Target Identification Using Machine Learning-Based Fusion of Hyperspectral and 3D Imagery

Target detection and identification in the Hyperspectral Imaging (HSI) domain relies on the uniqueness of spectral signatures, whereas classification in the spatial domain relies on morphological attributes of the target. Therefore, a detection and classification approach that fuses information from the spectral and spatial domains has the potential to considerably increase target detection confidence and reduce false positives. In Phase I, we will prototype a new spectral-spatial fusion processing chain that represents a significant advance over existing methods, which were developed mainly for surface and terrain classification. Target Bidirectional Reflectance Distribution Function (BDRF) and shadow effects are accounted for using advanced HSI target detection algorithms to produce provisional detection scores. Potential target-containing regions of interest (ROIs) are further identified using 3-D shape information contained in Digital Surface Models. Target detections and identifications are made with a machine learning algorithm followed by fusion with HSI scores. The Phase I work, conducted in collaboration with the RIT Center for Imaging Science, will provide a proof-of-principle for the development of deliverable software in Phase II.