Study On Recognition Of Tomato Diseases Based On Convolutional Neural Network

Tomato is widely cultivated in China as a kind of vegetable and fruit used in daily life.The quality and yield of tomatoes are two important factors that farmers pay attention to at present.In the whole growth cycle of tomatoes,many diseases will occur.If it is not found in time,it will directly affect the yield and quality of tomatoes and bring serious economic losses.Therefore,tomato disease identification is particularly important.Earlier,people judged the type of disease by eyes based on experience,which not only wasted time,but also prone to misjudgement.However,the generalization ability of traditional image recognition methods is poor.With the continuous development of deep learning technology,it has a high recognition rate in the task of image recognition.Based on the above analysis,this paper selected 6 datasets of tomato leaves and 5 diseases which are easy to occur,and proposed a method of tomato disease recognition based on Convolutional Neural Network(CNN)The following research has been done1.Comparing with the existing classification networks,the advantages of using short connection and cross-layer connection are analyzed to solve the problems of gradient disappearance,parameter quantity,training speed and so on in the process of changing the depth and width of each network structure.The role of different core sizes in each network layer,especially the use of 1*1 convolution core,can increase and decrease the dimensions of the network layer,while reducing the amount of parameters and computation2.An improved VGG16 network is proposed,in which core 1*1 convolution layer is used to connect across layers and remove some convolution layers.The aim is to provide the feature information of the front layer network to the back layer for learning as far as possible,while the number of convolution layers is maintained at 16 layers.In order to prevent over-fitting,BN layer and dropout layer were added to the improved network and the threshold of parameters was set to 0.5,while the data set was enhanced.Using the network training data set,the obtained model is compared with Alexnet and VGG16 network models.Experiments show that the recognition rate of the improved network model is higher than that of Alexnet and VGG16 networks3.Tomato disease recognition based on feature extraction combination network is proposed.The traditional feature extraction algorithm and convolution network feature extraction algorithm are combined to extract sample features.Compared with the traditional feature extraction algorithm,this paper chooses the Histogram of Oriented Gradient(HOG)algorithm.Because the feature dimension of samples extracted by HOG algorithm is too large,it needs dimension reduction processing,and chooses autoencoder to reduce the dimension of feature information.The improved convolution network is fine-tuned,then SGD is optimized.After the output of convolution layer,the dimension of convolution layer is combined with the dimension-reduced feature vectors of HOG.Finally,the convolution network is connected to the full connection layer and classified by using the Softmax classifier The parameters of HOG algorithm are set up,and a number of experiments are carried out to analyze and compare each model,and the model of the optimal parameter group is selected for predictionFinally,the advantages of the proposed method in identification of tomato diseases are analyzed and compared.Experimental results show that the proposed feature extraction combination network can effectively improve the recognition rate of tomato diseases.