Procymidone residue detection in Allium tuberosum using feature-level data fusion of colorimetric sensor arrays and electronic nose sensor arrays: Comparison of heterogeneous data fusion methods.

Procymidone residues threaten Allium tuberosum food safety. This study combines an electronic nose (E-nose) with a color sensor array (CSA) for qualitative and quantitative detection of its procymidone residues. By exploring architectural and performance differences between deep learning and traditional machine learning in feature-level data fusion, it achieves more precise detection of procymidone residue concentrations. Experimental results show that, vs. traditional feature-level fusion, the multi-source data fusion method based on Convolutional Neural Networks and Long Short-Term Memory Network Attention Fusion Network (CLAFNet) distinguishes low, medium, and high procymidone concentrations in Allium tuberosum samples with 100 % accuracy. Quantitative results indicate that combining Fully Connected Neural Networks (FCN) and CLAFNet achieves the best prediction of procymidone concentration. Thus, E-nose and CSA are effective tools for detecting procymidone residue concentrations in Allium tuberosum, and CLAFNet is a powerful tool for their data fusion.