Consensual Regression of Soluble Solids Content in Peach by Near Infrared Spectrocopy

Area of Science:

• Agricultural Science
• Analytical Chemistry
• Spectroscopy

Background:

• Accurate measurement of soluble solids content (SSC) is crucial for peach quality assessment.
• Traditional methods for SSC determination are often destructive and time-consuming.
• Near-infrared (NIR) spectroscopy offers a rapid, non-destructive alternative for quality analysis.

Purpose of the Study:

• To develop a robust NIR spectral calibration model for SSC in peaches.
• To reduce the uncertainty and spectral information loss associated with genetic algorithm (GA) optimization.
• To propose a consensus model fusion strategy for improved prediction accuracy.

Main Methods:

• Collected 266 peach samples and scanned their interactance NIR spectra.
• Measured SSC destructively using standard refractometry.
• Pre-processed spectra and used GA for variable selection to build partial least square (PLS) models (PLS_GA and PLS_RV).
• Developed a consensus model by fusing PLS_GA and PLS_RV with optimized weightings.

Main Results:

• The PLS_RV model retained useful spectral information and performed comparably to full-spectral models.
• Consensus models achieved a lower average root mean squared error of prediction (RMSEP) of 1.106% (SD=0.0068) compared to PLS_GA models (average RMSEP=1.116%, SD=0.0097).
• The fusion strategy effectively reduced model uncertainty and improved spectral information utilization.

Conclusions:

• The proposed consensus model fusion strategy enhances the reliability of GA-optimized NIR models for SSC prediction in peaches.
• This method maximizes the use of available spectral information, leading to more accurate and stable quality assessments.
• The strategy enables rapid, non-destructive detection of internal peach quality.