Remote Sensing Image Feature Selection Based on Rough Set Theory and Multi-agent System

Remote sensing image classification is a very important method to obtain the geographic information. For a better land cover classification, it is necessary to bring in more spatial information as auxiliary. While more spatial information may also lead to the over-fitting of the classifier algorithm, which, especially under the circumstance of few samples, will in return devalues classification quality. Select useful features are very important for remote sensing classification. The traditional rough set based feature selection algorithms utilize greedy search method which unstable and relay on initial feature input sequence. This study presents a classification method based on rough set and multi-agent system. Experiments show that, compared to the traditional way, the proposed method can be used to optimize the spatial attributes better for classification and improve the classification accuracy, with a high application value for the remote sensing image supervised classification.