3D point cloud semantic segmentation,as a critical technology for robots to attain cognitive intelligence,currently faces the challenge of low classification accuracy in edge features.This issue primarily arises from two limitations in existing semantic segmentation networks:the neglect of deep feature correlations among neighboring points during feature extraction,and the insufficient fine grained representation capability of global features during feature sampling.To address these issues,the proposed method introduces a topology aware mechanism during the feature extraction stage,enabling the network to capture similarity relationships between neighborhood point features to a greater extent.In the semantic information generation stage,a global feature enhancement module based on the U net architecture is developed,which utilizes feature tensors with identical dimensionality to those from the feature extraction stage to reinforce upsampled features.This dual strategy approach allows the network to focus on fine grained local information while preserving global consistency,thereby achieving enhanced semantic discrimination.Experimental results demonstrate that the method achieves an accuracy of 86.7% accuracy and a mean Intersection over Union (mIoU) of 60.7% on the Area 5 test set of the S3DIS dataset,along with an mIoU of 85.1% on the ShapeNet dataset.compared to classical methods,the proposed point cloud semantic segmentation algorithm achieves superior semantic segmentation performance.