Research On Apple Leaf Disease Diagnosis Based On Convolution Neural Network

China is the world’s largest producer and consumer of apples,with more than 40% of the planted area and production in this world.Currently,the apple industry has become the third largest growing industry in China,after grain and vegetables.However,apple diseases have a serious impact on the yield and quality of apples.Therefore,accurate identification of apple foliar diseases is a priority for timely control measures to effectively stop damage.In response to the inability of existing disease identification methods to accurately diagnose apple leaf spots at different scales,this paper explores the visual patterns of disease characteristics and carries out the following research work.(1)A novel method for apple leaf disease identification is proposed.To address the problem of low accuracy of existing apple leaf disease recognition models,a novel convolutional neural network recognition model,Apple＿Incep Net,is constructed in this paper.firstly,a pre-module consisting of alternating 3×3 convolution and pooling is designed to extract the underlying features.Then a multiplexed branching module was constructed based on the dilated convolution to achieve the extraction of multi-scale lesion features.Finally,a dense connection strategy is used to achieve information fusion of the underlying local features and the higher-level semantic features for accurate classification.The disease recognition model proposed in this paper can effectively extract the disease spot features and make accurate diagnosis of five apple leaf diseases.The experimental results show that the accuracy of Apple＿Incep Net on the apple leaf disease dataset can reach 96.63%,and the method is of great importance for improving apple quality.(2)A hierarchical apple leaf disease classification model was constructed.In order to solve the problem of significant differences in the visual characteristics presented by large and small spots,a hierarchical multi-task classification model is proposed in this paper to achieve more specific depth features to guide the classification according to the different feature types of spots.Firstly,the dataset is repartitioned based on the publicly available small target dataset standard(32×32 pixels);secondly,a predictive probability fusion strategy is proposed to construct a hierarchical cascade classification structure that effectively exploits the similarity between the spot features.The cascade classification model proposed in this paper is able to classify apple leaf diseases at a fine-grained level.Experimental results show that the classification accuracy of the hierarchical structure can reach 97.61%,which provides a new idea for crop disease identification.(3)A diagnosis system for common apple leaf diseases was developed.In view of the lack of practical methods to diagnose apple leaf diseases in actual production scenarios,the system was combined with the constructed hierarchical apple leaf disease identification model to develop a diagnosis system for apple leaf diseases.The two main functions of the system are the identification of five common apple leaf diseases and the introduction of disease control measures towards different degrees,making it easier for growers to diagnose apple leaf diseases in a timely and accurate manner and effectively reduce economic losses.