RPSR based multi-feature fusion grading for apple

Current research on apple grading mainly focuses on internal Near-Infrared spectra (NIRS) data or external image information, multi-feature fusion is seldom been considered. Therefore, the random permutation set reasoning (RPSR) is introduced in the comprehensive evaluation of internal and external quality of apples, which can handle the sequential relationship of features and fuse the multiple features at the decision level. In details, the internal features (sugar content and moisture content) are predicted by partial least squares (PLS) model based on the NIRS, external features (roundness and color) are calculated by using apples’ RGB images. Meanwhile, aim to the problems of low accuracy and poor applicability in single-feature prediction models, the RPSR model is proposed to predict the comprehensive grade by considering the fusion of the mix features with sequential relationships at the decision level. Finally, experimental simulation is done by comparing the RPSR model with the eXtreme Gradient Boosting (XGBoost) model. The experimental results show that the classification accuracy of XGBoost is 80.8% and the accuracy of RPSR is 83.2%. The validity of RPSR model is fully confirmed.