IRO Journals
Journal of ISMAC

Hybrid Graph-Kernel Fusion Network for Heritage Point Cloud Segmentation

Open Access
Published: 23 July, 2026
Pages: 264-276
  • Geetha Ramani R. , Geetha Ramani R.
    Department of Information Technology, Anna University, Chennai, India
    Department of Information Technology, Anna University, Chennai, India
  • Mirsha A K. , Mirsha A K.
    Department of Artificial Intelligence and Data Science, Anna University, Chennai, India
    Department of Artificial Intelligence and Data Science, Anna University, Chennai, India
  • Piruthviraj V S. Piruthviraj V S.
    Department of Artificial Intelligence and Data Science, Anna University, Chennai, India
    Department of Artificial Intelligence and Data Science, Anna University, Chennai, India
view PDF
view PDF

How to Cite

R., Geetha Ramani, Mirsha A K., and Piruthviraj V S. 2026. “Hybrid Graph-Kernel Fusion Network for Heritage Point Cloud Segmentation”. Journal of ISMAC 8 (3): 264-76. https://doi.org/10.36548/jismac.2026.3.005.
ACM ACS APA ABNT Chicago Harvard IEEE MLA Turabian Vancouver AMA

Download Citation

Endnote/Zotero/Mendeley (RIS) BibTeX

Keywords

Semantic Segmentation
Heritage Building Information Modeling (H-BIM)
Point Cloud Processing
Hybrid Graph–Kernel Fusion
3D Cultural Heritage Analysis

Abstract

Heritage point cloud semantic segmentation is generally used for digital heritage preservation, Heritage Building Information Modeling (H-BIM) and intelligent architectural analysis. However, it is difficult to identify various architectural elements because of the non-uniformity of the point clouds, complexity in geometry and structural occlusions. This research work proposes a novel architecture called the Hybrid Graph-Kernel Fusion Network (HKGFNet), which combines the graph and kernel approaches in terms of contextual information and feature extraction respectively, via a bi-directional cross-attention and gating technique for performing discriminative segmentation. The proposed framework is tested using the Images & PointClouds Cultural Heritage dataset containing labeled 3D point clouds of heritage buildings. The proposed architecture is designed using a systematic procedure consisting of data preprocessing, training and evaluation with a 70:15:15 dataset split. The experiment results yielded a mean Intersection over Union (mIoU) of 90.79% and overall accuracy of 96.15%, while the ablation study validated the efficacy of the hybrid fusion method proposed. The resulting segmentation revealed accurate identification of the various architectural elements and maintained the structural boundaries of the building, implying the suitability of HKGFNet in heritage documentation and Scan-to-BIM processes.

References

  1. Aziz, Abbas, Huma Shaikh, Amna Abbas, Kissa E. Zehra, and Bakhtawar Javed. "Microscopic Techniques for Nanomaterials Characterization: A Concise Review." Microscopy Research and Technique 2025, vol 88, no. 5: 1599-1614.
  2. Qi, Charles Ruizhongtai, Li Yi, Hao Su, and Leonidas J. Guibas. "Pointnet++: Deep Hierarchical Feature Learning on Point Sets in a Metric Space." Advances in Neural Information Processing Systems 2017:30.
  3. Wang, Yue, Yongbin Sun, Ziwei Liu, Sanjay E. Sarma, Michael M. Bronstein, and Justin M. Solomon. "Dynamic Graph CNN for Learning on Point Clouds." ACM Transactions on Graphics (ToG) 2019, vol 38, no. 5: 1-12.
  4. Thomas, Hugues, Charles R. Qi, Jean-Emmanuel Deschaud, Beatriz Marcotegui, François Goulette, and Leonidas J. Guibas. "Kpconv: Flexible and Deformable Convolution for Point Clouds." In Proceedings of the IEEE/CVF International Conference on Computer Vision 2019: 6411-6420.
  5. Lai, Xin, Jianhui Liu, Li Jiang, Liwei Wang, Hengshuang Zhao, Shu Liu, Xiaojuan Qi, and Jiaya Jia. "Stratified Transformer for 3d Point Cloud Segmentation." In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition 2022: 8500-8509.
  6. Yu, Xumin, Lulu Tang, Yongming Rao, Tiejun Huang, Jie Zhou, and Jiwen Lu. "Point-bert: Pre-Training 3d Point Cloud Transformers with Masked Point Modeling." In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition 2022: 19313-19322.
  7. Hu, Qingyong, Bo Yang, Linhai Xie, Stefano Rosa, Yulan Guo, Zhihua Wang, Niki Trigoni, and Andrew Markham. "Randla-net: Efficient Semantic Segmentation of Large-Scale Point Clouds." In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition 2020: 11108-11117.
  8. Matrone, Francesca, Eleonora Grilli, Massimo Martini, Marina Paolanti, Roberto Pierdicca, and Fabio Remondino. "Comparing Machine and Deep Learning Methods for Large 3D Heritage Semantic Segmentation." ISPRS International Journal of Geo-Information 2020, vol 9, no. 9: 535.
  9. Pellis, Eugenio, Andrea Masiero, Michele Betti, Grazia Tucci, and Pierre Grussenmeyer. "A Deep Learning Multiview Approach for the Semantic Segmentation of Heritage Building Point Clouds." International Journal of Architectural Heritage 2025, vol 19, no. 12: 3117-3139.
  10. Lu, Dening, Jun Zhou, Kyle Yilin Gao, Jing Du, Linlin Xu, and Jonathan Li. "Dynamic Clustering Transformer Network for Point Cloud Segmentation." International Journal of Applied Earth Observation and Geoinformation 2024, vol 128: 103791.
  11. G. Pellis, A. Masiero, M. Betti, G. Tucci, and P. Grussenmeyer, Images & PointClouds Cultural Heritage Dataset, Zenodo, 2025. Available - https://zenodo.org/records/14626611