Research And Application Of Apple Leaf Diseases Identification Based On MobileNet

China is the largest apple producer and consumer in the world,with planting area and output accounting for more than 40% of the world’s total output.Apple trees are often affected by various diseases and pests in the growth process,resulting in the decline of apple yield and quality,which seriously affects the economic benefits of fruit farmers.Therefore,timely and accurate identification of apple tree diseases is of great significance for the prevention and management of apple tree diseases.Apple tree diseases and pests often occur in leaf organs.The traditional image recognition methods need to invest a lot of energy in extracting the characteristics of diseases of leaves,which is not only time-consuming and labor-consuming,but also misjudged due to human subjective factors.Deep learning neural network can learn image features autonomously,which is very prominent in the field of image identification.Therefore,this study selects the lightweight convolution neural network MobileNetV3 to identify the types of diseases of apple leaves(Apple scab leaves,Apple black rot leaves,Apple rust leaves and healthy apple leaves).The main work is as follows:(1)The MobileNetV3 network is built using Keras.By modifying the structure of the network to make it suitable for the task of apple leaf diseases identification.The network is trained with 2040 collected leaves.The experimental results are compared with three other classical neural network models(VGG16,ResNet50 and Inception-V3)to prove the advantages of MobileNetV3 network.(2)In order to further improve the accuracy of model,an improved MobileNetV3 model structure is proposed.The attention mechanism module used by MobileNetV3 is SE Networks(Squeeze-and-Excitation Networks),which does not consider the space information of the image,so it is decided to introduce two hybrid mechanisms BAM and CA into MobileNetV3 to design two improved neural network BAM-MobileNet and CA-MobileNet.In order to further improve the generalization of the recognition model,the method of rotation and adding Gaussian noise was used to expand the disease data set to 7771 pieces.The extended apple tree disease leaves data set was used for training.(3)In order to meet the needs of practical application,a common disease identification system of apple leaf is developed,including Web terminal and mobile APP.The system is based on Django framework and mainly includes image preprocessing and disease identification functions.By uploading the pest pictures to the server,calling the trained CA-MobileNetV3 model,returning the identification results,displaying the interface,integrating the identified disease characteristics and manage methods,providing farmers with accurate disease information and guidance,and providing a fast and accurate way for disease manage.Through experiments,it is finally proved that the MobileNetV3 neural network model has the advantages of small model,high recognition accuracy and fast training speed compared with the classical networks such as VGG16,ResNet50 and Inception-V3.The improved MobileNetV3 neural network model has higher recognition accuracy when the model size is almost unchanged,which shows that hybrid attention is effective for improving the performance of MobileNetV3 neural networkThe above research is helpful to realize the rapid identification of apple leaf diseases and insect pests.The apple leaf diseases identification system developed according to the experimental results can provide help and guidance for the majority of fruit farmers.