Electronic nose combines an effective deep learning method to identify the rice quality under different storage conditions and storage periods

Abstract

Temperature is a key factor affecting the rice quality during storage, and an effective method to detect the rice quality during storage period is crucial. Gas information is an intuitive reflection of changes in rice quality. In this work, a deep learning algorithm, combined with an electronic nose (e-nose), provides a rapid detection method for rice quality. First, using a PEN3 e-nose system, gas information from rice stored under two different temperatures and periods is collected. Second, a Multi-branch Self-attention Module (MSAM) is proposed to focus on important gas features, enhancing the e-nose's classification performance. Third, MSAM-Net is established to identify the rice gas information under various storage conditions and periods. Ablation analysis and comparisons with state-of-the-art gas classification methods show that MSAM-Net delivers superior performance. At a temperature of 25 ℃ and a relative humidity of 35RH %, MSAM-Net achieves an accuracy of 96.25 % and an F₁-score of 96.84 %. At a temperature of 40 ℃ and a relative humidity of 35 RH%, MSAM-Net achieves an accuracy of 97.42 % and an F₁-score of 97.64 %. In summary, the combination of artificial intelligence and gas sensing provides an effective technical approach for rice quality detection.