Abstract
Abstract
Remote sensing classification is an important part in the process of extracting effective image information and research the foundation of land cover change. While traditional remote sensing image classification methods have some problems on low accuracy and uncertainty, machine learning algorithms are gradually applied to remote sensing classification. In this paper, support vector machines (SVM) method with high training speed and low computation burden is adopted to classify land cover based on GF-2 image, which is the domestic optical remote sensing satellite with high spatial resolution. The results show that: The overall classification accuracy by SVM is achieved 72.59% and the coefficient of Kappa is 0.65. The classification map is highly consistent with the original image, especially higher classification accuracy of cropland and tree. Partial regions were misclassified as shadow that didn’t reflect the real land objects. As a whole, there is favorable classification quality using SVM method and GF-2 multispectral bands.