Investigation And Study Of Ginkgo Forest Stand Factors Based On 3D UAV Remote Sensing Modeling

Forest is a major component of the terrestrial ecosystem as well as the important natural resources on which human beings depend.The forest spatial structure information plays an important role in ecological research and resource monitoring.The individual tree includes such structural parameters as crown width(CW),tree height(H),and diameter at breast height(DBH)and so on,which are the basic work for investigating forest resource and the basic parameters for evaluating forest aboveground biomass.The traditional forest survey is usually realized by manually checking the length of each tree,which is labor-intensive and time-consuming.In recent years,the rapid development of UAV remote sensing,which has the advantages of“flexible operating methods and low cost of data acquisition”,has attracted widespread attention in related industries.The UAV image processing technology based on computer vision technology continues to mature,which provides a new development opportunity for forest remote sensing research based on the UAV images.In this paper,we choose Ginkgo,basketball hall and Euonymus japonicus in the North Campus of Northwest A&F University as the research object,we studied the application of UAV image in the investigation of stand factors,and combined artificially measured data(Crown Width,CW;Tree Height,H;and Diameter at breast height,DBH),and selected several commonly used vegetation indices(r,g,b,RGRI,NGRDI,EXGR and VARI),used a variety of combination methods and mathematical modeling methods to establish different types of DBH prediction models,and comprehensive evaluation of the model according to different evaluation indicators.In addition,the 3D reconstruction methods of different objects based on UAV remote sensing are also compared and analyzed.The main conclusions are listed as follows:(1)For the two 3D reconstruction methods of Context Capture and Agisoft Photoscan,both in the three-dimensional reconstruction of basketball hall and the Euonymus japonicus,the former outperformed the latter;for the two objects of basketball hall and Euonymus japonicus,whether using Context Capture or Agisoft Photoscan generated two objects'3D models,the error of the basketball hall is smaller than that of the Euonymus japonicus,indicating that the basketball hall is more recoverable for 3D reconstruction.(2)Obtained the orthophoto the three-dimensional point cloud of the study area and extracted the crown as well as the tree height of the single tree.For images captured by a UAV equipped with a camera,the structure from motion(SFM)algorithm and the regional network adjustment technology were used to generate the two-dimensional orthophoto and three-dimensional point cloud data of the study area.The appropriate scale of the multi-scale segmentation algorithm was adjusted to effectively segment the ginkgo canopy and extract the single wood crown width.The crown width is obtained by Arc GIS using visual interpretation method;ENVI Li DAR is used to process the point cloud data to generate three-dimensional models.The sliding box algorithm is used in the three-dimensional point cloud model to extract the height of the single tree.(3)Completed the estimation of the crown width and height of the single tree canopy.Comparing the extracted values of UAV of crown width and tree height with artificially measured data,the results show that crown width and tree height are mostly underestimated.By establishing different mathematical models and analyzing evaluation indicators,the optimal crown width and tree height extraction are selected.Among them,the difference of the crown amplitude estimates is within±0.5 m.The quadratic polynomial has the best prediction effect,the determination coefficient(R~2)reaches 0.749,the Root mean square error(RMSE)is 0.369 m,and the mean absolute error(MAE)is 0.295 m;The difference between the estimated values of the tree height is within±3 m,the prediction effect of the exponential function prediction model is the best,the determination coefficient(R~2)reaches0.694,the minimum root mean square error(RMSE)is 0.528 m,and the mean absolute error(MAE)is 0.335 m.(4)Established a one-dimensional DBH prediction model(CW--DBH,H--DBH),a binary DBH prediction model(CW&VI--DBH,H&VI--DBH,CW&H--DBH)and a ternary DBH prediction model(CW&H&VI--DBH).According to different analysis and evaluation indicators,the best prediction model for DBH is the ternary model(CW&H&VI--DBH):DBH=1.207*CW+0.728*H+12.367*VI+4.098,where the vegetation index represents the r,the prediction coefficient(R~2)of this prediction model reaches 0.888,the root mean square error(RMSE)is 0.433 cm,and the mean absolute error(MAE)is 0.325 cm.In addition,the accuracy of the ternary prediction model has also been greatly improved compared to the accuracy of the ternary prediction model.R~2 is increased from the lowest 0.630 to 0.888.The research showed that the UAV images can effectively extract the crown width and the tree height,and the vegetation index can be used to predict the DBH with high accuracy,so as to realize the biomass estimation,and it provides new ideas for the automatic investigation and monitoring of forest resources.