摘要：In this paper we investigate a method for the classification of high-dimensional image data using a multi-stage classifier structured as a binary tree, and employing a reduced number of features at each node, in order to mitigate the Hughes phenomenon. The selected method for feature reduction is based on the optimization of Bhattacharyya distance at each individual node of the tree. As the Bhattacharyya distance is defined for a pair of classes, the binary tree approach allows the extraction of an optimal sub-set of features at each individual node. Experiments were performed using an AVIRIS image data set, varying the number of training samples and also the number of selected features at each node. The results have shown an improvement in the accuracy of the thematic image, as compared to more traditional methods for feature selection and extraction.
其他摘要：In this paper we investigate a method for the classification of high-dimensional image data using a multi-stage classifier structured as a binary tree, and employing a reduced number of features at each node, in order to mitigate the Hughesphenomenon. The selected method for feature reduction is based on the optimization of Bhattacharyya distance at each individual node of the tree. As the Bhattacharyya distance is defined for a pair of classes, the binary tree approach allows the extraction of an optimal sub-set of features at each individual node. Experiments were performed using an AVIRIS image data set, varying the number of training samples and also the number of selected features at each node. The results have shown an improvement in the accuracy of the thematic image, as compared to more traditional methods for feature selection and extraction.
关键词：Reconhecimento de Padrões;Imagem Hiperespectral;Árvores de Decisão;Distância de Bhattacharyya;Pattern Recognition;Hyperspectral Image;Decision Trees;Bhattacharyya Distance
