Shape Analysis and Target Detection from Multi Resolution Remote Sensing Satellite Images

Abstract

Researchers are working on developing algorithms to automatically identify targets in an optical images. The thesis discusses problems and issues from real world scenarios and addresses shape analysis, classification task and target detection. Shape is an important feature for target detection.In this thesis we have presented a method for getting optimal rectangle whose area is equal to number of pixel inside the region. This thesis will present the new and distinctive definition measure for two dimensional shape detection. This measure will help us in identifying different object shapes based on their fitness value. Effectiveness was demonstrated by the suggested definition. Here we have presented a new supervised classification method, MSBSVM which scan the entire data set only once and provide a high quality training sample which has a greater probability of becoming a support vector for SVM classification. The remarkable feature is that the method can be used in both circular and elongated training data sets. newline newline Target recognition has a vital issue concerning the detection of compound object, Buildings and other rectangular structures form an important subclass of man made features. Due to the resolution of the sensors, adjacent buildings in the scenes appear fused in the images and seem like a single compound rectangle. There is significance of separating the individual buildings from the resulting compound objects in a segmented image, nevertheless it is difficult. Oil tank is a significant target and automatic detection has remained an interesting study of research in remote sensing. Oil tank detection consists of area filtering, circular fitting, measure based filtering and supervised classification. We have tested the algorithm on several panchromatic imageries. The suggested approach has been beneficial in the precise of oil tanks from satellite images. newline newline