Enhancement of Motion Blurred Crack Images Based on Conditional Generative Adversarial Network

Crack is one of the most critical damage types in concrete structures, which requires regular inspection to ensure structural safety. Unmanned aerial vehicles (UAVs) have made it easier to acquire crack images of high-rise concrete structures, and computer vision models can be further trained to detect cracks automatically. However, motion blur of the UAVs may cause image quality degradation, leading to a decrease in crack detectability. To address this issue, we propose a crack image deblurring model based on a conditional generative adversarial network. The model uses U-Net with skip connections as the generator and the PatchGAN model as the discriminator to distinguish between synthetic and real images. To evaluate the performance of our deblurring model, we trained a segmentation model and tested it on both blurred and sharp images. Test results show that our model can effectively restore the global crack structure and edge texture features from blurred images, thereby improving the crack Intersection over Union by 14.98%.