A New Synthetic Aperture Radar (SAR) Image Classification Framework Model using SAR Despeckling

Abstract

In recent years, the popularity of the synthetic aperture radar (SAR) imaging system is growing exponentially. The SAR images are mainly used for the identification or classification of objects for various purposes. The classification of objects from SAR images can be difficult due to the multiplicative noise present in the images and the high dimensionality. In this manuscript, a machine learning based approach to classify images labeled with multiple classes has been proposed. Proposed approach first deals with speckling the images using a multi-objective enhanced Fruit Fly optimization method. Next, it applies a principal component analysis-based method for feature extraction. Finally, the despeckled and extracted features are used to evaluate the Support Vector Machine classifier. The proposed approach has performed well with the training and testing accuracy of 99.73% and 98.10% respectively. The experiments show that the despeckling has improved the performance of the proposed classifier to a great extent. The proposed approach also performed better than some other machine learning classifiers as well as some existing literature in terms of different performance measures.