当前位置: X-MOL 学术Autom. Constr. › 论文详情
Our official English website, www.x-mol.net, welcomes your feedback! (Note: you will need to create a separate account there.)
Automated portfolio-based strategic asset management based on deep neural image classification
Automation in Construction ( IF 9.6 ) Pub Date : 2022-07-13 , DOI: 10.1016/j.autcon.2022.104481
Zigeng Fang , Tan Tan , Jiayi Yan , Qiuchen Lu , Michael Pitt , Sean Hanna

Despite the popularity of image-based classification techniques in identifying building materials and elements in building and construction sites, the real feasibility of it accommodating the needs of operation and maintenance (O&M) inspection-repair processes in the real project scale is still awaiting testing due to the vast number of project assets, hundreds of asset categories, and requirements of portfolio-based strategic asset management (SAM) tasks. The building's O&M phase does not yet have a large-scale inspection-repair dataset created on images based on the actual workflow of building and infrastructure projects. This paper described a MobileNet1.0-based image classification method for optimising and automating a series of portfolio-based SAM service processes, including condition surveying, data validation, standardisation and integration. By examining the performance of image classification in the building O&M phase's SAM tasks, the paper examines whether the constructed MobileNet1.0 model can achieve satisfactory built assets category classification performance based on commercial surveying data with the enhancement of the online image training dataset. The constructed MobileNet1.0 model achieved satisfactory built assets category classification results for both building asset type category classification (with 85.6% level-1 test accuracy) and different building data attributes (e.g., built assets' condition, activity cycle, failure type, etc.). The experiment results proved the online enhanced image training dataset can further boost the on-site image classification performance. The results of this paper demonstrate a broader application of image-based technologies in portfolio-based SAM and other building and infrastructure projects' O&M phase applications. Future research can be conducted to exploit the performance of image classification methods over more building asset categories and building and building asset data attribute types.



中文翻译:

基于深度神经图​​像分类的自动化基于投资组合的战略资产管理

尽管基于图像的分类技术在识别建筑和建筑工地的建筑材料和元素方面很受欢迎,但它在实际项目规模中适应运维(O&M)检查-修复过程需求的真正可行性仍有待检验。大量的项目资产、数百种资产类别以及基于投资组合的战略资产管理 (SAM) 任务的要求。建筑的运维阶段还没有根据建筑和基础设施项目的实际工作流程在图像上创建大规模的检查修复数据集。本文描述了一种基于 MobileNet1.0 的图像分类方法,用于优化和自动化一系列基于投资组合的 SAM 服务流程,包括状态调查、数据验证、标准化和集成化。通过考察建筑运维阶段SAM任务中的图像分类性能,考察了构建的MobileNet1.0模型是否能够通过在线图像训练数据集的增强,基于商业测绘数据实现令人满意的建筑资产类别分类性能。构建的MobileNet1.0模型在建筑资产类别分类(一级测试准确率为85.6%)和不同建筑数据属性(如建筑资产状况、活动周期、故障类型等)上均取得了令人满意的建筑资产类别分类结果.)。实验结果证明,在线增强图像训练数据集可以进一步提升现场图像分类性能。本文的结果展示了基于图像的技术在基于组合的 SAM 和其他建筑和基础设施项目的运维阶段应用中的更广泛应用。可以进行未来的研究,以利用图像分类方法在更多建筑资产类别以及建筑和建筑资产数据属性类型上的性能。

更新日期:2022-07-14
down
wechat
bug