In autonomous driving systems and related fields, foggy weather has a significant impact on the imaging capabilities of cameras and other sensors, often manifesting itself as obstruction in the near field and blurring in the far field. This paper proposes an image defogging algorithm (DCGNet) based on an enhanced convolutional block attention module and a dual-scale gated feedforward network. The algorithm adopts the U-Net architecture as the backbone network, combining the enhanced convolutional block attention module (Enhanced CBAM), the dual scale gated feedforward network (DGFF) and residual connections to effectively enhance the image defogging performance. The enhanced CBAM module integrates three channels’ attention mechanisms, maximum pooling, average pooling, and variance pooling, and employs multiscale spatial attention to capture feature information at different scales. The DGFF achieves multiscale feature extraction through parallel deep convolution and dilated convolution, and adaptively fuses features of different scales via a gating mechanism. Experimental results demonstrate that the proposed DCGNet achieves outstanding defogging performance on the Haze4K dataset, outperforming existing mainstream defogging algorithms in evaluation metrics such as PSNR and SSIM.