Study on the Quantitative Damage of Apple Based on Convolutional Neural Network Combined With Mass Compensative Method

Abstract

Nondestructive quantitative analysis of fruit damage can not only provide technical support for fruit quality testing, but also provide the theoretical basis for the improvement of fruit packaging and transportation conditions. However, the models of quantitative prediction of fruit damage are susceptible to influence by own factors (size). Therefore, in order to improve the accuracy of quantitative prediction of fruit damage, one-dimensional convolutional neural network (1D-CNN) combined with the mass parameter method was proposed. The study results show that the performances of the 1D-CNN models are improved by 3.4%–7.0% compared to the traditional models. The performances of 1D-CNN prediction models based on the mass compensation have been improved by 7.5%–10.3% compared with the precompensation. In conclusion, the 1D-CNN models based on the masscompensation have positive effects in eliminating the influence of apple size on the quantitative prediction models of apple damage.