Bottled Water Identification & Fraud Detection Using Spectroscopy & Convolutional Neural Network

Abstract

In this paper, a sensing system using absorption spectroscopy and convolutional neural network to identify and classify bottled water was realized and demonstrated. With proper system design and measurement method, both systematic error and random noise were mitigated. Moreover, the implemented convolutional neural network was able to identify highly collinear samples with correct prediction probability of more than 99%. Consequentially, the system could accurately identify counterfeit bottled water samples.