CN-122020431-A - Hidden danger identification method, device, equipment and medium for slope geological disasters

CN122020431ACN 122020431 ACN122020431 ACN 122020431ACN-122020431-A

Abstract

The invention discloses a hidden danger identification method, device, equipment and medium for slope geological disasters. The method comprises the steps of obtaining multi-source remote sensing data of a target slope area, conducting layered pretreatment on the multi-source remote sensing data, conducting fusion wave band self-adaptive screening on the multi-source remote sensing data after time-space registration treatment, conducting feature enhancement treatment on the fusion wave bands after multi-scale decomposition, obtaining terrain perception information of the target slope area, constructing a heterogeneous feature matrix based on the multi-source remote sensing data and the terrain perception information after the feature enhancement treatment, inputting the heterogeneous feature matrix into a deep learning model, extracting a slope hidden danger feature vector, inputting the slope hidden danger feature vector into a pre-trained hidden danger detection model, locating a hidden danger area of a geological disaster, and identifying the type of the geological disaster of the located hidden danger area to obtain a hidden danger identification result of the slope type geological disaster. The invention improves the recognition precision of the hidden danger of the slope under the complex terrain.

Inventors

WANG YAN
WANG LEI
WU HUILONG
MA XIAOQIN
ZHANG GUIXIANG
LU XUELI

Assignees

深圳市勘察研究院有限公司

Dates

Publication Date: 20260512
Application Date: 20260413

Claims (10)

