波段最大筛选法及其在高光谱目标探测中的应用  

Maximum band screening and its application to hyperspectral target detection

作　　者：王静荔[1] 

机构地区：[1]空军工程大学理学院,陕西西安710051

出　　处：《红外与激光工程》2012年第6期1514-1519,共6页Infrared and Laser Engineering

基　　金：国家自然科学基金(41174093)

摘　　要：提出了一种面向目标探测的高光谱图像波段选择方法—波段最大筛选法(MBS),它将每个波段的图像看成一条波段向量,以两个最不相似的波段作为初始波段,每次从剩余波段中选取一个和已选波段最不相似的波段,通过对波段相似性阈值的合理调节,保证了目标探测算子在所选波段上探测效果最佳。为了验证MBS的有效性,对机载可见光/红外成像光谱仪(AVIRIS)获取的两幅真实高光谱图像进行了实验,结果表明,MBS选出的波段分别占据全部波段的15%和9%,从而使目标探测算子ACE和AMF在其上的探测性能有了明显改善。A new band selection method for target detection in hyperspectral imagery, namely maximum band screening（MBS）, was proposed. Considering each band image as a band vector, two most dissimilar bands were selected as the starting bands and the band most dissimilar with the bands selected was chosen from the remaining bands at each step. By properly adjusting the band similarity threshold value, the final band selection result was obtained on which the target detector could achieve best detection performance. In order to demonstrate the utility of MBS, two real AVIRIS hyperspectral data sets were tested for target detection. The experimental results show that less than 15% and 9% of the full bands are selected by MBS respectively, on which the detection performance of adaptive cosine estimator（ACE） and adaptive matched filter（AMF） has been improved greatly.

关 键 词：波段选择 高光谱目标探测 波段最大筛选法 光谱相似性度量 ROC曲线 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]