algorithm based on supervised learning has achieved great success in related tasks. However, compared with coarse-grained label data sets, the classification of fine-grained label data sets is still a difficult point. Geographic images are widely used in all aspects of society. Researchers often need to classify large-scale geographic image data. However, due to the small feature difference of geographic images, automatic classification is relatively difficult. In this paper, the fine-grained features of geographic images are labeled, and trained and learned through the deep convolution network, which greatly improves the classification accuracy of geographic images.