Abstract:

This paper proposes a laser point cloud segmentation algorithm for power distribution lines based on a fused Transformer, aiming to enhance the precision and efficiency of segmenting and extracting key modules such as power lines, towers, and insulators. A dual-channel parallel architecture feature extraction module is constructed to separately capture high-frequency and low-frequency features, with low-frequency features using average pooling and a fused Transformer extractor, and high-frequency features using max pooling and an MLP module that includes convolutional layers. The feature vectors from both channels are fused to enhance detail extraction. Incorporating an MLP module further refines feature processing for accurate point cloud target segmentation. Extensive experiments validate the algorithm’s accuracy.The algorithm proposed in this paper has the potential advantages of improving accuracy, enhancing automation, increasing robustness, integrating multi-source data, and reducing costs in UAV inspection.

Key words: Transformer, distribution lines channel, feature fusion, laser point cloud, neural network