Selective denoising autoencoder for classification of noisy gas mixtures using 2D transition metal dichalcogenides

Abstract

Electronic nose (E-nose) technology, which is composed of an array of chemical sensors and pattern recognition, has been widely utilized for the quantitative classification of gas mixtures. However, for the practical use of E-nose in real-industry, advanced algorithms are necessary to handle the noise in sensing data caused by various environmental variables. In order to achieve precise measurements even in real-world environments, it is necessary to denoise and classify noisy sensing data. To address these challenges, we have developed a novel deep learning approach, called selective denoising autoencoder (SDAE), which intelligently leverages both clean and noisy data gathered from real-world environments for mixed gas classification. Two-dimensional transition metal dichalcogenides (2D TMDCs) were used for the sensing channel. Raman spectroscopy, X-ray photoelectron spectroscopy, and scanning electron microscopy were used to characterize the TMDC sensing channel. Additionally, we conducted an analysis of the gas sensing properties towards NO2, NH3, and their mixtures (at ratios of 1:1, 1:2, and 2:1), and performed gas classification using our proposed SDAE model. The result achieved more than 95 % accuracy in all cases even in the noisy environment, which could be practically utilized in the industry.