An Object-Based River Extraction Method via Optimized Transductive Support Vector Machine for Multi-Spectral Remote-Sensing Images

The accurate extraction of rivers is closely related to agriculture, socio-economic, environment, and ecology. It helps us to pre-warn serious natural disasters such as floods, which leads to massive losses of life and property. With the development and popularization of remote-sensing and information technologies, a great number of river-extraction methods have been proposed. However, most of them are vulnerable to noise interference and perform inefficient in a big data environment. To address these problems, a river extraction method is proposed based on adaptive mutation particle swarm optimization (PSO) support vector machine (AMPSO-SVM). First, three features, the spectral information, normalized difference water index (NDWI), and spatial texture entropy, are considered in feature space construction. It makes the objects with the same spectrum more distinguishable, then the noise interference could be resisted effectively. Second, in order to address the problems of premature convergence and inefficient iteration, a mutation operator is introduced to the PSO algorithm. This processing makes transductive SVM obtain optimal parameters quickly and effectively. The experiments are conducted on GaoFen-1 multispectral remote-sensing images from Yellow River. The results show that the proposed method performs better than the existed ones, including PCA, KNN, basic SVM, and PSO-SVM, in terms of overall accuracy and the kappa coefficient. Besides, the proposed method achieves convergence rate faster than the PSO-SVM method.