Research on Citrus Quality Detection Method Based on Machine Learning

  • Li Shi-Hong, Ni Kong-Shi, Hu Hui-Pu

Abstract

The traditional destructive testing methods for the internal quality of citrus are mainly carried out by random sampling. The sample preparation and process are cumbersome, and the citrus has been destroyed to a certain extent, affecting the secondary sales. In order to improve the efficiency of non-destructive testing of the internal quality of citrus, this paper proposes a spectral analysis method based on machine learning. First, a test model for the internal quality of citrus was established and applied to the quality test of Wenzhou Yongjia mandarin oranges. Non-destructive testing was performed on 207 citrus samples. Secondly, using cross-validation and integrated learning techniques, the LightGBM algorithm is used to classify citrus samples, and the classification accuracy is around 90.61%. In addition, the citrus spectral data was clustered into three categories: excellent, good, and medium, and the V-Measure reached 0.7273. Finally, the accuracy and feasibility of the model are verified through experiments. The research results show that it is feasible to use the spectral data of citrus, using supervised learning methods as the main method, and unsupervised learning methods as a supplement, to quickly and non-destructively detect the quality of citrus, and the established prediction model is stable and accurate. High, the research results can provide a certain technical reference for non-destructive testing of citrus.

Published
2020-12-30
How to Cite
Li Shi-Hong, Ni Kong-Shi, Hu Hui-Pu. (2020). Research on Citrus Quality Detection Method Based on Machine Learning. Design Engineering, 955 - 966. Retrieved from http://thedesignengineering.com/index.php/DE/article/view/1051
Section
Articles