Boosting Ship Detection in SAR Images With Complementary Pretraining Techniques

Deep learning methods have made significant progress in ship detection synthetic aperture radar (SAR) images. The pretraining technique is usually adopted to support deep neural networks-based SAR detectors due the scarce labeled However, directly leveraging ImageNet hard obtain a good detector because of different imaging perspectives and geometry. In this article, resolve problem inconsistent between earth observations, we propose an optical (OSD) transfer characteristics ships observations images from large-scale aerial image dataset. On other hand, handle geometry images, optical-SAR matching (OSM) technique, which transfers plentiful texture features by common representation on OSM task. Finally, observing that OSD pretraining-based SSD has better recall sea area while can reduce false alarms land area, combine predictions two through weighted boxes fusion further improve results. Extensive experiments four datasets three representative convolutional network-based benchmarks are conducted show effectiveness complementarity proposed detectors, state-of-the-art performance combination detectors. method won sixth place 2020 Gaofen challenge.