An Adaptive Gas Feature Calculation and Classification Method Combined With an Electronic Nose for Origin Identification of Agricultural Products

A rapid, objective, and effective method for determining the origin of agricultural products is crucial for ensuring food safety and protecting the rights of geographical indication products. In this study, we propose an adaptive gas feature calculation and classification method, integrated with an electronic nose (e-nose), to identify the origin of agricultural products. Specifically, we propose a global synthetic attention (GSA) that adaptively focus on important gas features across both the time-series and sensor dimensions. Furthermore, peak factor, integral value, and steady-state mean are calculated to enhance temporal feature representation. Gas information from rice and soybean sourced from different origins is collected using PEN3 e-nose. The average classification results of five-fold cross-validation show that the adaptive gas feature calculation and classification method (LeNet+GSA) achieves 98.71% accuracy, 98.89% precision, and 98.67% recall for rice origin identification, while attaining 98.94% accuracy, 99.13% precision, and 98.87% recall for soybean origin identification.