Research On Landslide Detection Method Of Aerial Ortho-image Combined With Digital Elevation Model

With the development of economy and society and the expansion of urban boundaries,the impact of natural landslide disasters on human activities is increasing,Human activity factors also make natural disasters of landslides occur more and more frequently.Therefore,identification,monitoring and quantification of landslide disasters Research is of great significance.Traditional landslide disaster identification often requires a combination of field surveys and manual methods,which is timeconsuming,labor-intensive,and has safety issues.The identification of landslides through remote sensing image technology is fast and efficient and has important practical significance.In view of the fact that some ground features and the landslide area are confused in the image recognition,resulting in low detection accuracy,this paper intends to propose a method of aerial image landslide extraction method based on digital elevation model(DEM),which comprehensively utilizes image features Terrain-related features,to reduce the false detection of non-slide land features such as building roads and bare surfaces in the ground as landslide types as much as possible,and at the same time improve the accuracy of the landslide boundary by the level set evolution method based on active contour model.The specific work content includes the following aspects:(1)Introduce the image acquisition and pre-processing methods and the digital terrain model acquisition method for the study area.The image characteristics of the landslide in this area are related to the terrain at the pixel scale and the object scale based on multi-scale segmentation.The qualitative analysis and quantitative analysis of the features were carried out,and five interpretation signs for the landslide were proposed.The analysis of the interpretation process and results showed that DEM data assistance played an important role in the interpretation of the landslide.(2)According to the object-oriented image features and elevation slope features,the suspected landslide area extraction method based on fuzzy rules classification is used to divide the study area into 4 categories,and the preliminary excluding landslide objects include vegetation areas,buildings and road areas,and other areas;In the range of suspected landslide area,on the pixel scale,the landslide seed area is extracted based on DEM terrain features.The rate of missed detection of the suspected landslide area extracted from the research area experiment is less than 10%,narrowing the area of the landslide area by more than 90%,and the accuracy of extracting the landslide seed area reaches 88%.(3)Based on the landslide seed area and active contour model,the landslide boundary is accurately extracted to obtain a landslide area that is more consistent with the interpretation result,and part of the incorrectly extracted seed area is automatically removed through the evolution process.Combined with post-processing,the experiment shows that the proposed process can effectively reduce some buildings and road areas,bare surface and some bare bedrock that are easily confused with landslides in the image.The final extraction results are smooth and the overall accuracy rate is89%.The proposed process reduces the workload while ensuring accuracy as much as possible,and has certain practical application value.