IterLUNet: Deep Learning Architecture for Pixel-Wise Crack Detection in Levee Systems

Deep learning has recently been extensively used for crack detection in structural health monitoring settings. However, detecting cracks levee systems have yet to receive considerable critical attention. Thus, this study presents a novel encoder-decoder-based fully convolutional neural network detect from images at pixel level automatically. We propose that the feature be strengthened using decoder and bottleneck maps by concatenating them back encoder blocks. The addition reinforcement U-Net-like architecture results loop-like structure exploit all encoders, bottlenecks, decoders. proposed architecture, Iterative Loop U-Net (IterLUNet), outperforms state-of-the-art architectures on image dataset of system, achieving an increment Intersection over Union (IoU) 10.32% average 10-Fold Cross-Validation (FCV) compared baseline model 11.00%, 7.65%, 7.43% with range latest models MultiResUnet, Attention U-Net, Unet++ respectively. In addition, IterLUNet least 63% fewer parameters trained than model, thus, allowing less space consumption pixel-wise AI-based inspection systems.