Electronic Nose-Based a Time-Sensor Feature Classification Network for Adzuki Bean Origin Traceability

Establishing a technique for identifying the geographical origin of adzuki beans is crucial for combating origin fraud, protecting regional brands. To achieve rapid and non-destructive discrimination of adzuki bean origins, this study proposes a method that integrates an electronic nose (e-nose) with a time-sensor feature classification network (TSFC-Net). First, gas data from adzuki bean samples spanning six distinct geographical origins are collected using a PEN3 e-nose system. Subsequently, a time-sensor feature attention module (TSFAM) is proposed to collaboratively calculate temporal dynamics and sensor cross-sensitivity features, enabling the adaptive fusion of bidirectional features via soft attention. Finally, based on TSFAM, a lightweight TSFC-Net is designed to accomplish gas identification of adzuki beans from different origins. Experimental results demonstrate that TSFC-Net achieves outstanding performance in the origin classification task (accuracy: 98.52%, F1-score: 98.33%), while also exhibiting favorable lightweight characteristics in terms of parameter count (0.0152 M) and computational complexity (3.2859 M). Through ablation experiments, the rationality and effectiveness of the TSFC-Net design have been demonstrated. Compared with state-of-the-art gas classification methods, TSFC-Net achieves the best classification results and stability. This study provides an efficient and reliable technical for the practical application of an e-nose system in tracing the geographical origin of adzuki beans.