An Electronic Nose Combined With DFCC-Net for Origin Identification of Mung Beans

Due to varying ecological factors such as climate, soil, temperature, and precipitation, the quality of mung beans from different origins exhibits significant differences. A fast and effective method for identifying mung beans origin is essential for protecting origin-specific products and safeguarding consumer rights. In this work, an electronic nose (e-nose) combined with a deep learning algorithm is proposed to identify the gas information of mung beans from different origins. First, gas information of mung beans in six renowned origins of China is detected using an e-nose system. Next, based on the time-series characteristics and cross-sensitivity in gas information, a Deep Feature Computing Module (DFCM) is proposed to adaptively compute the deep gas features along both the time and sensor directions. Finally, a Deep Feature Computing and Classification Network (DFCC-Net) is designed to identify the gas information of mung beans at different origins. Through visual analysis of the gas information, ablation studies, and comparison with state-of-the-art gas classification methods, DFCC-Net demonstrates superior performance, achieving an accuracy of 97.93%, a precision of 98.09%, and a recall of 98.09%. Meanwhile, the Gradient-weighted Class Activation Mapping (Grad-CAM) visualization method is employed to highlight key gas features, further validating the effectiveness of feature computation and classification by DFCC-Net. In conclusion, the integration of the e-nose system with DFCC-Net offers an effective approach for accurately identifying mung beans origins and protecting origin-specific products.