Abstract: Four-side trees are an important part of forest coverage calculation. However, due to its sporadic distribution, the current statistical method is still a field research method with long statistical time and high calculation cost, and there is a lack of research on the use of remote sensing images and automatic extraction. Therefore, this paper takes the multi band Sentinel-2 remote sensing image as the data source for screening, combination and classification. It combines the definition of four-side tree and spatial analysis method to realize the automatic extraction of four-side tree. The experimental results show that the overall accuracy of the classification results of remote sensing images combined with 9, 6, 4 bands by using the support vector machine method of radial basis kernel function is 93.673 5%, the Kappa coefficient is 0.918 1, and the extraction accuracy of four side tree is 90%. The experimental accuracy is the highest. Compared with the field investigation method, this extraction method has high precision and faster speed, and is more suitable for a wide range of four-side tree information extraction and change detection.

Key words: four-side trees, image classification, spatial analysis, automatic extraction, support vector machines