Enhanced Wavelet Convolutional Neural Network And Its Application In Apple Tree Disease Control

Apple is a common economic crop in our country,with a plantation area of more than 6million hectares across the country.In the process of apple planting,diseases have a great influence on its yield and quality.If the disease can be accurately diagnosed and identified early in the onset of apples,and timely measures are taken to treat diseased plants in a targeted manner,the loss of apple yield can be reduced to a certain extent.The focus of this process is on how to quickly and accurately diagnose apple leaf diseases and insect pests.In order to improve the accuracy and efficiency of apple leaf disease and insect pest identification,this thesis takes the images of apple leaves grown in the natural environment as the research object,and proposes a new model I-VGG16 to improve the classification accuracy.First,the classification performance of four classic convolutional neural networks on apple leaf images is studied,and the neural network model with the highest accuracy is selected.Next,IVGG16 designed a new network structure based on this model,combined with the prior knowledge of wavelet transform,so that the influence of noise on apple leaf classification was suppressed.Finally,a detailed study on the identification and classification of common foliar pests and diseases on apple leaves was carried out in this thesis.The main research conclusions of the thesis are as follows:(1)This thesis compares the apple leaf image classification model based on convolutional neural network.First,using the convolutional neural network VGG16,Inception V3,Res Net50 and Dense Net121 models pre-trained on Image Net to perform transfer training and testing on the apple leaf image data set.The experiment shows that the VGG16 model has more advantages in both accuracy and processing speed.(2)This thesis studys the classification method of apple leaf image based on the combination of wavelet and VGG16 model.Convolutional neural networks are related to the robustness of weak noise,and the noise of image data is mostly high-frequency components.In order to suppress the influence of noise on image classification,this thesis uses discrete wavelet transform to replace maximum pooling,step convolution and average pooling to enhance the convolutional neural network.Finally,a new model named I-VGG16 model was designed,and this model was used to train on the apple leaf dataset.The experimental results show that the IVGG16 model can better handle the task of apple leaf image classification and can improve the accuracy of apple leaf classification.