Tree Parameters Extraction Using UAV Remote Sensing Image And SFM Point Cloud

Tree parameter information is widely used in the fields of forestry resource management,ecological environment protection,and physical geography process simulation.In recent years,the rapid development of UAV remote sensing can obtain fine information on surface features,and provide new technical means for the extraction of tree parameters.Firstly,based on UAV remote sensing image and SFM solution point cloud data,the paper extracted information of different vegetation regions by using object-oriented classification method.Aiming at the vegetation area composed of a single tree species,a single tree identification method based on image template matching was established in this paper.For the vegetation areas of different tree species,the regional maximum watershed labeling algorithm was used to extract the individual tree information and the UAV hyperspectral data was used to classify on the result of the individual tree extract.Finally,the paper evaluated the extraction accuracy of single tree canopy,tree height and tree species.The main research contents and conclusions include the following four aspects:(1)Research on vegetation area extraction method based on UAV remote sensing image and SFM point cloud.The paper integrated the DEM and nDSM data from the SFM point cloud and NGRDI,VDVI and other spectral information,adopted an object-oriented image analysis method,and established classification rules for different vegetation types in the study area on the basis of multi-scale segmentation to achieve the rapid extraction of different vegetation types of the district.The experimental results show that it is feasible to extract vegetations of different heights by using the nDSM information and image spectral information obtained from the image reconstruction point cloud,and the overall classification accuracy reaches 91.6%.(2)Research on single tree identification method based on image template matching.In this paper,a template matching correlation coefficient measurement algorithm based on high resolution imagery of UAV was constructed for the vegetation area of a single tree species.The correlation coefficient between the matching template group generated by single tree sample and the image band was calculated,and the maximum correlation coefficient was determined according to the region.For single tree,the effects of different spectral bands,different template groups,and different correlation coefficient thresholds on the recognition of single plant trees were analyzed.Experiments show that using a single template in the red band,the correlation coefficient threshold is set to 0.1 for better template matching,the complete rate of tree recognition in the study area reaches 85.9%,and the recognition accuracy rate reaches 95.7%.(3)The single-plant tree identification based on the regional maximum watershed labeling algorithm and the accuracy analysis of single tree parameter extraction.Based on the CHM data of different tree species,the paper used the region-maximal watershed marker algorithm to achieve single tree segmentation and canopy extraction,and analyzed the height of single tree and canopy extraction accuracy.The experiments show that individual tree identification accuracy and complete rate is 85.9%,89.3% respectively,individual tree crown to extract the overall accuracy can reach 83.76%,individual tree height inversion of the average relative error is 8.05%,based on the regional maximum watershed marker algorithm.(4)Study on the method of tree species identification based on UAV hyperspectral data.Based on the high-spectral orthophotos acquired by UAV,the recognition results of single tree and the basis of the minimum noise separation of hyperspectral data,the paper established the method of tree species identification based on spectral angle matching method.The experimental results show that the minimum noise separation after transform spectral Angle matching the overall accuracy is 54.05%,the matching precision of different tree species differences,white pine and prunus cerasifera accuracy can reach 95% and 96.43% respectively,similar spectral characteristics of tree type identification precision is low.