One-class Classification Based River Detection in Remote Sensing Image

Target detection is a fundamental problem in remote sensing images analysis. Multi-class classifiers are usually used in target detection. However, one-class classifier requires only the training samples of positive class, which has obvious advantages in specific target extraction. Based on one-class classification, the river target detection in remote sensing image is studied in this paper. The target detection process is divided into two phases: coarse screening and fine detection. In the screening phase, most non-target areas are excluded based on one-class classification. The fine detection phase extracts complex features from the target candidate regions and detects the river target by feature matching method. Based on one-class classification, the proposed method reduces the time complexity in target detection.