Applications of machine learning in pine nuts classification.

Area of Science:

• Agricultural Science
• Food Science
• Computational Biology

Background:

• Pine nuts are nutritionally valuable but difficult to distinguish due to morphological similarities among species.
• Accurate and non-destructive identification methods are crucial for quality improvement and preventing pine nut adulteration.

Purpose of the Study:

• To develop rapid, non-destructive methods for classifying seven species of pine nuts.
• To evaluate the effectiveness of machine learning algorithms applied to near-infrared (NIR) spectroscopy and image data for pine nut species identification.

Main Methods:

• Collected 210 near-infrared (NIR) spectra from seven pine nut species.
• Applied five machine learning models (Decision Tree, Random Forest, Multilayer Perceptron, Support Vector Machine, Naive Bayes) for spectral data classification.
• Utilized 303 images for morphological data collection and constructed classification models using five convolutional neural network (CNN) architectures (VGG16, VGG19, Xception, InceptionV3, ResNet50).

Main Results:

• The Multilayer Perceptron (MLP) model achieved an accuracy close to 0.99 for pine nut classification using NIR spectra.
• The InceptionV3 model demonstrated the highest accuracy at approximately 0.964 for classification based on morphological image data.
• Identified four key NIR waveband ranges (951-957 nm, 1147-1154 nm, 1907-1927 nm, 2227-2254 nm) critical for species classification.

Conclusions:

• Machine learning techniques are highly effective for the rapid, non-destructive, and accurate classification of different pine nut species.
• This study provides robust scientific methods to address pine nut adulteration and enhance quality control in the food industry.