This study presents a label-efficient segmentation method built on MinkUNet, a sparse voxel convolutional network enhanced with self-attention modules in its encoder–decoder for better spatial reasoning over corridor objects (e.g., trees, buildings, towers).
Abstract
Routine inspections of transmission lines are essential for maintaining the reliability of the power grid. Airborne LiDAR technology provides detailed 3D corridor data for automated hazard detection, such as vegetation encroachment and structural anomalies. However, manually analyzing large point clouds is inefficient, and current segmentation methods struggle with scene complexity, scale variation, and the high cost of annotation. In this study, we present a label-efficient segmentation method built on MinkUNet, a sparse voxel convolutional network enhanced with self-attention modules in its encoder–decoder for better spatial reasoning over corridor objects (e.g., trees, buildings, towers). To further handle structural diversity and class imbalance, we adopt task-specific data augmentations and focal loss. A multi-stage pseudo-labeling strategy is then employed to enable effective cross-scene generalization with minimal labeled data. We validate our method on three real-world transmission line datasets. On the Foshan dataset, it achieves a mean Intersection over Union (mIoU) of 0.740 with an inference time of 1.31 s. Cross-scene tests at two other locations, Shumuyuan and Langwang Village, yield mIoUs of 0.762 and 0.757, respectively. These results confirm robust performance even with limited annotations. Overall, our findings demonstrate the practicality of our approach for routine power line inspections, enabling reliable hazard detection with minimal annotation effort.
Abstract. Semantic classification is a fundamental step in Mobile Laser Scanning (MLS) point clouds processing, and remains a non-trivial task. In this work, we propose a classification framework based on a 3D Sparse Convolutional Neural Network (SparseCNN) for efficient processing of large-scale MLS data. A coarse-to-fine two-stage pipeline is introduced, where an essential model performs a classification for the entire scene, followed by a refinement stage for detailed ground-surface classes. To enhance robustness under diverse acquisition conditions, both point-wise and scene-wise data augmentation strategies are employed during the training, including rotation, jittering, density perturbation, noise injection, and patch swapping. To account for environmental and sensor variations, wavelength-specific models are trained for both urban and highway scenes. Experimental results on urban and highway datasets demonstrate strong performance, achieving over 90% accuracy for major classes, while ablation studies show that radiometric features are critical for distinguishing material dependent classes, such as traffic signs, and that the proposed augmentation strategies improve performance for challenging object categories, such as pedestrian, which is dynamic and structurally ambiguous.
Nan-Feng Li, H. Teufelsbauer, F. Pöppl et al.· The International Archives o...· 0 citations
LiDAR semantic segmentation is significant in applications such as autonomous driving and robot navigation, as it greatly improves scene perception and object detection. However, the existing methods face the challenges of achieving high segmentation accuracy while maintaining low computational cost and complexity. In this paper, we propose a new, to our knowledge, efficient and accurate semantic segmentation network for LiDAR, called Range-FDSeg. To reduce the risk of information compression and loss when projecting 3D point cloud data onto 2D range images, we design a multi-channel fusion interactive learning (FIL) module. This module effectively integrates multimodal channels, such as coordinates, depth, and reflectivity, for interactive learning. As a result, FIL module can reduce the noise interference inherent in individual channels and capture the underlying relationships between different physical quantities. To further improve the performance, we introduce a lightweight and dynamic upsampler, called Dysample-S+. It effectively resolves the inherent challenges of traditional sampling methods through its adaptive weighting mechanism, which dynamically adjusts to local geometric patterns and density variations in raw point clouds. Extensive evaluations on publicly available benchmark datasets, including SemanticKITTI, SemanticPOSS, and NuScenes, demonstrate that the proposed Range-FDSeg outperforms most existing state-of-the-art methods.
