多方向投影寻踪与高光谱遥感图像特征提取

The multi-directional projection pursuit and the feature extraction of hyperspectral remote sensing image

介绍了投影寻踪算法的基本原理,构造了面向易混分类别的高光谱遥感数据投影寻踪指标,在单方向投影寻踪算法的基础上,提出了序贯多方向投影寻踪算法.在该算法提取出的特征方向上,易混分地物类别的分类精度提高6%左右,但存在压制其他地物的矛盾.

The basic concepts of projection pursuit were introduced and the projection index oriented to easy mixed classified objects was constructed. Based on the single directional projection pursuit algorithm, the multi directional projection pursuit algorithm was proposed . On the feature directions extracted by the algorithm, the classification accuracy of easy mixed classified objects was improved about 6%, but the other objects were suppressed slightly.