Artificial Intelligence‐Enhanced “Photonic Nose” for Mid‐Infrared Spectroscopic Analysis of Trace Volatile Organic Compound Mixtures

Abstract Molecular identification of volatile organic compounds (VOCs) plays an important role in various applications including environmental monitoring and smart farming. Mid‐infrared (MIR) fingerprint absorption spectroscopy is a powerful tool to extract chemical‐specific features for gas identification. However, the detection and recognition of trace VOC gas mixtures remain challenging due to their intrinsic weak light–matter interaction and highly overlapped absorption spectra. Here, an artificial intelligence‐enhanced “photonic nose” for MIR spectroscopic analysis of trace VOC gas mixtures is proposed. To enhance the sensing performance by increasing bandwidth and sensitivity, the “photonic nose” is designed to employ coupled multi‐resonant plasmonic nanoantennas to cover MIR molecular fingerprints, coated with metal–organic frameworks as the gas enrichment layer. Low limits of detection are achieved (IPA: 1.99 ppm, ethanol: 3.43 ppm, and acetone: 9.82 ppm). With machine learning, a high classification accuracy of 100% is realized for 125 mixing ratios (IPA, ethanol: both 5 concentrations, 0–130 ppm; acetone: 5 concentrations, 0–201 ppm), and low‐deviation component concentration predictions of root‐mean‐squared error within 10 ppm are achieved for IPA and ethanol (both 0–130 ppm) under interference from 50 ppm acetone. The work paves the way for intelligent sensing platforms for environmental monitoring and smart framing.