Research On Fruit Grading Method Based On Computer Vision

Fruit grading is the key step of fruit commercialization.The backward fruit grading technology seriously restricts the development of fruit industry.At present,China's fruit grading is mainly based on artificial labor,with high labor intensity and low efficiency,which has been unable to meet the needs of industrial processing.Fruit grading based on computer vision has been widely concerned and studied for its high accuracy and speed.In this paper,the key technology of fruit grading based on computer vision is systematically studied.The main work is as follows:(1)This paper studies the theory and method of image classification,and the image classification technology based on digital image processing mainly introduces image segmentation,edge detection,morphological processing,etc.,and expounds its implementation principle and steps.The basic theory of image classification based on deep learning mainly introduces convolutional neural network,including its model structure and training algorithm.(2)The traditional fruit classification based on digital image processing is realized.Litchi and apple are taken as the experimental objects.By obtaining the characteristic parameters such as fruit surface color,fruit diameter size and defect area,a model for evaluating fruit maturity,size and defect degree is established.The structure and training process of deep network are studied,and the structure and characteristics of the classic pre training model vgg16 are studied With vgg16 as the pre training network,the classifier of the target network is fine tuned by using less data,and the fruit classification of migrating learning based on the classical classification model vgg16 is realized.By comparing and analyzing the fruit classification of traditional methods and the fruit classification of deep learning,the superior performance of deep learning algorithm is verified.(3)This paper studies and expounds the training and optimization methods of deep learning network.At the level of network training,migration learning technology is introduced to solve the problems of resource consumption of training deep learning model and insufficient samples of data set;at the level of data,data enhancement and batch normalization are introduced to solve the problems of insufficient data and uneven samples;at the level of network structure,data preprocessing operations are introduced In this paper,the different properties of the features extracted from each convolution layer of vgg16 are studied.The network structure of vgg16 is improved by combining the characteristics of fruit classification data.By deleting the higher convolution module,the network can obtain more local information and detailed features,and improve the accuracy of fruit classification.The influence of each convolution layer on the accuracy of fruit classification is analyzed through experiments.(4)The experiment of fine-tuning vgg16 fruit classification verifies that the network has difficulties in detecting the local features on the fruit surface in a small range.Aiming at the problem that the deepening of vgg16 network layer will cause the loss of detail information,an improved feature fusion network based on VGg isproposed for fruit classification.The vgg16 network is improved by removing the highest convolution module of the network and adding a convolution layer The experimental results show that the improved method has higher accuracy than the fine tuned vgg16 network.The innovations of this paper are as follows:(1)Compared with the traditional fruit classification method based on digital image processing,it greatly simplifies the steps of fruit classification and improves the accuracy of classification;(2)An improved feature fusion network based on VGg is proposed for fruit classification.A convolution layer is added to the vgg16 network to fuse the features of the middle layer of the network,improve the recognition ability of local features of the image,and improve the accuracy of classification;...