Self-Attention Approach for Inter-class Similarities of Grocery Product Classification

Artificial Intelligence plays a crucial role throughout the ages, specifically in smart retail. However, it is still flawed when differentiating similar patterns in various products. This paper aims to classify and analyze grocery products using ResNet50 alongside a self-mechanism algorithm. The dataset is developed internally and focuses on 1 brand of a product with 8 variations representing the product’s flavor. There are a total of 4000 images which are split into train, validation, and test. An experiment is then carried out by implementing a self-attention block after the backbone of the ResNet50 architecture. The performance result is evaluated by accuracy, loss, and confusion matrix. The results of the experiment yield an accuracy value of 96.8% and loss of 2%. This indicates that the combination of deep learning has the potential in supporting classify similar products in smart retail.