CN-121746939-B - Highway slope landslide debris flow disaster identification method, device, equipment and medium

CN121746939BCN 121746939 BCN121746939 BCN 121746939BCN-121746939-B

Abstract

The invention relates to the field of geological disaster identification, in particular to a method, a device, equipment and a medium for identifying landslide debris flow disasters of a highway slope. According to the invention, through the high spatial resolution optical remote sensing image of the highway side slope in the mountain canyon area and the DEM data in the same area, the depth fusion of the spectral features of the optical image and the elevation features of the DEM is realized through the cross-modal cyclic fusion module, and the texture damage and the earth surface deformation information of the landslide debris flow can be captured at the same time. Meanwhile, when the optical image is seriously covered by cloud and snow, the DEM data branch can be used as an independent data source for identification, so that the continuity and stability of identification work are ensured. According to the invention, two remote sensing data with different modes and complementary information are organically combined, so that the accuracy of identifying landslide detritus geological disasters is improved, the requirements of rapid general investigation on disaster points along highways in mountain canyons and valley areas and in major engineering areas can be met, the artificial workload can be reduced, and the operation efficiency is improved.

Inventors

WANG JUNLU
DONG CHANGSONG
CAO SHENGLIANG
FU ZHIPENG
LI ZHEN
TIAN QINGZHEN
YANG YUN

Assignees

中交第一公路勘察设计研究院有限公司

Dates

Publication Date: 20260512
Application Date: 20260228

Claims (7)

1. The method for identifying the landslide debris flow disasters of the highway slope is characterized by comprising the following steps of: S1, inputting remote sensing data of landslide debris flow to be identified in a high mountain gorge valley region into a pre-trained landslide debris flow disaster area identification network, wherein the remote sensing data comprises high-spatial resolution optical images and DEM data in the same period, and the high-spatial resolution optical images are of which the spatial resolution is less than or equal to 5 meters; S2, outputting the landslide and detritus flow identification result to be identified by the landslide and detritus flow disaster area identification network, wherein the landslide and detritus flow identification result comprises the boundary and the area of the landslide and detritus flow corresponding to the disaster area; the landslide debris flow disaster area identification network comprises a double-branch encoder, a cross-modal cyclic fusion module and a residual error attention decoder; The dual-branch encoder is used for respectively extracting the characteristics of the high-spatial resolution optical image and the DEM data, and comprises two parallel gating circulation context modules, wherein the gating circulation context modules comprise a gating unit and a context sensing unit which are sequentially connected and are used for adaptively fusing multi-scale context information of remote sensing data through a gating mechanism; the cross-modal cyclic fusion module is used for fusing the characteristics of the high-spatial resolution optical image and the DEM data to generate bimodal characteristics; the residual attention decoder is used for identifying the landslide debris flow to be identified according to the bimodal characteristics; the cross-modal cyclic fusion module comprises a contrast sharing fusion module and a consistent feature representation module, and comprises the following processing steps: the contrast sharing fusion module: Wherein, the For the fused multi-modal weight information, C () is a shared convolution function, W i and W j are input ith and j modal parameters, E () is a fusion metric function; The consistent feature representation module: Wherein W is the extracted contrast characteristic, For consistent feature maps, sigma and Is the function of the activation and, And Is a training consistency parameter, F () is a feature extraction processing function, and R () is an information reconstruction function; The contrast sharing fusion module further comprises a bidirectional cross attention mechanism; The bidirectional cross attention mechanism comprises two attention mechanism branches, weight extraction is carried out by taking the high spatial resolution optical image and the DEM data as main factors, fusion processing is carried out on the extracted weight graphs, and the expression is as follows: branch one: Wherein, the And The characteristic diagrams of the ith mode and the jth mode are processed by the shared convolution module respectively, Is a linear projection matrix of modality i on the query matrix, 、 The linear projection matrix of the mode j in the key matrix and the value matrix respectively; is the query matrix generated by modality i, Is a key matrix generated by modality j, Is a value matrix generated by modality j; Representative pair Performing matrix transposition; is a scaling factor; Is a normalization function; is the attention weight matrix of branch one; Is the attention output feature sequence of branch one, Is a function of the layer normalization, Features representing an enhanced ith modality; branch two: Wherein, the Is a linear projection matrix of modality j on the query matrix, 、 Respectively linear projection matrixes of the mode i in the key matrix and the value matrix; is the query matrix generated by modality j, Is a key matrix generated by modality i, Is a value matrix generated by modality i; Representative pair Performing matrix transposition; Is the attention weight matrix of branch two; Is the attention output feature sequence of branch two, Features representing an enhanced j-th modality; Feature fusion: Wherein, the Representing feature stitching, E () is a fusion metric function.
2. A method of identifying a landslide debris flow disaster of a highway slope according to claim 1, wherein said pre-training of a landslide debris flow disaster area identification network comprises the steps of: Remote sensing data of highway slopes in the mountain gorge valley areas are obtained and subjected to geographic registration, and the remote sensing data are output as marked sample data after manual marking; Performing data enhancement processing on the marked sample data to generate a landslide detritus flow marked sample data set, wherein the data enhancement processing comprises any one or more of random overturning, random rotation, random cutting, brightness random adjustment and color random enhancement; And carrying out model training on the landslide and debris flow disaster area identification network through the landslide and debris flow labeling sample data set, and outputting the landslide and debris flow disaster area identification network as a pre-training after model training is finished.
3. The method for identifying a landslide debris flow disaster of a highway slope according to claim 2, further comprising S3: Verifying the identification result of the landslide chip flow to be identified, marking the remote sensing data of the landslide chip flow to be identified according to the verification result, and storing the remote sensing data into the landslide chip flow marking sample data set; and after the landslide chip flow labeling sample data set is increased by a set number of labeling sample data, performing model training on the landslide chip flow disaster area identification network through the current landslide chip flow labeling sample data set.
4. The method for identifying the landslide debris flow disaster of the highway slope according to claim 1, wherein the landslide debris flow disaster area identification network comprises the following working steps: the double-branch encoder extracts the characteristics of the remote sensing data respectively; the cross-modal circulation fusion module fuses the extracted features to generate a self-adaptive attention weight graph of the landslide debris flow disaster area; And the residual attention decoder decodes according to the self-adaptive attention weight graph and outputs landslide debris flow identification results.
5. A highway slope landslide debris flow disaster identification device, comprising: The system comprises a data input module, a data processing module and a data processing module, wherein the data input module is used for inputting remote sensing data of landslide debris flow to be identified, and the remote sensing data comprises high-spatial resolution optical images and DEM data in the same period; The landslide and chip flow disaster identification module is used for executing the landslide and chip flow disaster identification method of the highway slope according to any one of claims 1 to 4 according to the remote sensing data, and outputting a landslide and chip flow identification result corresponding to the landslide and chip flow to be identified.
6. A highway slope landslide debris flow disaster identification device comprising at least one processor and a memory communicatively coupled to said at least one processor, said memory storing instructions executable by said at least one processor to enable said at least one processor to perform a highway slope landslide debris flow disaster identification method according to any one of claims 1 to 4.
7. A computer readable storage medium, characterized in that the computer readable storage medium stores a computer program which, when executed by a processor, implements a method for identifying a highroad slope landslide debris flow disaster according to any one of claims 1 to 4.

