IMPROVEMENT OF CRACK REGION DETECTION AND WIDTH ESTIMATION BY SEMANTIC SEGMENTATION FOR DEVELOPMENT OF INSPECTION WORK SUPPORT SYSTEM

Automatic detection of cracks using deep learning has been studied as one of the measuresto cope with the aging of concrete structures, which is a problem throughout Japan. Research is also being conducted on estimating the width of cracks to reduce the burden on inspectors. In this paper, we examin...

Full description

Saved in:
Bibliographic Details
Published inArtificial Intelligence and Data Science Vol. 3; no. J2; pp. 631 - 641
Main Authors YAMAMOTO, Yohei, HASHIMOTO, Takeshi, TAKEI, Yuma, KIKUCHI, Kohei, HASHMOTO, Tomohiro, YAMAMOTO, Shigehiro, IZAWA, Daisuke, NAKAJIMA, Norito, TAKANO, Yoshiyuki, ABE, Masato, SUGISAKI, Koichi, CHUN, Pang-jo
Format Journal Article
LanguageJapanese
Published Japan Society of Civil Engineers 2022
Online AccessGet full text

Cover

More Information
Summary:Automatic detection of cracks using deep learning has been studied as one of the measuresto cope with the aging of concrete structures, which is a problem throughout Japan. Research is also being conducted on estimating the width of cracks to reduce the burden on inspectors. In this paper, we examine the accuracy of crack detection and width estimation for the development of an inspection support system. Specifically, we applied SegFormer, which has recently attracted attention for its effectiveness in image recognition using Transformer, to crack area detection and compared its accuracy with that of DeepLab v3+, showing that SegFormer was effective in certain areas. For width estimation, we improved the method of width estimation from binary images of cracks detected by region detection, and compared its accuracy with that of existing methods using sinusoidal images, showing that the proposed method ismore effective.
ISSN:2435-9262
DOI:10.11532/jsceiii.3.J2_631