A point cloud segmentation approach driven by multiple geometric features, referred to as GeoFD, was proposed, to address the challenge of accurately segmenting diverse geometric forms encountered during ship assembly precision inspection. The inherent geometric attributes within point clouds were thoroughly excavated and leveraged by the method. A dual-stream feature representation was strategically designed, in which geometric position encoding was combined with Fast Point Feature Histogram descriptors, and robust feature representations were established across both local and global scales. Subsequently, a weighted linear combination mechanism based on optimal transport theory was incorporated. By this mechanism, the feature fusion process was guided in a geometry-driven manner, and more rational and interpretable integration of contextual information was promoted. For the final inference stage, a segmentation head utilizing similarity matching was adopted to accomplish the semantic labeling of individual assembly components. Experimental validation was performed on actual ship assembly point cloud data, and an Intersection-over-Union score of 89.2% was achieved by GeoFD. This performance exceeded that of several mainstream point cloud segmentation models. Moreover, the method demonstrated a strong sensitivity to intricate geometric structures and exhibited enhanced generalization capability. These findings confirm the method’s practical effectiveness and significant potential for application in precision inspection tasks within ship assembly workflows.