1. The hidden danger identification method for the slope geological disasters is characterized by comprising the following steps of: Acquiring multi-source remote sensing data of a target slope area, carrying out hierarchical pretreatment on the multi-source remote sensing data, and carrying out space-time registration treatment on the multi-source remote sensing data subjected to the hierarchical pretreatment; Performing adaptive screening of fusion wave bands on the multisource remote sensing data subjected to time-space registration processing, performing multiscale decomposition on the screened fusion wave bands through an image decomposition algorithm, and performing feature enhancement processing on the fusion wave bands subjected to multiscale decomposition; acquiring terrain perception information of the target slope region, constructing a heterogeneous feature matrix based on the multi-source remote sensing data and the terrain perception information after feature enhancement processing, inputting the heterogeneous feature matrix into a deep learning model, and extracting a slope hidden danger feature vector; And inputting the slope hidden danger feature vector into a pre-trained hidden danger detection model, positioning a hidden danger area of the geological disaster, and identifying the type of the geological disaster of the positioned hidden danger area to obtain a hidden danger identification result of the slope type geological disaster.
2. The method for identifying hidden danger of a slope-like geological disaster according to claim 1, wherein the steps of obtaining multi-source remote sensing data of a target slope region, performing hierarchical preprocessing on the multi-source remote sensing data, and performing space-time registration processing on the multi-source remote sensing data after the hierarchical preprocessing comprise the steps of: Acquiring optical remote sensing data, synthetic aperture radar interferometry data and laser radar data of a target slope area; Carrying out speckle reduction processing on the synthetic aperture radar interferometry data by adopting an enhanced direction smoothing filtering algorithm to obtain speckle-reduced synthetic aperture radar interferometry data; Performing radiation calibration processing on the optical remote sensing data, and performing atmospheric correction on the optical remote sensing data subjected to radiation calibration based on a radiation transmission model to obtain optical remote sensing data subjected to atmospheric correction; denoising the laser radar data based on an elevation anomaly detection algorithm to obtain denoised laser radar data; Taking a digital elevation model generated by the denoised laser radar data as a reference, and performing secondary polynomial transformation on the optical remote sensing data after the atmospheric correction and the synthetic aperture radar interferometry data after the speckle reduction to obtain preliminary geometric correction multi-source remote sensing data; And registering the multi-source remote sensing data after the preliminary geometric correction by adopting a feature matching model to obtain registered multi-source remote sensing data.
3. The method for identifying hidden danger of slope geological disasters according to claim 2, wherein the adaptive screening of fusion bands is performed on the multisource remote sensing data after time-space registration processing, multiscale decomposition is performed on the screened fusion bands through an image decomposition algorithm, and feature enhancement processing is performed on the fusion bands after multiscale decomposition, and the method comprises the following steps: Performing adaptive screening of fusion wave bands on the multisource remote sensing data subjected to time-space registration processing, and performing multiscale joint decomposition on the screened fusion wave bands by adopting downsampled contourlet-free transformation to obtain low-frequency subband coefficients and high-frequency subband coefficients of each wave band; weighting and fusing the low-frequency sub-bands based on the region correlation coefficient, and performing self-adaptive fusion on the high-frequency sub-bands based on the local direction energy; Carrying out downsampled contourlet inverse transformation processing on the fused low-frequency sub-band and the fused high-frequency sub-band, and reconstructing a primary fused image of the multi-source remote sensing data; And constructing a characteristic enhancement weight matrix based on gradient information in the laser radar data and deformation rate information in the synthetic aperture radar interferometry data, and carrying out self-adaptive enhancement on the primary fusion image based on the characteristic enhancement weight matrix to obtain multi-source remote sensing data after characteristic enhancement.
4. The method for identifying hidden danger of a slope-like geological disaster according to claim 3, wherein the obtaining of the terrain awareness information of the target slope region, the constructing of a heterogeneous feature matrix based on the multi-source remote sensing data and the terrain awareness information after the feature enhancement processing, and the inputting of the heterogeneous feature matrix into a deep learning model, the extracting of the hidden danger feature vector of the slope, comprises: Carrying out heterogeneous dimension mapping on the multi-source remote sensing data with the enhanced characteristics, and fusing the mapped fusion characteristics with the terrain perception information to construct a heterogeneous characteristic matrix; Feature screening is carried out on the heterogeneous feature matrix through a terrain self-adaptive attention gating mechanism, and a screened feature matrix is obtained; Constructing a terrain perception position coding vector, and fusing the terrain perception position coding vector with the screened feature matrix to obtain a fused feature vector; constructing a transducer model comprising a multi-scale window attention module and a topographic feature enhancement module, and inputting the fusion feature vector into the constructed transducer model to extract multi-level coding features; And performing cross-layer weighted fusion processing on the multi-layer coding features, performing dimension reduction on the multi-layer coding features subjected to the cross-layer weighted fusion processing through attention pooling, and outputting the slope hidden danger feature vector.
5. The method for identifying hidden danger of a slope type geological disaster according to claim 1, wherein the step of inputting the slope hidden danger feature vector into a pre-trained hidden danger detection model, locating a hidden danger area of the geological disaster, and identifying a geological disaster type of the located hidden danger area to obtain a hidden danger identification result of the slope type geological disaster comprises the following steps: generating a candidate hidden danger area by adopting a self-adaptive anchor point generation strategy based on the slope hidden danger feature vector; extracting the feature vector of each candidate hidden danger area through a target detection algorithm, inputting the feature vector of each candidate hidden danger area into a pre-trained hidden danger detection model, and respectively calculating the category probability and boundary coordinate offset of each candidate hidden danger area; Determining an initial hidden danger area and the confidence coefficient of the initial hidden danger area of the target slope area based on the class probability and the boundary coordinate offset of each candidate hidden danger area; And screening the initial hidden danger area by adopting a flexible non-maximum suppression algorithm, and correcting the confidence coefficient of the screened hidden danger area to obtain the hidden danger area of the target slope area and the corresponding geological disaster type.
6. The method for identifying hidden danger of a slope-type geological disaster according to claim 5, wherein the step of inputting the slope hidden danger feature vector into a pre-trained hidden danger detection model, locating a hidden danger area of the geological disaster, and identifying a geological disaster type of the located hidden danger area, and after obtaining a hidden danger identification result of the slope-type geological disaster, the method further comprises: Constructing a multi-dimensional accuracy verification index containing classification accuracy, space accuracy and feature consistency of the hidden danger areas, quantitatively evaluating hidden danger identification results of the slope geological disasters, and determining hidden danger areas of which the accuracy verification index of each dimension does not reach a preset accuracy threshold; Screening samples to be supplemented and marked by adopting an uncertainty sampling strategy for hidden danger areas of which the precision verification indexes of all dimensions do not reach a preset precision threshold; Adding the labeling sample to be supplemented to a training set of the hidden danger detection model, and adjusting the class weight of the loss function; And performing iterative optimization on the hidden danger detection model with the class weights of the loss function adjusted until the precision indexes of each dimension meet the preset precision threshold value, and obtaining an optimized slope type geological disaster hidden danger identification result.
7. The method for identifying hidden danger of a slope-type geological disaster according to claim 6, wherein the hidden danger detection model after the class weight of the loss function is adjusted is iteratively optimized until the precision index of each dimension meets a preset precision threshold, and after the optimized hidden danger identification result of the slope-type geological disaster is obtained, the method further comprises: extracting multi-dimensional core parameters related to the topography and deformation of the hidden trouble area and the environment based on the optimized hidden trouble identification result; Constructing a hidden danger susceptibility evaluation index according to the multi-dimensional core parameters, and generating a hidden danger distribution diagram and a risk level demarcation diagram of a target slope area; Fusing a digital elevation model generated by the denoised laser radar data with a risk level demarcation graph to construct a three-dimensional visualization model of the target slope area; And generating a standard visual result for identifying hidden dangers in the target slope area based on the hidden danger distribution map, the risk level demarcation map and the three-dimensional visual model.
8. Hidden danger recognition device of slope class geological disasters, characterized by, include: the acquisition module is used for acquiring multi-source remote sensing data of a target slope area, carrying out layered pretreatment on the multi-source remote sensing data, and carrying out space-time registration treatment on the multi-source remote sensing data after the layered pretreatment; the decomposition module is used for carrying out self-adaptive screening on fusion bands of the multisource remote sensing data after the time-space registration processing, carrying out multiscale decomposition on the screened fusion bands through an image decomposition algorithm, and carrying out characteristic enhancement processing on the fusion bands after the multiscale decomposition; the construction module is used for acquiring the terrain perception information of the target slope area, constructing a heterogeneous feature matrix based on the multi-source remote sensing data and the terrain perception information after the feature enhancement processing, inputting the heterogeneous feature matrix into a deep learning model, and extracting a slope hidden danger feature vector; the positioning module is used for inputting the slope hidden danger feature vector into a pre-trained hidden danger detection model, positioning hidden danger areas of geological disasters, and identifying the type of the geological disasters of the positioned hidden danger areas to obtain hidden danger identification results of slope type geological disasters.
9. The utility model provides a hidden danger identification equipment of slope class geological disasters which characterized in that includes memory and treater, wherein: the memory is used for storing a computer program; the processor is configured to read the computer program in the memory, and execute the steps of the method for identifying hidden danger of a slope geological disaster according to any one of claims 1 to 7.
10. A computer readable storage medium, characterized in that a readable computer program is stored thereon, which program, when being executed by a processor, implements the steps of a method for identifying potential hazards of a slope-like geological disaster according to any of claims 1 to 7.

