Complex Valued and Layer Based Categorization of Synthetic Aperture Radar Patches

Abstract

This paper presents a comparison between a complex valued layer based sparse coding and complex valued convolutional neural network for Synthetic Aperture Radar (SAR) patch categorization. Recent progress in convolutional neural networks made a classification and categorization of SAR patches very attractive. Layered sparse coding is based on an optimal dual based l1 analysis that can be applied to the problems of SAR image patch classification. In this paper a sparse coding approach is designed in multi layered architecture. This this paper proposes a layered based sparse coding using approaches introduced within deep learning architecture, therefore, we incorporated spatial pooling layer, normalization layer, map reduction layer and a classification layer into the process of sparse coding. In this paper a layer based convolutional neural network consisted of convolutional, drop out, Relu, fully connected, soft max and classification layers were combined to achieve the best classification accuracy. Experimental results showed that the CNN based classification achieved better results compared to the layered based sparse classification.