Ordinal Regression - Extreme Learning Machine based Apple Grading Method

In view of the fact that the traditional apple quality grading without making full use of the ordinal class label information between apple quality grades. In order to improve the prediction accuracy of the near-infrared technical grading model, this paper applies a combination of ordinal regression and an extreme learning machine to apple grading. In this paper, Yantai Red Fuji apples were classified into three classes according to their quality and applied to the ordinal regression extreme learning machine model. Firstly, a near-infrared spectral detector is used to detect the spectral information of apples, and the spectral feature information is extracted using principal component analysis and mahalanobis distance method. Secondly, the soluble solid content of apples was detected using a sugar meter, and apples were classified into classes based on the soluble solid content values. The extracted spectral information and the rank labels of apple classification were used as sample information input, and the ordered labels of apple classification ranks were used as prediction results output. The results showed that the grading accuracy of the ordinal regression extreme learning machine model was higher than that of the traditional grading model, and the grading accuracy reached 92.5%, which provided a research basis for non-destructive testing and grading of internal quality of apples.