The image deblurring algorithms are commonly used for restoring blurred images. Traditional image deblurring algorithms complete image deblurring tasks based on prior knowledge of the image, but the model often generates incomplete detail information and unclear image edge contours. This work studies an image deblurring method based on feature pyramid construction of generative adversarial networks. Using a feature extraction network, the features are fed into a pyramid network to construct a generator, which can effectively focus on key information, integrate low-level and high-level information, and then extract features from multiple scales. In addition, a dual scale discriminator network is used to provide more accurate gradient feedback on local information and a complex backbone network is used to address image deblurring tasks.