Research On Fruit Classification Based On Deep Learning

Fruit classification is of great significance in the fruit industry and development of social and economic.Currently,fruit classification methods mainly include manual classification and machine classification.Manual classification not only wastes a lot of manpower and resource but also has low efficiency.And the performance of machine classification needs to be improved.With the rapid development of artificial intelligence and machine vision,research on fruit classification methods based on deep learning is very important significance and good application value for improving the accuracy and efficiency of fruit classification.To address the above problems,this paper uses deep learning to study fruit classification methods.The main contents are as follows.(1)Research on quality classification based on Convolutional Neural Network(CNN)for navel orange.Firstly,a dataset of navel orange quality classification was constructed according to the national navel orange grading standard,in where the navel orange were divided into four categories: super grade fruit,first grade fruit,second grade fruit and other exogenous fruits.Then,based on Le Net-5 neural network,this paper proposes an improved CNN network model,which includes 2 convolutional layers,2 pooling layers,and a fully connected layer.Experimental results show that the overall accuracy rate of proposed convolutional neural network model reaches 94.8%,and the classification performance is better than the existing traditional fruit classification methods.This verifies that convolutional neural network method has better performance than traditional methods.(2)Research on fruit category classification based on deep learning and data augmentation.Because the traditional fruit classification method is complicated and its performance should be improved,we choose convolutional neural network to classify the fruit.Based on the ALex Net network,this paper propose an improved CNN modes with optimization named IANet,which has 17 layers and can improve the performance of fruit classification.We use the public fruit database Fruit-360 as the network training data set,and further augment the data set by image flipping,image rotation,contrast enhancement,brightness enhancement,and adding Gaussian noise.Comparing with Alex Net,13-Layer Net and Horea CNN network models,the experimental results show that the overall accuracy of the IANet network model reaches 98.60%.The performance is better than the current convolutional neural network model for fruit classification,and the performance gain is 0.62% over the best performance of Alex Net.(3)Research on fruit classification based on data augmentation with generative adversarial networks(GAN).In order to address the problem of insufficient samples in the fruit dataset,a method of fruit image augmentation based on generative adversarial networks is proposed.Compared with traditional data augmentation technology,the process is more complicated and the samples generated are more diverse.Furthermore,in order to achieve better accuracy of fruit classification,this paper proposes an improved VGG16 network based on transfer learning.Comparing to Alex Net,13-Layer Net,Horea CNN,and IANet networks,the experimental results show that the overall accuracy of the improved VGG16 network model is 99.37%,which is better than other network models.