基于局部联合偏度-峰度的高光谱图像波段选择方法  

Band Selection Based on Local Joint Skewness and Kurtosis for Hyperspectral Image

作　　者：王琪[1] 杨桄[1] 张俭峰[1] 向英杰[1] 

机构地区：[1]空军航空大学,吉林长春130022

出　　处：《激光与光电子学进展》2017年第11期115-122,共8页Laser & Optoelectronics Progress

基　　金：吉林省教育厅十二五科研项目(2015448)

摘　　要：偏度和峰度能够较好地表达高光谱图像的非高斯性,突出目标、纹理等异常信息,很好地应用于波段选择。为了更好地突出局部异常信息,在全局联合偏度-峰度指数模型基础上,提出了局部偏度-峰度的高光谱图像波段选择方法。利用全局联合偏度-峰度指数对原始图像进行波段子空间划分,然后选择适当大小的模板窗口,计算窗口内的局部联合偏度-峰度指数,并以此方法遍历所有波段,求出累积局部联合偏度-峰度指数,最后进行波段选择。波段选择结果表明,局部联合偏度-峰度指数方法所选择波段分布更加广泛,效果更好。异常检测实验结果和融合结果表明,本文方法所得图像在客观指标评价中具有较大优势。The non-Gaussian of the hyperspectral image can be well expressed by skewness and kurtosis, which highlight the target, texture and other anomaly information. They can be well applied to the band selection. In order to stand out the partial anomaly information better, the local joint skewness and kurtosis-based band selection for hyperspectral image is proposed on the basis of the global joint skewness-kurtosis figure. The bands of the original image are divided by the global joint skewness-kurtosis index into several subspaces. Then the template window of appropriate size is chosen and the local joint skewness-kurtosis index is calculated. All bands are traversed by this method. Finally, the accumulated local joint skewness-kurtosis index is calculated in order to complete the band selection. The band selection results show that the bands selected by the local joint skewness- kurtosis method are more widely distributed and the effect is better. The anomaly detection and fusion results show that the image obtained by the proposed method has great advantages in the evaluation of objective indicators.

关 键 词：图像处理 高光谱图像 波段选择 峰度 偏度 局部异常 

分 类 号：TP751.1[自动化与计算机技术—检测技术与自动化装置]