Underwater bridge pier structures are frequently exposed to stressors like water flow scour, ship collisions, and wave forces, leading to defects such as cracks, exposed rebar, and holes. Timely detection of these defects is critical to preventing catastrophic failures. Recent advancements in sonar imaging have improved underwater inspections, but the resulting images often suffer from noise and lower quality compared to optical images, complicating defect detection. Additionally, the scarcity of publicly available sonar datasets presents challenges in training accurate detection models.This paper addresses these challenges by proposing a novel data augmentation method using CycleGAN, which converts optical images of underwater piers into sonar images. This approach leverages the power of Generative Adversarial Networks (GANs) for effective sample augmentation, overcoming the limitations of traditional data augmentation techniques. This method significantly improves the robustness and accuracy of deep learning models for detecting defects in underwater bridge piers, providing a novel solution for augmenting limited sonar image datasets and advancing automated defect detection.