View in EDS

A spherical-harmonic deep learning framework for generic tree morphology and species identification.

Saved in:
Bibliographic Details
Title: A spherical-harmonic deep learning framework for generic tree morphology and species identification.
Authors: Jiang, Kang1 (AUTHOR), Yun, Ting1 (AUTHOR) yunting@njfu.edu.cn, Weng, Qihao2,3,4 (AUTHOR)
Source: ISPRS Journal of Photogrammetry & Remote Sensing. May2026, Vol. 235, p599-617. 19p.
Subjects: Spherical harmonics, Deep learning, Environmental monitoring, LIDAR, Biological classification, Forest biodiversity, Crowns (Botany), Point cloud
Abstract: Tree biodiversity is declining at an unprecedented pace, threatening forest ecosystems and their ecological functions. Reliable tree species identification is essential for monitoring biodiversity change and developing strategies to mitigate further losses; however, spectral intra-class variability, inter-class similarity, and the structural complexity of trees continue to hinder tree species identification by intelligent algorithms. Herein, we constructed a generalized framework from a transdisciplinary perspective to characterize tree morphology, based exclusively on LiDAR data for tree species identification. First, we devised multiple point cloud projection strategies and radial manifold-derived geometric features in spherical space to comprehensively depict tree morphogenesis. Second, we developed an innovative deep learning model based on spherical harmonics and the rotation group SO(3) to explore the structural commonalities for each tree species and identify tree species across mesoscale forest landscapes. By analysing point clouds of nearly 200,000 trees from 59 species, we demonstrated that our framework has the ability to capture the spatial clustering behaviour of foliage clumps, the uneven appearance of tree crowns, and the canopy architectures of diverse trees. Our framework achieved species identification overall accuracies of 85.42% on the Chinese dataset and 85.03% on the FOR-species20K dataset, surpassing state-of-the-art deep learning approaches by 5.19% and 5.02%, respectively. The results present the effectiveness of our framework as a robust tool for assessing forest diversity at both community and ecosystem levels and have the potential to be applied to larger geographical regions. [ABSTRACT FROM AUTHOR]
Copyright of ISPRS Journal of Photogrammetry & Remote Sensing is the property of Elsevier B.V. and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)
Database: Engineering Source
Description
Abstract:Tree biodiversity is declining at an unprecedented pace, threatening forest ecosystems and their ecological functions. Reliable tree species identification is essential for monitoring biodiversity change and developing strategies to mitigate further losses; however, spectral intra-class variability, inter-class similarity, and the structural complexity of trees continue to hinder tree species identification by intelligent algorithms. Herein, we constructed a generalized framework from a transdisciplinary perspective to characterize tree morphology, based exclusively on LiDAR data for tree species identification. First, we devised multiple point cloud projection strategies and radial manifold-derived geometric features in spherical space to comprehensively depict tree morphogenesis. Second, we developed an innovative deep learning model based on spherical harmonics and the rotation group SO(3) to explore the structural commonalities for each tree species and identify tree species across mesoscale forest landscapes. By analysing point clouds of nearly 200,000 trees from 59 species, we demonstrated that our framework has the ability to capture the spatial clustering behaviour of foliage clumps, the uneven appearance of tree crowns, and the canopy architectures of diverse trees. Our framework achieved species identification overall accuracies of 85.42% on the Chinese dataset and 85.03% on the FOR-species20K dataset, surpassing state-of-the-art deep learning approaches by 5.19% and 5.02%, respectively. The results present the effectiveness of our framework as a robust tool for assessing forest diversity at both community and ecosystem levels and have the potential to be applied to larger geographical regions. [ABSTRACT FROM AUTHOR]
ISSN:09242716
DOI:10.1016/j.isprsjprs.2026.03.024