[CS15-08] Conclete crack segmentation accuracy enhanced by synthetic image augmentation
Keywords:Semantic segmentation, Bridge, Diagnosis, Concrete crack, Conditional GAN, Synthetic data augmentation
In bridge maintenance, it is important to inspect condition and to make accurate diagnosis and judge the urgent repairs. In order to inspect every five years, the use of segmentation algorithms is expected for monitoring deterioration efficiently.
This paper aime to improve accuracy of crack segmentation algorithms for automatic crack detection on concrete bridges. We generate synthetic crack images using Conditional GAN and create the augmented data space of raw images and labels to enhance crack feature of interest. We demonstrate that the accuracy indices are improved on four architectures.
This paper aime to improve accuracy of crack segmentation algorithms for automatic crack detection on concrete bridges. We generate synthetic crack images using Conditional GAN and create the augmented data space of raw images and labels to enhance crack feature of interest. We demonstrate that the accuracy indices are improved on four architectures.
Please log in with your participant account.
» Participant Log In