02748nas a2200265 a 450000500170000000800410001703500150005810000260007324500720009926000670017130000220023849000260026050001260028650200590041252016060047165000370207765000310211465000330214570000370217870000420221570000480225771000460230581000580235185600730240920260817162314.0240223s2023    th u  m  tt   000 a eng d  a.b124205540 aPasuthep Vannaburana 10aAnalysis of building facade defect using deep segmentation models    aPathum Thani, Thailand :bAsian Institute of Technology,c2023  a78 leaves :bill.1 aThesis ;vno.CS-23-01  aA thesis submitted in partial fulfillment of the requirements for the degree of Master of Engineering in Computer Science  aThesis (M. Eng.) - Asian Institute of Technology, 2023  aBuilding crack identification and structural crack assessment is an important task for structural inspection, since the early detection helps reduce the risk of loss. The traditional crack detection can be difficult, time consuming, and in case of high-rise buildings, dangerous, not to mention costly. Since detection relies on the experience of specialists, manual inspection can also be biased or inaccurate, therefore a method that utilizes automated machine vision for detecting and evaluating building cracks has been proposed as one way of improving or complementing traditional manual inspection for detecting the cracks in buildings. Because cracks on different structural elements have different causes and effects, each input image is first fed to a semantic segmentation model that outputs an image in which each building component such as column, wall, or beam, is labeled. Each output is then fed to an instance segmentation model that outputs an image in which each crack is labeled and masked with a color. The models will not only detect cracks in images but also give the instance mask segmentations of the cracks along with the building elements the cracks occurred on, which will help provide the useful information such as width, position, configuration, severity level and corrective action of cracks as a report. This study demonstrates that one can incorporate deep learning in an automated or partially automated system to evaluate cracks in building surfaces. It offers a fresh outlook on the possibility of using images automatically for building inspection and monitoring.  0aDeep learning (Machine learning) 0aComputer-aided engineering 0aBuilding materialsxCracking1 aDailey, Matthew N.,eChairperson1 aAnwar, Naveed,eExamination Committee0 aMongkol Ekpanyapong,eExamination Committee2 aRoyal Thai Government,eScholarship Donor2 aAsian Institute of Technology.tThesis :vno.CS-23-01403Full-Textuhttp://203.159.5.9/ait-thesis/Viewer/viewer.php?id=B20401