Landslides detection: a case study in Conghua city of Pearl River delta

Abstract

Landslide is a typical geological disaster that has adverse effect on lives and properties, generating both direct and indirect economic losses in mountainous regions every year. Comparing to other geological disasters, landslides are considerably smaller in scale and more dispersed. The characteristics of landslide render detection and identification of landslides challenging. In this paper, object-based image analysis is used to detect landslide sites using remote sensing images. Firstly, multi-scale image segmentation was performed on the 0.61-meter Quickbird (QB) image of the study area and over tens of spatial, spectral, shape and texture features were extracted based on the segmented image objects. Secondly, 11 optimized features for landslides classification was selected using genetic algorithm (GA), which gives the best fitness value for landslides classification. Thirdly, in-situ landslides observation results were used as typical cases and cased-based-reasoning (CBR) classification was applied on all segmented image objects, from large scale to small scale. Finally, classification accuracy was evaluated over the whole study area. In conclusion, CBR method is able to detect landslides successfully using high resolution images. The CBR method proposed in this paper could achieve better classification accuracy than traditional supervised classification.