Research Interests
Large-scale 3D point cloud understanding is essential for urban and environmental remote sensing, enabling automated interpretation of complex scenes acquired by diverse LiDAR platforms. As urban landscapes evolve and sensing conditions vary, there is a growing need for methods that can discover novel structures and land cover types beyond predefined taxonomies, while remaining robust to environmental variability.
My research focuses on semantic segmentation, instance-level modeling, and novel class discovery in city-scale LiDAR point clouds. I develop scalable learning frameworks that address challenges posed by heterogeneous sensing conditions, domain shift, and adverse weather. My work combines geometry-aware representation learning, hybrid supervision strategies, and domain generalization techniques to advance urban mapping, land cover discovery mapping, and scene understanding from large-scale LiDAR and multisource geospatial data.
Education
- PhD in Systems Design Engineering, University of Waterloo, Canada, 2026
- MSc in Information and Communication Engineering, JiMei University, China, 2022
Publications
1. Du, J., Xu, L., Ma, L., Gao, K., Zelek, J., Li, J. 3D semantic segmentation: Cluster-based sampling and proximity hashing for novel class discovery. ISPRS Journal of Photogrammetry and Remote Sensing 223 (2025): 274–295.
2. Du, J., Zelek, J., Zhang, D., Li, J. Gated multi-source fusion with geometric sequence modeling for novel urban structure discovery. ISPRS Journal of Photogrammetry and Remote Sensing 230 (2025): 495–523.
3. Du, J., Zelek, J., Li, J. Weather-aware autopilot: Domain generalization for point cloud semantic segmentation in diverse weather scenarios. ISPRS Journal of Photogrammetry and Remote Sensing 218 (2024): 204–219.
4. Du, J., Zelek, J., Zhang, D., Li, J. 3DLCDM: Hybrid supervision for land cover discovery mapping of emerging urban structures in 3D remote sensing. Remote Sensing of Environment 331 (2025): 115018.
5. Du, J., Ma, L., Li, J., Qin, N., Zelek, J., Guan, H., Li, J. RdmkNet & Toronto-Rdmk: Large-scale datasets for road marking classification and segmentation. IEEE Transactions on Intelligent Transportation Systems 25.10 (2024): 13467–13482.
6. Du, J., Cai, G., Wang, Z., Su, J., Huang, M., Zelek, J., Marcato Junior, J., Li, J. MTCloud: Multi-type convolutional linkage network for point cloud instance segmentation. Expert Systems with Applications 270 (2025): 126432.