MORSECORP, INC — Department of Defense SBIR Phase II: SOCOM224-D001

Description

MORSE proposes Human-machine Ensembling for Track Entropy Exploitation, HEFTEE, a revolutionary, scalable, state-of-the-art machine learning-based system to automatically identify and combine the appropriate mix of sensors for an objective area. HEFTEE delivers de-duplicated, accurate, continuous tracks for unique entities, producing a clean and confident view of mission areas for SOCOM analysts. HEFTEE leverages MORSE’s established machine-learning ensembling techniques developed for Project Maven to combine the outputs of all the intelligent sensor systems. These proven techniques and solutions support live drone video, sporadic ELINT, periodic, and sporadic HUMINT. They are lightweight, with minimal processing overhead (in real-time, 30Hz) and can be deployed on CPU-only systems. The solution includes: Data Sources, Integration: Large sets of labeled synthetic and real data matching sensor characteristics of SOCOM track data will be generated to train ML models. • Stage 1: Identify redundant tracks that represent the same single physical object. • Stage 2: Consolidate and combine redundant tracks on the same physical object. • Stage 3: Post-Process tracks, resampling and connecting gaps. • The end result is a vastly improved user experience that enables analysts to move quickly and better focus on mission objectives. HEFTEE will easily scale by integrating new sensors, identifying important aspects of each sensor and learning to combine them in the most optimal way.