Terahertz Microfluidic Metasensors for Precise Determination of Organic Aqueous Solutions Based on Wavelet-Assisted Convolutional Neural Networks

Precise and nonlabel determination of organic aqueous solutions in terahertz (THz) bands is essential for chemical analysis and process optimization, where the picosecond-scale dynamics of hydrogen-bond network in organic aqueous solutions is involved. However, the weak interaction between the analyte and THz waves, coupled with the strong absorption of THz waves by liquid samples, significantly limits the development of high-sensitivity detection of liquids in THz bands. Here, by the device fabrication of microfluidic-integrated metasurfaces (MIMs), we effectively confine the liquid within a restricted space and strengthen the interaction of THz waves with analytes. The sequential extraction approach to the temporal data displays tiny changes in the effective refractive index of the MIMs. Results from the wavelet transform algorithm and the oscillator-Debye field model reveal that the changes of the hydrogen-bond networks in solutions significantly influence the properties of localized surface plasmons, demonstrating the coupling of the localized field energy to the liquids and a critical state in the hydrogen-bond networks. Moreover, a wavelet-assisted convolutional neural networks method was developed for precise classification of organic aqueous solutions, enabling the successful detection of 20 μL water in 1 mL organic solutions. Our work demonstrates that based on the MIMs platform, traditional THz spectroscopy exhibits great potential in the precise detection of organic aqueous solutions. The underlying physical implications offer valuable references for understanding LSPs-liquid interactions and the dynamic characteristics of water-organic mixtures.