In the intelligent recognition of construction drawings, architectural components significantly vary in size, and small objects are easily overlooked. To address these two challenges, this paper proposes a multi-scale feature fusion method based on channel and spatial attention mechanisms. Specifically, the method constructs a hybrid feature pyramid structure (ArchFuse-Net) to enhance multi-scale feature representation. Subsequently, a small-object perception module is introduced to improve the recall rate of small components. Finally, the bounding box regression loss function and post-processing mechanisms are refined to boost detection accuracy and localization precision. Experimental results demonstrate that the proposed method effectively improves overall detection accuracy across varying sizes and significantly boosts the recall of small objects, providing a robust solution for architectural drawing analysis.
Loading....