Crackwave R-convolutional neural network: A discrete wavelet transform and deep learning fusion model for underwater dam crack detection

Crack detection is an essential part of structural health monitoring (SHM) for underwater dams, which is crucial for preventing potential structural failures and ensuring the long-term stability. Deep learning-based image processing algorithms have become a research hotspot in the field of crack detection. However, the complex underwater environment has posed challenges to underwater dam crack detection. To address these issues, we propose CrackWave R-convolutional neural network (CW R-CNN), a novel underwater dam crack detection model that fuses discrete wavelet transform (DWT) and deep learning. The proposed model introduces a novel backbone network, DwtResNet, which incorporates DWT to comprehensively extract frequency-domain features from underwater crack images. To overcome the limitations of Intersection over Union (IoU), particularly when predicted and ground truth bounding boxes do not overlap, we employ the generalized IoU (GIoU) function. Furthermore, we apply the soft nonmaximum suppression (NMS) algorithm to reduce the risk of missing fine cracks. In addition, we utilized a self-developed underwater dam image acquisition robot to capture a large number of underwater dam crack images, forming the self-acquired dataset. Evaluating the proposed model on this dataset showed that its MAP_0.5 outperformed SSD, YOLOv5, and the conventional Faster R-CNN. The proposed model proved more effective than other models, especially in detecting fine cracks and handling complex backgrounds. These experimental results not only demonstrate the effectiveness of CW R-CNN in underwater dam crack detection but also highlight its potential application in SHM. It provides essential technical support for the safe monitoring of underwater dam structures.