Automatic target detection using multispectral imaging

We propose using multispectral imaging for on-the-fly target detection and classification instead of hyperspectral imaging. We initially pose the target detection problem as a classification problem with classes identified as target and clutter. The classification data consists of multispectral observations of the region of interest, focusing on visual and infrared wavelengths. We then solve this classification problem using nearest neighbor rule, support vector machines, and maximum likelihood classification. Simulation results on real data indicate that information from a multispectral sensor can offer better performance than both single band and hyperspectral sensors, also showing that costly hyperspectral analysis performance can be attained onboard a small airborne platform such as a smart missile using cost-effective multispectral sensors.