Research And Design Of Ginger Disease Recognition System Based On Optimal Convolutional Neural Network

Ginger,alias Ginger Root or Dixin,is a perennial herbal rhizome.It not only has edible value,but also softens blood vessels and promotes blood circulation in medicine.It has significant therapeutic effect on various discomforts caused by body cold.At present,ginger has been processed into sauce ginger,distiller's grains ginger,dried ginger slices,sour ginger and other forms.It has been widely circulated in major agricultural markets and has become an indispensable part of the production and life of the vast number of residents in China.At the same time,ginger disease is becoming more and more serious,which not only brings a devastating economic blow to the vast number of agricultural producers,but also has a huge impact on the agricultural market.Therefore,in view of the importance of ginger in China's agricultural market and the complexity of disease identification in the field environment,based on the picture of ginger disease collected in the natural environment,the anthrax,ginger blast,root-knot nematode and white star disease were studied and analyzed.A ginger disease identification system based on convolution neural network was proposed.The system collected ginger diseases firstly.In order to enhance the reliability of the data,binarization and contour segmentation are used to preprocess the images.Secondly,the processed image data are handed over to the optimized convolutional neural network model for analysis and learning,and simulated in the framework of Caffe.The results show that the recognition rate of the optimized model has been significantly improved.Finally,on the basis of the trained network model,we use Qt software to design human-computer interaction interface,so as to achieve data visualization and improve the convenience of the system.The main contents of the system are as follows:(1)In order to improve the validity and accuracy of the training neural network model,it is necessary to pre-process the training neural network model,such as removing redundant information and noise,such as binarization of disease image and segmentation and preservation of target area.(2)Based on the traditional LeNet-5 network model,a convolution neural network consisting of three layers of convolution layer,three layers of pooling layer and three layers of full connection layer is designed,and features are fused with the high-dimensional features extracted from Incepetin structure in Google LeNet model,fully considering the characteristics of ginger disease image.In addition,in order to further improve the recognition rate of the model,a new Swish activation function and BN layer are also introduced.(3)After setting up the model and related parameters,the simulation is carried out under the Caffe framework.In this study,comparative experiments are used to compare the original LeNet-5 model with the optimized hybrid convolution neural network model under the same conditions.The results show that the recognition rate of the optimized model is increased by 92%,which is significantly higher than that before optimization.(4)In order to improve the usability and reliability of the system,based on the Caffe framework,the human-computer interaction interface is designed by using Qt software.By feeding back the types of ginger diseases and their matching degree,users can quickly obtain effective information and make the most effective treatment based on their own experience.In addition,the interface also provides corresponding prevention and treatment measures for the types of ginger diseases with high matching degree.In this paper,a timely and effective identification system of ginger diseases is designed by using convolutional neural network algorithm in deep learning technology.Combining with data visualization interface,the problem of disease prevention encountered by ginger growers in production is alleviated.