A deep learning method combined with an electronic nose for gas information identification of soybean from different origins

In the soybean market, it is a common phenomenon that low-quality soybeans replace high-quality soybeans. At the same time, due to the influence of the natural growth environment, the quality of soybeans produced in different origins is different. An effective method for detecting the quality of soybeans should be proposed. In this work, a deep learning method is proposed to classify the gas information of soybeans from different origins, providing an effective detection method for the quality supervision of the soybean market. Firstly, a PEN3 electronic nose (e-nose) is used to collect the gas information of soybeans from five different origins. Secondly, a multi-kernels channel attention mechanism (MKCAM) is proposed to focus on the key features of deep gas information. Thirdly, this paper proposes a residual dense block-MKCAM (RDB-MKCAM) to avoid feature degradation, improving feature expression capability. The results show that compared with other gas information classification methods, RDB-MKCAM has achieved 97.80% accuracy, 97.86% precision, 96.82% recall, and 97.33% F1-score, obtaining the best classification performance. In conclusion, RDB-MKCAM combined with a gas sensor array effectively realizes the classification of soybean gas information from different origins and provides an effective detection method for the quality supervision of the soybean market.