Landslide Automatic Recognition Method Based On Remote Sensing Image Recognition

The research on automatic landslide identification has always been a hot and difficult issue in the field of intelligent interpretation of geological disasters.At present,the relevant research focuses on the automatic identification of new landslides or seismic landslides.Such landslides have obvious contrast with the background spectral characteristics and are easy to identify.However,the age of occurrence is relatively long,and the vegetation on the old landslide that is still threatened has been restored.The spectral characteristics of remote sensing images are not obvious.There is still a lack of systematic research on the division of such landslide recognition unit,the determination of segmentation scale,the construction of classification feature set and the selection and optimization of recognition model.After the occurrence of landslide,the change of microtopography is its essential feature.Positive and negative terrain,platform and slope turning are important geometric boundaries of landslide elements and their boundaries.In addition,the anomalies of slope vegetation and slope hydrological characteristics are also reflected in remote sensing images.Therefore,it is an effective way to identify such landslides by dividing recognition units according to landslide geometric features,quantifying remote sensing feature indexes of landslides and constructing classification decision model.Based on Sentinel-2A multispectral image and ASTER DEM terrain data,this paper takes Fengjie County as the research area to analyze the feasibility of slope unit division based on terrain factors for landslide identification.The selection method of the optimal segmentation parameters and the selection strategy of the optimal feature subset are determined,and the recognition accuracy of different machine learning models in Fengjie County landslide is evaluated.Finally,the overall framework of landslide automatic recognition is determined.The main work and innovation are as follows :1.Considering the geometric characteristics of landslide,using multiscale segmentation algorithm,slope and aspect as the input layer,the slope unit for landslide identification is extracted,and the optimal parameter combination is determined based on the global variance of the unit and spatial autocorrelation.2.The physical characteristics of landslide and its reflection on remote sensing images are analyzed,and 16 kinds of valuable basic spectral and topographic indexes are calculated,and the automatic identification index system of landslide is constructed.Based on e Cognition platform,the set of layer value,geometry and texture features is constructed,and the scale effect of different feature information gain rates is evaluated by using information entropy.3.The Person correlation coefficient between the basic features is calculated,and the redundant features are removed.Based on the correlation-based feature selection(CFS)and the best first search algorithm,the optimal object feature subset is determined,and the scale effects of CFS algorithm selection results based on different scale slope units is analyzed.4.The slope units are divided based on the optimal scale,and five machine learning classification models are trained by using the selected optimal feature combination : Naive Bayes(NB),Decision Tree(CART),Support Vector Machine(SVM),Random Forest(RF),Artificial Neural Network(ANN).The classification accuracy is evaluated,and the optimal classification model for landslide identification is determined.5.The random forest model is used to test the generalization performance of the model in Fengjie County.The application results show that the recognition accuracy of the model for landslide is above 80%,indicating that the model can be better used for the identification of landslide range.