[3Rin4] Interactive 1

[3Rin4-15] Verification of generalization capability of concrete crack detection corresponding to background differences

〇Junichiro Fujii¹, Ryuto Yoshida¹, Masazumi Amakata¹ (1.Yachiyo Engineering Co., Ltd.)

Keywords:deep learning, semantic segmentation, crack detection, generalization capability

A large number of concrete revetments are installed in urban rivers, and its efficient management is required. Therefore, a deep learning model for detecting cracks in concrete revetments by image recognition has been proposed. However, there are various types of river revetment depending on the block loading method and block design.
In this study, we focused on the feature of the concrete block and compare the accuracy of crack detection by the combination of block pattern for the purpose of improving the capability of crack detection for unknown block pattern. Based on the result, we propose the dataset necessary to implement a crack detection model applicable for multiple block patterns.