Study On Extraction Of Landslide Based On Fractal Theory And Saliency Analysis

The landslide has the characteristics of regionality,mass occurrence,multiple occurrence,and great harm.It is extremely destructive to bridges,roads,and other facilities,and poses a huge threat to the safety of the people and property.How to quickly and comprehensively identify and extract information such as landslide distribution characteristics,nature,and scale has very important practical and scientific significance.At present,the development of remote sensing technology provides a new direction for the identification and extraction of landslides.Remote sensing images have been used in all aspects of landslide disaster prevention.However,landslide identification methods based on remote sensing images are mainly aimed at earthquake-type landslides,and are not fully applicable to small-scale,rainfall-type landslides.Moreover,in the previous studies,the small-scale landslide boundary extraction was mainly carried out,but the large-scale landslide extraction and investigation were few.This paper takes Taining County,Fujian Province as a research area,uses 2019 Gaofen No.1 and Resource No.3 satellite remote sensing images to carry out research on landslide fractal features,and combines the fractal texture features of the images with saliency detection models,based on MATLAB and ArcGIS Platform to extract landslides in the study area.The main conclusions are as follows:(1)The landslide in Taining County is a typical area in the southeastern coastal area of China.Most of them are rainfall-type landslides,which are widely distributed and small in area.In this paper,the interpretation signs of landslides are established by analyzing the characteristics of color,texture and shape.(2)Using the fractal theory,the box dimension method is used to calculate the fractal dimension of the landslide obtained by human-computer interaction interpretation;the network coverage method is used to calculate the fractal dimension of the landslide spatial distribution.It is concluded that the boundary trajectory and spatial distribution of the landslide have good fractal characteristics.(3)From the perspective of computer vision,in the saliency detection model GBVS algorithm,the fractal texture of the remote sensing image is added as a new underlying feature,that is,the color,brightness,direction and fractal texture features of the image are extracted to extract the landslide.Experimental results show that the extraction accuracy of the proposed method is improved compared with the original GBVS method.In the large-scale landslide extraction in the county,the accuracy measurement value is 0.9091,which is feasible.Comprehensive research shows that the landslide extraction method combined with fractal theory and saliency analysis is feasible for the identification and extraction of large-scale small-scale,rainfall-type landslides,and can provide important information for landslide disaster investigation,monitoring,evaluation and treatment in related areas The technical support has great application value in landslide extraction.