Transfer learning-enabled computer vision in concrete technology: fundamentals, applications, and best practices
TL;DR
This review examines the transformative potential of combining TL and CNN to automate assessment and optimization in different areas of concrete technology, contributing to more cost-effective and sustainable life-cycle performance of concrete infrastructure.
Abstract
The deployment of new artificial intelligence (AI) methods is pivotal in driving innovative automation systems within the construction sector, with growing relevance for improving material sustainability assessment and decision-making. Among different methods, transfer learning (TL) has recently emerged as a key enabler of deep learning success in construction engineering, especially in computer vision applications using convolutional neural networks (CNNs). Recognizing the central role of concrete materials in construction and their significant life-cycle environmental impacts, this review examines the transformative potential of combining TL and CNN to automate assessment and optimization in different areas of concrete technology. It begins by introducing the concept of TL, highlighting prominent off-the-shelf image datasets and CNN models employed in concrete research. The review then showcases the potential of TL-enabled CNN computer vision systems across different stages of the concrete life-cycle, including material selection, construction, quality control, and maintenance, supporting data-driven and resource-efficient practices. Lastly, the review proposes future research directions to foster the integration of these AI-based automation systems into the concrete industry, contributing to more cost-effective and sustainable life-cycle performance of concrete infrastructure.