Research On The Application Of Deep Convolutional Neural Network In Leaf Disease Identification Of Apple Trees

Apple is one of the main cultivated crops in China.Apple trees planting area is large,high yield,related to the development of people’s livelihood.However,in the process of plant growth,due to factors such as climate and soil,the influence of apple tree leaves by pests and diseases is also expanding.If the problem of apple tree leaves and pests cannot be detected and solved in time,the yield and quality of apples will be greatly reduced,causing certain losses to farmers and consumers.At present,most of the methods for diagnosing crop diseases and pests still require professional and technical personnel to sample and measure,and judge based on personal experience,which consumes more manpower and material resources,and the efficiency of identifying diseases and pests is low.In recent years,with the continuous development of image processing and deep learning technology,image recognition and classification based on convolutional neural networks have been widely used in the control methods of apple tree leaf diseases and pests.Convolutional neural networks have the characteristics of automatic extraction of feature maps,strong generalization ability,and high accuracy of image recognition.Aiming at the problem that the recognition accuracy of traditional methods is not high,the main research contents of this paper are as follows:(1)11 kinds of apple leaf disease images including black spot,black spot,cedar rust and apple leaf disease were collected and downloaded from the public data set through the network.A total of 35017 images of apple leaf disease were obtained after the image was expanded by rotating,adjusting the contrast and brightness.It is divided into two groups of data sets.The small data sets are divided into six categories,with a total of 8640 images;The large data set is divided into 5 categories,with a total of 26377 images.(2)This paper studies the effect of improving the network structure on the training accuracy based on the original VGG-16 network model.Replace the activation function Re LU in the original network model with the P-Re LU activation function,and add a batch normalization operation after each convolution module to solve the problem of gradient disappearance in the training process and increase the training speed of the network.At the same time,the influence of different optimization algorithms on the convergence speed and training accuracy of the network is explored,and the original SGD random gradient descent algorithm is replaced by the Adam optimization algorithm.Six kinds of apple leaf disease images in the open Plant Village data set were selected in the experiment.The experiment showed that the training accuracy of the improved VGG-16 convolution neural network model reached 98.578%,which was better than the recognition accuracy of the original model(91.134%).(3)Under the background of the popularity of mobile terminals,the research applies the Mobile Net series lightweight convolution neural network to identify the diseases and pests of apple leaves.Through the analysis of the structure of the lightweight network model,the Mobile Net-V2 network was selected to classify and recognize the five kinds of apple leaf disease and pest images collected.There are more than 20000 pictures in the dataset,and the relevant parameters of Mobile Net-V2 network have been fine-tuned in the experiment.The results show that compared with other commonly used convolutional neural network models,the training time of the model is significantly shortened when using Mobile Net-V2 network.The memory space and training time used by using Mobile Net-V2 network are better than other large convolutional neural networks without affecting the recognition accuracy as much as possible.