摘要
针对高光谱小目标探测问题,利用两类高光谱目标探测算法对研究对象进行计算,包括完全自适应探测算法与半监督目标探测算法,并分析比较算法各自的性能。对于完全自适应探测算法,计算结果表明异常探测算法(RXD)与基于数据白化距离探测算法(WAAD)的性能要优于低概率目标探测算法(LPTD)和均衡目标探测算法(UTD)。对于半监督目标探测算法,从接收机操作特性(ROC)曲线评价,基于椭圆轮廓分布的双曲线门限型目标探测算法(ECDHy T)的性能最优,这是因为基于椭圆轮廓分布的算法能够更准确地表征各种因素的影响。通过对两类算法的类间对比分析,证实即使少量有关目标光谱的先验信息都能够极大地提升高光谱目标的探测效率。
In terms of small target detection in hyperspectral imagery, two classes of small target detection algorithms in hyperspectral imagery are studied, analyzed and compared, which are categorized into fully self-adaptive detectors and semi-supervised detectors. To fully self-adaptive detectors, calculations show that anomaly detector and whited- distance abnormity anomaly detector （WAAD） are better than low probability target detector （LPTD） and uniform target detector （UTD）. To semi-supervised detectors, elliptically contoured distributions detector with hyperbola threshold （ECDHyT） is the best judging from the receiver operating characteristic （ROC） curve. The reason is that ECDHyT is based on the elliptically contoured distributions which can characterize target and background influenced by many factors more precisely. Comparison between the two classes of the algorithing is made. Target detection efficiency can be improved remarkably with even a little prior information about the target.
出处
《激光与光电子学进展》
CSCD
北大核心
2015年第9期302-308,共7页
Laser & Optoelectronics Progress
关键词
遥感
目标探测
虚警概率
接收机操作特性曲线
remote sensing
target detection
false alarm rate
receiver operating characteristic curve
