SHS Web Conf.
Volume 144, 20222022 International Conference on Science and Technology Ethics and Human Future (STEHF 2022)
|Number of page(s)||9|
|Section||Application of Artificial Intelligence Technology and Machine Learning Algorithms|
|Published online||26 August 2022|
YOLO V5s-based Deep Learning Approach for Concrete Cracks Detection
School of Software, Jiangxi Normal University, Nanchang 330027, China
* Corresponding author. Email: email@example.com
Complex environmental conditions can lead to a variety of cracks in concrete engineering structures, and if these cracks are not promptly investigated and repaired, it is likely to lead to serious engineering accidents. Most of the traditional crack detection method is by manual exclusion, which overly relies on the knowledge and experience of inspectors, and different inspectors have different definitions of crack detection standards, so it lacks a certain objectivity in quantitative analysis, and the efficiency of manual detection will gradually decrease as the workload rises. In recent years, deep learning networks have made many developments in the field of computer vision by their strong feature extraction ability and autonomous learning capability. The main objective of this paper is to detect crack information in crack images using YOLO V5s-based deep learning algorithm. Considering the complexity of the crack image background, the author adopted the threshold segmentation method based on the Otsu maximum inter-class variance to achieve the purpose of removing the background noise from crack images by constructing a connected domain for grayscale change points so as to fuse the noise points with the background. After that, the author used the YOLO V5s model to train and test the 3500 manually labeled crack images, and adopted the K-Means method to calculate the optimal initial anchor box size and pass it to the model for training, so as to improve the model’s detection of cracks. The evaluation index of the model after these two optimization methods was 84.37% for average precision (AP), 76.01% for average recall (AR), and 79.97% for average F1-score.
© The Authors, published by EDP Sciences, 2022
This is an Open Access article distributed under the terms of the Creative Commons Attribution License 4.0, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.
Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.
Initial download of the metrics may take a while.