Detection Method for Walnut Shell-Kernel Separation Accuracy Based on Near-Infrared Spectroscopy.

Area of Science:

• Agricultural Engineering
• Spectroscopy
• Machine Learning

Background:

• Accurate separation of walnut shells and kernels is crucial for processing and quality control.
• Traditional methods for assessing separation accuracy can be time-consuming and labor-intensive.
• Developing rapid, non-destructive techniques is essential for the food industry.

Purpose of the Study:

• To establish a Near-infrared (NIR) spectroscopy-based detection model for walnut shell-kernel separation accuracy.
• To compare the performance of Support Vector Machine (SVM) and Extreme Learning Machine (ELM) classification models for this task.
• To investigate optimization techniques for enhancing model performance.

Main Methods:

• Collected 1200 NIR spectra from three types of walnut materials.
• Applied de-trending (DT) preprocessing.
• Developed SVM and ELM classification models using principal component factors.
• Optimized SVM parameters (C, g) and ELM parameters (W, B) using a genetic algorithm (GA).

Main Results:

• Achieved high classification accuracy for shell, kernel, and chimera detection.
• SVM model achieved 97.78% accuracy.
• ELM model achieved 97.11% accuracy.
• Optimized models demonstrated effectiveness in distinguishing between shell and kernel material.

Conclusions:

• NIR spectroscopy combined with SVM and ELM offers a robust and accurate method for detecting walnut shell-kernel separation.
• The optimized models provide a reliable reference for automated quality assessment in walnut processing.
• This approach has the potential to improve efficiency and reduce costs in the walnut industry.