Multi-category Construction Scene Element Detection Method Based on Improved YOLOv11

Existing machine vision algorithms for element detection in multi-category complex construction scenarios suffer from poor robustness and high miss rates. On the one hand, these methods rely on large amounts of labeled data to improve their adaptability to specific scenarios; on the other hand, they lack sufficient generalization capability, making it difficult to apply them to element detection tasks across different construction environments. This severely limits the practical deployment of visual perception technologies on smart construction sites. To address these issues, this paper proposes a construction element detection method for complex construction scenarios based on an improved lightweight YOLOv11 model. Specifically, DCNv4 deformable convolution is introduced to enhance adaptability to irregular targets, the SimAM parameter-free attention mechanism is embedded to suppress background interference, and Focal Loss is adopted to mitigate class imbalance, thereby establishing a detection model capable of handling multi-scale construction elements across diverse and complex construction scenarios. The proposed method is validated using datasets collected from multiple complex construction environments, including building sites, roadbed construction, and drill-and-blast tunneling. Ablation experiments demonstrate that SimAM reduces the false detection rate by 29.7%, while Focal Loss improves the recall rate of small targets by 11.5%. Compared with the baseline YOLOv11 model, the improved model achieves an average performance gain of 10.2% in mAP. The proposed method effectively balances detection accuracy and computational efficiency, providing a feasible solution for visual detection of construction elements in construction environments.