Laplace & LBP feature guided SAR ship detection method with adaptive feature enhancement block

Deep learning (DL) improves the accuracy of SAR ship detection a lot in comparison with traditional methods. However, most of the DL-based methods abandon traditional feature of ships and only use abstract feature derived from neural network, which decreases the interpretability and may hinder the upper limit of detection accuracy. Hence, to settle the problem, we propose a Laplace & LBP feature guided SAR ship detection method with adaptive feature enhancement block. Specifically, the proposed method consists of traditional feature extraction block (TFEB) and adaptive feature enhancement block (AFEB). The function of the TFEB is to extract traditional features, which put more prior knowledge into neural network and boost the training. However, different types of features have different degree of importance for SAR ship detection. Hence, we propose the AFEB that enhances more important features among traditional feature and original SAR images according to the weights learned adaptively. The experimental result on the open SSDD dataset reveals that our method gets a better performance in accuracy than other three detectors.