the confidence coefficient is increased to be above the confidence coefficient upper limit threshold value, and otherwise continuing iteration until the maximum iteration times are reached.
10. The intelligent side slope deformation interpretation system based on the InSAR time sequence image is used for realizing the intelligent side slope deformation interpretation method based on the InSAR time sequence image as set forth in claim 9, and is characterized by comprising the following steps: The InSAR image data acquisition and preprocessing module is used for acquiring multi-phase SAR images and generating a deformation rate field and a deformation time sequence; The deformation signal self-adaptive enhancement module is used for carrying out space-time double-domain self-adaptive enhancement processing on deformation data of a low coherence region; The multi-scale space-time feature extraction module is used for extracting deformed space-time features through a multi-scale space-time diagram attention-transducer coupling network; The multi-mode risk intelligent evaluation module is used for fusing deformation characteristics, geological topography factors and environment triggering factors to conduct risk level prediction and confidence calculation; The closed-loop parameter optimization module is used for reversely optimizing key parameters of the front-end module according to the risk assessment confidence level to realize closed-loop collaborative optimization; the output of the deformation signal self-adaptive enhancement module is used as the input of the multi-scale space-time feature extraction module, the output of the multi-scale space-time feature extraction module is used as the input of the multi-modal risk intelligent assessment module, the confidence level output of the multi-modal risk intelligent assessment module triggers the closed-loop parameter optimization module, and the parameter adjustment signal of the closed-loop parameter optimization module is fed back to the deformation signal self-adaptive enhancement module and the multi-scale space-time feature extraction module.

Description

InSAR time sequence image-based intelligent interpretation method and system for slope deformation Technical Field The invention relates to the technical field of geological disaster monitoring and prevention, in particular to a slope deformation intelligent interpretation method and system based on a synthetic aperture radar interferometry time sequence image, and belongs to the fields of remote sensing monitoring, deep learning and geological disaster early warning intersection. Background Slope instability is a common type of geological disaster in mountainous areas and reservoir areas, and forms a serious threat to infrastructure safety, lives, properties and ecological environments of people. Traditional slope monitoring mainly relies on ground measurement equipment such as total stations, GPS (global positioning system), inclinometers and the like, and although the accuracy of the method is high, the method has the problems of difficult arrangement, limited coverage range, high cost and the like, and the real-time monitoring requirement of a large-scale slope group is difficult to meet. The synthetic aperture radar interferometry technology can extract earth surface deformation information from satellite remote sensing images, has the advantages of large-scale coverage, all-weather observation, centimeter-level precision and the like, and becomes an important means for monitoring geological disasters. In particular to a permanent scatterer interferometry technology and a distributed scatterer interferometry technology, and a high-density deformation monitoring point and a continuous deformation time sequence can be obtained through the combined processing of long-time-sequence SAR images. However, the slope area is usually located in a mountain area covered by vegetation, so that serious space-time decoherence problem exists, the InSAR deformation measurement accuracy is reduced, deformation signals are submerged in noise, and potential landslide hidden danger is difficult to accurately identify. The existing InSAR deformation interpretation method mainly has the following defects: chinese invention CN114119643a discloses an automatic extraction method of deformation region boundary based on ALPHASHAPE algorithm, which screens PS points with abnormal deformation rate through statistical analysis, constructs Delaunay triangulation network, and then adopts ALPHASHAPE algorithm to extract deformation region boundary. The method is characterized in that firstly, the method is used for screening abnormal points only based on a single deformation rate index, evolution characteristics and spatial distribution modes of deformation time sequences are not considered, so that the recognition capability of complex deformation types such as acceleration deformation, periodic deformation and the like is insufficient, secondly, the method adopts a fixed statistical threshold to judge abnormal points, cannot adapt to the differences of deformation characteristics under different geological conditions and environmental factors, a large number of false recognition or omission can be easily generated in a low-coherence mountain area, thirdly, the method is used for extracting the spatial boundaries of the deformation areas, the deep interpretation of deformation mechanisms, risk grades and development trends is lacked, hierarchical early warning and risk management decisions of landslide hidden danger are difficult to support, and finally, the method is a unidirectional processing flow, and lacks a feedback optimization mechanism, so that processing parameters cannot be dynamically adjusted according to the reliability of recognition results, and stability and accuracy are insufficient in complex scenes. In the prior art, a deep learning method is also studied to interpret InSAR deformation, but most of the methods take deformation data as a single input characteristic, and space-time structural characteristics of deformation signals are not fully mined. In low coherence mountainous areas, inSAR deformation data contains a large amount of noise and errors, and direct application of a deep learning model easily causes poor overfitting and generalization capability. In addition, the deformation of the side slope is comprehensively influenced by various factors such as geological structures, topography, rainfall, reservoir water level change and the like, the risk of instability of the side slope is difficult to accurately evaluate by simply relying on deformation data, and multi-source heterogeneous data are required to be fused for comprehensive analysis. The existing method still has obvious defects in the aspects of multi-mode data fusion, long time sequence deformation mode identification, space dependency modeling and the like, and limits the application effect of the InSAR technology in large-scale slope monitoring. Therefore, development of a slope deformation intelligent interpretation method capable of adaptively