Research And Design Of Recognition Algorithm For Wolfberry Pests And Diseases

Accurate identification of crop pests and diseases is an important prerequisite for pest and disease prediction and prevention.Image recognition has become a research hotspot and main technical means for the prevention and control of pests and diseases in recent years due to its advantages of high efficiency,low cost and easy operation.As a characteristic advantage industry in Ningxia,the wolfberry industry plays an important role in the rural economy in all areas of Ningxia.In addition to the impact of market prices,there are two factors that have the greatest impact on the yield of Ningxia wolfberry:one is climate,and the other is pests and diseases.The experimental results show that the improved MobileNet V2 target recognition algorithm proposed in this paper can effectively identify 11 species of wolfberry pests and diseases.In this paper,the research on the image recognition algorithm of wolfberry pests and diseases will lay a certain foundation for the prevention and control of wolfberry.The main work and research results of this paper include the following aspects:(1)Collect and make image data sets of wolfberry pests and diseases.According to the research needs of this paper,a total of 1955 image sample sets containing 11 types of wolfberry pests and diseases were collected.In order to alleviate the over-fitting phenomenon,this paper uses the partial image space transform algorithm and the addition of Gaussian noise to expand the training set of the original sample.The expanded sample set number is 19115.(2)Select a network model suitable for the research on image recognition of wolfberry pests and diseases in this paper.From the theoretical and experimental aspects of analysis and comparison,MobileNet V2 was selected as the network model for classification and identification of wolfberry pests and diseases images.The network model achieved a good balance between recognition accuracy,model parameters and calculation time,and selected VggNet,Inception and ResNet,the three classical convolutional neural networks as the comparative network model,and related experiments and analysis were carried out using the dataset of the wolfberry pest and disease images.(3)In this paper,an improved MobileNet V2 target recognition algorithm is designed,which embeds the SE module in the MobileNet V2 network to obtain the new network model SE-MobileNet V2,which realizes the adaptive calibration of the feature channel.The experimental results show that the improved network model converges faster and improves the classification and recognition accuracy of wolfberry pest and disease images by about 2.46%.(4)Based on the image dataset of wolfberry pests and diseases,on the improved network model SE-MobileNet V2,different optimization methods are selected for multiple training to determine the optimization algorithm,network parameters and activation function suitable for the network model,which further improved the recognition accuracy by 1.02%and finally reached 97.67%.The training model was used to test the classification and identification of wolfberry pests and diseases on the webpage.The recognition time of single images was less than 15ms.