Machine learning-enhanced flexible metamaterial biosensor for pesticide qualitative and quantitative analysis

In recent years, Terahertz (THz) biosensors have achieved substantial and rapid development through the exploration of various electromagnetic modes, materials, and the combination of integrated microfluidic designs. However, existing THz biosensors are still mainly focused on the directional detection of single or specific substances, and most ultra-high sensitive sensing accuracies are inseparable from the targeted calibration of analytes. In practical applications, we often face the challenge of detecting multiple substances. The simultaneous qualitative and quantitative detection of multiple components in complex samples such as pesticide residues is still a gap in current research. Here, we proposed a flexible umbrella-shaped metamaterial (USM) biosensor and conducted modeling analysis for six pesticides (including 2,4-dichlorophenoxyacetic acid, carbendazim, thiophanate-methyl, 2-(4-thiazolyl)benzimidazole, Triazolo, and tebufenzide), simultaneously. To improve the contrast of the sensor, we introduced triple-resonance sensing based on high-order modes, taking into account both the resonance depth and the Q-factor. During the model-building process, we adopted a two-stage cascade ensemble learning framework, which significantly improved the accuracy, scalability, and convenience of qualitative and quantitative sensing results. The qualitative accuracy of the six pesticides reached up to 100%. For the quantitative analysis, the root mean square error (RMSE) was only 2.21 μ g, and R 2 reached 0.9952, showing excellent sensing performance. This study provides an alternative solution for the qualitative and quantitative detection of multiple pesticides through multi-resonance sensing, pesticide fingerprint, and data-driven analysis.