BAGAN: Semantic Segmentation of Oil Spills in Optical Remote Sensing Images Based on Bottleneck Attention Module and Generative Adversarial Networks

To address the serious threat that marine oil spills pose to the ecological environment, satellite remote sensing technology for monitoring these spills is especially important. Optical satellite remote sensing, with its benefits of high spatial and temporal resolution, high spectral resolution, and autonomous operation, can effectively track and monitor marine oil spill incidents. However, the acquisition of optical satellite data is highly susceptible to climate conditions, making it difficult to obtain sufficient data for efficient information extraction. Therefore, this article proposes a bottleneck attention generative adversarial network (BAGAN) that can accurately extract oil spill areas from visible remote sensing data. The addition of data preprocessing and bottleneck attention module (BAM) in the model has improved the accuracy of oil spill segmentation. Through the optical remote sensing data of GF-1 satellite in the Bohai Bay from 2022 to 2023, the average accuracy reached 89.11%, verifying that this model is superior to other models in semantic segmentation.