Hong Tan, Fen Chen, Tingna Liu et al.· Applied Optics· 0 citations
3D scene understanding is increasingly important in construction, yet most methods are developed on curated datasets that do not fully reflect real site sensing conditions. In many workflows, individual LiDAR scans provide rapid local updates rather than complete scene representations, producing limited surface coverage, acquisition-driven density variation, and severe imbalance between dominant planar surfaces and sparse construction elements. Because large point clouds must be downsampled, sampling resolution and point allocation directly affect the balance between geometric detail and spatial context. This study evaluates these effects under a fixed per-fragment point budget and introduces an incidence-aware sampling strategy for individual LiDAR scans. The method maps points to a geometry-normalized manifold space for voxel-based selection while preserving original Euclidean coordinates for downstream learning. It requires only point coordinates and normals and no backbone modification. Using the Site in Pieces (SIP) benchmark, experiments with Point Transformer and PointNeXt show improved resolution-averaged segmentation performance, especially for non-planar elements and ladders, while reducing sensitivity to sampling resolution. The results show that acquisition-aware sampling can provide a more stable geometric representation and should be treated as an active component of individual-scan 3D segmentation rather than generic preprocessing.
Seongyong Kim, Jingdao Chen, Y. K. Cho· 0 citations
With advantages in anti-interference and active sensing, Light Detection and Ranging (LiDAR) has become a primary modality for environmental perception in underground coal mines characterized by inadequate illumination and high dust concentrations. However, raw point clouds lack semantic attributes, which limits their direct application in scene understanding for digital mines. Although fully supervised deep learning methods for point cloud semantic segmentation achieve high performance, they rely heavily on labor-intensive point-wise annotations, incurring high costs and potential human error. To address this, this paper proposes a weakly supervised semantic segmentation framework specifically designed for coal mine roadways. The framework establishes a comprehensive workflow spanning from data acquisition to sparse-supervised training. By leveraging sparse labels to drive model optimization, the method maintains segmentation performance while significantly reducing annotation costs, achieving a strategic balance between engineering precision and deployment overhead. Experimental results demonstrate that using only 0.1% sparse annotations, the proposed method achieves an Overall Accuracy (OA) of 91.64% and a mean Intersection over Union (mIoU) of 78.83%. Compared to fully supervised models, it achieves a 2-3 times improvement in annotation efficiency with a precision loss of approximately 7%. The segmented structural point clouds are utilized for roadway deformation monitoring and parametric geometric reconstruction. This supports a comprehensive digital twin framework that achieves high-fidelity modeling and centimeter-level deformation early-warning. This research confirms that the proposed technology provides a cost-effective and efficient path for scene understanding in intelligent mines, demonstrating substantial engineering value and potential for industrial promotion.
Quanyi Xie, Yalin Zhang, Lizhi Zhou et al.· Measurement science and tech...· 0 citations
Point cloud semantic segmentation is fundamental for 3D scene understanding and has been widely used in autonomous driving and infrastructure inspection applications. However, its performance is often limited by insufficient representation of local geometric structures and the high cost of point-wise annotation. To address these issues, this paper proposes GeoFuse-AL, a label-efficient segmentation framework that integrates a geometry-guided point–voxel network with multi-cue active learning. The base model, GeoFuseNet, builds on a hybrid point–voxel backbone and incorporates a Local Geometry Prototype Attention module to enhance object boundaries, fine-grained structures, and local geometric patterns. An Adaptive Channel Fusion module is further designed to improve feature interaction between point-level details and voxel-level context. To reduce annotation dependence, a Multi-Cue Diversity Active Sampling strategy combines prediction uncertainty, color-gradient variation, geometric curvature, and feature-space clustering to select informative and diverse samples. Experiments on S3DIS and SemanticKITTI demonstrate that the proposed model achieves mIoU scores of 63.3% and 61.7%, respectively, outperforming several representative methods. Under limited annotation settings, the proposed strategy reaches 99.2% of fully supervised performance with only 15% labeled data on S3DIS and 97.6% with only 5% labeled data on SemanticKITTI. These results demonstrate that GeoFuse-AL improves segmentation accuracy while substantially reducing annotation requirements.
Validated on real tunnel data, PDAPointNet++ outperforms models such as PointNet, PointNet++, PointMLP, and dynamic graph convolutional neural network (DGCNN) and the impact of different WDFL balancing coefficients on model performance is discussed.
Xin Peng, Mingnian Wang, Hao Zhong et al.· Journal of computing in civi...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
