Assessment of SVM classification process for landslides identification

Assessment of SVM classification process for landslides identification
Luiz Augusto Manfré1, Eduardo Jun Shinohara , Janaína Bezerra Silva, Raquel Nogueira Del Pintor Siqueira and José Alberto Quintanilha

GIS Lab - EPUSP, Av. Prof. Almeida Prado, Travessa 2, n° 83, Cidade Universitária – São Paulo – SP, CEP: 05508-900 (,,,,

Abstract: The Support Vector Machines (SVM) algorithm has been used for landcover classifications. The theoretical assumption of SVM indicates that the quality of the results increases with the use of more bands. This paper aimed to evaluate the accuracy of SVM algorithm applied over several bands compositions for the identification of landslides at Sao Paulo State Coast. LANDSAT images for the year 2000 were used. To minimize the effect of the shadows, the Normalized Difference Vegetation Index (NDVI) enhancement was calculated. We applied the SVM to the NDVI enhancement and to the following compositions: bands 1, 2; 3, 4; 1, 2, 3; and 1, 2, 3, 4. The NDVI based classification presented the highest Overall Accuracy and the Kappa Index. A huge difference in the Debris Flow areas was found, except NDVI based classification, all other overestimated this class. The NDVI presents the smallest percent of commission errors, and the best results for user accuracy. Therefore, depending on the natural conditions of the area, there are factors that are more important for the classification process with the SVM algorithm. The use of enhancements may facilitate the classification process and also produce better results than the use of a many of bands.

Keywords: Classification Assessment, Omission and Comission Errors, User Accuracy, Landslides Mapping.

manfreAccuracy2012.pdf87.71 KB