Landslide susceptibility assessment based on multi-scale segmentation of remote sensing and geological factor evaluation

Abstract

The prediction and prevention of landslide is an important issue, and the study of regional landslide susceptibility is one of the core of landslide spatial prediction. Based on the multi-scale segmentation and object-oriented classification theory, four parameters including entropy, energy, correlation and contrast of remote sensing image are selected as the texture factor to extract the susceptibility features. the four types of geological factors including the reservoir water impact rating, slope, slope structure and engineering rock group were adopted to analyze the geological background, finally the C5.0 decision tree model was constructed to predict the four types of landslide-prone units in the study area. The results show that the engineering rock group of the high-susceptibility unit usually develops into soft rock group and soft-hard interphase group, and the slope was mostly between 15° to 30° in these units. The average correct rate of training samples and test samples is 91.64%, the Kappa coefficients are 0.84 and 0.51, respectively. Therefore, this kind of landslide susceptibility classification based on image multi-scale segmentation and geological factor rating has certain applicability.

Keywords

• landslide
• susceptibility
• remote sensing
• multi-scale segmentation
• c5.0 decision tree