Near-infrared spectroscopy combined with pattern recognition algorithms to quickly classify raisins

Area of Science:

• Agricultural Science
• Analytical Chemistry
• Computer Science

Background:

• The proliferation of counterfeit goods, particularly in the food industry, poses significant risks to consumers and legitimate businesses.
• Adulteration of raisins with lower-quality or fake products necessitates reliable detection methods to ensure product integrity.

Purpose of the Study:

• To develop and evaluate a classification method for identifying adulterated raisins.
• To assess the effectiveness of near-infrared (NIR) spectroscopy combined with pattern recognition algorithms for raisin authentication.

Main Methods:

• Collected near-infrared spectral data from three distinct types of Xinjiang raisins.
• Applied principal component analysis (PCA) for spectral data dimensionality reduction.
• Developed and compared classification models: Support Vector Machine (SVM), Multiscale Fusion Convolutional Neural Network (MCNN), and an improved AlexNet.

Main Results:

• The SVM model achieved a perfect classification accuracy of 100%.
• The improved AlexNet model demonstrated high accuracy at 97.78%.
• The MCNN model achieved 92.83% accuracy in identifying adulterated raisins.

Conclusions:

• Near-infrared spectroscopy, coupled with advanced pattern recognition techniques, provides a feasible and accurate approach for raisin inspection and authentication.
• The developed methods offer a robust solution for detecting counterfeit raisins in the market.