耿修瑞,
1.中国科学院空间信息处理与应用系统技术重点实验室 ??北京 ??100190
2.中国科学院电子学研究所 ??北京 ??100190
3.中国科学院大学 ??北京 ??100049
详细信息
作者简介:孟令博：女，1989年生，博士生，研究方向为高光谱图像特征提取及异常检测
耿修瑞：男，1965年生，研究员，研究方向为高光谱图像处理技术
通讯作者:耿修瑞 xrgeng@mail.ie.ac.cn
中图分类号:TP75计量
文章访问数:871
HTML全文浏览量:346
PDF下载量:48
被引次数:0
出版历程
收稿日期:2018-03-26
修回日期:2018-10-18
网络出版日期:2018-10-24
刊出日期:2019-01-01
A Hyperspectral Imagery Anomaly Detection Algorithm Based on Cokurtosis Tensor
Lingbo MENG,Xiurui GENG,
1. Key Laboratory of Technology in Geo-spatial Information Processing and Application System, Beijing 100190, China
2. Institute of Electronics, Chinese Academy of Sciences, Beijing 100190, China
3. University of Chinese Academy of Sciences, Beijing 100049, China
摘要
摘要:高光谱图像中的异常像元往往具有在图像中出现的概率低和游离于背景数据云团之外的特点，如何“自动”确定这些异常像元是高光谱遥感图像处理中的一个重要研究方向。经典的高光谱异常检测方法一般从图像的统计特性入手，广泛应用的RXD异常检测算法通过计算图像的2阶统计特征，可以直接给出异常点的分布情况，算法复杂度低，但缺点是没有考虑到图像的高阶统计信息。基于独立成分分析的异常检测算法虽然考虑了高阶统计量对异常点的敏感性，但需要反复迭代提取异常成分后，再对提取后的成分进行异常检测。该文提出一种基于协峭度张量的异常检测算法，该算法不需要事先提取异常成分，可以直接对观测像元进行逐一检测，从而给出异常点的分布情况。基于模拟数据和真实数据的实验结果表明，该方法能够在检测出异常像元的同时更好地压制背景信息、减小虚警率，从而提高异常检测精度。
关键词:高光谱图像/
异常检测/
高阶统计/
协峭度张量
Abstract:The abnormal pixels in hyperspectral images are often have the characteristics of low probability and scattered outside the background data cloud. How to automatically detect these abnormal pixels is an important research direction in hyperspectral imagery processing. Classical hyperspectral anomaly detection methods are usually based on statistical perspective. The RXD algorithm which is widely used can give the anomalies distribution directly through the second order statistical feature of the image, but the disadvantage is that it does not take into account the higher order statistics of the image. Anomaly detection algorithm based on Independent Component Analysis (ICA) considers the sensitivity of higher order statistics to outliers, but it needs iteration process to extract abnormal components first. And then the extracted components is used for anomaly detection. A method based on cokurtosis tensor for anomaly detection is proposed. This method does not need to extract anomaly components first. It can directly detect the observed pixels and give the distribution of abnormal pixels. Experiments results on both simulated and real data show that it can detect abnormal pixels while suppressing the background information better. Therefore, it can reduce false alarm rate and improve detection accuracy.
Key words:Hyperspectral imagery/
Anomaly detection/
Higher-order statistical/
Cokurtosis tensor
PDF全文下载地址:
https://jeit.ac.cn/article/exportPdf?id=5af5e8e2-a9af-427a-a419-f17ee63d6d87
