|
|
 |
| |
|
|
|
Advanced Grocery Store Classification Using Deep Transfer Learning and CNNs |
|
|
|
PP: 667-682 |
|
|
|
doi:10.18576/isl/130317
|
|
|
|
Author(s) |
|
|
|
Walid Dabour,
|
|
|
|
Abstract |
|
|
| This paper presents a system employing advanced techniques, including convolutional neural networks (CNNs),
transfer learning, fine-tuning, batch normalization, data augmentation, and dropout, to categorize grocery store images.
Approximately 285 million visually impaired individuals worldwide face challenges in grocery shopping. The goal of this
study is to aid visually impaired individuals in their shopping tasks. Unlike previous methods, which encountered
limitations, our approach combines deep learning with data augmentation. This approach utilizes three CNN architectures
(VGG19, MobileNet, and Extreme Inception Xception) trained on the ImageNet dataset and evaluate performance using a
grocery store image dataset. The hybrid model, particularly Xception, achieves a remarkable F1-score of 98% for overall
product recognition. Xception excels in fruit recognition (98% F1-score), MobileNet in vegetables (94% F1-score), and
VGG19 in packages (97% F1-score). Our model outperforms existing methods in classification accuracy, precision, recall,
and F1-score. |
|
|
|
|
|
|
|
