Applied Research In Acoustics LLC — Department of Defense SBIR Phase I: SCO182-006

Description

To support the detection and identification of high-value targets in mission-critical applications, specifically those for which there are few or no sample images, ARiA will develop and demonstrate the feasibility of LIFT (Learning Imagery for Few-Shot Training), a training-data augmentation tool for use in few-shot learning scenarios that: (1) intelligently applies image-processing functions to existing data to generate new training examples, (2) generates realistic synthetic training data given a geometrical model of the target, and (3) learns to select images that help the deep-learning model correctly classify rare targets and generalize to new environments and configurations.The Phase I effort will (1) design and develop a deep reinforcement learning framework that demonstrates a proof-of-concept system capable of improving object detection performance on objects with limited training data and the ability to improve generalizability of object detection to new environments; (2) demonstrate that LIFT can feasibly meet DoD needs through improvements in detection of rare objects within commercial high-resolution satellite imagery as assessed by performance metrics defined in Phase I; and (3) establish that LIFT can be developed into a useful product for DoD that is compatible with existing decision chains and workflows across multiple ISR systems and interfaces.