Description

Highway slope landslide debris flow disaster identification method, device, equipment and medium Technical Field The invention relates to the field of disaster detection, in particular to a method, a device, equipment and a medium for identifying landslide debris flow disasters of a highway slope. Background Landslide debris flow is one of the most common and highly dangerous geological disasters in China, and occurrence of the landslide debris flow not only can cause serious threat to life safety of people, but also can cause serious property loss. After a disaster occurs, how to accurately locate and identify the disaster area in time by utilizing informatization means and an intelligent technology becomes a key link in disaster prevention and auxiliary early warning. With the rapid development of remote sensing earth observation technology, the spatial resolution and the time resolution of remote sensing images are continuously improved, the data acquisition period is obviously shortened, and the cost is continuously reduced. Compared with the traditional manual investigation mode, the remote sensing technology has gradually become a main technical means for developing the identification and monitoring of the regional scale landslide by virtue of the advantages of large range, rapidness, objectivity and high efficiency. Currently, existing landslide identification research mostly adopts a single-source data supported artificial intelligence method to realize large-scale identification. The method generally takes a high-resolution optical remote sensing image or a Digital Elevation Model (DEM) as a main data source, constructs a landslide training sample database, and directly realizes intelligent identification of a landslide target through a deep learning network model. However, there are significant limitations to the single source data approach in mountain canyon areas, i.e., mountain areas with average elevation above 3500 meters, steep hills, large exposed bedrock areas, and deep valley areas with "V" shaped cross sections, brae steep (typically >25 °), narrow valleys, and large longitudinal ratio drops. Taking the Tibetan autonomous region of China as an example, the geographical environment of the region is complex, the topography fluctuation is severe, the main traffic trunk lines such as the Sichuan line, the Qinghai-Tibet line and the new Tibet line pass through the mountain gorges Gu Ou, are densely distributed along landslide disaster points, and particularly are prominent in steep topography regions such as Linzhi and Changdu. Because snow is accumulated in the high-altitude areas throughout the year and the topography shadows are strong, in the optical remote sensing images, the snow, the shadows and the landslide are very similar in tone and texture characteristics, and if only a single optical image is relied on for feature extraction and training, missed judgment and false judgment are very easy to cause. On the other hand, if single Synthetic Aperture Radar (SAR) data are adopted for identification, huge terrain height differences of the Tibetan autonomous region of China can cause serious problems of overlapping and masking, shadow, perspective distortion and the like of radar side view imaging, so that ground surface information is distorted, and accurate identification of landslide features is affected. Therefore, whether a single optical image or a single radar image is used, the common problem caused by special topography effect of the mountain gorge valley area is difficult to overcome, and large-scale, accurate and stable identification of landslide debris flow cannot be realized in the complex area. Therefore, there is an urgent need for a method, device, equipment and medium for identifying landslide and debris flow disasters of highway slopes, which can combine different characteristic cross-modal remote sensing data. Disclosure of Invention The invention aims to solve the problems that single source data in the prior art is difficult to overcome special topography effects of a mountain gorge region, so that generalization capability is weak, and landslide and debris flow characteristics cannot be accurately identified, and provides a method, a device, equipment and a medium for identifying landslide and debris flow disasters of a highway side slope. In order to achieve the above object, the present invention provides the following technical solutions: The method for identifying the landslide debris flow disasters of the highway slope is characterized by comprising the following steps of: S1, inputting remote sensing data of landslide debris flow to be identified in a high mountain gorge valley region into a pre-trained landslide debris flow disaster area identification network, wherein the remote sensing data comprises high-spatial resolution optical images and DEM data in the same period, and the high-spatial resolution optical images are optical remote sensing images with s