基于主成分建模的SVDD高光谱图像异常检测

PDF下载HTML

Title:

An SVDD algorithm for hyperspectral anomaly detection based on principal component modeling

作者:

曾现灵¹, 张立燕², 胡荣华¹

1. 首都师范大学 资源环境与地理信息系统北京市重点实验室, 北京 100048;
2. 首都师范大学 三维信息获取与应用教育部重点实验室, 北京 100048

Author(s):

ZENG Xianling¹, ZHANG Liyan², HU Ronghua¹

1. Key Laboratory of Resource Environment and GIS of Beijing, Capital Normal University, Beijing 100048, China;
2. Key Laboratory of 3D Information Acquisition and Application of the Ministry of Education, Capital Normal University, Beijing 100048, Chi

关键词:

主成分建模; SVDD; 局部邻域聚类; 光谱角余弦; 高光谱异常检测

Keywords:

principal component modeling; SVDD; local neighborhood clustering; spectral angle cosine; hyperspectral anomaly detection

分类号:

TP751

DOI:

10.3969/j.issn.1673-4785.201309081

摘要:

针对SVDD背景建模时混入异常点造成的检测率下降的问题, 提出了基于主成分建模的SVDD方法并应用于高光谱图像异常检测。利用高光谱图像的光谱特征提取背景的主要成分, 并分别对不同成分构建超球体, 形成单种背景成分SVDD模型, 最后利用综合决策函数对单个SVDD背景模型进行综合判断待检测像元, 从而实现高光谱图像异常像元的检测。用仿真数据和真实数据对算法的性能进行验证, 并将其与SVDD方法进行性能比较。结果表明, 新算法在低虚警概率下较之SVDD模型有更高的检测概率, 实验结果证明了算法的有效性。

Abstract:

An SVDD algorithm based on the principal component modeling is presented for hyperspectral anomaly detection, in order to solve the problem of its low detection rate caused by mixing abnormal points in the process of modeling background. This method extracts the principal components of the background samples by using the hyperspectral image’s spectral signature, and then uses these different components to build different super spheres respectively, forms different single background component SVDD models by these super spheres, finally uses the integrated decision function to judge these SVDD background models to detect any anomalies. The performance of the algorithm is verified by simulated and real data. The results show that the proposed method can obtain a higher detection rate under low false rate than the algorithm based on SVDD, verifying the effectiveness of this proposed method.