首页|基于YOLOv5s户外场景下混凝土裂缝识别模型改进

基于YOLOv5s户外场景下混凝土裂缝识别模型改进

扫码查看
近年来,机器视觉算法的发展推动了基于深度学习的裂缝检测技术在混凝土建筑物中的应用.该技术利用小米13 手机和大疆无人机在自然环境下采集了3 134 张包含裂缝、腐蚀、坑洞的图像,以增加模型的泛化能力,提高裂缝检测的准确性和效率,特别是在非固定端的实际应用中,为了实现这一目标,文中引入了Mosaic—9数据增强和ECA及CA注意力机制,优化了YOLOv5s模型.使用YOLOv5s目标检测网络,并引入GhostNet网络以优化计算效率,通过消融实验验证了轻量化优化策略的有效性.GhostNet的引入在略微增加模型参数和计算量的情况下,F1 分数提高了113.17%,模型识别速度提高了 62.3%,召回率提高了 23%.研究表明,结合数据增强、注意力机制和轻量化网络优化,可以有效地提高混凝土裂缝检测的准确性和效率.
Enhancement of Concrete Crack Identification Model in Outdoor Environment Utilizing YOLOv5s
In recent years,the development of machine vision algorithms has promoted the application of crack detection technology based on deep learning in concrete buildings.The technology uses Xiaomi 13 mobile phones and DJI drones to collect 3 134 images containing cracks,corrosion,and pits in natural environments to increase the generalization ability of the model.Improve the accuracy and efficiency of crack de-tection,especially in practical applications of non-fixed ends.To achieve this goal,the research introduced Mosaic-9 data enhancement and ECA and CA attention mechanisms to optimize the YOLOv5s model.Use the YOLOv5s target detection network and introduce the GhostNet network to optimize computational efficiency.The effectiveness of the lightweight optimization strategy is verified by an ablation experiment.The introduction of GhostNet increases the F1 score by 113.17%,the model recognition speed by 62.3%,and the recall rate by 23%,with a slight increase in model parameters and computation.The research shows that combining data enhancement,attention mechanism,and light-weight network optimization can effectively improve the accuracy and efficiency of concrete crack detection.

Fissures in concreteMachine vision technologyYOLOv5s

邱锐、孙勇、解可悦、刘松林、尹萌、杨安琪、汤丰莹

展开 >

黑龙江科技大学建筑工程学院,黑龙江 哈尔滨 150022

龙建路桥股份有限公司,黑龙江 哈尔滨 150006

哈尔滨工业大学交通科学与工程学院,黑龙江 哈尔滨 150090

混凝土裂缝 机器视觉 YOLOv5s

2024

江西建材
江西省建材科研设计院

江西建材

影响因子:2.247
ISSN:1006-2890
年,卷(期):2024.(9)