Golf is a luxury sport, and can promote physical and mental well-being for players and managers. However, it poses signi icant environmental and social issues, including chemical pollution, land use destruction and habitat loss, soil erosion, limited participation for af luent groups, and land use conlicts. Thus, effective golf course management is necessary at local and national scales. To support this effort, we proposed a convolutional neural network (CNN) approach to accurately segment golf courses in South Korea from the Sentinel-2 images. We ine-tune the U-Net, LinkNet, FPN (Feature Pyramid Network), and DeepLabV3Plus models, conigured with encoders of the Ef icientNet-b family (0, 3, 5, and 7). The CNN models performed outstandingly well and achieved a mean accuracy of around 90%. We found the combination of DeepLabV3Plus with Ef icientNet-b7 encoder was the most accurate structure (F1 = 0.98, IoU=0.97). The golf courses in South Korea are detected and segmented in the model inference and nicely mapped in GIS software. The deep learning modeling approach is a potential method recommended for replication. Future work can focus more on developing large training datasets to leverage for a higher model precision.