根据新的多阶段YOLOV10-ViT框架改进混凝土裂分类
Ali Mahmoud Mayya1, Nizar Faisal Alkayem2
1Computer and Automatic Control Engineering Department, Faculty of Mechanical and Electrical Engineering, Tishreen University, Lattakia 2230, Syria.
本研究介绍了一种新的多阶段深度学习框架,使用YOLOV10和视觉变压器 (ViT) 进行早期混凝土裂检测和分类. 先进的模型准确地识别了裂类型,有助于防止结构崩.
科学领域:
- 土木工程 土木工程是指土木工程.
- 计算机视觉 计算机视觉
- 人工智能的人工智能
背景情况:
- 传统的混凝土裂检测是劳动密集型和耗时的.
- 基于视觉的深度学习为结构健康监测提供了高效的自动化解决方案.
- 早期识别混凝土裂对于防止结构故障至关重要.
研究的目的:
- 开发和评估一种新的多阶段深度学习框架,用于混凝土裂检测和多类类型分类.
- 为了提高识别正常,简单裂和多分支裂类型的准确性和效率.
- 提供一个自动化系统,用于预警混凝土潜在的结构变形.
主要方法:
- 一个多阶段的框架,将YOLOV10用于缺陷区域检测和修改的视觉变压器 (ViT) 用于裂分类.
- 培训YOLOV10在1116个具体图像和边界框的数据集上.
- 在12,000张图像数据集上训练ViT模型,用于分类三个裂类型 (正常,简单,多分支).
主要成果:
- 拟议的多阶段YOLOV10-ViT模型实现了90.67%的精度,90.03%的回忆,以及裂纹类型分类的90.34%F1得分.
- 多阶段框架显著优于单个ViT模型,在回忆中显示了高达19.99%的改进.
- 该模型在检测和分类各种混凝土裂类型方面表现出高度准确性.
结论:
- 开发的多阶段深度学习框架为混凝土裂检测和分类提供了高度准确和高效的解决方案.
- 这个YOLOV10-ViT模型可以集成到建筑系统中,用于实时结构健康监测和早期预警.
- 这些发现有助于推进土木工程和结构维护中的自动化检查技术.
更多相关视频
04:58Mechanoluminescent Visualization of Crack Propagation for Joint Evaluation
Published on: January 6, 2023
09:00Visualization of Failure and the Associated Grain-Scale Mechanical Behavior of Granular Soils under Shear using Synchrotron X-Ray Micro-Tomography
Published on: September 29, 2019
相关概念视频
Types of Non-structural Cracks in Concrete
Plastic shrinkage cracks typically form within hours after the concrete is poured. The concrete's surface dries faster than the bottom, creating tensile stress that the still-plastic concrete cannot withstand, leading to diagonal or randomly patterned cracks on the concrete surface.
Microcracking in Concrete
Structural Classification of Joints
A fibrous joint is where the adjacent bones are united by fibrous connective...
Non-destructive Tests for Concrete Strength
Aggregates Classification
Petrographic classification groups aggregates based on common mineralogical characteristics. Some of the common mineral groups found in aggregates are...
Workability of Concrete
Concrete's workability is determined by its resistance to internal forces that arise...
