SBIR-STTR Award

Precision weapons using Ladar seekers with automatic target recognition (ATR) will be challenged by an enemy’s efforts to hide, and it will be necessary for these Ladar systems to have enhanced target detection capability. The design features that show the most potential for providing enhanced “hidden” target detection include: multi-pulse detection logic and the use of polarized multi-spectral laser transceivers. The more ways a sensor samples the environment, the harder it is to effectively camouflage an object. However, more complex sampling leads to a greater volume of data, and more dimensionality. This generally means that the discrimination / recognition problem requires more technical sophistication. In recognition of this fact the Air Force Research Lab has expressed a desire to apply recent advances in non-linear dimensionality reduction analysis to the multi-discriminant Ladar ATR problem. The objective of this proposed research is to exploit the use of multi-discriminant Ladar and non-dimensionality reduction analysis to develop algorithms for improving the detection of hidden objects and targets employing camouflage, concealment, or deception. The results of this research could lead to a significant improvement in weapons system effectiveness.

Keywords:
Ladar, Multi-Discriminants, Object Detection, Automatic Target Recognition, Manifold Learning, Nonlinear Dimensionality Reduction

The objective of this work is to develop ATR algorithms that use the spectropolarimetric information in multi-discriminant ladar images to segment, detect and identify hidden targets. Hidden targets using camouflage, concealment and deception are difficult to detect with standard weapons sensors including basic ladar systems. A multi-discriminant ladar having polarized, multi-spectral, and multi-pulse features is expected to provide a significant enhancement to the target detection process, but ATR algorithms must be developed to take advantage of these multi-discriminant attributes. The Phase I effort, which used synthetic ladar images, has demonstrated several promising algorithms based on novel formulations of spectropolarimetric feature vectors used with manifold learning/clustering techniques; however, these algorithms must be refined, extended and verified by application to measured multi-discriminant ladar images, which is the goal of the Phase II program. The use of multi-discriminant LEAP ATR algorithms tailored to AFRL experimental ladar systems will facilitate the rapid assessment of the compatibility of the ATR algorithms contemporaneously with the laboratory and field tests.

Keywords:
Ladar, Multi-Discriminant, Object Detection, Automatic Target Recognition, Manifold Learning, Non-Linear Dimensionality Reduction