摘要
为了提高房屋危险性等级鉴定工作的效率,避免人为干预,使鉴定结果更客观,设计DC-YOLO模型。该模型是将深度学习与目标检测算法结合,设计的一种砌体构件危险性等级自动化鉴定方法。采用K-means聚类获得最佳先验框数量和尺寸;扩大网格尺寸以提高对小目标的识别能力;引入可变形卷积,以适应砌体构件裂缝形状不规则的特点。实验结果表明,DC-YOLO模型与常规方法比较,对各类目标的识别率均有不同程度的提高,精确率、召回率、F1值均达到了较好的效果,对于建筑物砌体构件危险性等级有较好的分类性能。
In order to improve the efficiency of the house hazard level identification, avoid human intervention, and make the identification results more objective, the DC-YOLO model was designed. By combining deep learning with target detection algorithms, the model was an automated identification method for the hazard level of masonry components. K-means clustering was used to obtain the best number and size of a priori frames. The grid size was expanded to improve the ability to identify small targets. We introduced deformable convolution to adapt to the irregular shape of the cracks in masonry components. The experimental results show that compared with the conventional method, DC-YOLO model has improved the recognition rate of various targets in varying degrees. The precision ratio, recall and F1 value have achieved good results. It has better classification performance for the hazard level grade of building masonry components.
作者
张洪瑞
卫文学
车吉鑫
邵婉露
Zhang Hongrui;Wei Wenxue;Che Jixin;Shao Wanlu(College of Computer Science and Engineering, Shandong University of Science and Technology, Qingdao 266590, Shandong, China)
出处
《计算机应用与软件》
北大核心
2019年第9期181-185,213,共6页
Computer Applications and Software
基金
国家重点研发计划项目(2016YFC0801408)
关键词
深度学习
砌体构件
危险性等级鉴定
可变形卷积
Deep learning
Masonry components
Hazard level identification
Deformable convolution