A modified kernel-RX algorithm for anomaly detection in hyperspectral images

作者：Khazai Safa; Mojaradi Barat^*

来源：Arabian Journal of Geosciences, 2015, 8(3): 1487-1495.

DOI：10.1007/s12517-013-1218-5

• Anomaly detection
• Euclidean distance
• Hyperspectral images
• Kernel-RX
• Mahalanobis distance

The use of kernel-based methods has recently become a popular approach to many hyperspectral image applications such as anomaly detection (AD). A well-known kernel-based anomaly detector in hyperspectral images is the Kernel-Reed-Xiaoli (KRX) algorithm. This study aims at improving the performance of the original KRX. To meet this purpose, a modified version of the KRX that assumes a spherical covariance matrix for the background class is proposed. Performance comparison is accomplished using two real hyperspectral data sets that contain subpixel and multipixel targets. Experimental results demonstrate that the use of the modified KRX, compared to the original KRX, improves the detection and computational efficiency for AD.

• 出版日期2015-3