Description

Hidden danger identification method, device, equipment and medium for slope geological disasters Technical Field The invention relates to the technical field of geological disaster identification, in particular to a hidden danger identification method, device, equipment and medium for slope type geological disasters. Background The slope type geological disaster hidden danger points are wide in distribution, strong in concealment and high in disaster risk, and accurate and efficient hidden danger identification is the basis of geological disaster prevention and control work. The traditional slope type geological disaster hidden danger identification is based on manual field investigation, is low in efficiency and high in cost, is limited by factors such as terrain complexity and climate conditions, is difficult to realize large-scale and fine hidden danger investigation, and has personnel potential safety hazards in dangerous area investigation. With the development of remote sensing technology, single remote sensing data sources such as optical remote sensing data, synthetic aperture radar interferometry data and laser radar data are gradually applied to geological disaster identification, but the single data sources have inherent defects that the optical remote sensing is easily influenced by weather and vegetation shielding and is difficult to capture topographic deformation information, the synthetic aperture radar interferometry data has speckle noise, the identification precision of microminiature hidden danger is insufficient, the laser radar data can acquire high-precision elevation data, but the spectral information is missing, and the environmental characteristics of the hidden danger cannot be comprehensively represented. Meanwhile, the existing multi-source remote sensing data fusion method is mostly of a simple pixel level, space-time isomerism of different data sources is not fully considered, and the fused image has the problems of characteristic blurring, spectrum distortion and the like, so that the fine recognition requirement is difficult to meet. In addition, in the slope hidden danger feature extraction, the traditional machine learning algorithm has weak capability of capturing heterogeneous features under complex terrains, has the problems of missed detection of small-scale hidden danger, inaccurate boundary positioning, large type discrimination error and the like, and lacks a perfect precision verification and model iteration optimization system, so that the reliability of a recognition result is insufficient. Therefore, how to improve the identification accuracy of hidden dangers of slopes under complex terrains is a technical problem to be solved urgently. Disclosure of Invention The invention provides a hidden danger identification method, device, equipment and medium for slope geological disasters, which are used for solving the technical problem of how to improve the identification precision of slope hidden dangers under complex terrains. In a first aspect, the present invention provides a method for identifying hidden danger of a slope geological disaster, including: Acquiring multi-source remote sensing data of a target slope area, carrying out hierarchical pretreatment on the multi-source remote sensing data, and carrying out space-time registration treatment on the multi-source remote sensing data subjected to the hierarchical pretreatment; Performing adaptive screening of fusion wave bands on the multisource remote sensing data subjected to time-space registration processing, performing multiscale decomposition on the screened fusion wave bands through an image decomposition algorithm, and performing feature enhancement processing on the fusion wave bands subjected to multiscale decomposition; acquiring terrain perception information of the target slope region, constructing a heterogeneous feature matrix based on the multi-source remote sensing data and the terrain perception information after feature enhancement processing, inputting the heterogeneous feature matrix into a deep learning model, and extracting a slope hidden danger feature vector; And inputting the slope hidden danger feature vector into a pre-trained hidden danger detection model, positioning a hidden danger area of the geological disaster, and identifying the type of the geological disaster of the positioned hidden danger area to obtain a hidden danger identification result of the slope type geological disaster. Optionally, the acquiring multi-source remote sensing data of the target slope area, performing hierarchical preprocessing on the multi-source remote sensing data, and performing space-time registration processing on the multi-source remote sensing data after the hierarchical preprocessing, includes: Acquiring optical remote sensing data, synthetic aperture radar interferometry data and laser radar data of a target slope area; Carrying out speckle reduction processing on the synthetic aperture rad