An optimized generic framework for multi-gas classification in electronic nose applications

Electronic nose has been widely used in sensing and classification of different gases in a number of safety and security related applications. Within the whole pipeline, various techniques have been proposed for data analysis and classification, leading to different results and even inconsistent findings. To tackle this challenge, we present in this paper an established Optimized Generic framework for Multi-Gas Classification (OGEGC). Rather than focusing on the improvements on one or two certain modules, we aim to deriving an optimal roadmap in OGEGC by comprehensively benchmarking various technical modules in the pipeline, including down-sampling, baseline correction, channel selection, and classification. To address the high dimensionality and redundancy inherent in multi-channel electronic nose signals, a channel selection module is incorporated to reduce the data by 30%-70% while maintaining the classification accuracy. Through a comparative study of various machine learning and deep learning models, Deep Convolutional Neural Networks (DCNN) was found as the best among a few other evaluated methods, achieving accuracies of 98.01% and 96.14% on two datasets, respectively. After applying Classifier Guided channel selection(CGCS), the classification accuracy and computational efficiency have been further improved.