Research On The Inventory Of Forest Structure Parameters Using UAV Stereo Imagery

Forest is a major component of the land ecosystem as well as important natural resources for the subsistence of human beings.The forest spatial structure information play an important role in ecological research and resource monitoring.The traditional artificial measuring method could obtain the information,but is labor-intensive and time-consuming.In recent years,with the rapid development of unmanned aerial vehicle(UAV)platform,UAV remote sensing has drawn wide attention with the advantages of flexible operation and low cost of data acquisition.In particular,the UAV image processing technology based on computer vision technology is becoming mature,which provides a new opportunity for the development of forest remote sensing research based on UAV images.In this paper,we have studied the methods and techniques of the investigation on the forest structure parameters from UAV photogrammetry data in the following four aspects:(1)The acquisition and processing of UAV images.The forest image data was obtained by UAV at the Inner Mongolia Greater Khingan Range region.UAV remote sensing data with the overlap RGB UAV images were processed by the PhotoScan oftware.The processed data include: image point cloud and digital orthophoto map(Digital Orthophoto Map,DOM),digital surface model(DSM),The point cloud were classified into ground and non-ground points by a filtering algorithm.The digital elevation model(DEM)was generated from the ground points,and then the canopy height model(CHM)was obtained from DSM and DEM.(2)Individual tree identification from stereo-imagery point cloud.We draw lessons from the individual tree identification algorithm used for the Li DAR data.Aiming at the error recognitions in the study area of tree recognition,a new algorithm based on UAV point cloud and single tree crown structure analysis was proposed.Through the analysis of the vertical structure of the tree crown,the algorithm can eliminate the false tree,thus to improve the accuracy of tree recognition.(3)Improvement of missed trees due to crown overlapping in dense forests.The existing recognition algorithm have serious leakage problem in dense forest.We proposed an single tree identification algorithm by the mathematical morphology method with the orthophoto(DOM)and digital surface model(DSM)data.The algorithm comprehensively uses DOM'S spectra and DSM's geometric information for extraction of tree canopy.Meanwhile the expansion and erosion operations and connected region labeling of mathematical morphology were used to overcome the tree crown overlapping problem in dense forest.Finally,the separation and identification of tree crowns were realized,and the tree crown center were obtained.(4)Extraction of individual structural parameters(tree height and tree crown)and inversion of diameter at breast height(DBH)and estimation of tree biomass.The tree height as extracted from CHM by the corresponding locations of recognized individual tree.The tree crown was extracted by the watershed segmentation using individual tree position as a mark to realize accurate segmentation of individual tree crown.Individual tree DBH is an important parameter for the estimation of tree biomass,as remote sensing data can not extract individual DBH directly.We use the tree structure parameters which could extract to inverte DBH by establishing related models,then the biomass of single tree biomass is calculated using biomass allometric equation,and the biomass of plots.