Point Cloud Semantic Segmentation Based on Degraded Feature Consistency Learning under Adverse Weather Conditions

AN Yi; LIU Shuxin; HE Lijun; LI Fan

doi:null

您当前的位置：

首页 >

文章列表页 >

Point Cloud Semantic Segmentation Based on Degraded Feature Consistency Learning under Adverse Weather Conditions

更新时间：2026-09-02

- Point Cloud Semantic Segmentation Based on Degraded Feature Consistency Learning under Adverse Weather Conditions
- JOURNAL OF XI’AN JIAOTONG UNIVERSITY (2026)
- 作者机构：
  
  1.西安交通大学电子与信息学部，710049，西安
  2.西南技术物理研究所，610041，成都
- 作者简介：
- 基金信息：
- DOI：
  CLC： TP391
- Received：01 February 2026，
  
  Revised：2026-08-04，
  
  Accepted：05 August 2026，
  
  Online First：10 September 2026，
- 稿件说明：
移动端阅览
AN Yi, LIU Shuxin, HE Lijun, et al. Point Cloud Semantic Segmentation Based on Degraded Feature Consistency Learning under Adverse Weather Conditions[J]. JOURNAL OF XI’AN JIAOTONG UNIVERSITY,2026.
DOI：

AN Yi, LIU Shuxin, HE Lijun, et al. Point Cloud Semantic Segmentation Based on Degraded Feature Consistency Learning under Adverse Weather Conditions[J]. JOURNAL OF XI’AN JIAOTONG UNIVERSITY,2026. DOI：

摘要

针对雨、雪、雾等恶劣天气造成激光雷达点云点丢失、遮挡和几何扰动，导致三维语义分割性能下降，尤其影响行人、自行车等小目标分割精度的问题，提出一种基于退化特征一致性学习的点云语义分割方法。首先，构建由细粒度语义实例增强和空间分区距离感知扰动组成的数据增强模块，通过几何扰动模拟和小目标实例扩充，增强模型对结构变化及小目标特征的学习能力；其次，利用点云体素特征提取网络对点云进行编码，获得体素特征表示；最后，对体素特征施加随机退化，构建“良好—退化双分支”，并通过一致性约束提升网络在特征受损条件下的稳健表征能力。使用公开SemanticSTF数据集对所提方法进行实验验证，结果表明：该方法能够有效改善恶劣天气下的点云语义分割性能；在性能指标方面，平均交并比较基准模型提高10.7%，并在行人、自行车等小目标类别上取得更优的分割结果。研究结果可为复杂天气环境下三维感知系统的安全可靠应用提供参考。

Abstract

To address the issues of point loss

occlusion

and geometric distortion in LiDAR point clouds caused by adverse weather conditions such as rain

snow

and fog

which lead to degraded performance in 3D semantic segmentation and particularly affect the segmentation accuracy of small objects like pedestrians and bicycles

a point cloud semantic segmentation method is proposed based on degraded feature consistency learning. First

a data augmentation module consisting of fine-grained semantic instance enhancement and spatially partitioned distance-aware perturbation is constructed. Through geometric perturbation simulation and small object instance expansion

the model's learning capability for structural variations and small object features is enhanced. Second

a point cloud voxel feature extraction network is used to encode point clouds to obtain voxel feature representations. Finally

random degradation is applied to voxel features to construct a "clean-degraded dual-branch" structure

and the robust representation capability of the network under feature corruption conditions is improved through consistency constraints. The proposed method is experimentally validated on the public SemanticSTF dataset. The results show that this method can effectively improve the performance of point cloud semantic segmentation under adverse weather conditions. In terms of performance metrics

the mean Intersection over Union (mIoU) is 10.7% higher than that of the baseline model

and better segmentation results are achieved for small object categories such as pedestrians and bicycles. The research results can provide a reference for the safe and reliable application of 3D perception systems in complex weather environments.

关键词

Keywords

references

朱安迪 , 达飞鹏 , 盖绍彦 . 对融合特征敏感的三维点云识别与分割 [J ] . 西安交通大学学报 , 2024 , 58 ( 5 ): 52 - 63 .

Zhu Andi , Da Feipeng , Gai Shaoyan . Recognition and segmentation of 3D point clouds sensitive to fusion features [J ] . Journal of Xi'an Jiaotong University , 2024 , 58 ( 5 ): 52 - 63 .

杨军 , 王连甲 . 结合位置关系卷积与深度残差网络的三维点云识别与分割 [J ] . 西安交通大学学报 , 2023 , 57 ( 5 ): 182 - 193 .

