Squeeze-Excitation Attention-Guided 3D Inception ResNet for Aflatoxin B1 Classification in Almonds Using...

Area of Science:

• Food Science
• Analytical Chemistry
• Machine Learning

Background:

• Almonds are nutritious but susceptible to aflatoxin B1 (AFB1) contamination, posing health risks and economic challenges.
• Current detection methods for AFB1 in almonds are often destructive or inefficient.
• Rapid, non-destructive methods are crucial for ensuring almond food safety and supply chain integrity.

Purpose of the Study:

• To develop and validate a fast, non-destructive method for detecting AFB1 contamination in almonds.
• To utilize hyperspectral imaging (HSI) combined with advanced 3D deep learning for precise AFB1 identification.
• To enhance the accuracy and efficiency of AFB1 detection in almonds for industrial applications.

Main Methods:

• Development of an attention-guided Inception ResNet 3D Network (AGIR-3DNet) for HSI data analysis.
• Integration of multi-scale feature extraction, residual learning, and attention mechanisms within the 3D deep learning model.
• Comparative analysis of AGIR-3DNet against conventional machine learning models and a 3D Inception network.

Main Results:

• AGIR-3DNet achieved a validation accuracy of 93.30%, an F1-score of 0.94, and an AUC of 0.98.
• The proposed model demonstrated superior performance compared to conventional machine learning and other deep learning architectures.
• The model exhibited enhanced processing efficiency, indicating suitability for real-time applications.

Conclusions:

• AGIR-3DNet offers a highly accurate and efficient non-destructive method for detecting AFB1 contamination in almonds using HSI.
• The developed deep learning approach significantly improves upon existing detection techniques for food safety applications.
• This technology holds promise for real-time industrial implementation to safeguard the almond supply chain.