Research On Multi-feature Fusion Method Based On Image Processing For Apple Grading

China is a big apple producer,but not a trade power.Apple exports account for only 3percent of the total output,mainly due to the low level of quality inspection and classification of apples after harvest.At present,most of the apples rely on manual grading,which has low work efficiency.Moreover,apples are graded mainly based on a single feature in automation,but the classification accuracy is low.Therefore,the paper takes Red Fuji Apple as the research object and proposes the research on apple grading based on multi-feature and multi-classifier of image processing.The recognition of defective fruits and the classification of intact fruits are completed,and the efficiency and accuracy of classification are improved.First of all,the apple images were obtained in this paper,and the RGB and HSI models were introduced to extract the apple component graphs under these two models.Through comparison,it is found that the R component map is best when using Otsu method for image segmentation.Target edge was smoothed by open operation.Target edge was extracted by canny algorithm,and the color image of fruit stem and defect area was acquired.The above laid a foundation for feature extraction.Then,texture and geometric features were extracted in defect and fruit stem area.For feature extraction of intact fruit,apple shape was described by calculating the minimum and maximum radius ratio and roundness of apple image.The color characteristics of apple were obtained through color quantitative analysis and red coloration rate.According to GB/T10651-2008 "the fresh apple",maximum cross sectional diameter as fruit diameter,a method for apple size detection was proposed.The minimum and maximum radius ratio of each apple surface was calculated.The side with the largest ratio was the fruit diameter surface.Then the minimum circumferential circle diameter of fruit diameter surface was taken as the size of apple and the experimental comparison was made.The experimental result shows that the method is in line with the actual sorting requirements.Finally,apple grading was carried out.The logic judgment method and support vector machine model were used to determine whether apple is defective.In the classification of intact fruit,the intact fruit was graded based on a single feature.The intact fruit was graded according to multiple feature combinations under the k-Nearest Neighbor BP neural network and Support Vector Machine.In order to improve the classification accuracy,a classificationmethod of multi-classifier fusion was proposed,which was to fuse the single classifier with high accuracy and great difference,and combine the results of each single classifier with weighted voting method.And the classification accuracy rate is as high as 90.2%.The result of this study provides a reference for apple automatic sorting.