Plasmonic microsphere lens arrays-integrated microfluidic SERS chip for mixed pesticides identification with machine learning

The use of multiple pesticides in agriculture has caused mixed residues, complicating food safety testing. Surface-enhanced Raman scattering (SERS) offers high sensitivity and fingerprinting capabilities for pesticide detection; however,its application is limited by spot size, uneven sample distribution, spectral analysis challenges in mixed samples, and the lack of integrated automated systems for sample preparation, detection, and analysis. Hence, we developed a SERS microfluidic chip integrated with plasmonic microsphere lens arrays, which integrates optical field-modulated microspheres and plasmonic nanostructures to realise the transition of SERS detection from point-hotspots to uniform area-hotspots, achieving high-sensitivity, high-stability Raman detection under wide-field laser excitation. Moreover, a comprehensive a sample-to-answer total analysis system was established with this chip as the core, incorporating the micro-QuEChERS pretreatment module for efficient extraction of pesticides in complex matrices and the random forest-dual annealing spectral parsing algorithm for mixed spectral analysis. The results indicate that the limits of detection for 2,4-dichlorophenoxyacetic acid, acetamiprid, and thiabendazole were 1.15 nM, 0.63 nM, and 0.69 nM, respectively. The linear ranges were 10⁻⁴–10⁻⁹ M, with R² values of 0.9982, 0.9989, and 0.9996, respectively. Average recoveries ranged from 90.8 % -104.66 %, 89.1 %-107.89 %, and 89.73 %-108.4 %, respectively, with relative standard deviations below 8 %. The algorithm quantified mixed pesticides with mean absolute percentage errors of 4.96 %, 5.35 %, and 7.72 %, respectively. The system showed 3 s detection response and 15-minute analysis time. Thus, the method provides a rapid, on-site solution for detecting mixed pesticides in complex matrices by exceptional optical performance and automated processing functionality.