A novel sensor system for food authentication is presented, which is based on computer vision and pattern recognition. The sensor system uses a smartphone to generate a sequence of light with varying colours to illuminate a food sample, and uses the smartphone camera to receive reflected light by way of recording a video. The video is processed using computer vision techniques and transformed into sensor data in the form of a data vector. The sensor data is analysed using pattern recognition techniques. The locally weighted partial least squares regression method is extended for classification to improve the modelling effectiveness and robustness. The sensor system is evaluated on the task of authentication of olive oil and milk – to verify how they are labelled. Large quantities of olive oil and milk were purchased from supermarkets, and sensor videos were created using the sensor system. Test accuracies of 96.2% and 100% were achieved for olive oil and milk authentication respectively. These results suggest the proposed sensor system is effective. Since the sensor system is built in a smartphone, it has the potential to serve as a low-cost and effective solution for food authentication and to empower consumers in food fraud detection.
Bibliographical noteFunding Information:
This work is supported by Ulster University Global Challenge Fund - Cloud Agri-Food Safety (70642R, PI Hui Wang). Hui Wang came up with the video-based imaging spectrometer concept used in this work, Jordan Vincent wrote the Android app, Weiran Song and Nanfeng Jiang conducted the experiments. Weiran Song wrote the first draft of the manuscript, and other co-authors reviewed the manuscript.
© 2019 Elsevier B.V.
Copyright 2019 Elsevier B.V., All rights reserved.
- Food authentication
- Partial least squares
- Smartphone based spectrometric sensor