OSAM: Outdoor 3D Point Cloud Semantic Segmentation using Attention and Multi-Scale Features

Semantic segmentation, as a key application in intelligent edge computing fields such as autonomous driving and robotics, is crucial for scene understanding and precise perception. Traditional point-based and voxel-based methods are computationally intensive, making them unsuitable for outdoor scene segmentation. To address this, we propose an outdoor point cloud semantic segmentation method based on attention mechanisms and multi-scale features (OSAM). OSAM incorporates multi-scale feature learning with a pyramid structure formed by dilated and branch strip convolutions during encoding. In the decoding phase, channel attention is used to capture global dependencies across different feature map resolutions. Experimental results on the SemanticKITTI dataset show that OSAM significantly improves segmentation accuracy and provides robust perception in complex driving scenarios.