Using Himawari-8 geostationary satellite data, this study makes sea fog/low stratus (SFLS) labels based on cloud base height from CALIPSO satellite and the characteristics of fog/cloud horizontal uniformity, uses a FCN-CRF (Fully Convolutional Network and Conditional Random Field) model, and proposes a nighttime SFLS detection method. After training, the SFLS observation of CALIPSO data is used to test the FCN-CRF model and the dual channel difference (DCD) method. The FCN-CRF model performs well with a probability of detection (POD) of Symbol`@@0.611, a false alarm ratio (FAR) of 0.174, a critical success index (CSI) of 0.541, Hanssen-Kuiper Skill Score (KSS) of 0.436, and Heidke Skill Score (HSS) of 0.577. Overall, it is better than the DCD method.