Addressing Class Imbalance in Semantic Segmentation of Rural Watershed Point Clouds Using Loss Re-Weighting

Minjae Kim; Jonghyuk Lee; Byunghun Seo; Dongsu Kim; Yejin Seo; Dongwoo Kim; Yerim Jo; Jimin Shim; Won Choi

doi:10.5389/KSAE.2026.68.2.033

All Issue

2026 Vol.68, Issue 2 Preview Page Next Page

Addressing Class Imbalance in Semantic Segmentation of Rural Watershed Point Clouds Using Loss Re-Weighting 농촌 유역 포인트 클라우드 의미론적 분할에서의 클래스 불균형 대응을 위한 손실 재가중 기법 연구

2026. pp. 33-44

PDF XML

Abstract

This study addresses the challenge of severe class imbalance in semantic segmentation of rural watershed point clouds. Due to the dominance of a few classes and the scarcity of minority classes, segmentation performance can degrade significantly. To mitigate this issue, we investigated loss function strategies based on class weighting and focal loss using the Point Transformer V3 architecture. The dataset, collected via LiDAR from Sanyang Reservoir watershed in Icheon, South Korea, was labeled into ten semantic classes and preprocessed using sub-sampling. Four loss function configurations were evaluated: baseline cross entropy, class-weighted cross entropy loss, focal loss, and a combination of focal loss with class weighting. Experimental results show that combining class weighting and focal loss achieves the best performance (mIoU = 0.3222; OA = 0.7464), outperforming the baseline, with clear IoU improvements in several minority classes such as Road, Water, and House, while Bridge and Reservoir remain challenging due to extreme data scarcity. Hyperparameter tuning further indicates optimal focal loss parameters of γ = 3.0 without class weighting (mIoU = 0.3311) and γ = 2.0 with class weighting (mIoU = 0.3222). These findings highlight the effectiveness and limitations of tailored loss re-weighting for addressing class imbalance in rural watershed point cloud segmentation.

Keywords

Rural watershed

point cloud segmentation

semantic segmentation

class imbalance

loss re-weighting

References

REFERENCES

J. Behley, M. Garbade, A. Milioto, J. Quenzel, S. Behnke, C. Stachniss and J. Gall, In Proc. IEEE/CVF International Conference on Computer Vision, SemanticKITTI: A dataset for semantic scene understanding of LiDAR sequences; 9297-9307 (2019)10.1109/iccv.2019.00939

A. Boulch, Computers & Graphics, ConvPoint: Continuous convolutions for point cloud processing, 88; 24-34 (2020)10.1016/j.cag.2020.02.005

CloudCompare, CloudCompare (Version 2.13), Accessed November 10, 2025. http://www.cloudcompare.org (2023)

Y. Cui, M. Jia, T. Y. Lin, Y. Song and S. Belongie, In Proc. IEEE/CVF Conference on Computer Vision and Pattern Recognition, Class-balanced loss based on effective number of samples; 9268-9277 (2019)10.1109/cvpr.2019.00949

DJI, DJI Terra user manual (Version 3.7.6), Accessed November 10, 2025. http://www.dji.com/dji-terra (2023)

H. J. Go, J. H. Hyun, J. H. Lee and J. Ahn, Korean Journal of Construction Engineering and Management, Development of a quantitative risk assessment model by BIM-based risk factor extraction: Focusing on falling accidents, 23(4); 15-25, (in Korean) (2022)10.6106/KJCEM.2022.23.4.015

T. Hackel, N. Savinov, L. Ladicky, J. D. Wegner, K. Schindler and M. Pollefeys, ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Semantic3D.net: A new large-scale point cloud classification benchmark, IV-1(W1); 91-98 (2017)10.5194/isprs-annals-iv-1-w1-91-2017

J. Hou, B. Graham, M. Nießner and S. Xie, In Proc. IEEE/CVF Conference on Computer Vision and Pattern Recognition, Exploring data-efficient 3D scene understanding with contrastive scene contexts; 15587-15597 (2021)10.1109/cvpr46437.2021.01533

Q. Hu, B. Yang, L. Xie, S. Rosa, Y. Guo, Z. Wang, N. Trigoni and A. Markham, In Proc. IEEE/CVF Conference on Computer Vision and Pattern Recognition, RandLA-Net: Efficient semantic segmentation of large-scale point clouds; 11108-11117 (2020)10.1109/cvpr42600.2020.01112

H. J. Kim, M. H. Kwon, W. H. Cho, G. C. Kim and J. G. Kim, In Proc. Korean Institute of Information and Communication Sciences Conference, Digital transformation in water management: Building a digital twin-based platform, 26(1); 284-287, (in Korean) (2022)

T. H. Kim, D. H. Lee, S. I. Bae, Y. K. Hong and Z. S. Park, Journal of the Regional Association of Architectural Institute of Korea, Proposal of field applied safety control process using BIM, 15(5); 243-250, (in Korean) (2013)

KOSTAT (Statistics Korea), Damage by cause of natural disaster, (in Korean), Daejeon, Korea. Statistics Korea. (2023)

J. Kwak, J. Kim, H. Lee, J. Lee, J. Cho and M. S. Kang, Journal of the Korean Society of Agricultural Engineers, An integrated flood simulation system for upstream and downstream of the agricultural reservoir watershed, 65(1); 41-49, (in Korean) (2023)10.5389/KSAE.2023.65.1.041

