一种基于目标与背景特征分离模型的高光谱目标检测修正算法  

A Revised Target Detection Algorithm Based on Feature Separation Model of Target and Background for Hyperspectral Imagery

作　　者：吴护林[1] 邓贤明 张天才 李忠盛[1] 岑奕 汪家辉 熊杰 陈知华 林牧春 WU Hu-lin;DENG Xian-ming;ZHANG Tian-cai;LI Zhong-sheng;CEN Yi;WANG Jia-hui;XIONG Jie;CHEN Zhi-hua;LIN Mu-chun(The 59th Research Institute of China Ordnance Industry,Chongqing 401329,China;Aerospace Information Research Institute,Chinese Academy of Sciences,Beijing 100101,China)

机构地区：[1]中国兵器工业第五九研究所,重庆401329 [2]中国科学院空天信息创新研究院,北京100101

出　　处：《光谱学与光谱分析》2024年第1期283-291,共9页Spectroscopy and Spectral Analysis

基　　金：科工局课题(HDH59030303);装发领域基金项目(61402080404)资助。

摘　　要：高光谱图像立方体数据可以提供成像场景中地物在可见光和近红外波长范围内的空间信息和地物属性诊断的光谱特征信息,在目标检测与识别方面拥有得天独厚的天然优势。然而,基于高光谱图像数据的目标检测也存在一定缺陷,如经典的高光谱目标检测算法仅利用光谱维度信息检测目标,检测模型要么对背景高维特征矩阵构建的准确度不足,要么对背景先验光谱特征的完备性要求较高,导致算法对不同复杂度的检测场景适应性不强。因此,基于计算复杂度较低、参数需求量较少且检测性能较为优异的经典多目标检测算法—多目标约束能量最小化(MCEM),提出了一种基于目标与背景环境特征分离模型的高光谱目标检测修正算法(R-MCEM)。首先,设计了一个与目标形状、尺寸相近的逐像元移动运算窗口,依次计算窗口中的每个像元与窗口内其他像元的光谱距离之和D1,像元与各类目标的光谱距离之和D2。其次,采用获得D1/D2最小值的像元替换窗口内的所有像元值。然后,自左向右、自上而下逐像元移动窗口,重复窗口内每一个像元与目标、背景像元的光谱距离运算,并确定窗口内与背景相似度最高、与目标相似度最低的像元。直到移动运算窗口遍历整个高光谱图像,大幅提升了基于目标与背景环境特征分离的背景高维特征矩阵准确度。分别设计了基于实测高光谱图像数据和模拟图像数据的修正检测算法性能验证试验,并采用三维操作特征曲线(3D ROC)结合目标与背景分离度(SDBT)开展修正算法的检测精度评估。试验结果表明,提出的修正算法有效减少了虚警率,提高了检测精度。基于实测数据的检测精度、目标与背景分离度由MCEM算法的0.937 7、 0.57提升到R-MCEM的0.993 5、 0.67,基于模拟数据的亚像元检测能力由MCEM的20%丰度提升到R-MCEM的15%丰度。Hyperspectral imagery cube Data can provide spatial information and diagnostic spectral characteristics,in the range of visible and near-infrared wavelength,about the attributes of materials in the scene,which contribute to the unique advantage of hyperspectral imagery for target detection.However,hyperspectral target detection has some shortcomings,such as the classical hyperspectral target detection algorithm only uses the information of spectral dimension to detect the target.The detection model either has insufficient accuracy for the construction of background high-dimensional f characteristics matrix or has high requirements for the completeness of background prior spectral characteristics,resulting in poor adaptability of the algorithm to different scenarios.Therefore,based on the classic multi-target detection algorithm-multiple target constrained energy minimization(MCEM),which has low computational complexity,fewer parameter requirements and better detection performance,this paper proposes a modified algorithm R-MCEM(revised MCEM)based on the separation model of target and background.First of all,this algorithm designs a pixel-by-pixel moving operation window that is similar to the shape and size of the target and sequentially calculates the spectral distances between each pixel and other pixels in the window,referred to as D1,and the spectral distances between the pixel and all targets,referred as D2.Next,all pixel values in the window are replaced with the pixel obtaining the minimum value of D1/D2.Then,move the window from left to right and from top to bottom and repeat the same calculation.Until the moving operation window traverses the entire hyperspectral image,all the interested targets in the hyperspectral image are eliminated as much as possible,and the accuracy of the background high-dimensional characteristics matrix is greatly improved.In this paper,the performance verification tests of the modified detection algorithm based on the true hyperspectral image data and the simulated image data a

关 键 词：高光谱目标检测 目标与背景特征分离模型 3D ROC SDBT 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置]