Change Detection Analysis in Multitemporal Satellite Image Series

Change Detection Analysis in Multitemporal Satellite Image Series
José Guilherme Fronza and Carlos Antônio Oliveira Vieira

Federal University of Santa Catarina – UFSC - Campus Universitário Trindade – CEP 88040-900 – Florianópolis, Santa Catarina. Phone: ++55 48 3721-8593 (guilherme.fronza@gmail.com, carlos.vieira@cfh.ufsc.br)

Abstract: The present work aims to perform ratio and difference techniques to detect changes in Nova Friburgo, Rio de Janeiro – Brazil caused by landslides. Using Landsat-5 images obtained before and after the landslides in January 2011 that was possible. These techniques allow identifying abrupt change on spectral response in remotely sensed images. There are many factors influencing the success of applying change detection techniques, among them, one could mention: atmospheric condition, sensors calibration procedures, water content in soil, differences in sun angle at the image obtaining. Neighborhoods and forested areas were destroyed causing major problems to citizens and slopes subject to new landslides. Given a set of spectral bands, it is difficult to define a good threshold for histogram and the best bands in order to use change detection analysis. The better result was obtained by using the band TM7 and TM5 (infrared) with difference between images technique. Kappa was set in 0,9255 and 0,9243, respectively to detect changed areas in landscape. Noise in result was frequently and its minimization was essential. To image ratio technique was obtained good results (TM3 with 0,8582 Kappa) but with great difficulties to define a good threshold cut-off in image histogram.

Keywords: Change Detection, feature selection, remote sensing, landslides.