Single Tree Extraction And Tree Species Classification Based On Airborne LiDAR Data And Aerial Images

The forest is known as the “Lung of the Earth”.It is rich in species,complex structures and diverse functions.Forest resources are of great significance for the protection of the global ecosystem.UAV aerial photography has the advantages of flexible image acquisition,fast image resolution,wide application range and low cost,which enhances the ability of forestry investigation work.LiDAR is an emerging remote sensing technology.It has powerful active earth observation function and can quickly and directly acquire three-dimensional coordinate data of the ground.It has been widely used in forestry scientific research.In this paper,the aerial image is used to study the single tree extraction of tree species,and the laser point cloud data is obtained by laser radar to study the classification of tree species.The main research contents are as follows:Using the forest aerial image,the image is subjected to simple linear iterative clustering(SLIC)superpixel segmentation.Each superpixel is used as a node,and the Euclidean distance of the adjacent superpixel RGB color average is used as the weight of the edge to construct the weight.For the undirected graph,the minimum spanning tree is constructed according to the weight of the edge.Finally,the algorithm of the graph is used to merge the superpixel to obtain the single-wood extraction result of the aerial image.In this thesis,four kinds of plants,including Metasequoia,Willow,Ligustrum lucidum and Bamboo,were selected for the test and statistics of crown width and quantity,and compared with the measured data on site,and the precision and quantity precision of the crown of the experiment were calculated.It can be seen from the results that the crown extraction accuracy is above 90%,and the extraction accuracy is above 80%.Taking the five dominant species of metasequoia,willow,privet,bamboo and apple tree in Qianjiang New City Forest Park of Hangzhou and Hongqipo Farm in Aksu Prefecture of Xinjiang as the research object,using the drone to obtain the airborne LiDAR data,the point cloud of the tree Features: structure features(SF),tree texture features(TF)and tree crown features(CF),tree species classification using support vector machine classifier,calculation tree species classification accuracy.Combining the seven combined parameters of SF,TF,CF,SF+TF,SF+CF,TF+CF and SF+TF+CF,the overall accuracy of the five species of plants is 58%,64%,60%,respectively.73%,70%,77% and 85%.Experiments show that using multiple types of characteristic parameters can significantly improve the classification accuracy.Combined with the three types of characteristic parameters,the overall accuracy of the five types of plants is correctly classified as 85%,and the Kappa coefficient is 0.81.The experimental results show that the single wood extraction and tree species classification algorithm proposed in this paper is effective and provides a good technical support for forestry science research.