样本加权约束能量最小化算法

Sample Weighting Constrained Energy Minimization Algorithm

针对高光谱图像小目标探测中约束能量最小化算法对同类地物光谱多样性敏感,且不能有效识别大目标的问题,提出了一种样本加权CEM目标探测算法.通过光谱单位化处理,减小了目标点所在环境不同而出现的光谱差异.为精确地确定目标物在所有像元中所占的比例,以光谱相关性作为权值的度量对样本进行加权处理,有效降低了目标像素在样本自相关矩阵运算中所占的比重,使算法对大目标探测同样有效.

Constrained Energy Minimization（CEM） algorithm is very sensitive to spectral difference of the same object and cannot detect the large targets.We proposed a sample weighting CEM algorithm.Through spectral vector unitization,the errors caused by different environment are decreased,and target recognition accuracy is increased.To decrease the proportion in the sample autocorrelation matrix,we use spectral correlation as a similarity measure to weight the samples.The modified algorithm acquired the satisfied effect for large targets.