Intelligent optical pH sensing via metasurface spectral encoding and machine learning

To overcome the subjectivity and inaccuracies associated with visual interpretation of pH test strips, this work proposes a compact and intelligent optical pH sensing system based on a metasurface. By employing metasurface-based random spectral encoding in combination with a neural-network-assisted reconstruction framework, accurate reconstruction of the transmission spectra of pH test strips is achieved. Leveraging the continuous correlation between color variations of the pH test strips and their characteristic spectral signatures, a three-dimensional dataset comprising superpixel intensities, reconstructed spectral data, and corresponding pH values is constructed. A fully connected neural network is then utilized to learn the nonlinear mapping from optical intensity to spectral information and pH values, enabling quantitative pH detection without the need for frequent on-site electrode calibration, a key limitation of electrochemical sensors. The proposed system provides a compact, robust, and efficient solution for rapid on-site pH analysis, demonstrating strong potential for portable chemical sensing applications.