摘要
为解决传统RX异常检测算法导致的星上高光谱图像异常检测准确率低、成本高等问题,对一种改进的星上高光谱异常检测算法进行了研究。在传统RX异常检测算法的基础上,用噪声自适应主成分变换对原始图像进行变换,选择变换后大于设定的合适信噪比阈值的数据;用局部正交子空间投影(LOSP)算法将相应数据投影到正交子空间获得图像残差数据,通过抑制背景等强干扰信息而突出小概率的异常目标信息;用空域滤波方法提取残差数据的特征,将大部分图像信息以某种标准集中于少数的主成分;对获得的波段子集用中值滤波器滤波,消除噪声干扰;用RX异常检测算子对滤波后的波段子集进行异常目标检测。试验结果证明了算法的有效性,算法通过数据量压缩,降低维度、抑制干扰信息,减少了异常检测处理的数据量以提高异常检测效率和精度。
To solve the problems of the low accuracy and high expense of hyperspectral anomaly detection onborne using traditional RX anomaly detection algorithm,an improved algorithm of hyperspectral anomaly detection was studied in this paper.Based on traditional RX anomaly detection algorithm,the original image was transferred by noise adaptive principal component transform.The data which single-to-noise ratio was bigger than threshold set would be selected.The selected data space was projected to orthogonal subspace to obtain the characteristics of the residual error of the image by local orthogonal subspace projection algorithm,so the anomaly target information with low probability could the outstand by suppressing the background and other strong interference.The characteristics of the residual error data were picked up by space-domain filtering method which could make majority information of the original image concentrate on few wave bands under certain standard.The subspace obtained was filtered to eliminate the noise by mean value filter.The anomaly detection of the filtered subspace was carried on using RX anomaly detection algorithm.An experiment showed the effectiveness of the method proposed.The method proposed can reduce the information in order to improve the accuracy and efficiency of anomaly detection onborne by data compressing,dimension reducing and disturbance suppressing.
出处
《上海航天》
CSCD
2017年第3期47-51,共5页
Aerospace Shanghai
基金
国家青年科学基金项目资助(F050405)
关键词
星载探测
高光谱
异常检测
局部正交子空间投影
RX异常检测算法
噪声自适应主成分变换
信噪比
阈值
on-borne detection
hyperspectral
anomaly detection
local orthogonal subspace projection(LOSP)
RX anomaly detection algorithm
noise adaptive principal component transform
signal-to-noise ratio
threshold
作者简介
李恪（1980-）,男,博士,主要从事光电遥感技术研究
