Apple Grading Method Based on Support Vector Ordered Regression

Abstract

The improvement of apple grading technology is conducive to the growth of apple exports, thereby boosting the development of the national economy. Considering the ordered labels in apple grading problem, in this paper, support vector ordered regression model is proposed to achieve the mapping relation between the near-infrared spectrum and soluble solid content of an apple. Given near-infrared spectra, features are extracted for them using partial least square method. Also, the value of soluble solid content is transferred to the label of an sample. To build the support vector ordered regression model, the kernel function is used to obtain the optimal fitting curve. The model is constructed on the training set containing 359 sample data and then verified on the test set containing 80 samples. The results of the experiment show that the accuracy of using the support vector ordered regression method to classify apples reaches 95%, fully verifying the effectiveness of support vector ordered regression.