基于低概率检测的高光谱异常目标检测算法研究

Anomaly detection based on low probability detection for hyper-spectral image

在深入分析高光谱数据特点的基础上,系统研究了基于低概率检测的高光谱图像异常检测方法。首先针对高光谱图像数据维数高的特点研究高光谱图像降维方法,重点研究自适应子空间分解(ASD)算法对高光谱图像进行降维;然后研究高光谱图像异常目标检测算法,异常检测算法能够在没有先验光谱信息的情况下检测到与周围环境存在光谱差异的目标,具有较强的实用性,成为一个重要的研究热点,重点研究低概率检测(LPD)算法,并用此算法对高光谱图像进行异常检测。此外,还研究了其它算法如RX算法,并与LPD算法进行比较,在此基础上对LPD算法进行改进,寻求以较高的鲁棒性进行高光谱异常目标检测,最终用基于特征融合的低概率检测算法对LPD算法进行改进。

Based on the analysis of data characteristics of hyper-spectral image, the method of anomaly detection based on low probability detection is studied systematically. First of all, because of the large data and high dimensions of hyper-spectral image, methods are studied to reduce dimension, in which adaptive subspace decomposition （ASD） algorithm is deeply studied. And then study anomaly detection algorithm. Anomaly detection algorithm has become an important research focus because of its practice use. This article focuses on the low probability detection （LPD） algorithm, and puts it into practice use. In addition, other algorithms are studied such as the RX algorithm, and then it is compared with the LPD algorithm. On the basis of this, low probability detection algorithm is improved for the purpose of finding a better algorithm, and finally use integration of the low probability detection algorithm to improve the LPD integration.