Research And Application Of Machine Vision-based For Grape Leaf Diseases

Grapes have a long history in China and are not only of high nutritional value and widely used in the winemaking industry,but their large-scale cultivation mode also gives an opportunity for the spread of infectious diseases.The traditional recognition of diseases requires growers to rely on experience,which is not only highly subjective and less accurate,but also prone to blindly spraying pesticides,causing environmental pollution and disease resistance,and missing the best time to prevent and control diseases.With the development and progress of artificial intelligence,the application of machine vision in the field of agricultural image processing is becoming more and more widespread,especially in the field of crop disease recognition.When grapes are infected with diseases,there are more characteristics reflected in the leaves,so we use this characteristic to take grape leaves as the research object and classify them using machine vision related technology.The experimental results show that the classification model proposed in this paper achieves good results in the classification of grape leaves diseases.In this study,three types of grape leaves diseases are investigated as follows: first,data augmentation is performed using a cycle consistent adversarial generative network.A suitable initial learning rate is set and the model is optimized iteratively by a cycle learning rate algorithm,and the resulting loss curves are analyzed and summarized.At the same time,the visu alization results of the cycle consistent adversarial generative network are compared with the visualization results of traditional image augmentation to analyze the quality of the generated images.4062 samples of the original data set are generated and e xpanded to 8124 samples as input for iterative optimization of the classification model to improve the performance of the subsequent classification model.Second,transfer learning is introduced.The pre-trained weight parameters of the ImageNet dataset are introduced to the classification model to reduce the model training parameters and improve the optimization efficiency of the model.The dropout strategy is added to randomly freeze the neurons of the network model and the diversity of sample data brough t by pre-training to suppress overfitting,and it can improve the average accuracy of the model classification recognition while reducing the number of samples required for the dataset.Finally,four typical convolutional neural network models were combine d for grape leaf image recognition,namely AlexNet,Vgg16＿bn,ResNet34,and DenseNet121 A suitable initial learning rate was set and a cycle learning rate algorithm was used for iterative optimization of the model,resulting in training loss,validation loss,accuracy images,and confusion matrix.A total of 2031 photos were identified using 584 photos of grape black rot leaves,694 photos of esca virus leaves,546 photos of leaf blight leaves and 207 healthy leaves,yielding an average accuracy of 97.96%,97.58%,98.42% and 98.76% for the four network models,respectively,which is a significant improvement over the current field of grapevine leaf disease recognition.By summarizing and analyzing the parameters such as accuracy,recall,precision,F1 score,confusion matrix and the recognition ability of each network model for the three disease types through iterative optimization,the optimal classification network model was selected as DenseNet121 by combining the hardware resource share,and it was deploye d on the web with the grape leaf data augmentation model,so that users can upload grape leaf images for The disease recognition and image augmentation were performed by uploading images of grape leaves.