Identification of Internal Defects in Potato Using Spectroscopy and Computational Intelligence Based on Majority...

Area of Science:

• Agricultural Engineering
• Spectroscopy
• Machine Learning

Background:

• Potatoes are a vital global food source, but internal defects reduce marketability and increase disease risk.
• Current quality assessment methods often overlook internal defects in visually healthy potatoes.

Purpose of the Study:

• To develop a non-destructive method for identifying internal defects in potatoes.
• To select optimal spectral wavelengths for defect detection using machine learning.

Main Methods:

• Utilized visible (Vis), near-infrared (NIR), and short-wavelength infrared (SWIR) spectroscopy to collect spectral data.
• Employed hybrid artificial neural networks (ANN) and cultural algorithms (CA) for optimal wavelength selection.
• Classified potato samples using an ensemble of classifiers (ANN-ICA, ANN-HS, LDA, KNN) with majority voting.

Main Results:

• Optimal wavelengths were identified in both Vis/NIR and SWIR regions.
• The ensemble method achieved high classification rates, with SWIR spectral data yielding 96.3% accuracy.
• Linear Discriminant Analysis (LDA) using selected SWIR wavelengths achieved the highest accuracy at 97.7%.

Conclusions:

• Spectral analysis combined with machine learning effectively detects internal potato defects.
• Selected optimal wavelengths and LDA provide a highly accurate and efficient method for potato quality assessment.
• This approach can significantly improve potato marketability and reduce post-harvest losses.