Novel Deep Learning-Based Technique for Sentinel-1 Ocean SAR Vignettes Classification

Ocean SAR scenes classification is a challenging task under the SAR image classification domain. The main problem is the wrong classification of similar SAR images, primarily due to speckle presence. This study proposes a deep learning model developed from scratch to automatically classify oceanic and atmospheric phenomena from Sentinel-1 SAR wave mode (WV) vignettes. The proposed preprocessing technique reduces the speckle and can improve the minor features for discriminating the SAR images. We have experimented using a publicly available hand-curated dataset that describes ten ordinarily occurring atmospheric or oceanic processes. The experimental results indicate that our proposed deep learning model with preprocessing technique has good classification performance.