Automated data collection and machine learning methods for civil infrastructure condition assessment in sparsely inhabited regions of Canada

加拿大是总面积第二大的国家，但它是世界上人口密度最低的国家之一。大约三分之二的人口居住在距该省南部边界100公里以内的地区，而这一地区仅占加拿大领土总面积的4%。尽管人口分布不均，但庞大的民用基础设施网络遍布加拿大人烟稀少的地区。这些资产对于土著和地方社区的公平获取以及社会和经济繁荣至关重要，许多农业和工业经营都依赖于它们的可靠表现。许多研究工作已发展出诊断和管理民用基础设施的方法，但这些项目大多数集中在人口密集地区的主要资产上。例如，道路状况评价领域的研究大多是针对铺装道路的评价方法。然而，在加拿大偏远地区，未铺设的道路在运输和服务活动中发挥着重要作用，该国约60%的公路网是未铺设的。除了道路，许多重要的基础设施资产，如水利设施，分散在加拿大的偏远地区。这些资产中有许多是几十年前建造的，其中相当一部分在加拿大基础设施报告卡中被评为公平或低于公平水平。传统的基础设施状况评估方法是劳动密集型和昂贵的，由于缺乏专家和需要覆盖的长距离，这些问题在偏远地区更加成问题。本项目旨在通过应用先进的数据收集和机器学习方法，开发提高偏远地区状态评估和数据分析自动化水平的方法。本研究计划的主要目标可以概括为：1)弥合最先进的计算机视觉和机器学习算法与基础设施状况评估领域实际应用之间的差距；2)加拿大偏远地区基础设施资产的数据中心管理。提出的研究计划包括几个研究项目，每个研究项目将调查一个具体的和相关的主题在这个领域。该研究项目将潜在地将基础设施资产的状态评估从传统转变为高度自动化的数据驱动实践，从而提高状态评估过程的效率和准确性。此外，所制订的方法将促进地方社区对基础设施管理的贡献，并减少将状况评估工作外包的需要。偏远地区状况评估自动化水平的提高将节省成本，并改善加拿大农村地区基础设施的可服务性。

Canada is the second largest country by total area, but it has one of the lowest population densities in the world. Approximately two-thirds of the population resides within 100 kilometers of the county's southern border, which accounts for only 4% of Canada's total territory. Despite this uneven distribution of the population, vast networks of civil infrastructure are spread throughout the sparsely inhabited regions of Canada. These assets are critical for equitable access and social and economical prosperity of Indigenous and local communities, and many farming and industrial operations depend on their reliable performance. Many research efforts have developed methods for diagnostics and management of the civil infrastructure, but the majority of these projects focused on major assets that exist in densely populated regions. For example, most of the research in the area of road condition assessment developed methods for assessment of paved roads. Unpaved roads, however, have a major role in transportation and service activities in the remote regions of Canada, where about 60% of the country's road network is unpaved. In addition to the roads, many vital infrastructure assets, such as hydraulic structures, are scattered in the remote regions of Canada. Many of these assets were built decades ago and a considerable portion of them are rated in fair or below fair levels by the Canadian Infrastructure Report Card. Conventional infrastructure condition assessment methods are labour-intensive and expensive, and these issues are more problematic in remote regions due to the shortage of experts and the long distances that need to be covered. This research program aims at developing methods to enhance the automation level of condition assessment in remote regions and data analysis by the application of advanced data collection and machine learning methods. The main objectives of this research plan can be summarized as: 1) bridge the gap between the state-of-the-art computer vision and machine learning algorithms and practical applications in the field of infrastructure condition assessment; 2) data-centric management of the infrastructure assets in the remote regions of Canada. The proposed research program consists of several research projects and each one will investigate a specific and related topic in this area. This research program will potentially transform condition assessment of infrastructure assets from conventional to data-driven practices with high level of automation, leading to improvements in efficiency and accuracy of the condition assessment processes. In addition, the developed methods will facilitate the contribution of local communities to the management of infrastructure and reduce the need for outsourcing condition assessment efforts. The increased automation level of the condition assessments in remote regions will result in cost savings and improvement in the serviceability of the infrastructure in the rural regions of Canada.