AI-Integrated Optoelectronic Platform for Low-Latency Classification of Toxic Industrial Chemicals

This study introduces an optoelectronic platform designed for high-accuracy detection and classification of Toxic Industrial Chemicals (TICs), addressing key limitations of conventional Leak Detection and Repair (LDAR) systems. The system integrates colorimetric sensor membranes (CSMs) with a 3 × 3 IGZO phototransistor array, enabling the conversion of gas-induced color variations into electrical signals for reliable TIC identification. Applying a multi-power sensing approach with three distinct laser intensities (0.3, 0.75, and 1.9 mW), 100% classification accuracy is achieved through K-means clustering, demonstrating the robustness of the sensing mechanism. In addition, a gas detection framework based on GRU modeling and vector quantization maintained 100% accuracy with reduced input conditions while reducing model size by 66.14%, supporting efficient, low-latency processing. The proposed system offers scalability, compactness, and compatibility with resource-constrained environments, representing a promising pathway for next-generation fugitive emissions management and LDAR implementation.