Detection of Pear Quality Using Hyperspectral Imaging Technology and Machine Learning Analysis.

Area of Science:

• Agricultural Science
• Food Science
• Spectroscopy

Background:

• Non-destructive fruit quality detection is crucial for the agricultural and food industries.
• Hyperspectral imaging (HSI) technology offers a promising approach for rapid, non-destructive quality assessment.

Purpose of the Study:

• To apply hyperspectral imaging (HSI) and machine learning for efficient pear quality assessment.
• To develop a robust technical method for non-destructive analysis of key pear quality parameters.

Main Methods:

• Six pear varieties were analyzed using hyperspectral data (398-1004 nm).
• Least Squares Support Vector Machine (LS-SVM) models were optimized with preprocessing (FD-SNV) and feature selection (CARS).
• Backpropagation Neural Network (BPNN) was used for variety classification.

Main Results:

• FD-SNV preprocessing and CARS feature selection significantly improved LS-SVM model performance for predicting firmness, SSC, pH, color, and maturity.
• Integrating data from multiple pear varieties (n=6) resulted in RPD > 2.0, indicating robust model performance.
• The BPNN model achieved >99% accuracy in distinguishing between the six pear varieties.

Conclusions:

• The combination of HSI and machine learning provides an efficient, rapid, and non-destructive method for pear quality detection.
• This approach has practical value for quality control and commercial processing in the fruit industry.
• Optimized preprocessing and feature selection enhance predictive accuracy and reduce computational load.