Eleventh International Conference on The Image

Abstract

Floods are one of the major troubles caused to human beings where there are lots of loss to livelihood and properties. This work deals with the automatic classification of synthetic aperture radar (SAR) images into highly flooded and low flooded regions, and a region-based identification which helps any disaster recovery team to work for the periods of disaster and to serve the affected areas quickly. First. the input synthetic aperture radar is de-speckled using a fuzzy discontinuity adaptive non local means approach, and it is segmented using a modified fuzzy c means in which the input feature vector is reduced by the usage of two techniques, namely quantization and aggregation, based on which there is a reduction in the algorithm convergence rate. This fuzzy c means algorithm is used to segment the input synthetic aperture radar image into three clusters, namely, flooded regions, permanent water bodies like rivers, ponds, etc., and land surfaces. After these clusters are obtained from the fuzzy c means segmentation, some statistical features are extracted from this segmented region by constructing a grey level co-occurrence matrix and gray level run length matrix. After the features are extracted, the neural network classifier is trained with these features to classify the input synthetic aperture radar image into high and low flooded regions, and based on the classification, the regions can be easily identified based on the coordinates of the input image. After the particular region is identified, the information can be given to the disaster recovery team.