论文部分内容阅读
分析了超谱数据处理过程中概率统计和相关分析典型的特征选择准则,指出了直接将其用做评价指数的弊病,进而提出了一类基于Tsallis熵冗余度的评价指数。该指数正比于多个变量间的相关信息含量,可以很方便地构造出适合超谱数据特征选择算法的评价方法。在AVIRIS数据的评价实验中,当两组相同数目波段集合的总体分类精度相差不小于2%时,基于2次Tsallis熵冗余度的评价方法正确率可达75%;当总体分类精度相差不小于8%时,正确率可达90%。
The typical feature selection criteria of probability statistics and correlation analysis in the process of hyperspectral data processing are analyzed. The disadvantages of using them directly as evaluation index are pointed out. Then a kind of evaluation index based on Tsallis entropy redundancy is proposed. The index is proportional to the relative information content among multiple variables, and the evaluation method suitable for the feature selection algorithm of hyperspectral data can be easily constructed. In the AVIRIS data evaluation experiment, when the overall classification accuracy of the same number of bands in two groups differs by no less than 2%, the correctness of the evaluation method based on the twice Tsallis entropy redundancy is up to 75%. When the overall classification accuracy is not the same Less than 8%, the correct rate of up to 90%.