Machine learning-enhanced on-chip micro-ring resonator platform for detection and recognition of low-concentration...

Micro-ring resonator (MRR) platforms based on silicon-on-insulator substrates have shown great potential for gas detection applications. However, challenges such as weak signal intensity and insufficient selectivity remain in the detection of low-concentration mixed gases. To overcome these limitations, this study proposes a machine learning-enhanced silicon nitride-based micro-ring resonator chip for the detection and recognition of methane (CH₄), carbon dioxide (CO₂), and hydrogen sulfide (H₂S) gas mixtures. By combining micro-ring resonator sensing data with machine learning models, the detection performance of the optical waveguide sensor was substantially improved. Experimental results show that the sensing chip can accurately identify CH₄, CO₂, and H₂S, with limits of detection (LODs) of 153 ppb, 184 ppb, and 83 ppb, respectively. With the aid of machine learning algorithms, the sensor achieves a classification accuracy of 91.4% in complex multi-component gas environments and can precisely determine methane concentration in unknown gas mixtures, with an average error of only 4.7%. This study not only provides an innovative solution for the detection of low-concentration gas mixtures but also demonstrates the broad application prospects of silicon photonics in the field of gas sensing.