An Improved Region Growing Algorithm for River Extraction from Remote Sensing Images

Abstract

Focusing on the problems of low extraction accuracy and difficult extraction of complex images in the existing remote sensing image river extraction methods, an improved region growing algorithm is proposed in this paper. Firstly, the image is fused with its edge feature by preprocessing, and then the edge feature images are corroded continuously to reduce the background influence. An improved K-Means algorithm is used to cluster corroded images to realize automatic selection of seed points in region growing algorithm. Finally, an adaptive threshold change method based on variation coefficient is proposed in region growing algorithm. In this paper, the remote sensing images are tested. The results show that the average of accuracy, precision and mIoU of our method are higher than those of other segmentation algorithms.