Maritime Target Automatic Target Recognition from Inverse Synthetic Aperture Radar (ISAR) Utilizing Machine Learning Navy SBIR FY2018.1 Sol No.: Navy SBIR FY2018.1 Topic No.: N181-029 Topic Title: Maritime Target Automatic Target Recognition from Inverse Synthetic Aperture Radar (ISAR) Utilizing Machine Learning Proposal No.: N181-029-0441 Firm: Systems & Technology Research 600 West Cummings Park Suite 6500 Woburn, Massachusetts 1801 Contact: Ron Dilsavor Phone: (937) 310-3240 Web Site: http://www.STResearch.com Abstract: Radio frequency missile seekers operating over the ocean against maritime targets must identify specific ship classes on a tight timeline while addressing the inherent complexity of low altitude over-ocean RF propagation and 6-DOF ship motion that occurs during the coherent processing interval of a range-Doppler image. Our Phase 1 program will develop a practical machine-learning based ISAR ATR system that addresses these challenges. The system will be designed to achieve high target recognition and confuser rejection performance, to be robust to target state and environmental conditions, to be easily extensible to classification of new targets, and so that online processing may be implemented on low SWAP hardware to meet demanding smart munitions timelines. To create this design, the Phase 1 effort will integrate and utilize high fidelity RF propagation and scattering simulations to drive trade study analyses that will identify deep learning architecture attributes, model training strategies, and sensing and signal processing approaches that address seeker-based ISAR target recognition challenges in maritime environments. Benefits: Our ISAR ATR design for missile seekers operating against maritime targets will achieve high accuracy using modern machine learning techniques and large training data sets generated via cost-effective high-fidelity signature simulation instead of expensive measured data collections. The design of the simulation engine uses best-of-breed models of maritime RF propagation and scattering as seen by low altitude cruise missile seeker geometries. It achieves fast online operation through single-pass implementation on embedded GPU-based processors; replacing the expensive database search required by classical template-based approaches. The design supports rapid addition of new targets using transfer learning techniques and includes special processing to reject confuser vessels that were not seen during training. Our commercialization approach pursues direct Government transitions through the Navy, STR product line enhancements, and co-funded partnership agreements with OEM system vendors working in the security and law enforcement markets. Return