Mediterranean Food Image Recognition Using Deep Convolutional Networks.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Food Science

Background:

• Accurate food recognition is crucial for dietary assessment and understanding eating habits.
• Existing datasets may not adequately represent diverse cuisines like Mediterranean and Greek food.
• Developing robust food recognition systems requires specialized, large-scale datasets.

Purpose of the Study:

• To introduce the Mediterranean Greek Food (MedGRFood) dataset, a novel resource for food image recognition.
• To evaluate the performance of deep learning models on Mediterranean and Greek cuisine recognition.
• To provide a benchmark for future research in dietary assessment systems.

Main Methods:

• Collected and curated 42,880 food images across 132 classes from Mediterranean and Greek cuisines.
• Developed a deep learning model based on the EfficientNetB2 architecture.
• Employed fine-tuning, transfer learning, and data augmentation techniques for model optimization.

Main Results:

• The proposed deep learning schema achieved 83.4% top-1 accuracy on the MedGRFood dataset.
• The model demonstrated high performance with 97.8% top-5 accuracy.
• The MedGRFood dataset proved effective for training and evaluating food recognition models.

Conclusions:

• The MedGRFood dataset is a valuable resource for advancing food recognition and dietary assessment research.
• The EfficientNetB2-based model shows significant potential for accurate recognition of Mediterranean and Greek dishes.
• The study highlights the importance of specialized datasets and advanced deep learning techniques for culinary domain AI.