Curvelet-based image fusion algorithm for effective anomaly detection in hyperspectral imagery

Anomaly detection is one of the most important applications for hyperspectral imagery. However, some technical difficulties haven't been effectively solved so far, such as high data dimensionality and high-order correlation between spectral bands. In this paper, a new curvelet-based image fusion algorithm is proposed for effective anomaly detection in hyperspectral imagery. In the proposed algorithm, the original hyperspectral data are firstly partitioned into several subspaces according to the correlation between spectral bands. In each subspace, curvelet transform is used to decompose hyperspectral images at different resolution. The decomposed coefficients are transformed by kernel principal component (KPCA) at corresponding frequency range from different images of same subspace. Using those first principal components transformed by KPCA at different frequency range is to reconstruct the fused image. The same operation is performed in each subspace, and those fused images are obtained to used in final detection with RX detector. The experimental results show that the proposed algorithm modifies the performance of the conventional RX algorithm.