This research presents a single-image dehazing method that integrates a Lightweight Vision Transformer (LVT) and U-Net to capture both local and global features. LVT enhances resolution, U-Net extracts local features, and LVT refines global dependencies before fusion. Evaluations on O-Haze and HSTS datasets show PSNR scores of 27.88 (O-Haze, ResNet-50) and 28.22 (HSTS, no backbone), outperforming existing methods while maintaining competitive SSIM. The results demonstrate effectiveness in real-world haze scenarios, such as wildfire-induced haze in Indonesia.