Deep learning enabled miniature mass spectrometer for rapid qualitative and quantitative analysis of pesticides on vegetable surfaces.

Excessive pesticide use poses a significant threat to food safety. Rapid on-site detection of multi-target pesticide residues in vegetables is crucial due to their widespread distribution and limited shelf life. In this study, a rapid on-site screening method for pesticide residues on vegetable surfaces was developed by employing a miniature mass spectrometer. A direct pretreatment method involves placing vegetables and elution solution into a customized flexible ziplock bag, allowing thorough mixing, washing, and filtration. This process effectively removes pesticide residues from vegetable surfaces with minimal organic solvent usage and can be completed within 2 min. Moreover, this study introduced a deep learning algorithm based on a one-dimensional convolutional neural network, coupled with a feature database, to autonomously discriminate detection outcomes. By combining full scan MS and tandem MS analysis methods, the proposed method achieved a qualitative recognition accuracy of 99.62%. Following the qualitative discrimination stage, the target pesticide residue and internal standard can be simultaneously isolated and fragmented in the ion trap, thus enabling on-site quantitative analysis and warning. This method achieved a quantitative detection limit of 10 μg/kg for carbendazim in cowpea. These results demonstrate the feasibility of the proposed analytical system and strategy in food safety applications.