Automatic Extraction of Shadow and Non-shadow Landslide Area from ADS-40 Digital Aerial Photographs

• Main Authors: Liao, Cheng-Sung, 廖晟淞
• Format: Others
• Language: zh-TW
• Published: 2012
• Online Access: http://ndltd.ncl.edu.tw/handle/17000495789452478034

碩士 === 國立屏東科技大學 === 熱帶農業暨國際合作系所 === 100 === Remote sensing image is usually used for the detection of landslide locations in disaster monitoring. However, the presence of shadows often disturbs image information, easily affecting classification of the results. Therefore, the objective of this study are to analyze landslide areas on shaded and non-shaded conditions, extracted by ADS-40 airborne multispectral image, and to design an effective and fast method for future monitoring and management. This study used the Jhuoshuei river forest working circles as study area. We used stratified classifications, filter vegetation image, and slope thresholds to discard false positive landslides among the several image classes. Moreover, histogram matching and linear-correlation corrections were used to restore shaded images. Results showed that shaded images are also suitable for classification, but that restoration by histogram matching or linear-correlation correction didn’t affect significantly classification results. Because non-shaded images don’t have shadow interference, non-shaded areas demonstrated good classification ability. In this study, slopes of 25° and 15°demonstrated higher accuracy for landslide detection on shaded and non-shaded areas, respectively. At these slopes, Kappa values of classification were 0.6204 and 0.8859, and overall accuracy were 83.07% and 94.42%, respectively. Finally, after subsuming the mean of textures into analysis of the non-shaded area, the Kappa value and overall classification accuracy increased to 0.9162 and 95.92%, respectively.