[Application of some different modeling algorithms to pear MT-firmness detection using NIR spectra]

Area of Science:

• Agricultural Science
• Spectroscopy
• Chemometrics

Context:

• Near-infrared (NIR) spectroscopy is a rapid, non-destructive technique for assessing agricultural product quality.
• Accurate statistical modeling is crucial for correlating NIR spectral data with internal quality attributes like firmness.
• Classical methods like Partial Least Square Regression (PLSR), Principal Component Regression (PCR), and Stepwise Multilinear Regression (SMLR) are commonly employed.

Purpose:

• To establish a nonlinear model for predicting the Magness Taylor (MT) firmness of 'Xueqing' pears using NIR spectra.
• To compare the performance of PLSR, PCR, SMLR, and a mixed SMLR-Artificial Neural Network (ANN) algorithm for firmness prediction.
• To evaluate the effectiveness of original versus derivative NIR spectra in the modeling process.

Summary:

• NIR diffuse reflectance spectra (800-2500 nm) of intact pears were analyzed.
• PLSR models using original spectra yielded the best prediction results (r=0.87 calibration, r=0.84 validation).
• The mixed SMLR-ANN algorithm showed promising results but was slightly outperformed by PLSR.

Impact:

• Demonstrates the potential of NIR spectroscopy for non-destructive, in-situ prediction of pear firmness.
• Highlights PLSR as a robust statistical method for developing accurate NIR-based quality assessment models.
• Indicates the need for further research in statistical modeling to enhance predictive accuracy for agricultural products.