Deep Learning-Based Food Image Classification Using Enhanced CNN Architecture and Data Augmentation Techniques

Abstract

Automated food image classification has emerged as a critical research area due to its applications in dietary monitoring, calorie estimation, and personalized nutrition. However, the variability in food appearances, including high intra-class variability and inter-class similarity, poses significant challenges. This study proposes an optimized deep learning-based solution using a custom convolutional neural network (CNN) architecture to address these challenges. The model incorporates advanced techniques such as batch normalization, dropout, and data augmentation to improve generalization and mitigate overfitting. The proposed model was evaluated on the widely used Food-101 dataset, consisting of over 100,000 images across 101 food categories. The images were preprocessed through resizing, normalization, and data augmentation to enhance diversity and robustness. The model was trained using the Adam optimizer with a learning rate of 0.001, and its performance was assessed using accuracy, precision, recall, and F1-score metrics. Results indicate that the proposed model achieved an accuracy of 93.2%, outperforming benchmark architectures such as VGG-16 (87.6%) and ResNet-50 (91.0%). It also demonstrated a balanced performance with a precision of 92.4%, recall of 93.1%, and an F1-score of 92.7%. These results underscore the effectiveness of the proposed architecture in capturing complex visual features inherent in food images. While the model performs exceptionally well, challenges remain in classifying visually similar food categories. Future work aims to address these limitations by integrating attention mechanisms and multi-modal data to enhance classification accuracy and real-world applicability. This research contributes to advancing the field of food image classification with robust and scalable solutions