Research And Implementation Of Intelligent Monitoring System Of Diseases And Pests On Rice Canopy

Diseases and pests of the rice canopy are one of the main reasons affecting rice yield and quality.Strengthening the monitoring of rice canopy diseases and pests is of great significance to guide the prevention and control of rice diseases and pests and food safety production.At present,the survey method of rice canopy diseases and pests is mainly based on manual survey.This model has problems such as time-consuming and labor-intensive,strong subjectivity,and difficult to trace historical data.It is difficult to meet the needs of real-time monitoring of diseases and pests on rice canopy in large areas.Satellite remote sensing technology does not have small-scale monitoring capabilities,and it is difficult to accurately detect the occurrence of diseases and pests in each rice plant.In this paper,three common diseases and pests on rice canopy,the Cnaphalocrocis medinalis,the Chilo suppressalis and the Ustilaginoidea virens were studied,and the detection model of diseases and pests on rice canopy based on deep learning was studied.Based on the model,an intelligent monitoring system for diseases and pests rice canopy that can be applied in natural scenes was built,realizing multi-point real-time monitoring and automatic and accurate diagnosis of diseases and pests on rice canopy.The main research content and results are as follows:(1)An intelligent monitoring system for diseases and pests on rice canopy was designed.The system mainly includes three parts: image acquisition end,application server end and clien.The image acquisition end is responsible for timing and fixed-point automatic capture of high-definition images of the rice canopy,and sends the images to the application server through the 4G network.The application server is responsible for loading the image detection service to automatically detect the diseases and pests in the image and store the detection data.The client is responsible for real-time video preview,setting device parameters and displaying detection results.According to the system design scheme,various functions of the system have been implemented.The application server is developed by using the Spring Boot framework,and the My SQL database is used to store various information.The image detection service realizes cross-language calls through the Django framework,and the Web client is developed by using the Vue framework.The functional integrity test and performance test of the system are carried out,and the test results show that the system meets the expected design goals.(2)Establish and compare three detection models of diseases and pests on rice canopy based on deep learning.Trained and tested the Two-stage target detection algorithm Faster R-CNN,the One-stage target detection algorithm Retina Net and YOLOv4.The test results show that the m AP value of YOLOv4 is 84.01%,which is0.88 and 12.10 percentage points higher than the detection accuracy of the Faster R-CNN and Retina Net models,respectively,and the FPS of YOLOv4 reaches 25.64 frames per second,much higher than the other two Model.Finally,the YOLOv4 model was selected as the basic network for diseases and pests on rice canopy detection,and it was improved.(3)A YOLO-DPD-based detection model for diseases and pests on rice canopy was established.In order to further improve the detection accuracy of the YOLOv4 model,an improved model YOLO-Diseases and Pests Detection(YOLO-DPD)was proposed to detect diseases and pests on rice canopy.The model uses the Residual Feature Augmentation method to narrow the semantic gap between different scale features of diseases and pests on rice canopy;the Convolution Block Attention Module is introduced into the backbone network,which significantly enhances the characteristics of diseases and pests on rice canopy,and suppresses the background noise.The test results show that the AP values of the model for the detection of the C.medinalis,the C.suppressalis and the U.virens are 92.24%,87.35% and 90.74%,respectively,which are 8.06%,5.50% and 4.73% higher than before the improvement.and the FPS of this model can reach 21.28 frames per second.