Machine learning approaches for large scale classification of produce.

Area of Science:

• Agricultural Science
• Analytical Chemistry
• Machine Learning

Background:

• Verifying produce authenticity (taxonomy, vendor, organic nature) is crucial in distribution networks.
• Existing analysis techniques have limitations in classifying produce attributes effectively.

Purpose of the Study:

• To present a novel data-centric approach for classifying produce attributes.
• To achieve high classification accuracies for produce taxonomy and farmer origin.

Main Methods:

• Utilized visible and near-infrared (NIR) spectroscopy on over 75,000 produce samples.
• Employed a support vector machine (SVM) model with optimized hyperparameters.
• Analyzed spectral data, focusing on color variations (chlorophyll, anthocyanins) and chemical content (water, sugar).

Main Results:

• Achieved 0.90-0.98 classification accuracy for taxonomy.
• Attained 0.98-0.99 classification accuracy for farmer origin.
• Identified produce color, water, and sugar content as key differentiating factors.

Conclusions:

• The data-centric spectroscopic approach offers high accuracy for produce attribute classification.
• Optimized SVM models and high-quality spectral data are essential for reliable results.
• Provided guidelines for data collection, experimental design, and machine learning optimization for future studies.