ISAR Image Classification with Wavelet and Watershed Transforms

Abstract

Inverse Synthetic Aperture Radar images are playing a significant role in classification of sea and air targets. First we acquire the ISAR images of targets using a sensor like radar and extract the characteristics of targets from the ISAR images in the form of feature vectors. The computed feature vectors are used for classification of targets. In this work, widely used and efficient segmentation tool Watershed transform and the multi resolution technique wavelet transform are explored to derive the target features. An artificial neural network based classifier is used for classification. The Wavelet analysis divides the information of an image into approximation and detail sub signals. The approximate and three detail sub signal values are taken as feature vectors and given as input to the classifier for ship ISAR image classification. The widely used segmentation technique, Watershed transform is applied to the ISAR images. The wavelet coefficients are computed for the segmented ISAR images and used as feature vectors for classification of the ISAR images. Also, the statistical moments mean and standard deviation are computed for the color ISAR images itself, taken in RGB format. These statistical color moments are used as feature vector. The classification accuracy is compared for the feature vectors.