基于张量Tucker分解的高光谱图像异常目标识别  

Hyperspectral image abnormal object identification based on tensor Tucker decomposition

作　　者：陈楚申 唐国吉[1] CHEN Chushen;TANG Guoji(School of Mathematics and Physics,Guangxi Minzu University,Nanning 530006,China)

机构地区：[1]广西民族大学数学与物理学院,广西南宁530006

出　　处：《现代电子技术》2024年第13期43-46,共4页Modern Electronics Technique

摘　　要：高光谱图像每个像素点的光谱信息包含数百甚至数千个波段,使得高光谱图像在维度上具有高度的复杂性,且由于光谱波段众多,其中存在大量的冗余信息,加大了异常目标识别计算的负担。为此,文中提出基于张量Tucker分解的高光谱图像异常目标识别方法。通过张量Tucker分解压缩高光谱图像后,采用依据高光谱图像数据样本学习的构造方法,构建压缩后高光谱图像的字典,获取高光谱图像数据的稀疏表示形式后,通过RX异常检测方法检测出高光谱图像中的异常目标。实验结果表明:所提方法张量分解重构高光谱图像后,可以缩短压缩时间,减少算法复杂度;重构后的高光谱图像清晰度高,且高光谱图像异常目标检测虚警率低。The spectral information of each pixel in hyperspectral images contains hundreds or even thousands of bands,making hyperspectral images highly complex in terms of dimensions.Moreover,due to the numerous spectral bands,there is a large amount of redundant information,which increases the computational burden of abnormal object identification.To this end,a method for hyperspectral image abnormal object identification based on tensor Tucker decomposition is proposed.After compressing hyperspectral images by tensor Tucker decomposition,a construction method based on learning from hyperspectral image data samples is used to construct a dictionary of the compressed hyperspectral images.After obtaining a sparse representation of the hyperspectral image data,the RX anomaly detection method is used to detect abnormal objects in the hyperspectral images.The experimental results show that the proposed method for tensor decomposition reconstruction of hyperspectral images can shorten compression time and reduce algorithm complexity.In addition,the reconstructed hyperspectral image has high sharpness and low false alarm rate for detecting abnormal objects in hyperspectral images.

关 键 词：张量Tucker分解 高光谱图像 异常检测 目标识别 稀疏表示 压缩图像 数据降维 

分 类 号：TN911.73-34[电子电信—通信与信息系统] TP751.1[电子电信—信息与通信工程]