Automated detection and recognition system for chewable food items using advanced deep learning models.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Signal Processing

Background:

• Food identification via eating sounds is crucial for allergy management, dietary restrictions, and cultural understanding.
• Existing methods lack robust accuracy in distinguishing food items based solely on auditory cues.

Purpose of the Study:

• To develop and evaluate a novel deep learning methodology for accurate food identification using eating sounds.
• To explore the efficacy of various deep learning architectures and feature extraction techniques for audio-based food classification.

Main Methods:

• Collected and analyzed 1200 labeled audio files across 20 food items.
• Employed signal processing techniques (spectrograms, MFCCs) for feature extraction.
• Trained and hybridized deep learning models including GRU, LSTM, InceptionResNetV2, CNN, Bidirectional LSTM+GRU, RNN+Bidirectional LSTM, and RNN+Bidirectional GRU.

Main Results:

• The Gated Recurrent Unit (GRU) model achieved the highest accuracy at 99.28%.
• Hybrid models like Bidirectional LSTM+GRU and RNN+Bidirectional LSTM demonstrated strong performance with accuracies of 97.7% and 97.45%, respectively.
• All evaluated deep learning models showed significant potential in associating specific sound patterns with food classes.

Conclusions:

• Deep learning models are highly effective for identifying food items based on their eating sounds.
• The proposed methodology offers a promising approach for applications requiring automated food recognition.
• Further research can explore larger datasets and diverse food categories for enhanced generalization.