Grocery Recognition Engine Using Convolutional Neural Network and Bayesian Inferencing

Abstract

This study presents the development and evaluation of a grocery recognition engine using Convolutional Neural Networks (CNN) and Bayesian inferencing, implemented on a Raspberry Pi platform. The system aims to automate the recognition and pricing of grocery items in supermarkets, addressing the inefficiencies of manual processes. The literature review explores the application of CNNs for image classification and the challenges of fine-grained classification in retail. The methodology involves designing and testing various CNN architectures, with the best model achieving 74.1% validation accuracy. A prototype was developed to evaluate the model with in real-world scenarios, achieving 91% accuracy. Bayesian inferencing was employed to enhance classification for items with subclasses, demonstrating the system's potential for improving grocery item recognition and pricing in retail environments. Recommendations for future work include further model optimization and data augmentation techniques.