Enhanced Detection of Hidden Targets Using Multi-Discriminant Ladar

To achieve the optimal detection and classification of hidden targets, a tightly linked sensor and algorithm design effort will be required. The Phase I work described here focuses on using data from an existing sensor to explore how high-dimensional spatial, spectral and polarimetric data can be used to detect and classify hidden targets. The general approach will be to apply specific dimensionality reduction techniques to experimental data to map the data into a lower dimension feature space. In this lower dimension feature space, the statistical properties will be examined to determine the effect of the dimension reduction techniques on detectability, separability, and robustness to noise. The following dimensionality reduction techniques will be explored: Diffusion Mapping, ISOMAP and Locally Linear Embedding.