<p>Single image dehazing is a critical task in computer vision. However, existing methods struggle with handling non-uniform haze distribution, capturing global contextual information, and filtering noise while preserving fine details. To overcome these limitations, we propose a Transformer-based dehazing network enhanced by attention and frequency domain techniques. A dual-path feature encoder is designed, incorporating an Optimized Self-Attention Transformer Block (OSATB) to improve global dependency modeling and a Feature Refinement Extraction Block (FREB) to extract fine texture details using attention mechanisms. Furthermore, a Feature Attention Fusion Block (FAFB) adaptively assigns weights to feature maps, facilitating the integration of global and local information. A Dual-Frequency Enhancement Block (DFEB) further refines pixel-level details and suppresses noise in the frequency domain. Finally, the decoder reconstructs a clear image. To comprehensively evaluate the performance of the proposed method, we conducted a systematic training and evaluation process on multiple large-scale synthetic and real-world hazy image datasets, including RESIDE and other widely used benchmarks. Experimental results demonstrate that the proposed method achieves superior performance on both synthetic and real-world hazy image datasets, effectively addressing the challenges mentioned above.</p>