Aflatoxin detection in naturally contaminated peanuts based on vision transformer and multi-scale convolutional...

Area of Science:

• Agricultural Science
• Food Science
• Computer Science

Background:

• Aflatoxins are toxic metabolites found in peanuts, posing significant risks to human health.
• Accurate and efficient detection of aflatoxin contamination is crucial for food safety.

Purpose of the Study:

• To develop an improved model for detecting aflatoxin-contaminated peanuts under natural conditions.
• To enhance the feature discrimination capability for improved detection accuracy.

Main Methods:

• An improved 1D-MCFViT model was developed, integrating Vision Transformer with multi-scale convolutional fusion.
• Spectral curves were extracted from cleaned RGB images, and data augmentation was performed using autoencoder networks and Gaussian resampling.
• The model's performance was evaluated against traditional machine learning and deep learning models.

Main Results:

• The 1D-MCFViT model achieved 92.6% accuracy and 94.4% recall on the validation set.
• The proposed method demonstrated a 1.23% improvement in accuracy compared to the 1D-ViT model.
• The approach significantly outperformed traditional machine learning and mainstream deep learning models.

Conclusions:

• The developed 1D-MCFViT model offers a robust and accurate solution for detecting aflatoxin in peanuts.
• The data generation techniques effectively enhanced feature discrimination, leading to improved detection performance.
• This research provides a strong foundation for the development of online aflatoxin detection devices for food safety applications.