Yang Jun , Wang Lianjia . Recognition and segmentation of 3D point cloud through positional relation convolution in combination with deep residual network [J ] . Journal of Xi'an Jiaotong University , 2023 , 57 ( 5 ): 182 - 193 .

薛豆豆 , 程英蕾 , 文沛 , 等 . MLMS-Net: 多层次多尺度点云分类网络 [J ] . 西安交通大学学报 , 2020 , 54 ( 12 ): 70 - 78 .

Xue Doudou , Cheng Yinglei , Wen Pei , et al . MLMS-Net: a point cloud classification network with multi-level and multi-scale [J ] . Journal of Xi'an Jiaotong University , 2020 , 54 ( 12 ): 70 - 78 .

李悄 , 李垚辰 , 张玉龙 , 等 . 采用稀疏3D卷积的单阶段点云三维目标检测方法 [J ] . 西安交通大学学报 , 2022 , 56 ( 9 ): 112 - 122 .

Li Qiao , Li Yaochen , Zhang Yulong , et al . A single-stage point cloud 3D object detection method using sparse 3D convolution [J ] . Journal of Xi'an Jiaotong University , 2022 , 56 ( 9 ): 112 - 122 .

仇志江 , 张林 , 姚垚 , 等 . 三维点云场景语义分割研究进展 [J ] . 中国图象图形学报 , 2025 , 30 ( 7 ): 2325 - 2342 .

Qiu Zhijiang , Zhang Lin , Yao Yao , et al . Survey on semantic segmentation in 3D point cloud scenes [J ] . Journal of Image and Graphics , 2025 , 30 ( 7 ): 2325 - 2342 .

武雨田 , 李擎 , 孙文蔚 , 等 . 复杂天气下车载激光点云目标检测方法综述 [J ] . 工程科学学报 , 2025 , 47 ( 5 ): 1081 - 1093 .

Wu Yutian , Li Qing , Sun Wenwei , et al . Overview of object detection methods based on LiDAR point cloud under adverse weather conditions [J ] . Chinese Journal of Engineering , 2025 , 47 ( 5 ): 1081 - 1093 .

李佳乐 . 面向自动驾驶的三维点云语义分割关键技术研究 [D ] . 杭州 : 浙江大学 , 2023 .

Zhao Haimei , Zhang Jing , Chen Zhuo , et al . UniMix: towards domain adaptive and generalizable LiDAR semantic segmentation in adverse weather [C ] // 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) . Piscataway, NJ, USA : IEEE , 2024 : 14781 - 14791 .

Singhal P , Walambe R , Ramanna S , et al . Domain adaptation: challenges, methods, datasets, and applications [J ] . IEEE Access , 2023 , 11 : 6973 - 7020 .

Zhou Kaiyang , Liu Ziwei , Qiao Yu , et al . Domain generalization: a survey [J ] . IEEE Transactions on Pattern Analysis and Machine Intelligence , 2023 , 45 ( 4 ): 4396 - 4415 .

范文辉 , 林茜 , 罗欢 , 等 . 面向三维点云的域自适应学习 [J ] . 遥感学报 , 2024 , 28 ( 4 ): 825 - 842 .

Fan Wenhui , Lin Xi , Luo Huan , et al . Domain adaptation learning for 3D point clouds: a survey [J ] . National Remote Sensing Bulletin , 2024 , 28 ( 4 ): 825 - 842 .

Jiang Peng , Saripalli S . LiDARNet: a boundary-aware domain adaptation model for point cloud semantic segmentation [C ] // 2021 IEEE International Conference on Robotics and Automation (ICRA) . Piscataway, NJ, USA : IEEE , 2021 : 2457 - 2464 .

Yuan Zhimin , Wen Chenglu , Cheng Ming , et al . Category-level adversaries for outdoor LiDAR point clouds cross-domain semantic segmentation [J ] . IEEE Transactions on Intelligent Transportation Systems , 2023 , 24 ( 2 ): 1982 - 1993 .

Muandet K , Balduzzi D , Schölkopf B . Domain generalization via invariant feature representation [C ] // Proceedings of the 30th International Conference on Machine Learning . Chia Laguna Resort, Sardinia, Italy : PMLR , 2013 : 10 - 18 .

Balaji Y , Sankaranarayanan S , Chellappa R . MetaReg: towards domain generalization using meta-regularization [C ] // Proceedings of the 32nd International Conference on Neural Information Processing Systems . Piscataway, NJ, USA : IEEE , 2018 : 1006 - 1016 .

Li Haoliang , Pan S J , Wang Shiqi , et al . Domain generalization with adversarial feature learning [C ] // 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition . Piscataway, NJ, USA : IEEE , 2018 : 5400 - 5409 .

