Machine Learning-Assisted Detection of Phosgene and Acetyl Chloride via a Dual-Probe Fluorescent Platform with Differential Reactivity

Phosgene and acyl chlorides are highly toxic chemicals that pose serious threats to human health and environmental safety, yet their rapid and reliable detection remains a major challenge due to their high reactivity and environmental interferences. Herein, we report the rational design and synthesis of two donor-π-acceptor (D-π-A) fluorescent probes, <b>TPA-APPA</b> and <b>TPA-HPO</b>, which incorporate excited-state intramolecular proton transfer and hybridized localized and charge-transfer characteristics. These probes exhibit fast, highly selective, and sensitive responses toward phosgene, with <b>TPA-APPA</b> showing distinct fluorescence enhancement and colorimetric changes for both phosgene and acetyl chloride at low detection limits. Benefiting from its excellent photostability, <b>TPA-APPA</b> was successfully applied to bioimaging in live cells and zebrafish. Furthermore, by integrating machine learning algorithms, including convolutional neural networks and the Swin Transformer, we developed a ceramic-fiber detection strip capable of intelligent recognition and classification of fluorescence signal variations. This hybrid system achieved high classification accuracy for toxicant vapors, demonstrating the strong potential of coupling advanced fluorescent probes with machine learning for next-generation toxic gas monitoring and bioimaging applications.