A Novel Approach to Toxic Gas Detection using an IoT Device and Deep Neural Networks

Smoking remains one of the top 3 causes of illness in the US, one of top 5 causes of fire hazards in a home and is the single most preventable cause of illness and premature death in the US. The use of Deep Neural Networks in tandem with advances in the sensitivity of gas sensor technology can enable detection of cigarette/Vape smoke much sooner and with much higher accuracy than conventional smoke/carbon monoxide detectors used today. We report a hardware demonstration and prototype that engages Classifiers to not only discriminate cigarette/vape emissions from other sources of smoke and carbon monoxide such as burning coal, wood or food - typically not possible with conventional smoke detectors, but also to accurately detect cigarette/vape smoke produced in a large space (7K ft3) from a single cigarette when concentrations of component gases of cigarette smoke are extremely low. Our prototype also demonstrates the opportunity to classify and discriminate different levels of toxicity and flammability for spaces used by different people enabling AI based sensing at scale where besides accuracy and speed, the utility of larges sets of sensor nodes are critical to enabling cost advantages over conventional detector technology.