Enhanced Long-Range Maritime Vessel Classification

RDRTec has made significant advancements in the automated classification of ships at long ranges using real time feature extraction from Inverse Synthetic Aperture Radar (ISAR) imagery. RDRTecÆs Maritime Classification Aid (MCA) can classify a ship to its fine naval class level using ISAR data collected at long ranges in all environments. While physical dimensions of major structural elements of the ship provide the primary features that feed the classification expert system, micro-Doppler based signatures associated with rotating antennas have been shown to provide important additional information to support separation between ship classes with similar physical features. The objective of this project is to improve classification performance through the exploitation of micro-Doppler signatures by (1) identifying ship structural vibrations due to movement of the ship through water and the ships power plant and (2) expanding the exploitation of rotating antenna signatures. In addition to improving classification of ship types that are difficult to separate from similar ship types (confusers), we will explore the use of vibration induced micro-Doppler artifacts for battle damage assessment, feature aided tracking, and fingerprinting of a specific hull. RDRTec proposes to continue to attack this topic in collaboration with a diverse set of partners to examine the problem from multiple viewpoints, leverage a broad range of expertise, and increase the likelihood of transitioning the results to the US war fighter. This team, successfully united in Phase I, is comprised of: RDRTec (small business research prime), Duke University (research institute for this topic), and the University of Michigan (additional subcontractor). Dr. Nickolas VlahopoulosÆ group at the Univ. of Michigan will apply Finite Element Analysis (FEA) methodsĀwhich he developed for vibro-acoustic analysis of complex Naval systems to generate simulated vibration data setsĀfor ship cases to use in assessing the feasibility of the extraction of micro-Doppler signatures and training machineĀlearning-based classifiers.Ā RDRTec will generate simulated ISAR returns from those datasets. Resulting radar measurements will be processed at RDRTec using modified versions of physics-based algorithms currently used to extract rotator micro-Doppler artifacts. Using physics-informed data-driven machine learning techniques, Dr. Jeffrey KrolikÆs group at Duke University will use a multi-time-scale neural network (NN) using micro-Doppler signatures from vibrating and rotating shipboard machinery for vessel classification. This will allow us to examine the feasibility of both physics based expert systems and machine learning approaches for exploiting vibration and rotator induced micro-Doppler.