C. J. Lee and S. I. Ham, Journal of the Architectural Institute of Korea, Design guide systems of safety handrail based on DfS and BIM for preventing of fall accidents, 36(10); 235-241, (in Korean) (2020)10.5659/JAIK.2020.36.10.235

B. J. Lee, N. Y. Kim and S. S. Yoon, Journal of the Korean Society of Agricultural Engineers, Exploration of water pumping station digital twin system development process according to software development Methodologies, 66(3); 53-62, (in Korean) (2024)10.5389/KSAE.2024.66.3.053

T. Y. Lin, P. Goyal, R. Girshick, K. He and P. Dollár, In Proc. IEEE International Conference on Computer Vision, Focal loss for dense object detection; 2980-2988 (2017)10.1109/TPAMI.2018.2858826

W. Liu, Z. Wang, P. Duan, X. Kang and S. Li, arXiv preprint, Agriculture-Vision Challenge 2024—The runner-up solution for agricultural pattern recognition via class balancing and model ensemble (2024)10.48550/arXiv.2406.12271

G. M. Morton, A computer oriented geodetic data base and a new technique in file sequencing, Ottawa, Canada. International Business Machines Company. (1966)

R. Muller, S. Kornblith and G. E. Hinton, In Advances in Neural Information Processing Systems, When does label smoothing help?, 32 (2019)10.48550/arXiv.1906.02629

OpenMMLab, MMSegmentation documentation—Focal loss, Accessed November 10, 2025. http://mmsegmentation.readthedocs.io (2023)

J. Park, M. S. Kang, J. H. Song and S. M. Jun, Journal of Korean Society of Rural Planning, Design and implementation of IoT-based intelligent platform for water level monitoring, 21(4); 177-186, (in Korean) (2015)10.7851/ksrp.2015.21.4.177

C. R. Qi, H. Su, K. Mo and L. J. Guibas, In Proc. IEEE Conference on Computer Vision and Pattern Recognition, PointNet: Deep learning on point sets for 3D classification and segmentation; 652-660 (2017a)10.1109/cvpr.2017.16

C. R. Qi, L. Yi, H. Su and L. J. Guibas, In Proc. 31st International Conference on Neural Information Processing Systems, PointNet++: Deep hierarchical feature learning on point sets in a metric space; 5105-5114 (2017b)10.48550/arXiv.1706.02413

H. Sagan, Space-filling curves, New York, NY. Springer Science & Business Media. (2012)

H. Son, C. Kim and Y. Turkan, In Proc. International Symposium on Automation and Robotics in Construction, Scan-to-BIM—An overview of the current state of the art and a look ahead, 32; 1-8 (2015)10.22260/isarc2015/0050

H. Thomas, C. R. Qi, J.-E. Deschaud, B. Marcotegui, F. Goulette and L. J. Guibas, In Proc. IEEE/CVF International Conference on Computer Vision, KPConv: Flexible and deformable convolution for point clouds; 6411-6420 (2019)10.1109/iccv.2019.00651

C. Thomson and J. Boehm, Remote Sensing, Automatic geometry generation from point clouds for BIM, 7(9); 11753-11775 (2015)10.3390/rs70911753

TorchVision, torchvision.ops.sigmoid_focal_loss—PyTorch documentation, Accessed November 10, 2025. http://pytorch.org/vision/stable/ops.html (2024)

X. Wu, L. Jiang, P. S. Wang, Z. Liu, X. Liu, Y. Qiao, W. Ouyang, T. He and H. Zhao, In Proc. IEEE/CVF Conference on Computer Vision and Pattern Recognition, Point Transformer V3: Simpler, faster, stronger; 4840-4851 (2024)10.1109/cvpr52733.2024.00463

S. Xie, J. Gu, D. Guo, C. R. Qi, L. Guibas and O. Litany, In Proc. European Conference on Computer Vision, PointContrast: Unsupervised pre-training for 3D point cloud understanding; 574-591 (2020)10.1007/978-3-030-58580-8_34

Z. Xiong and T. Wang, IOP Conference Series: Earth and Environmental Science, Research on BIM reconstruction method using semantic segmentation point cloud data based on PointNet, 719(2); 022042 (2021)10.1088/1755-1315/719/2/022042

H. Zhao, L. Jiang, J. Jia, P. H. S. Torr and V. Koltun, In Proc. IEEE/CVF International Conference on Computer Vision, Point Transformer; 16259-16268 (2021)10.1109/iccv48922.2021.01595

Information

Publisher :The Korean Society of Agricultural Engineers
Publisher(Ko) :한국농공학회
Journal Title :Journal of Korean Society of Agricultural Engineers
Journal Title(Ko) :한국농공학회논문집
Volume : 68
No :2
Pages :33-44
Received Date : 2025-09-01
Revised Date : 2025-12-09
Accepted Date : 2025-12-09
DOI :https://doi.org/10.5389/KSAE.2026.68.2.033

Journal of Korean Society of Agricultural Engineers ISSN:1738-3692(Print) 2093-7709(Online) 한국농공학회논문집

All Issue

REFERENCES