Hybrid wavelength selection technique and spectral binning for wheat protein estimation using hyperspectral imaging

Area of Science:

• Agricultural Science
• Spectroscopy
• Data Science

Background:

• Hyperspectral imaging offers potential for estimating wheat protein content.
• Current methods are limited by expensive equipment and high-dimensional data.
• Reducing data complexity is crucial for developing cost-effective systems.

Purpose of the Study:

• To identify a minimal set of informative wavelengths for wheat protein estimation.
• To reduce computational complexity in spectral data analysis.
• To facilitate the development of low-cost spectral imaging systems.

Main Methods:

• Employed thirteen wavelength selection algorithms on spectral data (5 nm resolution).
• Utilized a two-step hybrid strategy combining Random Forest, Genetic Algorithm, and Support Vector Regression.
• Investigated the impact of reduced spectral resolution (10 nm) using spectral binning.

Main Results:

• Identified an optimal set of 6 wavelengths yielding high accuracy (R² = 0.9790, RMSE = 0.2104).
• Comparable accuracy was achieved with reduced 10 nm resolution (R² = 0.9688, RMSE = 0.2564).
• Validated the effectiveness of the selected wavelengths and reduced resolution for wheat protein estimation.

Conclusions:

• Six wavelengths at 10 nm resolution are sufficient for accurate wheat protein content estimation.
• This research paves the way for developing inexpensive multispectral imaging devices.
• The findings support practical applications in precision agriculture and crop quality assessment.