[Study on variable selection of NIR spectral information based on GA and SCMWPLS]

Area of Science:

• Utilizes Vis/NIR spectroscopy for quantitative analysis of fruit quality parameters.
• Focuses on spectral data compression and informative variable selection techniques.

Context:

• Near-infrared (NIR) spectroscopy is increasingly applied for rapid quality assessment in agriculture.
• Accurate prediction of soluble solid content (SSC) is crucial for fruit quality evaluation and marketability.

Purpose:

• To investigate the effectiveness of spectral data preprocessing and variable selection methods for improving NIR prediction models.
• To simplify complex spectral data and enhance the accuracy of soluble solid content (SSC) prediction in apricots.

Summary:

• Employs pretreatment methods including second derivative, normalization, and orthogonal signal correction (OSC) to filter irrelevant spectral data.
• Compares SCMWPLS and Genetic Algorithm (GA) coupled with Multiple Linear Regression (MLR) for variable selection, demonstrating superior performance over full-region PLS models.
• Achieves high prediction accuracy for SSC with selected spectral regions, showcasing the efficacy of the proposed methods.

Impact:

• Demonstrates that OSC effectively filters irrelevant signals and reduces the number of latent variables required for accurate modeling.
• Highlights the capability of SCMWPLS and GA in identifying optimal informative variables for NIR analysis.
• Establishes the universal significance of these low-dimension, high-precision modeling approaches for developing express NIR analysis models.