Scholarlink

<p>Automated identification of material defects plays a crucial role in ensuring the safety, durability, and cost-effectiveness of construction projects. Conventional computer-vision methods often struggle to recognize fine-scale flaws on uneven or reflective surfaces, particularly under varying lighting conditions. To overcome these challenges, this research introduces a hybrid CSO–YOLOv8–EVC framework, which integrates Multi-Objective Chicken Swarm Optimization (CSO) with an Enhanced Visual Center (EVC) module inside the YOLOv8 detection network. The CSO algorithm automatically refines model hyperparameters to maintain an effective balance between exploration and exploitation, while the EVC module improves multi-scale feature learning through a combination of global and local attention. The proposed model was trained and validated on an extensive image dataset containing multiple types of construction defects, including cracks, spalling, stains, and irregular surface textures. When benchmarked against established detectors such as Faster R-CNN, RetinaNet, EfficientDet-D3, YOLOv5, and YOLOv8, the hybrid model achieved superior outcomes including 96.1% precision, 94.8% recall, an F1-score of 0.95, and mAP@0.5 of 96.7%, while operating at real-time speed of 48 FPS with an inference latency of 21 ms. These findings confirm the framework’s suitability for high-speed, high-accuracy inspection of construction materials across complex textures and lighting conditions. The results demonstrate that the proposed approach offers a reliable and scalable foundation for intelligent vision-based quality-control systems applicable to civil and industrial inspection environments.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Deep learning approaches for automated defect detection and quality control in construction materials

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月2日

查看全文
  • Huaxia Zhuang,
  • Weijun Jiang

摘要

Automated identification of material defects plays a crucial role in ensuring the safety, durability, and cost-effectiveness of construction projects. Conventional computer-vision methods often struggle to recognize fine-scale flaws on uneven or reflective surfaces, particularly under varying lighting conditions. To overcome these challenges, this research introduces a hybrid CSO–YOLOv8–EVC framework, which integrates Multi-Objective Chicken Swarm Optimization (CSO) with an Enhanced Visual Center (EVC) module inside the YOLOv8 detection network. The CSO algorithm automatically refines model hyperparameters to maintain an effective balance between exploration and exploitation, while the EVC module improves multi-scale feature learning through a combination of global and local attention. The proposed model was trained and validated on an extensive image dataset containing multiple types of construction defects, including cracks, spalling, stains, and irregular surface textures. When benchmarked against established detectors such as Faster R-CNN, RetinaNet, EfficientDet-D3, YOLOv5, and YOLOv8, the hybrid model achieved superior outcomes including 96.1% precision, 94.8% recall, an F1-score of 0.95, and mAP@0.5 of 96.7%, while operating at real-time speed of 48 FPS with an inference latency of 21 ms. These findings confirm the framework’s suitability for high-speed, high-accuracy inspection of construction materials across complex textures and lighting conditions. The results demonstrate that the proposed approach offers a reliable and scalable foundation for intelligent vision-based quality-control systems applicable to civil and industrial inspection environments.