关键词: 遥感图像, 随机森林, 决策树, 相关性, 人工免疫

Abstract: It is difficult to get training samples of remote sensing, Random Forests Algorithm（RFG） is introduced in the classification of remote sensing image. In order to improve the accuracy and stability of RFG on the case of small training samples, Firstly, a high random features selection method is proposed to build decision trees which have lower correlation, consequently a lower generalization error of random forests; secondly, Artificial Immune System is used to optimize and compress Random Forests, which can resolve the conflict between the accuracy and scale of Random Forest. The experiments are performed on the UCI data sets, and the results show that the efficiency and feasibility of the improvement algorithm and the optimization processing. After optimization, the forests have lower scale but higher accuracy. Finally, the improved RFG is used to remote sensing image and the results are analyzed by comparing to traditional methods.

Key words: remote sensing image, random forest, decision tree, correlation, artificial immune system