结合导向滤波和最大概率的高光谱图像分类

A Hyperspectral Image Classification Method Combined with Guided Filter and Maximum Probability

针对利用滤波器提取高光谱图像的空间特征辅助光谱信息来提高高光谱图像分类精度的不足,提出导向滤波提取的空间纹理信息和最大概率结合的高光谱图像分类算法(SGD-SVM-GD)。鉴于空间纹理信息挖掘不足,该方法首先利用导向滤波提取由主成分分析降维后的高光谱图像空间纹理特征,然后将空间信息与光谱信息结合,交由支持向量机完成分类得到初始分类结果,最后结合导向滤波和概率最大化优化分类结果。实验表明,相比单纯使用光谱信息、纯空间信息和空谱结合的SVM分类方法以及边缘保持滤波的方法,所提出的SGDSVM-GD方法对高光谱图像的分类精度有较大提高,表明了该方法的有效性。

Aiming at the deficiency of supplementing spectral information with spatial information obtained by filter to improve the classification of hyperspectral image,an improved scheme was put forward according to existent methods.In view of the insufficient of spatial texture information extraction,the paper proposed an algorithm of supervised classification which was combined with spatial information obtained by guided filter and the maximum probability.Firstly,the method extracted the spatial feature of hyperspectral image whose dimensionality was reduced by PCA.Secondly,the spatial information and spectral information were combined together,which were classified by SVM.Finally,the classification was optimized by the guided filter and the maximum probability.The experiments show that the BS SVM algorithm is better than original SVM with the pure spectrum information,the pure spatial information,the spatial spectral information,and the method of edge preserving filtering.The performance of the classification with SGD SVM GD algorithm for hyperspectral image is greatly improved,and the effectiveness of the method is fully verified.