Lightweight Convolutional Neural Network for False Alarm Elimination in SAR Ship Detection

SAR ship detection is an important task of SAR image interpretation and plays a significant role in global marine surveillance. However, since there are usually many false alarms in the detection results, a desirable performance is rarely achieved. Aiming at this problem, this paper proposes an improved method for the ship detection process, using the lightweight convolutional neural network EfficientNet-B0 as a false alarm elimination network, cascaded after the detection stage, to reduce the false alarm rate in the detection results and improve the overall detection accuracy. The results on the MSTAR data set show that EfficientNet-B0 has a good classification effect, and the experiments on the custom data set also verify the effectiveness of the network.