高光谱遥感图像特征提取及分类研究——基于离散余弦变换(DCT)及支撑向量机技术

高光谱遥感图像具有波谱连续，维数高的特点．当样本较少时，在原始特征空间采用传统的统计识别方法分类达不到理想的效果．经研究发现有两种方法可以解决小样本高维的非线性分类问题．一是将原始空间通过离散余弦变换（DCT）压缩到低维空间，再用统计识别方法分类；二是利用支撑向量机的内积函数，将原始空间映射到高维空间，使其在新的特征空间线性分类．实验表明，这两种方法比利用马氏距离判别法直接对原始图像分类有更好的分类效果．

Picking up Character and Classification of the Hyper-spectral Remote Sensing Image Based on DCT and SVM

Hyper - spectral remote sensing possesses the character of spectral characteristic curve series and the high dimensions. Under a few training samples condition, the classification result will be bad if we classify image in original feature space with traditional statistical recognition. Research shown that there are two methods can solve this nonlinear classification problem of a few training and high dimensions. One is to cast the original dimensional vector space into the low dimensional vector space through DC transformation. And then, use statistical recognition to classify the image; Another is to change input space into high - dimensional one by non - linear variable defined by accumulating function of SVM. The remote sensing image will be llne classify in the new feature space. Test result indicates that both methods have better classification and recognition than Mahalanobis Distance classification for original image.