Transmission Line Point Cloud Segmentation Method Based on Dynamic Dual-Branch Encoding

To address the low segmentation accuracy of small targets in transmission line point cloud semantic segmentation, this paper proposes a dynamic dual-branch encoding model. A dynamic dual-branch encoding module, utilizing local dynamic convolution and adaptive attention, is constructed to enhance small target feature extraction. Additionally, a feature fusion module integrates multi-scale features, and a class-weight-based hybrid loss function alleviates insufficient training. Experiments show the model achieves 97.2% overall accuracy, 71.5% mean intersection over union, and 66.5% insulator segmentation accuracy, providing reliable support for transmission line point cloud segmentation.