Huang Jiaxing , Guan Dayan , Xiao Aoran , et al . FSDR: frequency space domain randomization for domain generalization [C ] // 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) . Piscataway, NJ, USA : IEEE , 2021 : 6887 - 6898 .

Huang Chao , Cao Zhangjie , Wang Yunbo , et al . MetaSets: meta-learning on point sets for generalizable representations [C ] // 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) . Piscataway, NJ, USA : IEEE , 2021 : 8859 - 8868 .

Huang Hao , Chen Cheng , Fang Yi . Manifold adversarial learning for cross-domain 3D shape representation [C ] // Computer Vision – ECCV 2022 . Cham : Springer Nature Switzerland , 2022 : 272 - 289 .

Park J , Kim K , Shim H . Rethinking data augmentation for robust LiDAR semantic segmentation in adverse weather [C ] // Computer Vision – ECCV 2024 . Cham : Springer Nature Switzerland , 2025 : 320 - 336 .

Park J , Lee H , Kang I , et al . No thing, nothing: highlighting safety-critical classes for robust LiDAR semantic segmentation in adverse weather [C ] // 2025 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) . Piscataway, NJ, USA : IEEE , 2025 : 6690 - 6699 .

Xiao A , Huang J , Xuan W , et al . 3D semantic segmentation in the wild: learning generalized models for adverse-condition point clouds [C ] // Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition . Piscataway, NJ, USA : IEEE , 2023 : 9382 - 9392 .

Tarvainen A , Valpola H . Mean teachers are better role models: weight-averaged consistency targets improve semi-supervised deep learning results [C ] // Proceedings of the 31st International Conference on Neural Information Processing Systems . Red Hook, NY, USA : Curran Associates Inc. , 2017 : 1195 - 1204 .

Liu Yuanan , Tian Yu , Chen Yuanhong , et al . Perturbed and strict mean teachers for semi-supervised semantic segmentation [C ] // 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) . Piscataway, NJ, USA : IEEE , 2022 : 4248 - 4257 .

Wu Yushuang , Yan Zizheng , Cai Shengcai , et al . PointMatch: a consistency training framework for weakly supervised semantic segmentation of 3D point clouds [J ] . Computers & Graphics , 2023 , 116 : 427 - 436 .

Wu Zhonghua , Wu Yicheng , Lin Guosheng , et al . Reliability-adaptive consistency regularization for weakly-supervised point cloud segmentation [J ] . International Journal of Computer Vision , 2024 , 132 ( 6 ): 2276 - 2289 .

Choy C , Gwak J Y , Savarese S . 4D spatio-temporal ConvNets: minkowski convolutional neural networks [C ] // 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) . Piscataway, NJ, USA : IEEE , 2019 : 3070 - 3079 .

van Erven T , Harremos P . Rényi divergence and Kullback-Leibler divergence [J ] . IEEE Transactions on Information Theory , 2014 , 60 ( 7 ): 3797 - 3820 .

Behley J , Garbade M , Milioto A , et al . SemanticKITTI: a dataset for semantic scene understanding of LiDAR sequences [C ] // 2019 IEEE/CVF International Conference on Computer Vision (ICCV) . Piscataway, NJ, USA : IEEE , 2019 : 9296 - 9306 .

Xiao Aoran , Huang Jiaxing , Guan Dayan , et al . PolarMix: a general data augmentation technique for LiDAR point clouds [C ] // Proceedings of the 36th International Conference on Neural Information Processing Systems . Red Hook, NY, USA : Curran Associates Inc. , 2022 : 11035 - 11048 .

Kong Lingdong , Ren Jiawei , Pan Liang , et al . LaserMix for semi-supervised LiDAR semantic segmentation [C ] // 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) . Piscataway, NJ, USA : IEEE , 2023 : 21706 - 21716 .

He Pei , Jiao Licheng , Li Lingling , et al . Domain generalization-aware uncertainty introspective learning for 3D point clouds segmentation [C ] // Proceedings of the 32nd ACM International Conference on Multimedia . New York, NY, USA : Association for Computing Machinery , 2024 : 651 - 660 .

Views

下载量

CSCD

Alert me when the article has been cited

提交

Tools

Publicity Resources

Topology-Optimized Liquid Cooling Combined with Phase Change Materials for Thermal Management of Large Cylindrical Batteries

Binocular Vision-Based Synchronous Measurement Method for Rotational Errors

Study on Electrothermal Characteristics of Dendrite-Induced Internal Short Circuit in Lithium-Ion Batteries

Interpretable Mechanical Fault Diagnosis Method Based on Dual-Tree Complex Wavelet Transform

Progress in Visualization Experimental Studies on Heat Transfer of Supercritical Fluids

Related Author

No data

Related Institution

No data

⁰