Abstract: For the problems such as the complexity of the river because of the bridge images being caused by terrain, weather and environment, and hardly collecting all river samples of the images, a segmentation of river based on self-supervised learning is proposed. The approach uses the part of the river area automatically extracted by combining the K-means clustering method with Harris corner method as river sample in self-supervised learning, according to the color and texture feature extracted from river sample, trains the sample with the one class support vector machine. Then the river is segmented by the trained classifier. The experimental results demonstrate the proposed method has good performance in automatically segmenting river and can adapt to bridge images in different scenarios.

Key words: self-supervised learning, segmentation of river, K-means clustering, Harris corner, support vector machine