Developing a more accurate method for individual plant segmentation of urban tree and shrub communities using LiDAR technology

Abstract Application of LiDAR technology has greatly enhanced tree segmentation and phenotypic analysis. There are few studies in urban green spaces using tree segmentation methods. Our aim is to improve the single-plant segmentation accuracy in tree and shrub communities through segmenting algorithm optimisation based on TLS LiDAR data of the urban green space. We developed a multi-round comparative shortest-path algorithm (M-CSP) to achieve the objectives: a) tree and shrub plant layer pre-division (TSPD); b) shrub type classifications (STC) into spherical, cylindrical, and rectangular shapes. The overall detection kappa value using M-CSP is 0.933, which is 18% higher than the CSP value of 0.790. M-CSP-based overall segmentation accuracy value (F-score) is 0.886, which is 13% higher than the CSP value of 0.783. The shrub F-score using M-CSP is 0.817, which is 26% higher than the CSP (0.646). M-CSP should provide a more accurate, faster, and less costly tool to study plant communities in urban green spaces.