A new YOLOv5s model incorporating Slim-neck and SimAM for detecting surface bubble defects of fair-faced concretes

Shi-Han Xu , Jing-Liang Liu , Wen-Ting Zheng , Zi-Xuan Chen

Abstract

Since traditional methods for detecting surface bubble defects usually suffer from drawbacks such as slow speed and low accuracy, a new You Only Look Once version 5s (YOLOv5s) model, incorporating Slim-neck and simple parameter-free attention module (SimAM), is proposed to address this challenge. Initially, data augmentation strategies are employed to expand the original dataset, which comprises 1,835 images, into a new dataset of 7,340 photographs, thereby enhancing the generalizability of the newly proposed YOLOv5s model. Subsequently, a lightweight Slim-Neck network, grounded in a group shuffling convolution (GSConv) module, is integrated into the backbone network to reduce the complexity of the newly proposed YOLOv5s model while maintaining a high level of target detection accuracy. Furthermore, a SimAM mechanism is embedded within the neck network, enabling the refined model to focus more intensively on the key characteristics of surface bubbles in fair-faced concrete. To validate the effectiveness and accuracy of the proposed YOLOv5s model, a dataset including 7,340 bubble defect images is investigated. The results demonstrate that the proposed model successfully strikes a commendable balance between the detection speed and accuracy. Additionally, it offers a promising and effective tool for detecting surface bubbles in real-world fair-faced concrete structures.

Key Words

attention mechanism; fair-faced concrete lightweight network; object detection; surface bubble defects; YOLOv5s

Address

(1) Shi-Han Xu, Jing-Liang Liu — College of Transportation and Civil Engineering, Fujian Agriculture and Forestry University, Fuzhou 350108, China
(2) Shi-Han Xu — College of Civil Engineering, Fuzhou University, Fuzhou 350108, China
(3) Wen-Ting Zheng — College of Civil Engineering, Fujian University of Technology, Fuzhou 350118, China
(4) Zi-Xuan Chen — Faculty of Science and Technology, University of Macau, Macau, China.

PDF Viewer

Preview is limited to the first 3 pages. Sign in to access the full PDF.

← Back to Volume 38, No. 2