量测重构线性混合模型高光谱图像目标检测  

Target Detection in Hyperspectral Imagery Based on Linear Mixing Model Reconstructed from Measurements

作　　者：贺霖[1] 潘泉[1] 赵永强[1] 

机构地区：[1]西北工业大学自动化学院,陕西西安710072

出　　处：《电子学报》2007年第1期23-27,共5页Acta Electronica Sinica

基　　金：国家自然科学基金(No.60172037);航空科学基金(No.03D53032);武器装备预研基金(No.51401040204HK0359);西北工业大学科技创新基金

摘　　要：针对高光谱图像中无背景和目标先验信息情况下的目标检测问题,给出了一种基于量测重构光谱混合模型的目标检测算法.通过构造投影算子削弱背景干扰,提高自动搜索目标光谱特征的准确性;对光谱空间进行估计后,构造量测重构光谱混合模型;以此量测重构混合光谱模型为基础,使用投影抑制背景并提高信噪比以改善检测效果.同时给出了目标信号与局部杂乱背景之间的均方根误差SLCR及目标信号峰值与局部杂乱背景均值的比例PSLCMR两个检测评价指标的定义.利用可见光/近红外波段高光谱图像进行了实验,实验结果和理论分析表明了算法的有效性.There presents a detection algorithm based on spectral mixing model reconstructed from measurement in this paper in order to detect unknown targets in unknown environment. Firstly, we project the hyperapectral imagery to suppress the background interference in order to search target spectral more accurately. Then, we estimate the spectral subspace and construct a spectral mixing model reconstructed from measurements, And based on the proposed spectral mixing modeling, we project the hyperspectral imagery, which suppress spectral signatures of background and improve the SNR, in order to increase the detection power. Finally, the Signal to Local Clutter RMSE （SLCR） and Peak Signal to Local Clutter Mean Ratio（ PSLCMR）, which is proposed, are used to evaluate the detection. Theoretic analysis and the results of experiment on visible/near-infrared hyperspectral imagery verify the effectiveness of the algorithm.

关 键 词：高光谱图像 目标检测 量测重构 光谱混合模型 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]