•  
  •  
 

Abstract

Landslides are natural phenomena for the dynamic balance of the earth's surface. Because of frequent occurrences of typhoons and earthquakes in Taiwan, mass movements are common threats to people's lives. In this paper, the interpretation of knowledge is quantified as recognition criteria. Multisource high-resolution data, for example, a SPOT satellite image, 20 m × 20 m Digital Terrain Model (DTM) reduced from Light Detection And Ranging (LiDAR) data, and aerial orthophotos were used to construct the feature space for landslide analysis. Landslides were recognized by an objectoriented method combining edge-based segmentation and a Supported Vector Machine (SVM) method. The classification results are evaluated in comparison with those by manual interpretation. Two cases from northern and central Taiwan are tested. Both cases show that the object-based SVM method is better than a pixel-based method in classification accuracy. The commission error of the proposed method is also smaller than that of the pixel-based method. Moreover, except for the spectral features, the slope and Object Height Model (OHM) characteristics are also important factors for improving landslide classification accuracy. Further study is required for assessing the mixed pixel effect when the resolution is as large as 20 m and for characterizing the effects of sampling rates or scaling caused by changes in resolution.

Download
COinS