Multi-Modal Detection of Chemical Threats Using Deep Learning

This SBIR Phase I project proposes development of deep learning based chemical threat detection and localization framework that exploits multiple sensor data streams including imaging and chemical sensors. The proposed technologies combine deep hypernetworks and reinforcement learning techniques to fuse sensor data streams in an online fashion. Novateur team will perform training of deep networks by collecting data in realistic scenarios and by performing simulation in a variety of configurations including different type of chemical threats with environmental factors into consideration. The proposed effort will build upon Novateur teams in the areas of deep learning networks for detection in visual sensors, sensor data fusion, sensor localization in unknown environments, optimization and development of deep learning algorithms for SWaP constrained mobile platforms and CBRNE.