Robust Feature Extraction and Sensor Fusion for Land Mine Detection

The overall goal of this effort is to develop an automatic mine detection algorithm based on fusion of information from multiple sensors. The sensor data requirement for the fusion process can be achieved under realistic operating conditions. Under PhaseI, we will focus our algorithm development on the GPSAR and acoustic data. The algorithm will be extended to incorporate other sensor platforms under Phase II. The fusion procedure differs from other decision level fusion methods in one important aspect.Under the new procedure, the individual classifiers are fused dependent on other classifiers during the training phase, whereas other methods combine the classifiers independently of one another. The end product of the Phase I effort will be software codethat implements the algorithms.The project team consists of SSCI, Planning Systems Inc. (PSI) and the University of Mississippi. SSCI brings to the team expertise in developing ATR capabilities for landmine detection technologies. PSI and the University of Mississippi will assist in theATR development through their insight into the workings of the sensors. The participation of PSI and the University of Mississippi also ensures a smooth transition of the techniques developed under this project. Immediate benefits arising from the researchare in the improvement of mine detection performance and speed of advancement for the future mine detection systems. The automatic nature of the detection process also help lower training cost and facilitate wide spread use of the new systems. Commercialapplications for the developed technology can be realized on any products that require automatic object recognition such as surveillance, land-use surveys, and resource management.