Real-time recognition model of 3D workpiece for transmission tower based on uniseg3d and vision-inertia fusion

Abstract

To tackle the challenges existing in point cloud segmentation for transmission tower components-specifically, weak small-target perception, blurred boundaries, poor cross-scene adaptability, and insufficient real-time performance-this study develops a real-time recognition model by integrating LiDAR-IMU with UniSeg3D. The core contribution lies in formulating a systematic solution architecture. First, to address the issues of small targets and boundary ambiguity, a multi-scale boundary-aware module is designed. By jointly optimizing classification and boundary losses, it substantially strengthens feature extraction for small components (e.g., insulators) and enhances localization accuracy along boundary regions. Second, to mitigate the reliability degradation of single-sensor systems under complex conditions (such as vegetation occlusion or illumination variation), a vision-inertial bimodal fusion strategy is introduced. High-frequency motion data from the IMU are employed to dynamically compensate for the spatial sparsity of LiDAR point clouds, thereby improving the model’s robustness across diverse scenes. Third, to meet the real-time demands of engineering applications, a feature contribution screening mechanism is incorporated. This enables structured pruning and model lightweighting-reducing processing time per single tower to within 3 seconds-while maintaining accuracy. Experimental results demonstrate that, through the synergistic integration of these techniques, the proposed model achieves an average F-score of 0.853 across six core component categories, with particularly notable advantages in small-target recognition and complex-scene adaptability compared to existing approaches. This model has realized high-precision real-time identification of components in complex scenarios, which provides a technical scheme for intelligent inspection of transmission lines, and promotes the upgrade of the inspection mode from manual to automatic. At last, the stable operation of power systems is guaranteed.