Marine oil spills cause great harm to the marine ecological environment; thus, an accurate detection of the oil-spill area is of great essential for a rapid emergency treatment. At this stage, synthetic aperture radar (SAR) provides an important data basis for marine oil-spill detection, but the widespread marine biological oil film, the presence of low wind areas, and the considerable speckle noise of SAR images are very likely to be oil spills, limiting the accuracy of marine oil-spill detection. Therefore, the present research proposes an improved fully-convolutional network (FCN)-based marine oil-spill intelligent detection framework in multi-polarized SAR image. The first step is to perform Pauli decomposition and Refined-Lee filtering preprocessing on the polarized SAR image to ensure the polarization characteristic information while reducing the effect of the suspected oil-spill noise on the detection accuracy. Secondly, considering the lack of consideration of the spatial information in the FCN model, the fusion mechanism of convolutional layers with different levels is used to realize the fusion of high-level semantic features and low-level spatial details, thereby improving the accuracy of marine oil-spill area detection. Experimental comparison and analysis show that the marine oil-spill intelligent detection framework, based on an improved FCN, can effectively reduce the effect of suspected oil-spill areas on detection accuracy, while considering multi-polarization and edge feature information to achieve pixel-based oil-spill area detection. The excellent detection accuracy can reach 95.7%.