Utilizing the spectral data, sea surface temperature, and other auxiliary data provided by Aqua/MODIS between June and September from 2014 to 2018, three categories of sample datasets are established, in which samples are labled as sea fog, cloud, and sea surface based on the cloud base height extracted from CALIOP/VFM. Combining K-fold cross-validation, grid search, and particle swarm optimization techniques, 4 different machine learning models are trained, including random forest (RF), support vector machine (SVM), multilayer perceptron (MLP), and fully convolutional networks (FCN). The performance of these models in the summer sea fog detection over the Arctic Ocean is compared. The results show that RF, SVM, MLP, and FCN all exhibit some capability of sea fog detection, with probability of detection (POD) exceeding 70%. Among them, FCN shows the best overall performance, achieving a POD of 79.91% and a low false alarm rate (FAR) of 24.90%, resulting in a critical success index (CSI）of 63.17%.