Multidimensional image processing for remote sensing anomaly detection

This paper presents a unique multidimensional image processing approach for autonomous detection of anomalous materials in unknown natural clutter scenarios. Scene anomaly detection has a wide range of use in remote sensing applications requiring no specific material signatures. The approach uses a repeated multisampling scheme to characterize the unknown clutter background and the most popular anomaly detection algorithm - the Reed-Xiaoli algorithm - for scoring. The approach requires only a small fraction of the data cube to characterize clutter, it does not perform segmentation, and it is invariant to objects´ scales (i.e., relative spatial sizes of objects in the imagery). Results using real multivariate spectral data are promising for autonomous manmade object detection tasks under different atmospheric conditions.