Generative AI for Automatic Target Recognition via Sparse, Multimodal Sensing

The SaraniaSat team is proposing to use Generative AI for Automatic Target Recognition via Sparse, Multimodal Sensing (GATR). In this effort, SaraniaSat will partner with SRC Inc., an independent, not-for-profit research and development company in the defense, environment, and intelligence space. Both our organizations collaborate with the Information Directorate of the Air Force Research Laboratory to develop novel computation technologies for advanced sensing.ááThe performance of conventional AI/ML models is highly dependent on the availability and quality of training data, which often is problematic to obtain. The purpose of our GATR solution to address the challenges posed by sparse and low-quality data and multimodal sensor fusion, and to ultimately lower false positives and false negatives in Automatic Target Recognition (ATR) programs.ááThe current-use technology for ATR is Neural Radiance Field (NeRF), a state-of-the-art Machine Learning Algorithm for generating a 3D representation with good fidelity to the actual target or ground truth. The difficulty that NeRF encounters is that it requires a large number of input images to generate 3D models, which is highly restrictive when sparsity and low quality limit the number of images available.ááááTo overcome these obstacles, SaraniaSat will first use NeRF techniques to generate 3D models from well characterized objects similar to the desired targets. Then, using the sparse target image inputs available, Generative AI will be applied in conjunction with the inital NeRF models or Priors to synthesize completely new views of the target.áThe application of Generative AI to Synthetic Aperture Radar (SAR) images is still in its infancy. SAR images are highly sensitive to observation configurations and exhibit significant variations across different viewing angles, making it challenging to represent and learn their anisotropic features.á