Study on starch content detection and visualization of potato based on hyperspectral imaging

Area of Science:

• Agricultural science
• Analytical chemistry
• Spectroscopy

Background:

• Starch content is a key indicator of potato quality, affecting taste and nutrition.
• Traditional chemical analysis for starch is slow and labor-intensive.
• Accurate and rapid starch determination is crucial for potato processing and grading.

Purpose of the Study:

• To develop a rapid and non-destructive method for predicting potato starch content.
• To evaluate the effectiveness of hyperspectral imaging combined with chemometrics.
• To identify optimal sampling sites and wavelength selection methods for accurate starch prediction.

Main Methods:

• Hyperspectral imaging was used to collect data from potato samples (Kexin No.1 and Holland No.15).
• Spectral data were preprocessed using Standard Normal Variate (SNV).
• Characteristic wavelengths were selected using Competitive Adaptive Reweighted Sampling (CARS), Iterative Variable Subset Optimization (IVSO), and Variable Iterative Space Shrinkage Approach (VISSA).
• Partial Least-Squares Regression (PLSR) and Support Vector Regression (SVR) models were built.
• Pseudo-color technology was used for visualization.

Main Results:

• The sampling site significantly impacted prediction model accuracy.
• The umbilicus region yielded the best results when combined with the CARS-SVR model.
• The optimal CARS-SVR model achieved high accuracy: Rc=0.9415, Rp=0.9346, RMSEC=15.9 g/kg, RMSEP=17.4 g/kg, RPD=2.69.
• Starch content distribution was visualized using pseudo-color imaging.

Conclusions:

• Hyperspectral imaging combined with chemometrics offers a viable method for rapid, non-destructive potato starch content determination.
• The umbilicus region and CARS-SVR model are recommended for optimal starch prediction.
• This approach supports efficient potato quality monitoring and grading.