Deep learning-enhanced ultra-sensitive fiber optic SPR sensor for mercury ion detection

Abstract

Mercury ions (Hg²⁺) are highly toxic heavy metal pollutants that pose significant risks to the environment and human health. In this paper, a fiber optic surface plasmon resonance (SPR) sensor is proposed for ultra-sensitive detection of Hg²⁺. The probe features Ag/ Ge₂Sb₂Te₅ (GST)/Ag composite film and metal–organic framework (MOF)-sensitized layer functionalized with probe DNA (pDNA). In the presence of Hg²⁺, pDNA interacts with target DNA (tDNA) carrying gold nanoparticles (AuNPs) in the solution through the T-Hg²⁺-T structure, immobilizing AuNPs on the sensing surface and inducing a spectral red shift. By combining GST’s high dielectric constant and MOF’s large surface area with strong pDNA affinity, the sensor achieves a maximum refractive index (RI) sensitivity of 11471 nm/RIU, detecting Hg²⁺ as low as 10 pM. Furthermore, the deep learning model based on the residual neural network (ResNet) is employed to classify and recognize the SPR spectra corresponding to six distinct concentrations of Hg²⁺, achieving a training accuracy of 99.88 % and a test accuracy of 97.5 %. Therefore, ultra-sensitive fiber optic SPR sensor with deep learning delivers high-performance sensing while enabling rapid, precise, and sensitive quantification of Hg²⁺ detection.