Research Of Fruit And Vegetable Images Classification In The Trading Environment Based On Convolutional Neural Network

With the development of science and technology,and commodities trading becoming more and more convenient,automatization infiltrate into every aspect of trading.Automatic farm produce classification and recognition technology is not mature in the process of producting and marketing.For example,at the trading stage,farm produce category is done mostly via manual sorting,which increases both labor costs for dealers and inconvenience for consumers shopping experience.It's the computer vision technology that provide solution to this problem,using in farm produce classification and quality testing,particularly fruit and vegetables.However,nowadays automatic identification usually via shallow learning techniques to achieve recognition of images,such as distinguish single or without background image,through texture or color of fruits and vegetables,which cannot fulfill the recognition of a variety of fruits and vegetables.According to the above background,this study focus on image recognition of the fruit and vegetable in trading process,using the deep learning method to improve image recognition rate.The main method is structuring the model based on CNN(convolution neural network)to classify and recognize the fruits and vegetables.Main tasks are as follows:1)Collect images of 15 kinds of vegetables and fruits,a total of 111309 samples.Through the binary conversion,form the fruits and vegetables agricultural image library which is used as a training set,validation set and test set.2)Design the structure optimization experiment of the CNN model LetNet-5 which be used to classify fruit and vegetable.By observing the effect of classification,it can be draw the conclusion that: increasing the number of convolution and convolution kernels,narrowing the scope of convolution kernels and pooling size,can improve recognition effect.Based on the above conclusion,optimize the initial model.And introduce Dropout to prevent over fitting.In the end.Design a CNN model DCNN-V which apply to classify fruits and vegetables under the complex background environment.3)Based on DCNN-V choose the fruit and vegetable images which collected under the different light intensity,background and has the different quantity as the validation set and test set.Do identify and evaluate Recognition effect.Compare with the other identification methods.Research results show that,by reasonable adjustment of parameters and methods,CNN can be applied to recognize fruits and vegetables image in complex background environment.Identification accuracy rate under different light intensities to newspapers,aluminum trays,shopping bags as a background image is(rise to)96.7%~98.4%.Model has solved the problems existing in traditional recognition methods that the Preprocessing operations is tedious and generalization ability is poor.At the same time,improved the recognition accuracy.There have certain research significance and practical value.