Model-Based Fusion of Multiple Look SAR for Automatic Target Recognition
Agency / Branch:
DOD / USAF
We propose to develop and evaluate robust model-based approaches to combat ID by fusing multi-look SAR imagery of ground vehicles. We have already developed a baseline decision-level multi-look fusion approach, based on the MSTAR system, that accumulatesevidence over target type. Extensive evaluation of this multi-look system has indicated significant target identification performance benefits. Under this proposed effort we will develop, implement, and analyze several improvements to the decision-levelfusion strategy: (1) hypothesis-level fusion, where we accumulate evidence not only over target type but also of target pose, thereby ensuring consistent interpretation across all the images; and (2) feature-level fusion, where we accumulate evidenceover parts of the model, thereby correctly accounting for model region visibility across the multiple views. As we increase the fidelity of the multi-look fusion approach, we also require finer image registration requirements. To support accurateregistration we propose to apply our hierarchical pixel/feature/region registration algorithms, which have proved to be effective on related applications. In order to analyze the performance tradeoffs of the different multi-look approaches and understandtheir benefits and limitations, we will perform extensive analysis on available in-house multi-look MSTAR SAR imagery covering a broad range of operating conditions.The technology developed under this program will contribute directly to the overallmilitary objective of improving automatic combat identification from SAR imagery for a variety of targets and under a variety of conditions. Specifically, multiple look fusion will improve identification performance, reduce false alarm rate, increaserobustness against variabilities of target and collection conditions, and achieve fine discrimination among similar targets. We anticipate that these methods could be used to automate SAR peacetime applications such as treaty compliance assessment andmonitoring. Moreover, model-based multi-look fusion techniques could be used for law enforcement applications as well as disease detection and diagnosis in 2D medical imagery.
Small Business Information at Submission:
50 Mall Road Burlington, MA 01803
Number of Employees: