Passive System for Detection and Identification of UAVs Using Multispectral/Hyperspectral Imaging Technologies

Toyon Research Corp. proposes to develop a robust passive system for detection and identification of UAVs using multispectral imaging by integrating physics-based image- and video processing algorithms into a deep learning feature extraction, tracking, and ATR system. The proposed approach involves embedding LWIR/MWIR optics models directly into a deep neural network (DNN) architecture to improve the visual clarity and signal-to-noise ratio (SNR) of targets prior to detection and subsequent classification. Improved focusing of imagery enables more consistent feature representations of targets, resulting in higher ATR robustness to environmental conditions and improved generalization capabilities across target types. In addition, incorporation of target tracking and dynamics models enables the use of spatiotemporal features for ATR, enabling classification to novel, previously-unseen targets based on a combination of their dynamics and multispectral signatures.