Research On Tree Crown Extraction And Tree Species Recognition Based On UAV Images

UAVs technique has become more and more widely used in forestry due to their low cost,strong timeliness,and high collection efficiency.UAV aerial photography can obtain forest canopy information and realize tree species identification by obtaining high-resolution UAV images.In this study,small UAV were selected to acquire images of the Northeast Forestry University Urban Forestry Demonstration Base,and the acquired images were spliced to generate orthophotos.Based on the orthophotos,the tree canopy and gaps in the sample plot were extracted,and a random forest model was established to identify and classify tree species based on the extracted tree crown information,and the accuracy of the tree crown segmentation and tree species identification classification results was verified through field survey data.The research results are as follows:(1)UAV forest stand image stitching based on different stitching algorithmsThe SIFT algorithm,SURF algorithm,ORB algorithm and KAZE algorithm were used to stitch the obtained UAV images of different canopy closure forest stands.For the stitching of two images,the feature matching rate of the ORB algorithm was significantly lower than other algorithms,but the time consumed was the shortest.Therefore,if the accuracy requirements are not high,the ORB algorithm can be used for rapid image stitching;the number of feature points extracted from low canopy closure images was lower than that of high canopy closure images.For the image stitching of medium and high canopy closure forest stands,the correct matching rate of SIFT algorithm and SURF algorithm was about 45%,and the correct matching rate of KAZE algorithm was about 55%.From the time consumption point of view,image stitching time for the SURF algorithm was 577.8s,which was significantly lower than the 876.3 s of the SIFT algorithm and the 806.8s of the KAZE algorithm.(2)Tree crown extraction based on watershed algorithmBy setting different UAV flight heights,and taking the Pinus sylvestris sample plot in the Harbin Urban Forestry Demonstration Base as the research object,the traditional and improved watershed algorithm were used to extract the single tree crown and forest gaps respectively.The crown width and crown projection area were estimated.Based on the traditional and improved watershed algorithm,the average single tree crown recognition rate is 51.11%and 80.74%,respectively.For the traditional algorithm,the average extraction accuracy of crown width and crown projection area were 69.72%and 53.59%,while the improved algorithm has The average extraction accuracy of crown width and canopy projection area was 79.84%and 76.04%,respectively,which indicated that the accuracy of the improved watershed algorithm was greater than that of the traditional watershed algorithm.The forest images acquired at a flight altitude of 50m can effectively extract information on the tree crown and gap area,and there were many unidentified gaps at a flight altitude of 70m and 100m.(3)Tree species recognition and classification based on UAV imagesThe random forest algorithm was used to establish a model to identify and classify dominant tree species.When the scale segmentation parameter was set to 400 and the second segmentation was performed,the occurrence of over-segmentation and under-segmentation can be reduced;the overall accuracy of tree species identification was 79.51%through the confusion matrix.,and the Kappa coefficient was 0.76.The tree species were arranged in descending order of producer accuracy:Pinus sylvestris>Ulmus pumila>Fraxinus mandshurica>Pinus koraiensis>Picea koraiensis>Abies nephrolepis>Larix gmelinii,and according to the user precision,the order was Fraxinus mandshurica>Pinus sylvestris>Pinus koraiensis>Picea koraiensis>Abies nephrolepis>Ulmus pumila>Larix gmelinii;It was found that each tree species performed better in the classification result map by dividing and mapping the dominant tree species in the sample plot and comparing them with field investigations.In summary,this research combined UAV aerial photography and image processing technology to obtain high-resolution orthophotos,which can provide effective guarantee for the later extraction of forest structure parameters.The tree crown information of i ndividual trees and the identification of different tree species also provide data sources for forest biomass estimation and forest species diversity monitoring.