Deep learning assisted detection of toxic heavy metal ions based on visual fluorescence responses from a carbon nanoparticle array

Abstract

Factual failure in decimating global concerns related to toxic heavy metal ions, even after a decent amount of research, calls forth the production of an inexpensive, intelligent sensing platform for fast, accurate, and automated detection of As(III), Cd(II), Hg(II), Cr(VI), and Pb(II). This is addressed herein through the development of a fluorescent sensor array with nine alizarin red S (ARS)-based fluorescent carbon nanoparticles (CNPs) through an economical and facile synthesis approach. The fluorescence responses of the array with the heavy metal ions under UV-light irradiation were digitally captured to use as features in finding out an intelligent machine learning-based computer model for toxic heavy metal detection without human mentation. Seven supervised classification algorithms were used to canvass the array data set, where augmented multi-layer perceptron (Aug-MLP) outdid the other algorithms with the help of generative adversarial nets (GANs) by artificially amplifying the data sets. The accuracy of Aug-MLP in identifying toxic heavy metal ions in laboratory samples as well as spiked river and sewage water defines the success of our approach which points towards possible automation of on-site identification. Overall, this work is a unique proof of concept of artificial intelligence (AI) integration with a visual sensor array extendable to other chemical entities.