高光谱遥感影像的广义判别分析特征提取

高光谱遥感影像具有丰富的光谱信息,在地物分类识别方面具有明显的优势。针对复杂高光谱影像分类问题,应用了一种广义判别分析特征提取技术。将输入样本通过非线性函数映射到特征空间,在特征空间中应用线性判别特征提取方法;算法求解过程中涉及到在特征空间的内积用核函数代替,简化计算的同时也使得算法与非线性函数的具体形式无关。通过影像分类试验表明,该方法较常用特征提取方法更有利于分类精度的提高。

Generalized Discriminant Analysis Feature Extraction of Hyperspectral RS Image

The hyperspectral remote sensing image is rich in spectrum information,so it can be better to carry on the ground targets classification.To deal with classification hyperspectral image with high dimension,an approach of generalized discriminant analysis feature extraction is applied.We map the input space into a high-dimensional feature space with linear properties.In the new space,one can solve the problem in linear discriminant analysis.In the processing of this method,the vector dot product in the feature space was replaced by kernel function,so it is easy to compute and don't have to know the nonlinear function.Compared with other common feature extract method,this method is more suitable to extract nonlinear feature of hyperspectral image,which is proved by the image classification test.