Research On Vegetable Disease Monitoring Method Based On Drone Image

In agricultural detection,there are many types of diseases,diverse expressions,and the same disease behaves differently on different species of plants.The disease can even cause crop yield reduction in severe cases,so it is very important to detect crop diseases.In recent years,manual identification of the degree of disease alone can no longer meet the needs of large-scale production.Early detection of disease and timely treatment to prevent the spread of disease is a key aspect of ensuring crop yield and quality,and is one of the effective means to ensure yield.Research work on crop diseases in the growing environment has been carried out because of the high incidence of diseases in crops that are not easily detected and the complex background environment in which they grow and the variety of diseases.The main research work is as follows:(1)For crop diseases,UAVs are used for field disease sample collection and to produce a crop disease dataset based on UAV images,containing healthy,mildly diseased,severely diseased and an atlas of healthy and disease-infected crops co-existing in the same camera.Supervised learning is used to process the crop disease dataset.(2)In the process of image acquisition by UAV,the best flight attitude of UAV is explored,and the best flight height,speed and path for UAV image acquisition are figured out.Also the problem of blurred images due to the UAV flight process.Super-resolution processing is used to refine the image details and enrich the image features.Ensure the accuracy of subsequent disease features.(3)To address the traditional defective problem of disease recognition,a disease detection algorithm based on improved SSD(Diseases Recognition Based on SSD,DRBS)is proposed in this paper.The algorithm incorporates an attention mechanism and adjusts the network structure of feature extraction.The problem that the algorithm does not pay enough attention to image detail information is improved,and the image feature extraction capability is optimized.Several improved methods are tried and various classical target detection algorithms are compared simultaneously.Experiments show that the improved algorithms have improved from robustness and accuracy.(4)A disease grading test is proposed to grade the disease of plants into healthy,moderate disease,and mild disease,and to focus on the early control of disease.This is because overly heavy diseases can be recognized by the naked eye even by inexperienced people.At the same time,heavy diseases have missed the best control period.So heavy diseases are instead less significant and the focus is on the control of lighter diseases for the purpose.(5)Develop a GUI interface for disease detection,visualize operation,and support target detection picture detection,video detection,and camera detection.Through this platform,even people who do not know about target detection can operate to achieve disease detection.