摘要
针对建筑物裂缝检测准确率不高、效率较低、检测结果易受环境因素干扰等问题,提出一种基于YOLOv5改进的建筑物裂缝检测模型.该模型增加了针对裂缝狭长特点的注意力机制以提高检测准确率;增加了Ghost模块替代部分传统卷积以提高检测效率.模型采用SDNET2018数据集进行验证测试,经过对比试验,该模型相比于YOLOv5精度提高10.1%,速率提高了13.3帧/s.该模型具有准确率高、轻量化的特点,可进行边缘端部署以实现复杂环境下的实时监测.
In response to challenges like subpar accuracy,inefficiency,and vulnerability to environmental disturbances in building crack detection,this study introduces an enhanced model based on YOLOv5.This upgraded model incorporates an attention mechanism tailored to the elongated and narrow attributes of cracks to boost detection precision.Additionally,it integrates Ghost modules in lieu of traditional convolutions to enhance detection efficiency.Validation of the model was conducted using the SDNET2018 dataset.Comparative analysis revealed a 10.1%increase in accuracy and a 13.3 frames/s improvement in speed compared to YOLOv5.Notably,this model offers high accuracy and is lightweight,enabling edge deployment for real-time monitoring in complex environments.
作者
王泽晖
武斌
赵洁
WANG Zehui;WU Bin;ZHAO Jie(School of Computer and Information Engineering,TCU,Tianjin 300384,China)
出处
《天津城建大学学报》
2025年第4期302-307,共6页
Journal of Tianjin Chengjian University
基金
国家自然科学基金(62204168)
天津市重点研发计划科技支撑重点项目(19YFZCGX00130)
天津市企业科技特派员项目(19JCTPJC47200)
天津市研究生科研创新项目(2021YJSS351)。
