[2K4-ES-2] Machine learning: GAN

[2K4-ES-2-04] Synthetic Image Augmentation for Damage Region Segmentation using Conditional GAN with Structure Edge

〇Takato Yasuno¹, Michihiro Nakajima¹, Seiji Sekiguchi¹, Kazuhiro Noda¹, Kiyoshi Aoyanagi¹, Sakura Kato¹ (1. Yachiyo Engineering, Co.Ltd.)

Keywords:Social infrastructure, Damage inspection, Sparse segmentation, Rare class generation, Conditional GAN

For inspection and diagnosis of social infrastructure, primary damage region are recognized for repair targets. But, the degradation at worse level has rarely occurred, and the damage regions of interest are often narrow. The both scarcity and imbalance property on the damage region of interest influences limited performance to detect damage. If additional dataset of damaged images can be generated, it may enable to improve accuracy in damage region segmentation algorithm. We propose a synthetic augmentation procedure to generate damaged images using the image-to-image translation mapping from the tricategorical label that consists the both semantic label and structure edge to the real damage image. Actually, in case of bridge inspection, we apply the RC concrete structure. We applied popular per-pixel segmentation algorithms such as the FCN-8s, SegNet, and DeepLabv3+Xception-v2. We demonstrates that re-training a dataset added with synthetic augmentation procedure make higher accuracy based on indices the mean IoU, damage region of interest IoU, precision, recall, BF score.