Deep Transfer Learning and Data Augmentation for Food Image Classification

The problem of food image classification has become a prominent topic that attracts many researchers due to its multiple benefits and applications in various aspects of life, from health to marketing. Image classification applications rely heavily on recent advancements in computer vision-based object recognition. In this paper, several deep transfer learning methods were investigated for food image classification. Furthermore, we applied a data augmentation approach to expand the Food-101 dataset. The impact of applying data augmentation and transfer learning was evaluated using five different deep learning models including Mobilenet, EfficientNetB1, and ResNet. It was noted that the EfficientNetB1 classifier achieved the best results with a score of 96.13%. In addition, we found that our data augmentation process was able to improve model performance.