Recognition Of Powdery Mildew Disease Of Strawberry Leaves Based On Convolution Neural Network

With the development of industrialized agriculture,artificial disease identification can not meet its needs,so it is urgent to solve the problem of automatic identification of crop diseases.The identification of strawberry leaf powdery mildew is a kind of crop disease identification.Through the characteristics and classification of strawberry leaf image,the disease types are identified.In recent years,the excellent performance of convolutional neural network in image recognition and classification has provided a new solution for the research of automatic identification of crop diseases.Convolutional neural networks do not need to select features manually,but automatically extract and classify features through network training.At the same time,by introducing convolution layer and sampling layer,and playing the role of two special strategies of local perception and weight sharing,the complexity of the network can be effectively reduced and the learning efficiency of the network can be improved.In this paper,the convolution neural network technology is applied to identify the powdery mildew of strawberry leaves.The main research work is as follows:(1)Establishment of image database of Strawberry Powdery Mildew disease.The image of strawberry leaf disease was collected in a strawberry planting research base.The image database of strawberry leaf powdery mildew disease was established after pretreatment of the collected image.Specifically,there are 3968 images in simple background and 8225 images in complex background.(2)Construction and optimization of strawberry leaf powdery mildew disease identification model based on convolution neural network.Firstly,the basic structure of convolution neural network for identifying powdery mildew in strawberry leaves was established.Secondly,three network depths(3,4,5 convolution operations)and three convolution cores(5×5,3×3 and 5×5 and 3×3 mixing)were designed.Nine convolution neural network structures were crossed and combined.Four sampling layer construction methods(mean pooling,maximum pooling,median pooling and mixed pooling)were selected.The model was optimized from three aspects: network depth,convolution core size and pooling method.Finally,the identification model of strawberry leaf powdery mildew disease based on convolution neural network was determined.The experimental results show that the CNN-9(5 convolution operations,5×5 and 3 ×3 mixing)model based on mixed pooling has the best recognition effect,and the overall recognition rate can reach 98.61%.(3)Research on image database expansion method.Based on the nine disease recognition models mentioned above,the effects of two operation methods(rotating image and mixed expansion)and the original image database on the recognition rate of the model in three aspects: network depth,convolution core size and pooling method were explored.The experimental results show that the maximum error of the two methods is 1.69% and the minimum error is 0.25% compared with the original image database.Specifically,the error of rotating extended database on the model with fewer network layers is lower,the lowest is0.7%;the error of mixed expanded image database on the model with more network layers is lower,the lowest is 0.42%.The two methods of database expansion have low errors on the models based on mixed convolution kernels(5×5 and 3×3)and mixed pooling.The proposed method based on convolution neural network can effectively and accurately identify powdery mildew in strawberry leaves,which has a certain reference value for other crop disease identification.