N0002424CS089

Small Business Innovation Research Program (SBIR) Phase I

Low-shot Algorithm for Mine Data Augmentation (LAMDA)

Advanced Artificial Intelligence/Machine Learning Techniques for Automated Target Recognition (ATR) Using Small/Reduced Data Sets

Mines are extremely inexpensive to implement, but effective at inflicting maximum damage on our Navy's assets. Thus, the Navy must make a concerted effort to ensure that its Mine Countermeasures (MCM) capabilities are sufficient to address the newest generation of mining technologies. Recently, Artificial Intelligence and Machine Learning (AI/ML) has been leveraged for Automated Target Recognition (ATR) to improve the efficiency of MCM. But training AI/ML algorithms (e.g., Deep Neural Networks) requires large amounts of data. Collecting training data across the entire range of potential mine threats, ocean bottom conditions, and sonar systems is an extremely costly and time-consuming effort. Potential solutions to this problem are: (1) make use of the latest generation of one-shot/low-shot/few-shot learning models, or (2) generate synthetic training data to augment the small amount of real data. The team of In-Depth Engineering Corporation (IEC) and Miami University of Ohio (MUOH) propose the Low-shot Algorithm for Mine Data Augmentation (LAMDA) solution a comprehensive approach to training an ATR algorithm that combines use of the latest low-shot ATR models with a robust data generation technique to mitigate the need to collect a large real-world training data set.

The goal of phase I is to establish the technical merit, feasibility, and commercial potential of proposed R&D efforts and determine the quality of performance of the small business awardee organization.

N241-025

N241-025-0617

24.1

Amanda L Shaw