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

In this study, Near-infrared (NIR) spectroscopy was adopted for the collection of 1200 spectra of three types of walnut materials after breaking the shells. A detection model of the walnut shell-kernel separation accuracy was established. The preprocessing method of de-trending (DT) was adopted. A classification model based on a support vector machine (SVM) and an extreme learning machine (ELM) was established with the principal component factor as the input variable. The effect of the penalty value (C) and kernel width (g) on the SVM model was discussed. The selection criteria of the number of hidden layer nodes (L) in the ELM model were studied, and a genetic algorithm (GA) was used to optimize the input layer weight (W) and the hidden layer threshold value (B) of the ELM. The results revealed that the classification accuracy of SVM and ELM models for the shell, kernel, and chimera was 97.78% and 97.11%. The proposed method can serve as a reference for the detection of walnut shell-kernel separation accuracy.