[Region optimization of SSC model for Pyrus pyrifolia by genetic algorithm].

Area of Science:

• Agricultural Science
• Analytical Chemistry
• Biotechnology

Context:

• Near-Infrared Spectroscopy (NIRS) is a valuable tool for non-destructively assessing fruit quality.
• Developing accurate calibration models for soluble solid content (SSC) in Pyrus pyrifolia is crucial for quality control.
• Traditional full-spectrum analysis can be computationally intensive and may include irrelevant variables.

Purpose:

• To apply a genetic algorithm-based region selection (R-SGA) method for optimizing NIRS calibration models for SSC in Pyrus pyrifolia.
• To reduce the number of variables used in NIRS calibration models while maintaining or improving prediction accuracy.
• To investigate the feasibility of building a universal NIRS model for different Pyrus pyrifolia varieties.

Summary:

• The study successfully implemented R-SGA to reduce variables from 2,075 to 690 for NIRS calibration models of pear SSC.
• Optimal R-SGA latent variables were identified for different pear varieties (Hosui, Wonhwang, Whangkeumbae).
• The R-SGA-based models demonstrated superior or comparable prediction accuracy (RMSEP) to full-spectrum analysis, proving its effectiveness in variable selection and model optimization.

Impact:

• This research validates R-SGA as an efficient method for NIRS data optimization, significantly reducing model complexity.
• The findings highlight the potential for developing robust and universal NIRS calibration models for diverse Pyrus pyrifolia varieties.
• The integrated approach of genetic algorithms and Partial Least Squares (GA-PLS) offers a powerful strategy for agricultural product quality assessment.