Deep learning-based detection of construction waste in complex scenarios with an improved lightweight algorithm

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

To tackle the challenges in construction waste detection under complex scenarios—such as insufficient recognition accuracy, significant feature variations among identical waste categories, limited publicly available CDW datasets, and excessive computational resource consumption that hinders real-time performance—this paper proposes a novel improved algorithm, GTS-YOLO. Built on YOLOv11, GTS-YOLO achieves model lightweighting by optimizing the C3K2 module in the backbone network, enhances detection accuracy effectively through integrating spatial attention mechanisms between the backbone and neck, and redesigns the detection head with reference to the task alignment principle to better handle object detection of construction waste under occlusion and deformation. On our self-constructed 10-category dataset, compared with YOLOv11n, GTS-YOLO increases mAP50 by 3.24% to 67.47%, improves Precision by 3.1%, and reduces the parameter count by 19.4%, In addition, the model size is only 4.23 MB, which is approximately 20.6% smaller than YOLOv11n, demonstrating the effectiveness of the proposed lightweight design.achieving an effective balance between accuracy and efficiency.We have also validated the model performance on public datasets, demonstrating its generalization ability across diverse scenarios.