Non-Destructive Detection of Elasmopalpus lignosellus Infestation in Fresh Asparagus Using VIS-NIR Hyperspectral...

Area of Science:

• Agricultural Science
• Spectroscopy
• Machine Learning

Background:

• Internal damage by Elasmopalpus lignosellus in asparagus presents a significant challenge for the agro-export industry.
• Traditional visual inspection methods lack the sensitivity for early detection of this pest, impacting quality control.

Purpose of the Study:

• To evaluate VIS-NIR hyperspectral imaging combined with machine learning for discriminating between infested and sound asparagus spears.
• To develop a non-destructive method for early detection of internal damage in asparagus.

Main Methods:

• Acquisition of a balanced dataset of 900 asparagus samples.
• Application of Savitzky-Golay and SNV for spectral preprocessing.
• Comparison of four machine learning classifiers: SVM, MLP, Elastic Net, and XGBoost.
• Spectrum reduction using LOBO and RFE techniques.

Main Results:

• The optimized Support Vector Machine (SVM) model demonstrated superior performance with cross-validation accuracy of 0.9889 and AUC of 0.9997.
• External validation on 3000 samples yielded an accuracy of 97.9% and an AUC of 0.9976.
• Spectrum reduction to 60 bands maintained high model performance.

Conclusions:

• VIS-NIR hyperspectral imaging integrated with machine learning is a viable non-destructive technology for asparagus quality control.
• This approach significantly improves the detection of internal damage caused by Elasmopalpus lignosellus.
• The findings support the implementation of these systems to enhance asparagus export quality.