VAD-MambaYOLO: improvement and application of MambaYOLO for weld detection on metal roofing systems

View at publisher Summarise with VIRAT
Authors: Jiankun Bai, Wanyi Yu, Zhiyong Xin, Dayang Wang, Zitao Huang
Cited by: 0
References: 37

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

Abstract The automated inspection of welds on metal roofing systems at large-scale infrastructures, such as airport terminals, is essential for long-term structural health monitoring. However, achieving a practical balance between high detection precision and computational efficiency remains challenging due to the complex visual characteristics of weld defects, including scale variation and low contrast under reflective metal surfaces. To address this problem, this study proposes VAD-MambaYOLO, a high-precision detection model that performs multi-dimensional optimizations on the MambaYOLO baseline architecture. It integrates an improved VisionClueMerge module, which combines multi-scale convolutions with attention mechanisms to strengthen feature extraction and contextual awareness. Additionally, a dynamic upsampling operator is introduced to adaptively refine feature reconstruction, simultaneously enhancing localization accuracy and processing efficiency. Evaluated on a real-world dataset of metal roof welds, VAD-MambaYOLO achieves a precision of 0.882, recall of 0.853, mAP50 of 0.892, and mAP50-95 of 0.432. Compared with various YOLO variants and other detectors, the proposed model achieves a significant improvement in detection accuracy while maintaining a favorable computational footprint, offering a robust and efficient solution for automated weld quality assurance in metal roofing construction.

Fields of Study & Subjects

Industrial Vision Systems and Defect Detection Infrastructure Maintenance and Monitoring Advanced Neural Network Applications

Publication details

Journal / source
Engineering Research Express
Volume
8
Issue
7
Pages
075106–075106
Publication date
27 March 2026
Language
EN