An Explainable CNN and Vision Transformer-Based Approach for Real-Time Food Recognition.

Area of Science:

• Artificial Intelligence
• Computer Vision
• Nutritional Science

Background:

• Food image recognition is essential for computational gastronomy and nutritional platforms.
• Convolutional Neural Networks (CNNs) excel at local feature extraction but struggle with long-range dependencies.
• Transformer architectures are needed to capture global context crucial for accurate food classification.

Purpose of the Study:

• To develop a robust food image classification model using a hybrid CNN-Transformer approach.
• To improve accuracy by handling both local and global features in food images.
• To enhance food image recognition for precise nutritional analysis.

Main Methods:

• Utilized a hybrid model combining RestNet50 (CNN) for local features and Vision Transformer (ViT) for global features.
• Employed multi-head attention and pre-trained weights in the ViT encoder for classification.
• Integrated explainable AI techniques (grad-CAM, LIME) to ensure model transparency.

Main Results:

• Achieved superior performance over state-of-the-art methods across five diverse food image datasets.
• Demonstrated enhanced generalizability and robustness by leveraging complementary features.
• Integrated the model into a mobile application for real-time food recognition and diet tracking.

Conclusions:

• The hybrid CNN-ViT model significantly advances food image recognition capabilities.
• This research enables practical applications in personalized nutrition and healthcare.
• Highlights the potential of AI in transforming nutritional sciences and dietary platforms.