Tongtong Lu

Mr. Tongtong Lu

orcid

Research Focus & Metrics

Academic Career Length

1Year (2026-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
27 March 2026 in Remote Sensing
Verified
Journal Article
Open access
PDF file
TLTongtong Lu
Tongtong Lu
FHFang Huang
Fang Huang
YDYuxin Ding
Yuxin Ding
QLQingzhe Lv
Qingzhe Lv
HGHao Guan
Hao Guan
GTGeer Teng
Geer Teng
Aiming at diverse urban tree structures and difficulties in vegetation point cloud extraction and utilization, this study proposed single-tree-scale biological parameter estimation methods for urban scenarios to enhance point cloud’s application value in urban greening management. For single-tree segmentation, it constructed a method based on the constraints of the trees’ geometric features and combined the gravitational modeling characteristics, called the CGF-CG single-tree segmentation method. This method (i) combines clustering and principal direction analysis to extract trunk points, (ii) introduces canopy segmentation based on trunk positions, (iii) optimizes edge point attributes via a gravitational model. Based on CGF-CG’s accurate results, an improved random forest method for single-tree biological parameter (IRF-BP) estimation (aboveground biomass, carbon storage, leaf area index, living vegetation volume) was proposed: (i) correlation analysis with variable screening, (ii) adaptive feature selection and pigeon-inspired optimization to enhance model generalization, (iii) adopting Shapley Additive Explanations (SHAP) to improve interpretability. Based on these, a complete model for different tree species was constructed. Validation showed that CGF-CG exhibited negligible over-segmentation and under-segmentation in the selected study areas, with overall average precision, recall, and F1-score over 98.5%. Additionally, on the selected overall region, the overall mF1 score, mPTP, and mPTR of our method are 99.13%, 99.15%, and 99.12%, respectively, which are superior to Forestmetrics, lidR, PyCrown, and DBSCAN methods. IRF-BP performed well, with a highest R2 of 0.81 and a lowest mean absolute percentage error of 7.5%, effectively surpassing the performance of traditional models such as RFR, GBR, KNN, and XGB. In summary, results provided theoretical and technical support for urban green resource management and evaluation.

20 April 2026 in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
Verified
Journal Article
Open access
TLTongtong Lu
Tongtong Lu
XQXiaoyong Qiang
Xiaoyong Qiang
JWJian Wang
Jian Wang
BGBinfu Ge
Binfu Ge
HGHao Guan
Hao Guan
FHFang Huang
Fang Huang
The task of registering heterogeneous point clouds of urban scenes is particularly challenging due to differences in the imaging mechanisms of sensors and the limitations imposed by the conditions under which these images are acquired. The relevant challenges include a large number of outliers, differences in the densities of point clouds of several orders of magnitude, low rates of overlap between them, data scarcity, and large rotational angles that yield severe mismatches in the data distribution. These issues create significant performance-related bottlenecks for traditional methods of point cloud registration as well as those based on feature extraction through deep learning. In response to these issues, this study proposes a framework for registering heterogeneous point clouds called the super-point registration network (SR-Net). SR-Net consists of a pyramid feature extraction module based on KPConv, a super-point feature extraction and matching module, a local feature extraction module, and a local-to-global registration method. To enhance the capability of the model to extract features from heterogeneous point clouds with different resolutions, the authors propose an adaptive Top-K method for point cloud padding. The model incorporates self-attention and cross-attention mechanisms to capture the geometric structure within point clouds and the geometric consistency between them. The geometric information contained in the super-points is encoded through geometric structure embedding. Finally, to compensate for the scarcity of annotated multimodal data, we pre-trained the model on large-scale homologous benchmarks. This enabled it to learn the underlying invariant geometric feature priors across domains that guided it to stably converge on a dataset collected by the authors that involved major differences in the distribution of the features. We comprehensively validated the proposed model through a progressive experimental system (EXP-I∼VII) that covered basic evaluation, ablation, cross-domain generalization, and stress testing for efficiency. SR-Net achieved a relative rotational error (RRE) of 0.24 ° and a registration recall (RR) of 99.8% on the KITTI benchmark. Moreover, it achieved an RR of 96.97%, an RRE of 0.56 °, and a relative translational error (RTE) of 0.92 m on a challenging dataset of heterogeneous point clouds that was collected by the authors, with an average overlap rate of only 7.62% that reflected optimal stability. Our model also yielded an RR of 92.6% and a RTE of 0.31 m in experiments on the Cross3DReg heterologous datasets, significantly surpassing mainstream methods. The final evaluation confirmed that it achieved a high RR of 91.2% on 3DMatch, while optimizing the inferential delay to 85.96 ms. This verifies the high robustness of SR-Net when handling complex tasks of multimodal point cloud registration.