关键词: 高光谱, 分类, 动态分类器集选择, 集成学习, SVM动态集成

Abstract: Based on Bagging Support Vector Machine (SVM), this paper applied dynamic ensemble selection technique to the SVM ensemble learning, and investigated the application of dynamic SVM ensemble to the classification of hyperspectral remote sensing images. Considering the characteristics of hyperspectral data, Bagging SVM was improved by selecting feature subspace randomly and feedback learning; the algorithm of computing local area of K nearest neighbors was ameliorated through adopting plus composite distance; the validation set samples were more representative by means of appending the misclassified training samples to the validation set. The experimental results show that in comparison to single optimized SVM and other popular SVM ensemble methods the improved dynamic SVM ensemble exhibits the highest classification accuracy, and it could effectively improve the classification precision of hyperspectral remote sensing images.

Key words: hyperspectral, classification, dynamic ensemble selection, ensemble learning, dynamic SVM ensemble