Point cloud denoising method for annular forgings based on dual-path feature fusion

Accurate dimensional measurement of annular forgings is crucial for ensuring product quality and operational safety. However, environmental interference and manufacturing uncertainties often result in non-uniform point cloud distributions and overlapping scales between structural features and noise, leading to conflicts between global and local geometric information that compromise measurement accuracy. To address these challenges, we propose what we believe to be a novel point cloud denoising framework-dual-path structured denoiser (DSD)-that establishes a collaborative global-local modeling framework. DSD simultaneously extracts local topological structures and captures global geometric patterns through a dual-path feature learning architecture while integrating a priori geometric constraints with confidence-based assessment to strengthen noise suppression. A multi-layer refinement structure enables progressive filtering of outliers. Experimental results on both annular forging samples and the Stanford dataset demonstrate that DSD surpasses state-of-the-art denoising methods in precision, recall, F1 score, and shape preservation, validating its effectiveness and practical applicability in high-noise industrial environments.