半导体光电, 2020, 41 (1): 141, 网络出版: 2020-04-13  

联合空间信息的改进低秩稀疏矩阵分解的高光谱异常目标检测

Joint Spatial Information and Improved Low-rank and Sparse Matrix Decomposition-based Anomaly Target Detection for Hyperspectral Imagery
作者单位
1 陆军工程大学石家庄校区 电子与光学工程系, 石家庄 050003
2 解放军31681部队, 甘肃 天水 741000
3 解放军68129部队, 兰州 730000
摘要
针对低秩稀疏矩阵分解的高光谱异常目标检测算法忽略了图像的空间信息, 导致检测精度低的问题, 提出了一种联合空间信息的改进低秩稀疏矩阵分解的高光谱异常目标检测算法。算法综合利用了高光谱图像的光谱信号与空间信号, 并与图像自身的稀疏性相结合, 对经典的基于低秩稀疏矩阵分解的目标检测算法进行改进, 该算法以待测像元为中心构建一定大小的空间窗, 计算中心像元与邻域内其他像元的空间相似度权值和光谱相似度权值, 通过计算邻域内其他像元对中心像元的比例权值得到了中心像元的重构光谱值并作差得到两者的残差矩阵; 最后基于低秩稀疏矩阵分解的高光谱异常目标检测算法得到图像的稀疏矩阵, 将代表异常目标信息的稀疏矩阵和残差矩阵相加并求解矩阵行向量之间的欧式距离得到像元的异常度, 设置阈值, 得到检测结果。为验证所提算法的检测性能, 采用了真实的高光谱数据进行仿真实验, 并与现有算法进行对比, 结果表明该算法能够得到更高的检测精度。
Abstract
The hyperspectral anomaly target detection algorithm for low rank sparse matrix decomposition ignores the problem that the spatial information of the image leads to low detection accuracy. In this paper, proposed is an improved low-rank sparse matrix decomposition method for hyperspectral anomaly target detection based on spatial information. The algorithm comprehensively utilizes the spectral signal and spatial signal of hyperspectral image, and combines with the sparsity of the image itself to improve the classical target detection algorithm based on low rank sparse matrix decomposition. The algorithm is built around the pixel to be measured. A spatial window of a certain size calculates both the spatial and spectral similarity weight of the central pixel and other pixels in the domain, and obtains the weight of the central pixel by calculating the proportional weight of other pixels in the domain to the central pixel. The spectral values are constructed and the residual matrix is obtained. Finally, the hyperspectral anomaly target detection algorithm based on low rank sparse matrix decomposition is used to obtain the sparse matrix of the image, and the sparse matrix and residual matrix representing the abnormal target information are added and solved. The euclidean distance between the matrix row vectors obtains the degree of abnormality of the pixels, sets the threshold, and obtains the detection result. In order to verify the detection performance of the proposed algorithm, real hyperspectral data is used for simulation experiments, and compared with the existing algorithms, higher detection accuracy can be obtained.

张炎, 华文深, 黄富瑜, 严阳, 王强辉, 索文凯. 联合空间信息的改进低秩稀疏矩阵分解的高光谱异常目标检测[J]. 半导体光电, 2020, 41(1): 141. ZHANG Yan, HUA Wenshen, HUANG Fuyv, YAN Yang, WANG Qianghui, SUO Wenkai. Joint Spatial Information and Improved Low-rank and Sparse Matrix Decomposition-based Anomaly Target Detection for Hyperspectral Imagery[J]. Semiconductor Optoelectronics, 2020, 41(1): 141.

关于本站 Cookie 的使用提示

中国光学期刊网使用基于 cookie 的技术来更好地为您提供各项服务,点击此处了解我们的隐私策略。 如您需继续使用本网站,请您授权我们使用本地 cookie 来保存部分信息。
全站搜索
您最值得信赖的光电行业旗舰网络服务平台!