Aflatoxin rapid detection based on hyperspectral with 1D-convolution neural network in the pixel level.

Area of Science:

• Agricultural Science
• Food Science
• Computational Biology

Background:

• Aflatoxins are toxic metabolites produced by certain molds, commonly found in contaminated food products.
• Classified as a Group 1 carcinogen by the World Health Organization, aflatoxins pose significant public health risks.
• Accurate and efficient detection of aflatoxins is crucial for food safety and international trade.

Purpose of the Study:

• To develop and evaluate a one-dimensional convolutional neural network (1D-CNN) for the pixel-level classification of aflatoxin contamination.
• To optimize 1D-CNN parameters for maximum detection accuracy across different food matrices.
• To compare the performance of the 1D-CNN approach against traditional feature selection methods.

Main Methods:

• Utilized a one-dimensional convolutional neural network (1D-CNN) architecture for image-based aflatoxin detection.
• Optimized key 1D-CNN parameters, including epochs (30), learning rate (0.00005), and activation function ('relu').
• Validated the model's performance on peanut, maize, and mixed food sample data.

Main Results:

• The optimized 1D-CNN model achieved high test accuracies: 96.35% for peanut, 92.11% for maize, and 94.64% for mixed data.
• Demonstrated superior detection efficiency compared to traditional feature selection methods.
• Visualizations confirmed the network's ability to accurately classify aflatoxin presence at the pixel level.

Conclusions:

• The 1D-CNN model provides a highly accurate and efficient method for aflatoxin detection in food.
• This research offers a core algorithm for intelligent sorting systems, enhancing food processing safety.
• The findings are significant for grain processing industries and pre-export detoxification in foreign trade.