Hyperspectral Algorithm Development for Military Applications: A Multiple Fusion Approach

Developments of long range target detection techniques have been a military priority, especially against very low observable targets like deeply hidden vehicles. Not surprisingly, the need for the development of a highly efficient passive detection system with sensitivity capable to detect extremely low cross-section objects like mines is even more demanding. In this paper, we summarise the achievement of a 3-year DTC hyperspectral algorithm development (HAD) project for the research of a near-real time long range detection technique using hyperspectral remote sensing technology. The main emphasis of this project has been on the detection of low contrast targets. One of the main deliverables from the project has been the demonstration of prototype software, which integrates various components developed within the programme, into a more robust, and, high performing anomaly detector. Unlike conventional multiple approach fusion [MAF] technique, this prototype fuses several detectors in the algorithm level as well as in the detector output. This mixture modelling and spectral unmixing fusion in 2-level [MUF2] algorithm, was found to be extreme efficient with demonstrated detection performance better than a conventional anomaly detector by orders of magnitude, even in its prototype form.. The nominal processing time of MUF2 for a 20K pixel imagery takes ~2minutes by using a 2.5G PC on the Matlab platform.