Research On Fruit Quality Grading Algorithm Based On Fruit Defect Multi-Feature Fusion And Convolution Neural Network

China is the largest fruit producer in the world,but its share in the international fruit market,especially the high-end fruit market,has been at a low level,among which the main reason is that the grading standards of fruit quality are not strict.Through computer vision and image processing technology,the method of improving the accuracy of fruit quality grading can reduce the subjectivity of artificial fruit quality grading,and has important social and economic significance.Fruit quality grades are difficult to distinguish due to inconsistent grading standards between different fruit types,small differences in standardization between different fruit grades,and similar fruit fruit stalk areas and fruit defect areas.In view of the above problems,this paper mainly studies the method of fruit grading based on defect multi-feature fusion and convolutional neural network.The main research contents and contributions are as follows:(1)Aiming at the inconsistency of fruit quality grading standards on different types of fruits,a fruit species identification method based on image HSV spatial color features,PLBP texture features and HOG features was proposed.The original LBP features can only reflect the overall texture features of the image,and can not describe the shortcomings of the local texture features.This paper uses the image pyramid idea to improve the LBP features,and then reflects the texture features of fruit images from different scales.The HOG features are used to describe the local target shapes.It is complementary to color features and texture textures,but it has the disadvantage of high feature dimension.In this paper,the PCA algorithm is used to reduce the dimension of HOG features,and the main components of HOG features are extracted to reduce redundancy.In this paper,a comparative experiment was carried out on the Fruit-360 test set to verify the effectiveness of the multi-feature fusion fruit type recognition method.(2)Aiming at the problem that fruit stem region and fruit defect area are difficult to distinguish and affect fruit quality classification,a defect fusion method based on decision layer is proposed.Firstly,the defect segmentation algorithm integrating threshold segmentation and k-means segmentation was used to segment the suspected fruit defect areas from the fruit images,and then the suspected defect areas were identified.In addition,two methods,defect fusion based on feature layer and defect fusion based on decision layer,were adopted to classify the fruit quality.The defect fusion algorithm based on decision-making layer effectively reduces the influence of the fruit stem region on the defect area,and proves that the defect fusion method based on decision-making layer is more effective.(3)Aiming at the small difference between different fruit quality grades and the poor design of manual design features,this paper proposes a fruit quality classification algorithm based on improved AlexNet convolutional neural network based on AlexNet convolutional neural network.The network optimizes the size of the convolution kernel receptive field while maintaining the convolutional layer of the AlexNet network,increases the pooling layer,and improves the AlexNet convolutional network optimizer,which is effective in reducing the overall network parameters.Improved recognition accuracy of fruit quality grading.