Preliminary hyperspectral band selection for difficult object detection

Automatic target detection has been a well-known topic since the early 1990's. With the development of digital photographic techniques it has became even more popular. So far, a lot of algorithms for artificial objects detection from natural backgrounds have been elaborated and developed. Unfortunately detecting difficult objects like military, camouflaged targets is still a hard task and it leads in many times to a big number of false alarms. The objective of this paper is to provide a comparative analysis of methods for object detection based on single hyperspectral band, two-band and multiple band. In this studies, to check the spectral contrast between objects and background, the algorithm based on mahalanobis distance has been used. All possible two-band and three band combinations have been checked and compared with single band. We analyzed data coming from hyperspectral ground based system consists of digital video camera and optoelectronic tuneable filter. Data was collected mostly in near infrared range with 10nm spectral resolution. Additional we took a challenge to classify the methods taking into account possibility of unknown object detection and detecting object with known spectral characteristic. Results obtained are interesting. They point to a need of proper band selection for difficult object detection and form the basis for the expansion of research.