Comparative Study of Feature Extraction Approaches for Ship Classification in Moderate-Resolution SAR Imagery

This paper presents a comparative study of existing feature extraction approaches for ship classification in moderate-resolution synthetic aperture radar (SAR) images. Ship classification is a key functionality in many maritime surveillance applications. For efficient ship classification, appropriate feature extraction is crucial. Most of existing studies have used high-resolution images. For maritime surveillance, however, wide-area coverage is essential whereas it inevitably reduces the spatial resolution. In this paper, we evaluate the applicability of representative methods to moderate-resolution images. The evaluated methods are hand-crafted feature extraction (HCF), principal component analysis (PCA) and autoencoder (AE) based on neural-network. The evaluation is done on the basis of accuracy for two-class ship classification into tanker and cargo. The experiments demonstrate that AE outperforms HCF and PCA in classification accuracy by 7.4% and 2.6%, respectively. Furthermore, AE performs best even in classification of challenging cases such as small ships.