Enabling Weather-Independent Gas Detection through Deep Learning on Light-Activated Sensors
Kichul Lee, Minhyun Kim, Yeongjae Kwon, Seyeon Park, Yunsung Lim, Do Y. Kwak, Jaeseok Jeong, Baul Kim, Jaewan Ahn, Jihan Kim, Yong‐Hoon Cho, Il‐Doo Kim, Inkyu Park
Abstract
Light-activated gas sensors offer a low-temperature, low-power approach for detecting target species, and their high-performance capabilities make them ideal for practical applications. The direct integration of Bi-doped In 2 O 3 nanofibers onto micro light-emitting diode (μLED) platforms enables high-performance sensors for simultaneous NO 2 and H 2 O detection. Introducing Bi into In 2 O 3 matrices facilitates the formation of oxygen vacancies and the dissociative adsorption of H 2 O, enhancing the adsorption and reactions with NO 2 . Under blue illumination, this μLED sensor system exhibits high NO 2 sensitivity, with a response value ( R g / R a ) of 264.9 at 1 ppm and 60% relative humidity and response and recovery times of less than 30 s. The use of μLEDs enhances light activation with a high energy transfer efficiency, resulting in outstanding NO 2 sensing characteristics. A convolutional neural network-based algorithm is employed to analyze transient sensing signals, accurately predicting with 99% classification accuracy and 10% regression error for both NO 2 and H 2 O, thereby demonstrating weather-independent sensing. This integration of Bi-doped In 2 O 3 nanofibers, which are specifically activated by blue illumination, μLEDs, and deep learning analytics, enables highly effective real-time environmental monitoring of NO 2 and humidity under environmentally variable outdoor conditions.