<p>Image dehazing is a fundamental challenging task in computer vision, aiming to restore clear and realistic scene information from hazy images. However, existing methods often overlook local details and fail to fully exploit multi-layer features, leading to restored images with color distortion and residual haze. To address these issues, we propose an enhanced Stacked Multi-input Multi-output Attentional Feature Fusion Network (SMIMO-AFFN) for single image dehazing. We first introduce a novel Attentional Feature Fusion Module (AFFM) to enhance MIMO-UNet, resulting in MIMO-AFFN. The AFFM can adaptively fuse features using channel and spatial attention mechanisms. This fusion highlights important information and effectively integrates features of inconsistent semantics and scales. Then, we employ a stacking architecture to construct the final SMIMO-AFFN model, which improves the model’s capacity to learn complex representations without significantly increasing the number of parameters. Extensive experiments on the RESIDE dataset demonstrate the excellent performance of the AFFM and the efficacy of stacking architecture. Additionally, we evaluate the impact of our method on downstream tasks using the YOLOv5 object detection model. The results demonstrate that our approach outperforms other image dehazing methods in terms of PSNR, SSIM, QSSIM, and visual quality, effectively addressing color distortion and residual haze, thus confirming its superior performance in the image dehazing task.</p>