Study On Tomato Leaf Disease Recognition And Classification Based On Convolution Neural Network

Tomato quarantine is an important process to protect tomato growth,which ensures that farmers can quickly and efficiently deal with tomato diseases,thus ensuring the yield and quality of tomato.There are many kinds of tomato diseases and the onset time is uncertain.The traditional computer vision method for disease identification is very dependent on artificial experience,which consumes a lot of manpower and material resources,and increases the production cost.At present,the recognition of tomato disease using deep convolution neural network is generally based on some simple data sets with clean background and single target,which often fail to achieve good results when put into practical application.At the same time,due to the large amount of calculation of the deep convolution neural network and the high requirements for equipment,it is not conducive to the real-time identification of tomato diseases by farmers.Therefore,this paper proposes a method based on the combination of lightweight target detection and target classification to achieve real-time recognition of tomato disease under complex background and multi-target conditions.The specific research contents are as follows:(1)Acquisition and production of data sets: The images of tomato diseases on the Internet are downloaded in batches through crawlers,and the data sets are integrated according to the existing tomato disease data sets on the Internet.The data sets are further expanded through image enhancement,and the tomato leaf area is marked with Label Img software to build the tomato leaf recognition data set.(2)Build tomato leaf disease identification network: This paper builds tomato leaf disease identification model based on YOLO v3,uses the lightweight network Mobile Net v2 to replace Darkenet53 as the backbone network of YOLO v3,adds SPP module,and modifies the size of Anchor box.The tomato leaf recognition data set is used for model training.The experimental results show the effectiveness of the improved model.Compared with other series of algorithms of Fast R-CNN,SSD,and YOLO,the improved YOLO v3 is proved＿mobv2＿Spp model has high accuracy and real-time.(3)Build tomato leaf disease classification network: based on the lightweight network Shuffle Net v2,this paper replaces all 3 * 3 convolution cores in Shuffle Net v2 basic module with 5 * 5 convolution cores,and adds SE module in the trunk branch,which controls the calculation amount while increasing the receptive field and improving the expression ability of the model.In order to reduce the training time,this paper uses the transfer learning to freeze part of the convolution layer,and uses the tomato leaves obtained from the target recognition for training.After verifying the effectiveness of the improved model,it carries out comparative experiments with Res Net 50,Mobile Net v2,Shuffle Net v2,which also have the residual module,and proves the reliability and superiority of the model.In this paper,the deep convolution neural network is used to complete the tomato leaf recognition and tomato disease classification network.The production of the data set is more close to the practical application.For the complex background and multi-objective,the optimized network has not only greatly improved the accuracy,but also greatly reduced the training time of the model.Therefore,this model can be better applied to mobile or embedded devices,and provide help for farmers to identify tomato diseases in time.