There are approximately 285 million visually impaired people worldwide. People with visual impairments have difficult performing shopping activities. In this paper, we proposed an image classification method that uses a convolutional neural network (CNN), transfer learning, fine tuning, batch normalization, data augmentation, and dropout to develop a system to classify multiclass images of grocery stores. The proposed method is intended to help visually impaired people perform shopping tasks.Several deep learning algorithms have been developed to solve the problem of grocery store product identification and classification. Such techniques, however, have limitations. As a result, this paper utilizes deep learning in combination with the concept of data augmentation, which is based on three well-known Convolutional Neural Network (CNN) architectures (Visual Geometry Group (VGG19), MobileNet, and Extreme Inception (Xception),ImageNet was used to train these three models. The suggested approach's performance is evaluated using the Grocery store dataset. Compared to this research's current techniques, The proposed method was achieved using a hybrid model illustrated as follows; on the base model, the Xception proved to be the best model, with an F1-score of 98% on the testing set. The Xception network was the most effective for the fruits model, obtaining an F1-score of 98% on the testing set. For the vegetables model, MobileNet was identified as the best model, obtaining an F1-score of 94% on the testing set. For the packages model, the VGG19 network obtained the best results, demonstrating an F1-score of 97% on the testing set. The proposed model outperformed existing state-of-the-art models in terms of classification accuracy, precision, recall, and F1-score.

convolutional neural network; data augmentation; transfer learning; fine-tuning