Improved YOLOX-S Marine Oil Spill Detection Based on SAR Images

Marine oil spill spreads rapidly and has a long-term impact. Once it occurs, it will cause severe damage to the ecological environment. Synthetic Aperture Radar (SAR) is widely used in marine oil spill monitoring due to its all-weather and all-day characteristics. However, the contrast of different SAR images is inconsistent, making it difficult for the network to learn valuable features. To address this issue, this paper proposes an improved YOLOX-S (IYOLOX-S) model for marine oil spill detection. The model enhances image contrast by a truncated linear stretch module, uses CspDarknet and PANnet to extract image features, and obtains oil spill detection results through Decoupled Head. First, a truncated linear stretching module is added, which can improve the image contrast. It also highlights the characteristics of oil spill areas to enhance the networks learning ability. Second, the proposed score loss into the global loss function enhances the learning ability of the model and improves the detection accuracy. Experiments are carried out on the collected oil spill dataset, and the test sets average precision (AP) is 90.02%. The experimental results show that the improved YOLOX-S model accurately identifies oil spill areas.