Investigation And Monitoring Technology Of Forest Resources In Karst Mountainous Region Of Guizhou Province By Integrating UAV LiDAR And Visible Light Camera

Forest resource investigation and monitoring is the basic work of forest resource management,a necessary link to promote the sustainable development of regional ecosystem,and an important guarantee for the safe and orderly operation of social production and life.Guizhou is a typical karst mountain,although forest coverage in the national top ten,but rocky desertification is severe,ecologically fragile,the environmental bearing capacity is low,frequent natural disasters,biomass,community is simple,therefore,do a good job for our province forest resource survey and monitoring of resource supply,industry development,ecological regulation,disaster reduction,disaster prevention,such as green development is of great significance.In view of the problems of high time complexity and low efficiency existing in the current forest resource survey,the investigation of stand height,stand density and stand species in the forest resource survey,In this paper,the watershed algorithm is improved,and the improved U-NET model and the improved 3D-CNN model are used to carry out forest survey based on uav airborne radar and UAV airborne visible camera.In terms of data acquisition,airborne laser radar of uav is used to collect data in the research area.At the same time,visible light camera was used to collect data in the study area.Two plots were divided by total station.Total station,RTK equipment,altimeter and other equipment were used for manual data survey.In terms of data processing,Canopy Height Model(CHM)was generated by point cloud data of airborne LIdar of UAV,and Digital Orthophoto Map(DOM)was generated by images taken by visible light camera.For the canopy height model generated from point cloud data,the improved watershed algorithm was used to segment single trees and extract tree height,plant number and other information.The improved 3D-CNN network was used to train the separated single trees and classify the tree species.It is compared with traditional 3D-CNN and random forest methods.The improved U-NET network was used for single tree segmentation and extraction of tree height,tree number and other information for digital forward photography images.The improved U-NET was used for single tree segmentation and classification,and compared with FCN and traditional U-NET algorithm.The model trained in plot 1 was used to separate individual trees and classify tree species in plot 2,so as to verify the robustness of the algorithm for different slope plots.In this study,single wood segmentation of Pinus huashan,Pinus massoniana,Cunninghamia lanceolata,Juniper and Platycladus orientalis was conducted.The overall accuracy of single wood segmentation is more than 60%,which can meet the accuracy demand of forest department for single wood acquisition.The effectiveness and advantages of the proposed method in the process of single wood segmentation and recognition are verified by experiments.