Novel Partial Least Squares-Based Hybrid Neural Network Application for Automatic Target Recognition

This project will evaluate existing automatic target recognition algorithms, including template generation and matching, neural network (NN) -based algorithms that use basis-set signal decomposition techniques, and NN-based algorithms that use target feature extraction and image compression techniques, for synthetic aperture radar (SAR) and forward looking infrared (FLIR) sensors. This project will also explore a novel target image classification algorithm that uses the powerful partial least squares (PLS) transformation for class-discriminative target feature (signature) extraction and image data compression and an optimized neural network classifier for target signature identification. This new algorithm also shows great potential for SAR and FLIR sensor fusion at both low and high levels. A Bayesian algorithm for high-level-only sensor fusion and a fast neural network training strategy using a proprietary pre-training mechanism will also be implemented in this project. This project will be beneficiary to automatic target recognition applications such as military radar imaging systems, military and commercial surveillance systems, law-enforcement activities, automatic vehicle navigation systems, and medical imaging systems. It will also be useful in the development of other advanced data compression and sensor fusion techniques.