Deep learning-assisted multicolor fluorescent probes for image and spectral dual-modal identification of illicit drugs

Abstract

llicit drug detection is an important part of drug problem management, and existing detection strategies suffer from inefficiency and narrow scope of application, which cannot fully meet the needs of practical work. Here, a multi-color fluorescent carbon dots was synthesized in one step and used as probes for the fluorescence sensing detection of three habitual illicit drugs, heroin, ketamine and methamphetamine. The fluorescence response results were recorded in the form of image fixation and spectral detection, which contained the subtle fingerprint information of these illicit drugs. Further, a multi-modal detector based on multi-task deep learning was constructed to achieve qualitative and semi-quantitative illicit drug detection, with accuracies of 98.4 % and 84.4 % for image and 99.9 % and 99.6 % for spectral, respectively, and a broad detection range from 1 ng/mL to 1 mg/mL. The process showed that deep learning significantly enhanced the detection sensitivity and range of fluorescent probes, and the dual-modal image-spectrum recognition met the practical needs of illicit drug on-site fast detection and accurate analysis, providing a new technical solution for illicit drug detection.