Dr. Bo Xiao wins 2023 ASCE Best Paper Award

Dr. Bo Xiao, a recent addition to the ICC and Michigan Tech in general, received recognition from the American Society of Civil Engineers (ASCE). His work, entitled “Deep Learning Image Captioning in Construction Management: A Feasibility Study”, received the 2023 ASCE Best Paper Award in the Journal of Construction Management and Engineering for decisive and impactful outcomes that aim to accelerate a crucial part of construction processes.

Pictures are taken constantly throughout the construction process as it is important to document all steps to track work progress and completion. However, it is hard to compile these pictures efficiently that also allows for efficient retrieval on the backend when someone wants a specific picture.

That’s where Dr. Xiao and his team come in. In collaboration with Dr. Shih-Chung Kang and Dr. Yiheng Wang from the University of Alberta – Edmonton, Dr. Xiao worked with six different deep-learning capturing methods to identify if leveraging artificial intelligence to automatically caption worksite images to simplify this tedious process was a feasible solution.

Spoiler alert – it was. Captions generated using deep-learning methods correctly labelled most images, with the higher-performing methods producing an accuracy rate around 90%. This demonstrates the strong base this technology already possesses for this specific and extremely useful application, as well as the exciting prospect at building on the outcomes to create even more accurate deep-learning image captioning systems.

The ASCE Editor-in-Chief Jesus de la Garza asserts that the work was selected from over 200 publications that were in the journal from July 2022 through June 2023, and endorsed Dr. Xiao and his team for clearly presenting an involved methodology that could have a significant impact on the operations of construction management entities and sets the stage for further developments.

Leave a Reply

Your email address will not be published. Required fields are marked *