Research And Implementation Of Tomato Disease Recognition Mechnology Based On Deep Learning

Tomato is an important vegetable crop in China,whose cultivation area and total yield rank first in the world.In the whole growth process of tomato plants,in addition to the influence of regional weather factors,various diseases have the greatest impacts on the quality and yield of tomato.If the disease category cannot be accurately judged after it infects the plant,it will reduce the yield of tomato and bring huge economic losses.Therefore,it is of great significance to identify the disease type of tomato quickly and accurately.Traditional manual detection requires the diagnostician to have abundant knowledge reserve and relevant experience,while the generalization ability of this method is poor,and the accuracy cannot be guaranteed.With the development of technology,deep learning has achieved good results in image recognition,and has been widely used in disease detection and other related fields.The deep convolutional neural network is applied to the recognition of tomato disease,and the effects of related models on tomato disease are studied.Five common tomato diseases including tomato brown spot,tomato mosaic,tomato leaf spot,tomato leaf mold and tomato early blight are collected from the ’ Global AI Challenge ’ as experimental data sets.Besides,the classical VGG16 and Inception V3 deep convolutional neural network models are constructed for testing.In order to prevent the over-fitting of the model from enhancing the data set,L2 regularization and Dropout are added.The experimental results show that the two kinds of deep convolutional neural network models have fine recognition effects on tomato diseases,but there are still shortcomings,such as slow convergence speed of VGG16 model,large number of parameters,and many Inception V3 modules.In view of the above shortcomings,a network model based on branch and structural parameter reconstruction is proposed.The overall architecture of VGG16 model is retained,and the branch structure is introduced to replace the original module to speed up the feature extraction.The decomposition convolution is used in the branch structure and 1 * 1 convolution is added between the decomposition convolutions for small-scale dimension reduction.The nonlinear characteristics and model depth are increased without losing the resolution of the feature map.The original 3 * 3structure is retained in the shallow layer of the network,and 1 * 1 parallel convolution and identity branches are added.The structural parameter reconstruction method is used to decouple training and reasoning,and the BN layer is added after each convolution layer of the model for normalization.The improved model is superior to the traditional model in terms of convergence speed,accuracy and memory occupation,and has a good classification effect.Based on the improved model,a tomato disease identification system is constructed.Users can upload the disease images to be detected,and the system will feedback the corresponding disease identification results,which is convenient for users to use.