| As a modern agricultural country,the history of grape cultivation in China can be traced back to the Jianyuan period of Emperor Wu of Han Dynasty.It not only has a long planting history,but also has the largest export volume of grape in the world.Grape diseases have seriously affected the yield of grapes,but also damaged the quality of grapes and their downstream food products to a certain extent,resulting in huge economic losses.Therefore,the purpose of this study is to alleviate the impact of grape diseases,design methods to achieve rapid and accurate detection in the early stage of the disease,so as to provide a basis for disease control.In this study,the focus of disease recognition is focused on the grape leaves.In the process of image preprocessing,this paper designs an image enhancement algorithm based on improved fractional differential mask;in the aspect of image recognition,this paper studies grape leaf disease recognition by two methods: traditional machine learning and deep learning.In this paper,in the use of traditional machine learning to identify grape leaf diseases,firstly,image processing technology is used to segment the foreground and background of grape leaves;secondly,image filtering template based on improved fractional differential is used to enhance image details and texture;thirdly,the color,texture,shape,LBP and hog features of the lesion area and foreground area are extracted and cascaded;finally,traditional classification algorithm is used Complete disease identification.In the process of using deep learning to recognize grape leaf disease,firstly,the image filtering template based on improved fractional differentiation and image affine transformation are used to expand the data set;secondly,the combined features are generated based on the joint model proposed in this paper;finally,the recognition of grape leaf disease is completed through Softmax layer.After verification,the improved fractional differential method designed in this paper can enhance the leaf image well,and enhance the image details,texture and other effective information.Compared with the original image features,the features extracted from the enhanced image are more distinguishable,and the model recognition accuracy is improved by about 1%.Compared with the single LBP feature and hog feature,the proposed feature fusion method improves the model accuracy by about 3-5%.The deep learning combination model proposed in this paper has achieved excellent results in grape leaf disease recognition,and can achieve 100% recognition accuracy in the grape disease test set of Plant Village data set.The first mock exam is based on the improved fractional differential algorithm.The proposed algorithm can enhance the image and enhance the distinguishability of the feature.The combined feature of the proposed algorithm is more helpful to the traditional classification model recognition.The deep learning combination model proposed in this paper has better recognition ability than the single model which is popular.Relatively speaking,both traditional machine learning method and deep learning method have good performance in grape leaf disease classification.Deep learning method has better recognition accuracy,while traditional machine learning method has lower time loss.Therefore,in the actual production,the platform with poor computing power can be identified by machine learning method.If the computing resources are sufficient,the combined neural network model with better performance can be used.In the actual production,the two methods can reduce the time for fruit farmers to distinguish and find grape diseases as soon as possible,which has high practical value. |
