DOM detection of water based on fiber SPR sensors and multi-classifiers

Abstract: Dissolved organic material (DOM) has posed adverse impacts on the detection of water quality between different water pollutants. Once the total amount of DOM reaches a critical level, the explosive growth of algae can be induced by eutrophication in the water, leading to a more complicated composition. There is a more serious interference in the detection, as the DOM aggravated during this time. The previous research also shows that the effect of DOM is closely related to the total amount, and the components. It is a high demand to accurately measure the DOM components for effective water quality monitoring. Particularly, the DOM component measurement is highly required to effectively implement, due to the complex organic structure. For this reason, it is difficult for a single sensor to complete the complicated test of the total amount and components of DOM in water. In this study, the fiber sensing array was proposed to detect the DOM components using the non-specific selectivity of the fiber SPR sensor. The cross-sensitivity analysis was carried out to obtain the different SPR sensing arrays using the fiber SPR sensor. A field test was been realized by the SPR sensing array in large-scale water bodies. Particle Swarm Optimization (PSO) was selected to optimize the artificial neural network (ANN). As such, effective predictions were obtained for the five DOM components and their concentrations in four kinds of measured water. The SPR sensors were then prepared with different optimal refractive indices using multimode fiber and gold film with seven thicknesses of 55-85 nm. The optimal refractive index of each sensor was effectively distributed in the range of 1.33 to 1.43, according to the design requirements. Correspondingly, each sensor presented excellent sensitivity and linearity in the best measurement interval. The sensitive crossing-response was achieved in the measurement interval corresponding to other sensors through the wavelength, spectrum width, and light intensity. In terms of the classifier and intelligent algorithm, the global search PSO was used to train the BP-ANN, in order to avoid the local search easy to fall into the local extremum. After that, the DOM water sample was prepared to determine the DOM components in the water body. The SPR effect was realized to measure the refractive index using a sensing array. The artificial intelligence network BP-ANN was trained by the PSO. Three classifiers were then constructed, including the PSO-BP (wavelength), PSO-BP (spectral width), and PSO-BP (light intensity). The comprehensive training was verified by the resonance wavelength, spectral width and light intensity of the SPR effect in the tested water. Therefore, five DOM components were tested, including the tyrosine proteins, tryptophan proteins, fulvic acid, soluble microbial metabolites, and humic acids of Outer Canal (A), Hongze Lake (B), Park Landscape Lake (C) and Campus Landscape Lake (D). The highest recognition rate was up to 85% from the samples of P2.n in Hongze Lake (B), indicating the excellent prediction of DOM components. Anyway, the PSO-BP multi-classifiers can be expected to mine the cross-sensitivity information by the SPR sensor. The finding can provide a new idea for the application of fiber SPR sensors and multi-classifiers using cross-sensitivity analysis.