基于高分辨率卫星影像的大区域松材线虫病预警监测

The pine wilt disease epidemic is a devastating disaster,characterized by a wide range of hazards,fast onset rate,multiple transmission routes,and difficult management.Traditional methods of investigation have high costs,huge workload,long time cycles,and are prone to omissions,making it difficult to meet the needs of emergency monitoring in large areas.Starting from three aspects:early warning,monitoring,and commercialization of pine wilt disease,we will construct a sky ground collaborative monitoring technology system for the prevention and control of pine wilt disease,in order to achieve the goal of pine wilt disease prevention and control.1)Firstly,in terms of early warning of pine wilt distribution,a maximum entropy distribution prediction model was constructed by considering natural requirements,human activities,and landscape elements from multiple perspectives in the disaster prone environment.Improvements were made to two of these variables.Firstly,as human activities are the main transmission route of pine wilt disease,this study used two Jilin-1 sub satellite images with a 1-year interval during the monitoring window.A twin neural network model was used to automatically identify construction sites within a 2km radius around the pine forest,which were input as one of the human activity factors into the model;The second method is to extract the dominant tree species as small groups of pine forests based on a forest map,and use MODIS inversion to input the model of pine canopy dryness.The monitoring area is divided into four categories of high,medium,low,and unsuitable areas for pine wilt disease occurrence based on forestry small groups,as an important basis for forest inspection and census.2)Secondly,based on the"sky ground"collaborative model,conduct a survey and monitoring of abnormal discolored pine trees.Among them,"empty"refers to the collection and construction of positive and negative sample datasets of centimeter level unmanned aerial vehicle(UAV)images of abnormally discolored Pinus massoniana trees,based on the layout of epidemic small classes and predicted areas with high survival suitability for pine wilt disease,uniformly distributed within the monitoring area,and downscaled sampling of the sample set;10%of the total sample size is sampled for on-site verification,and the sample set is subjected to correction evaluation until the downscaled samples meet the accuracy requirements;"Tian"refers to the 0.5m resolution satellite image and 10 meter multispectral image synchronously obtained during the detection window period.The YOLOv8 small target recognition model is trained using existing downscaled samples.When the confidence threshold is set to 0.3,the sub meter level satellite image recognition color change wood recall rate exceeds 90%.The sensitive spectral features filtered from multispectral images are enhanced,and then the pseudo color change points in target recognition are distinguished using a time series classification algorithm based on deep learning.The final color change wood discrimination accuracy reaches 85%.3)In order to solve the practical business application problems of multiple grassroots forestry units in large regions with a large number of color changing points,a supporting sky ground collaborative platform and mobile verification terminals have been developed to solve the business work of point map spot task distribution,data synchronization,multi-phase data management,zoning statistics,forest area verification and reporting under no signal mobile terminals,sampling and removal,forming a complete collaborative monitoring technology system,and building a universal technical and business support for protecting forest resources and ecological security.