高斯最大似然分类在高光谱分类中的应用研究

Applied research of Gaussian maximum likelihood classification in hyperspectral classification

分析了高斯似然分类错误率和Bhattacharyya距离的关系,同时推导出在独立特征条件下Bhattacharyya距离具有相加的性质,并在这些基础上提出了一种新的特征选择算法。该算法以各特征的相对Bhattacharyya和作为准则函数选择能有效降低分类错误率的一组特征,最后利用这组特征进行高斯似然分类。实验采用AVIR IS数据,结果证明了该算法的有效性。

The relationship between Gaussian maximum likelihood classification error and Bhattacharyya distance was analyzed, and the addition property of Bhattacharyya distance was enumerated under uncorrelated features condition. Based on such analyses, a new feature selection algorithm was derived. This algorithm adopted the relative Bhattacharyya distance summation of each feature as the criterion function to select the features which contributed more to the reduction of classification error. These features then could be used for Gaussian maximum likelihood classification. Adopting AVIRIS data, the experimental results verify the effectiveness of this algorithm.