Multitemporal remote sensing image classification by nature- inspired techniques

Multitemporal remote sensing image analysis is very important and thus requires trustable tools to analyze land cover changes. In this chapter, we present nature inspired techniques for flood damage evaluation using multitemporal MODIS data. The primary focus of the proposed approach is to combine the spectral and spatial features for the proper assessment of flood prone regions. Multitemporal MODIS data are analyzed in three stages: before, during, and after flood. Methodology involves two steps: in the first step, image clustering is carried out using the genetic algorithm, particle swarm optimization, and the firefly algorithm to distinguish water regions from nonwater using pixel-based spectral information. Most of the multimodal optimization problems are often solved by these three algorithms quite effectively. Subsequently, these clustered images are segmented using geometric features to extract water-covered regions. From the results obtained, a comprehensive performance evaluation of different algorithms is carried out, which leads to the conclusion that integrating region-based image segmentation along with clustering algorithms proves to be a very efficient tool for flood mapping.