Abstract: Using high-resolution remote sensing data with high spatial resolution characteristics, typical natural resource elements are extracted based on traditional convolutional neural network deep learning algorithms using multi-source high-resolution remote sensing images of 0.3 and 1m in Xining, Qinghai province as data sources. The results show that the accuracy of extracting farmland and forest land from 0.3m remote sensing images is over 85%, with a recall rate of over 89%. The accuracy of extracting farmland and forest land from 1m remote sensing images is over 90%, with a recall rate of over 91%. The research results can be used for intelligent extraction of typical elements of natural resources in Xining.

Key words: high-resolution, convolutional neural networks, deep learning, remote sensing interpretation