Prediction of Apple Quality Indicators Under Different Bagging Treatments Using Hyperspectral Imaging Integrated With...

Area of Science:

• Agricultural Science
• Computer Science
• Spectroscopy

Background:

• Apple quality indicators like color indices (L*, a*, b*) and soluble solids content (SSC) are crucial for sorting.
• Traditional destructive methods are inefficient for rapid sorting of apples with varied bagging treatments.

Purpose of the Study:

• To develop a novel stacking model (SDAE-PLSR-RR) integrating hyperspectral imaging and deep learning for non-destructive apple quality detection.
• To analyze spectral-spatial features of Fuji apples subjected to different bagging treatments (non-bagged, mesh-bagged, paper-bagged).

Main Methods:

• Hyperspectral imaging was used to capture data from 307 Fuji apples.
• A stacked denoising autoencoder (SDAE) extracted non-linear deep features.
• Partial Least Squares Regression (PLSR) processed linear trends in original wavelengths.
• Ridge Regression (RR) fused predictions from SDAE and PLSR models.

Main Results:

• The SDAE-PLSR-RR model achieved high prediction accuracy for quality indicators (R²p > 0.84).
• This model outperformed full-spectrum (R²p > 0.73) and feature-wavelength-based models (R²p > 0.75).
• The model demonstrated effectiveness for apples with different bagging treatments.

Conclusions:

• The integrated hyperspectral imaging and neural network model (SDAE-PLSR-RR) is effective for non-destructive quality detection.
• This approach offers a robust solution for rapid sorting of apples based on quality indicators.
• The study validates the synergy of deep learning and hyperspectral imaging for agricultural applications.