Using convolutional neural networks for hierarchical grocery store product classification

• Main Author: Antioquia, Arren Matthew C.
• Format: text
• Published: Animo Repository 2021
• Subjects: Computer vision; Image processing; Neural networks (Computer science); Computer Sciences
• Online Access: https://animorepository.dlsu.edu.ph/faculty_research/9103
• Institution: De La Salle University

Convolutional Neural Networks have been used to solve various computer vision problems due to its success in classifying common objects. These models are now being adapted to numerous products and devices, including visual support systems which provide assistance to people with visual impairments. These systems help by reading texts, recognizing people, and describing scenes, among others. However, these products do not have capability to provide visual support in certain scenarios performed regularly, such as grocery shopping. In this work, we adapt various modern Convolutional Neural Networks to develop classifiers for common grocery store products such as fruits, vegetables, and various refrigerated products. We train the classification architectures on a hierarchical grocery store dataset with fine-grained and coarse-grained labels. Our implementation achieves superior classification accuracy on the Grocery Store dataset, with 86.04% and 91.99% on the fine-grained and coarse-grained labels, respectively, exhibiting dominant performance against current state-of-the-art classification methods. Our code and trained models will be made publicly available upon acceptance.