Research On Grape Leaf Disease Identification Based On Deep Convolutional Generative Adversarial Network

Grapes are one of the top four fruits are popular with the public because of their blood vessel softening and aging-delaying properties,compared with traditional African growing countries,there are some difference between quality and quantity.Leaf is highly susceptible to diseases and their symptoms are similar as well.Lack of accurate disease identification techniques by farmers can lead to irrational use of pesticides,result in problems such as poorer quality and lower yields.Therefore,we build an accurate and effective grape leaf disease identification model,which can play an important role in improving the yield and soil protection of farmers.Details of the research are as follows:(1)The study of grape leaf disease data enhancement algorithms based on deep convolutional generative adversarial networks.We proposes a new DCGAN(I-DCGAN)data enhancement method,which achieves fast convergence of the model by expanding the resolution of the network structure,improving the activation function,and optimizing the learning rate.According to the characteristics of small size,low pixel size and difficulty in the cutting of disease images generated by the original deep convolutional generative adversarial network(DCGAN),Residual Blocks are introduced into the model to realize the deep extraction of grape leaf disease features and two evaluation metrics,FID and Avg-SSIM,are used to verify the validity and diversity of the generated images.The experimental results show that our algorithm can solve well the problems of the low resolution of the original network,small image size,and not easy to cut,meanwhile,the generated data is conducive to the fast extraction of disease features,thus achieving fast convergence of the network.(2)Research on Dense Net121 grape leaf disease identification.To solve the problems of low efficiency,slow convergence,and large errors in traditional grape leaf disease recognition,We proposes an improved Dense Net121 model for grape disease recognition based on IDCGAN data enhancement techniques.By adding SK to the original Dense Net121 model,each SK is embedded into each dense block to achieve deep extraction of the input disease feature regions.To further optimize the structure of the network model and improve the accuracy of the model,the Efficient Channel Attention is also introduced.The experimental results show that our improved model has a certain advantage over the original network in terms of storage footprint and recognition speed,and the recognition accuracy is 96.53%,which is higher than the 92.88% of the original Dense Net121.(3)Development of the grape leaf disease recognition system.In this paper,the Py Qt5-based grape leaf disease identification system is developed to visualize,accurately,and quickly.The interface of the system includes seven interfaces: disease information,model selection,model identification,control suggestions,disease statistics,user records,and help.The system is convenient and easy to use,and help farmers to effectively achieve rapid identification of grape leaf diseases.