Creating a Dataset to Boost Civil Engineering Deep Learning Research and Application
| Autor: | Murad Al Qurishee, Ignatius Fomunung, Babatunde Atolagbe, Joseph Owino, Mbakisya Onyango, Weidong Wu |
|---|---|
| Rok vydání: | 2020 |
| Předmět: |
Government
business.industry Computer science Deep learning 02 engineering and technology Pedestrian Benchmarking 010402 general chemistry 021001 nanoscience & nanotechnology 01 natural sciences Civil engineering Convolutional neural network 0104 chemical sciences Work (electrical) Structural health monitoring Artificial intelligence 0210 nano-technology business Intelligent transportation system |
| Zdroj: | Engineering. 12:151-165 |
| ISSN: | 1947-394X 1947-3931 |
| DOI: | 10.4236/eng.2020.123013 |
| Popis: | With cutting edge deep learning breakthrough, numerous innovations in many fields including civil engineering are stimulated. However, a fundamental issue that civil engineering research community currently facing is lack of a publicly available, free, quality-controlled and human-annotated large dataset that supports and drives civil engineering deep learning research and applications on such as intelligent transportation including connected vehicle, structural health monitoring, and bridge inspection. This paper is a general discussion about demanding needs and construction of a long-anticipated dataset for researchers and engineers in civil engineering and beyond for providing critical training, testing and benchmarking data. The establishment of such a free dataset will remove a major hurdle and boost deep learning research in civil engineering and we hope this work will urge researchers, engineers, government agencies and even computer scientists to work together to start building such datasets. A framework has been developed for the proposed database. Also, some pilot study databases were developed for concrete crack detection, pavement crack detection using normal and infrared thermography, as well as pedestrian and bicyclist detection. A convolution neural network model called Faster RCNN was deployed to check the detection accuracy and a 98% detection accuracy of the proposed datasets was obtained. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|