关键词: K-means聚类, 皮尔逊相关系数, 多特征融合, 改进MLP, 苹果分级

Abstract: In order to accurately realize apple classification with multi‑feature fusion，a classification method of Fuji apples based on K‑means clustering and improved MLP was proposed. The method mainly included image preprocessing，brightness equalization，background segmentation，feature weighting and improved MLP classification network. Firstly，the image quality of apple was improved by means of mean filtering algorithm and brightness equalization operation；then background segmentation was performed by means of K‑means clustering algorithm；on the basis of fruit body and background segmentation，the fruit diameter，fruit shape，color，defect and texture features of apple were extracted in turn；then Pearson correlation analysis and artificial preference weight were used to comprehensively weight the feature data set to simulate the artificial grading scene；finally，the feature data was sent to the improved MLP neural network to complete the apple grading.Through the grading test of 400 graded apples，the accuracy rate reached 94.25%，which verified the feasibility and accuracy of the grading method.This method corresponds to the systematic apple grading standard，and has the advantages of strong timeliness and complete detection indicators.

Key words: K?means clustering, Pearson correlation coefficient, Multi?feature fusion, Improved MLP, Apple classification