Research On Citrus Disease Recognition Based On Convolutional Neural Network

China is a large citrus-growing country,with a wide area of citrus cultivation and huge production,which is an important contribution to the country’s economic development.However,due to climate change and poor manual management,citrus planting is usually accompanied by various disease problems,which lead to a decline in yield and quality,and in serious cases,the death of the fruit tree.Therefore,early detection and management of diseases is very effective.However,traditional machine vision methods for identifying citrus leaf diseases are mainly based on manual selection to extract information about the feature areas,which is time consuming.This makes it difficult to achieve accurate identification of leaves with complex disease areas in the field environment.With the development of computer technology,deep learning has been widely used in various fields.Convolutional neural networks,as one of the algorithms of deep learning,are widely used in the field of image recognition due to their strong ability to automatically extract feature information.Therefore,in order to achieve efficient and accurate recognition of citrus diseases,this paper investigates the recognition of citrus diseases based on convolutional neural networks.The main research work is as follows:(1)The problem of inadequate feature extraction capability and insufficient feature information extracted by a single network,so a fusion model based on Inception V3 and Res Net34 was constructed.The CBAM attention mechanism is embedded respectively to focus on the useful information of the input features from the channel and spatial levels.The improved models are fused to improve the expressiveness and recognition accuracy of the models.This paper uses data augmentation techniques to pre-process experimental data with the aim of enhancing the experimental results and meeting the requirements of deep learning on data volume.This experiment expands the dataset by data enhancement methods such as rotating,panning,changing brightness and adding noise.Finally,experiments are conducted on the dataset using the pre-processed data.The results show that the proposed method in this paper achieves an accuracy of98.49% on the test set,which is better than other methods in terms of recognition.(2)Deep learning networks have problems such as long training time,large model size and consideration of practical application deployment,so this paper uses migration learning and lightweight convolutional neural networks for citrus disease leaf recognition research.This paper uses a model-based training approach for migration learning,where the Mobile Net V2 model is trained on the source domain to obtain a pre-trained model.The structure of the pre-trained model was optimised and loaded onto the target domain for retraining,and the optimal number of freezing layers was obtained by fine-tuning,resulting in a citrus leaf disease recognition model.The results show an accuracy of 94.69% on the test set and a model size of 43.6MB.This can be applied to citrus leaf disease recognition in a real-world environment.(3)Lightweight citrus disease network recognition model deployed in the cloud with the aim of obtaining a fast method for citrus leaf disease identification.This paper uses the Flask framework to build a cloud server-based online citrus disease detection and control platform.The system functions include registration,login,query,image upload,stage warning,relevant control recommendations and modification of user information.Different functional modules are responsible for different user requirements,and users perform relevant operations via mobile terminal,which is of great significance to help growers in citrus orchard control.