Humidity-Counteraction Identification of VOCs Driven by Semiconductor Sensor Temperature Modulation Response Data

The accurate and robust identification of volatile organic compounds (VOCs) poses significant challenges within the existing identification models. In this study, a humidity-counteraction identification scheme based on semiconductor sensor is developed for accurate and robust VOCs identification. Specifically, cyclic temperature modulation (CTM) measurements are utilized to improve sensor selectivity. The sensor response data undergo data augmentation to expand sample size and augment feature representation. Utilizing augmented samples, an accurate and robust species classification model is developed by integrating 2-D convolutional neural network (2D CNN) and generalized learning vector quantization (GLVQ), mitigating interference from concentration variations and ambient humidity changes. To counteract the ambient humidity drift for concentration calibration, the scaling factor method is developed to contract the distribution variance of sample modules. Subsequently, gas concentration calibrations are conducted using the scaled modules, showcasing remarkable accuracy and generalization performance. This study effectively mitigates the influence of ambient humidity on VOCs identification, offering a strategic and theoretical framework for developing high-performance identification models, with the expectations to play essential roles in smart medical care, intelligent driving, and industrial Internet of Things (IoT).