针对当前特征提取方法不能充分挖掘高光谱影像稀疏特性的问题,提出一种基于稀疏判别分析的高光谱影像特征提取方法.首先,在线性判别分析的系数向量中引入稀疏正则项来捕获具有更强判别能力的特征,将高光谱影像映射至低维稀疏的子空间;然后,利用迭代优化方法对模型进行求解.利用Salinas和Pavia University高光谱影像进行对比实验,所提方法与分类方法结合用于影像分类时,其分类精度优于其他方法,总体分类精度分别达到97.42%和97.64%.%To overcome the problem that current feature extraction methods cannot fully exploit the sparse character of hyperspectral image, sparse discriminant analysis is proposed for hyperspectral imagery feature extraction in this paper. First, L1 penalty is applied to the optimal scoring formulation for liner discriminant analysis to capture more discriminative features, which can project the hyperspectral image to lower dimensional sparse subspace. Then, the iterative algorithm for finding a local optimum is used in sparse discriminant analysis. The experiments on the Sali-nas and Pavia University hyperspectral images are performed, experimental results indicate that the proposed meth-od has better calssifcation accuracy than other algorithms when it is applied to the classification images, and the o-verall classification accuracies reach 97.42% and 97.64%, respectively.
展开▼
