Tools for Test & Evaluation on Incompletely-Labeled Satellite Imagery Datasets

The Department of Defense (DOD) is seeking innovative solutions through the Small Business Innovation Research (SBIR) program for the development of tools aimed at evaluating AI/ML models on incompletely-labeled satellite imagery datasets. The primary objective is to create a flexible system that can reliably estimate the performance of AI/ML models in the presence of label noise, which is a common issue in overhead imagery datasets, thereby enhancing confidence in model performance post-deployment. This opportunity is particularly significant for organizations like the National Geospatial-Intelligence Agency (NGA) that require robust evaluation methods for various computer vision tasks, including object detection and automated target recognition. Interested parties should note that the solicitation is currently in the pre-release phase, with the open date set for April 23, 2025, and a close date of May 21, 2025, by which applications must be submitted. For further details, applicants can visit the DOD SBIR website at https://www.dodsbirsttr.mil/topics-app/.