We consider the generalized robust PCA problem, for more details please refer to the [slides].

Demo: Target detection in Hyperspectral imaging

We have here a demo showing the performance of the proposed algorithm on the Indian Pines dataset. The video shows the recovery of class 16, the stone-steel towers, for increasing values of the regularization parameter, which penalizes the sparsity of the coefficient matrix.



Interestingly, at about 5 sec. into the video, a diagonal line appears at the top right corner. This line corresponds to the rail-tracks in the scene, that is mostly composed of agricultural land, and are closely related to the spectral signatures of the stone-steel towers.