Abstract:

In the anomaly detection for hyperspectral image, the false rate is higher because that the original data sources can not correctly represent the distribution of the background data. To overcome this problem, a new weighted KRX algorithm based on target orthogonal subspace projection (OWKRX) is proposed. Starting with the estimation of background covariance matrix, the algorithm projects each pixel into the target orthogonal subspace, and self-adaptively gives each pixel a proper weight, thereby, diminishes the influence of background estimation because of the existence of target information. Numerical experiments were conducted on real hyperspectral images collected by AVIRIS. The results prove that the proposed algorithm outperforms existing algorithms, obtains better detection effect and lowers false rate.

Key words: information processing, hyperspectral image, endmember extraction, orthogonal subspace, kernel RX algorithm