Construction Garbage Classification Detection Using an Enhanced YOLOv12 Framework
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
Construction waste detection is challenged by mixed material composition, significant scale variation, complex backgrounds, and the prevalence of small targets. To address these issues, this paper proposes an enhanced construction waste classification detection model, termed BWM-YOLO, based on the YOLOv12 framework. A weighted bidirectional feature pyramid network (BiFPN) is introduced in the neck to enable efficient crossscale feature fusion through fast normalized weighting, thereby improving multi-scale feature representation. In addition, the original CIoU loss is replaced with Wise-IoU v3, which employs a dynamic non-monotonic focusing strategy to suppress low-quality anchor interference and enhance small-object localization accuracy. Furthermore, Mixup data augmentation is adopted to expand the training data distribution and mitigate overfitting under complex construction waste scenarios. Experimental results on the CODD dataset demonstrate that the proposed model achieves mAP50 and mAP50-95 values of 89.5% and 85.3%, respectively, outperforming the baseline YOLOv12 by 3.4 and 1.9 percentage points. Notably, these improvements are obtained with only a marginal increase in model parameters and no additional computational cost. Comparative and ablation experiments further verify the effectiveness of the proposed modules, indicating that the model satisfies the accuracy and real-time requirements of intelligent construction waste sorting applications.
Fields of Study & Subjects
Advanced Neural Network Applications
Industrial Vision Systems and Defect Detection
Infrastructure Maintenance and Monitoring
Publication details
- Journal / source
- International Conference on Electrical Automation and Artificial Intelligence (ICEAAI)
- Pages
- 1–4
- Publication date
- 16 January 2026
- Language
- EN