DocumentCode :
1791361
Title :
A RX-based hyperspectral target detection method by fusing two kernels
Author :
Xiangwei Wu ; Baofeng Guo ; Chunzhong Chen ; Honghai Shen
Author_Institution :
Sch. of Autom., Hangzhou Dianzi Univ., Hangzhou, China
fYear :
2014
fDate :
14-16 Oct. 2014
Firstpage :
536
Lastpage :
540
Abstract :
The paper discusses a kernel RX algorithm for hyperspectral target detection. Because it is difficult to estimate the covariance matrix accurately for background areas, directly using the RX Algorithm for hyperspectral target detection is not a good choice in many cases. Therefore, we apply a kernel RX algorithm to our application. The kernel RX algorithm has good nonlinear anomaly detection ability due to its nonlinear mapping from the low dimensional data space to a high dimensional feature space. On the basis of a Gaussian kernel function, we propose a hybrid kernel RX (H-KRX) algorithm by adding a modified spectral angle kernel function to the original Gaussian kernel. Experiments are put into effect based on our tested hyperspectral data and the public AVIRIS 92AV3 data sets. The results indicate that the proposed method can improve the hyperspectral target detection accuracy by 5% with a similar false alarm rate.
Keywords :
Gaussian processes; covariance matrices; geophysical image processing; object detection; Gaussian kernel function; RX-based hyperspectral target detection method; covariance matrix; kernel RX algorithm; low dimensional data space; modified spectral angle kernel function; nonlinear mapping; Algorithm design and analysis; Hyperspectral imaging; Kernel; Polynomials; Signal processing algorithms; Vectors; hybrid kernel; hyperspectral imagery; nonlinear mapping; spectral angle kernel;
fLanguage :
English
Publisher :
ieee
Conference_Titel :
Image and Signal Processing (CISP), 2014 7th International Congress on
Conference_Location :
Dalian
Type :
conf
DOI :
10.1109/CISP.2014.7003838
Filename :
7003838
Link To Document :
بازگشت