The potential of Chinese high-spatial satellite imagery and deep learning models in coastline extraction were explored in this work. The performances of three deep learning models including U-Net, ResUnet, and SegNet were compared using 2-m resolution pansharpened products of GaoFen-1(GF1), and Ziyuan-3 (ZY3) imagery. The prediction results of ResUnet were significantly more accurate than those of U-Net and SegNet. The trained ResUnet model was then used to predict and extract the coastlines of Haikou City, Hainan Province. The 2-m resolution coastline products of Haikou City in 2016, 2018, and 2019 were obtained. The results showed that no significant changes in the coastline of Haikou City from 2016 to 2019.