Toward an uncertainty-guided closed-loop synthetic data generation pipeline for UAS classification

This work introduces a synthetic data generation pipeline designed to synthesize the information gaps in UAS and bird detection and conduct desired data augmentations. By controlling various parameters such as UAS type and environmental conditions, we generate target images that closely resemble real-world scenarios. Our ongoing uncertainty analysis aims to explore the potential benefits of incorporating synthetic data for improving model generalizability, reducing confusion between UAS and avian species, and increasing robustness against adversarial attacks against existing models. We propose an explorative pipeline to guide data generation, targeting model weaknesses and advancing the state of the art in synthetic data applications across commercial and defense sectors.

SNL is managed and operated by NTESS under DOE NNSA contract DE-NA0003525