Track: Graduate Student Paper Competition

Abstract

The ever-growing link between food waste and climate change has led the European Commission (EC) to introduce Industry 5.0 framework, aiming to meet the Sustainable Development Goals (SDGs) by fostering sustainable, human-centric, and resilient systems. Along these lines, Artificial Intelligence (AI) is considered a blueprint for this change, with great potential for the agri-food industry. To accelerate the shift towards Industry 5.0, an Edge-AI system is proposed. This system captures timely data through visual recognition of the On-Shelf Availability (OSA) index at supermarket shelves, monitoring which products are in high or low demand for efficient distribution. For this purpose, a bi-objective model is designed to: 1) identify the type of agri-food shelf detected, and 2) determine whether the OSA is fully stocked. An original dataset was compiled, comprising up to 16,000 imagery samples for these objectives. For the first one, Deep Learning (DL) was employed through a custom Convolutional Neural Network (CNN) with reasonable computation time, achieving a testing accuracy of up to 99.78% in identifying eight products. For OSA detection, Transfer Learning (TL) using the Residual Network (ResNet) yielded a testing accuracy of up to 98.76%. To blur the lines among different Agri-Food Supply Chain (AFSC) facilitators, a web platform will be developed to represent this real-time information in an intuitive manner. This platform will serve as an interactive decision-making tool to simulate “what if” scenarios. From a holistic standpoint, this approach contributes to building an equitable food security construct while upholding environmental responsibility to meet Industry 5.0 pillars.