Indefinite OCSVM method for object detection in hyperspectral imagery

高斯径向基核函数是基于光谱向量间欧氏距离的度量,对于因光照强度变化而引起的地物光谱变异敏感,当同类地物光谱发生变异时,基于高斯径向基核的高光谱影像地物检测算法的性能下降。为了解决该问题,基于光谱曲线形状相似性描述提出了光谱角度余弦核测度这一非正定核函数,并应用于一种非正定OCSVM方法的高光谱影像地物检测。最后利用两幅高光谱影像进行了实验分析,实验结果证明了本文算法的有效性。

As the Gaussian radial basis function （RBF） is based on the Euclidean distance of two spectral vectors, it is sensitive to spectral curve variations resulted from radiation intensity variation. When the spectral curves of same materials are different, the detection performance of the RBF based OCSVM objective detector will degradate. In order to solve this problem, a non- positive definite kernel, named as the spectral Angel Cosine Kernel Measure （SACKM）, is proposed based on the spectral curves similarity description, and was applied to object detection based on an indefinite OCSVM method in hyperspectral imagery. Fi- nally, the experiments were carried out with two hyperspectral images, which are used to validate the proposed method.