Sustainable deep vision systems for date fruit quality assessment using attention-enhanced deep learning models.

Area of Science:

• Computer Vision
• Agricultural Technology
• Machine Learning

Background:

• Automated fruit classification is crucial for modern agriculture but challenging due to appearance variations.
• Existing methods often struggle with the nuances of fruit imagery, necessitating improved techniques.

Purpose of the Study:

• To develop and evaluate a novel deep learning model for enhanced date fruit image classification.
• To improve feature representation and model generalization for accurate agricultural product identification.

Main Methods:

• Integration of DenseNet121 pre-trained on ImageNet with Squeeze-Excitation (SE) Attention blocks.
• Utilized data augmentation and Nadam optimization for improved model performance and generalization.
• Compared the proposed DenseNet121+SE model against YOLOv8n for date fruit classification.

Main Results:

• The DenseNet121+SE model achieved 98.25% accuracy, 98.02% precision, 97.02% recall, and 97.49% F1-score.
• YOLOv8n achieved 96.04% accuracy, 99.76% precision, 99.7% recall, and 99.73% F1-score.
• The proposed model demonstrated superior performance in accuracy and F1-score compared to YOLOv8n.

Conclusions:

• The DenseNet121+SE model offers a robust and effective solution for automated date fruit classification.
• The integration of SE attention significantly enhances feature representation for improved classification accuracy.
• This approach provides a practical tool for quality control in the agricultural and food industries.