To achieve effective restoration of a single turbulence-degraded image, this study presents an image restoration methodology grounded in an enhanced Generative Adversarial Network (GAN). The proposed approach employs a GAN framework that incorporates multi-scale attention feature extraction units and feature dynamic fusion units within the generator. The feature extraction unit combines cascade blocks and content-guided attention modules, enabling the extraction of features across multiple scales while enhancing the efficacy of the feature information after downsampling. Furthermore, the feature dynamic fusion module adaptively enhances the output features from the extraction unit. Simultaneoustly, the discriminator utilizes a dual discriminator architecture that integrates both global and local components, enhancing the network"s capability to restore image details and colors. Experimental results demonstrate that, compared to the CLEAR, BSRGAN, SwinIR, and PiPN methodologies, the enhanced GAN exhibits superior restoration performance for turbulence-affected degraded images.