Research On Apple Appearance Quality Grading Based On Image Processing

In China,apple is one of the most popular fruits because of its rich nutritional value.Its output is still at the peak in the world,but the proportion of apple export in China is very small.Take the data of 2019 as an example,the export volume is only 2.3% of apple production.One of the reasons for this situation is not only the large domestic demand market,but also the insufficient work in apple export.Take some countries with developed apple export industry as an example,the apple they export has a complete set of technical process.However,this process is less in China,which makes it difficult for China’s apple to form a strong competition in the international market.Therefore,this paper will take the apple grading process as the research object,and take Shaanxi Red Fuji as the research object to classify the apple appearance by manual feature extraction and convolution neural network feature extraction.The main contents of this paper are as followsIn the aspect of artificial feature selection,this paper first preprocesses the apple image to be classified,mainly including graying to reduce the amount of computation;Then,by comparing mean filter,median filter and BM3 D filter,BM3 D filter with better noise reduction effect is selected;Then the mask is formed by binarization and closing operation,which is used to mask the original color image and get the apple image without background.After the pretreatment,based on the grading requirements of Fuji apple in GB / T 10651-2008 "fresh apple",the red coloring rate,fruit diameter,calyx,stem and defect of apple surface were extracted;Finally,the extracted features are put into libsvm for classification and recognition,and the classification results are evaluated.In addition,for the rapid development of convolutional neural network in recent years,and its good performance in many aspects,this paper proposes the method of feature extraction of apple image by using alexnet,and classification of Apple by using libsvm.The main step is to unify the size of the image as 227 * 227,and then use the eight layer alexnet convolution neural network to extract the features of the apple image set to get the features of the training set and the test set,and then put the extracted features into libsvm for training and testing,and then evaluate the accuracy.Finally,through the comparison of the two classification methods,the results show that the combination of alexnet and libsvm is 6.5% higher than the former method,and 3% higher than the result of small sample of alexnet class.