基于张量径向基核函数支持向量机的高光谱影像分类

Hyperspectral image classification based on tensor-based radial basis kernel function and support vector machine

针对如何利用高光谱影像的空间和光谱特征实现地物目标的精确分类,本文在径向基核函数(RBF)的基础上,提出一种基于张量径向基核函数(Tensor-RBF)和支持向量机(SVM)的高光谱影像分类算法。首先,用像素及其空间邻域像素的光谱向量组成的三阶空-谱张量块表达该像素空-谱信息,并作为后续高光谱影像分类的基本处理单元;然后,定义作用在张量数据上的Tensor-RBF核函数;最后,设计基于Tensor-RBF核函数SVM的多分类器,实现结合空-谱信息的高光谱影像多类地物目标分类。为了验证提出算法的有效性,分别对3幅高光谱影像进行实验,将本文算法与3种对比算法的分类结果进行定性和定量对比分析。实验结果表明,本文算法对3幅高光谱影像的总体精度分别为93.10%、93.43%和86.38%,相对3种对比算法具有更高的总体精度。

Aiming at how to use the spatial and spectral features of hyperspectral images to achieve accurate classification of ground objects, presents a hyperspectral image classification algorithm based on tensor-based radial basis kernel function(Tensor-RBF) and support vector machine(SVM). Firstly, the three-order spatial-spectral tensor block is composed with the spectral vectors of a pixel and its spatial neighborhood pixels to express the spatial and spectral information of the pixel, and the generated spatial-spectral tensor block is used as the basic processing unit in subsequent hyperspectral image classification. Secondly, Tensor-RBF kernel function on tensor data is defined to express of the image. Finally, the SVM multi-classifier is designed based on Tensor-RBF kernel function to achieve multi-class ground object classification of hyperspectral images based on spatial-spectral information. In order to verify the effectiveness of the proposed algorithm, the classification experiments were conducted on three hyperspectral images, and the classification results were compared with those using other three algorithms qualitatively and quantitatively. The experiment results show that the overall accuracies of the proposed algorithm are 93.10%, 93.43% and 86.38% for the three hyperspectral images, respectively, which are higher than those of other three algorithms.