Electronic Nose System Combined With a Deep Learning Network (AGFC-Net) to Identify Quality Adulteration of Mung Beans

Fresh mung beans are rich in nutrients and typically command higher prices than aged beans. Detecting adulteration between fresh and aged mung beans is therefore of significant importance in areas such as food safety, market regulation, and consumer rights protection. In this work, an electronic nose (e-nose) system combined with a proposed adaptive gas feature classification network (AGFC-Net) is used for the identification of mung bean adulteration. First, six adulteration scenarios are created by mixing fresh and aged mung beans in varying proportions, and gas data are collected using the e-nose system. Then, a gas feature selection module (GFSM) is designed, incorporating a spatial attention mechanism to compute deep gas features along both the time-series and sensor directions. Finally, AGFC-Net is constructed by integrating a residual dense block (RDB) with GFSM to classify the adulteration conditions of mung beans. Compared with state-of-the-art gas information classification methods, AGFC-Net achieves an accuracy of 98.50%, a precision of 98.55%, and a recall of 98.39%. These results demonstrate that the combination of the e-nose system and AGFC-Net offers an effective and reliable approach for ensuring fair trade in the mung bean market and safeguarding food safety.