Abstract: Optical satellite remote sensing has the characteristics of high spatial and temporal resolution and high spectral resolution, which can realize the effective monitoring of sea oil spill disasters. However, optical images are easily affected by weather, resulting in a great lack of data for intelligent extraction and training using optical satellite image data. Aiming at the above problems, this paper designs an intelligent extraction method for sea oil spills based on optical satellite remote sensing data and constructs a small sample generation adversarial network (SSGAN) to perform semantic segmentation of oil spill areas in optical satellite images. It solves the limitation of the lack of optical satellite image data and fuses the spatial attention mechanism in the model to improve the accuracy of oil spill intelligent extraction. Through the long-term oil spill monitoring data of the GF-1/2/6 satellite from 2020 to 2023, the model verification and comparison are carried out. The experimental results show that the proposed SSGAN model is superior to other models in terms of oil spill semantic segmentation, and a number of Gaofen series of optical satellite oil spill monitoring special product images are produced according to the extraction results.