SPARC: Sparse-data Perception And Recognition Classifier

Commercial AI image analysis and object detection methods are insufficient for the U.S. Army mission, due to a lack of mission-relevant training data. Traditional labeling efforts are costly and time-consuming, and military applications require higher accuracy and robustness than general-purpose models. To address these issues, Boston Fusion Corp (BFC) proposes Sparse-data Perception And Recognition Classifier (SPARC), an automated object detection framework for EO, IR, and/or SAR images. SPARC will reduce the workload on human analysts and provide faster actionable information to decision-makers by providing reliable, accurate and explainable object detection. The system will leverage BFC’s expertise in few-shot machine learning techniques, uncertainty tracking, and AI explainability to support military analyst workflows, and enhance user adoption and trust. SPARC will utilize a multi-algorithm approach with a small-target object detection model in parallel with a large and medium object detection model. Additionally, SPARC will leverage an ensemble of algorithms for feature embedding and clustering to enable refinement of object classification and confidence values. SPARC leverages the Boston Fusion Engine (BFE), an AI-driven, sensor-agnostic fusion framework, developed through multiple DoD projects for rapid testing and training of ML pipelines. Integrating with Project Linchpin, SPARC will meet the Army's need for efficient, accurate, and agile image analysis systems that can operate effectively with limited labeled training data while maintaining the necessary performance metrics for personnel safety and mission success.