【摘要】 针对基于高斯径向基核函数的OCSVM等异常检测算法,对地物光谱变异极为敏感,导致算法异常检测性能不稳定的问题,根据光谱角度余弦测度对光谱形状相似性的描述不受地物光谱辐射强度变异影响的特性,将具有非正定核特性的光谱角度余弦核测度引入非正定SVM算法中,提出一种基于非正定OCSVM的高光谱影像地物异常检测算法。利用四组模拟数据进行目标异常检测实验,结果表明,该算法能够有效检测出高光谱影像数据中的目标地物,检测精度提升明显。更多还原

【Abstract】 The anomaly detection performance of OCSVM based on Gaussian Radial Basis Function(RBF)is sensitive to the spectral variation coming from radiation intensity variation.As the description of spectral shape similarity based on spectral Angel Cosine Measure is not affected by the variation of spectral radiation intensity,a non-positive definite kernel named as spectral Angel Cosine Kernel Measure is proposed,and applied to anomaly detection based on an indefinite SVM method in hyperspectral imagery.The experiments are carried on with four groups of simulated hyperspectral image.Experimental results have shown that this method can effectively detect the target in hyperspectral image data and the accuracy of detection is improved significantly.更多还原