Rice Pest And Disease Monitoring Based On Multi-scale Remote Sensing Technology

Rice is one of the main food crops in China and is of great significance to the survival and development of our people.A major disturbance factor affecting rice growth is rice pests and diseases.A variety of rice pests and diseases not only affect rice yield,but also affect the quality of rice.Therefore,accurate monitoring and scientific control of rice pests and diseases are critical.This thesis takes rice as the research object,and takes Jianli County of Hubei Province as the research pilot.With the Tiangong-2 wide-band imager in the study area,the near-infrared image,the sentinel 2A image and the hyperspectral non-imaging monitor onboard the drone and Rice canopy spectral data and high-definition images acquired by high-definition imager,using convolutional neural network algorithm(CNN),vegetation index algorithm and probabilistic neural network algorithm(PNN)from regional scale,The field scale and canopy scale are used to quantitatively study the spatial distribution information of different types of rice and the monitoring of rice pests and diseases.The main results obtained from the following are as follows:(1)The regional scale relies on the large-scale multi-spectral remote sensing image,using the convolutional neural network algorithm,the modeling accuracy is 95.72%,the overall accuracy of the classification result is 97.01%,and the Kappa coefficient is 0.96,which is the accurate extraction of spatial distribution of different types of rice planting.A good foundation has been laid.(2)The field scale analysis of hyperspectral data was carried out by means of hyperspectral non-imaging monitors onboard with drones,spectral analysis under spectrum stress,pest stress,and different interpolation methods.The algorithm research was integrated.Vegetation index algorithm and probabilistic neural network algorithm establish a correlation model between hyperspectral data and crop growth conditions.(3)The canopy scale uses the high-definition imager of the low-altitude drone platform to obtain the high-resolution image of the canopy rice and the ground measured data,and uses the agricultural expert visual interpretation and computer vision method to diagnose the true state of the damage severity of rice,and Accuracy verification is carried out with hyperspectral prediction data.The verification method adopts ROC curve accuracy evaluation method.Finally,the detection rate of the sample trained by the neural network model is 95.8%,and the false alarm rate is 17%.(4)In the course of the research,the system design and operation process analysis of the intelligent agriculture inspection and integration machine system were explored,which opened up a new path for intelligent monitoring and prevention of pests and diseases.