Intelligent crack detection in underwater concrete structures: An image enhancement and deep learning fusion-driven approach

The cracks in underwater concrete structures represent a typical form of long-term service damage. Such initial damage is highly susceptible to propagating into severe defects, impairing the durability of the engineering structure. However, the complex underwater environment and imaging conditions often lead to blurred crack images, color distortion and low contrast, which constrain the performance of existing visual inspection methods. Accordingly, this paper proposes a two-stage detection method based on fused image enhancement and an improved YOLOv8 object detection algorithm. First, an underwater concrete crack image enhancement model is constructed by combining an improved CycleGAN network and the Deep Inception and Channel-wise Attention Module (DICAM), effectively improving image quality. Meanwhile, an underwater concrete imaging test platform is established, and a crack image database covering multiple angles and operating conditions is built. Subsequently, an improved YOLOv8 object detection model is proposed, achieving multi-scale feature fusion to enhance the model’s ability to recognize crack features. The proposed image enhancement method significantly improves underwater crack image quality. Compared with the original images, the enhanced images achieve improvements in UCIQE, UIQM and UIConM by 118.35%, 162.17% and 14.93%, respectively. Furthermore, integrating image enhancement with the improved YOLOv8 model for joint detection reduces the model size to 9.44% of the original, increases Precision by 10.12%, and raises detection speed by 20 frames per second. This method reliably detects cracks in underwater concrete structures and provides a theoretical basis for digital damage assessment of such structures, including bridge piers.