Detection in IR with Incremental Low-shot Learning (DIRILL)

Physical Sciences, Inc. (PSI) proposes to develop the Detection in IR with Incremental Low-shot Learning (DIRILL) algorithm, which integrates state-of-the-art machine-learning object detection and classification capabilities with thermal IR sensors using low-shot training techniques.  DIRILL will interface with existing Army platforms to support Aided Target Recognition (AiTR) on deployed combat vehicles with modern thermal IR sensors.  The DIRILL algorithms will be natively trained for operation in thermal IR imagery to yield optimal detection and classification performance for that sensor modality.  DIRILL will be able to accurately detect and identify targets even with a training set limited to tens of images of new targets.  The algorithm will support an easily-updated target set, which enables targets to be rapidly customized on a per-mission basis without requiring retraining on previously learned targets. DIRILL will be optimized for embedded processing platforms and support real-time detection and classification of identified targets.  The locations of detected targets will be conveyed to slewable platforms to maintain targeting as the objects move around the vehicle.  The overall system will reduce operator burden and maximize the effectiveness of deployed sensors by automatically detecting and classifying potential targets in real-time.