<p>Haze in remote sensing imagery diminishes contrast and obscures details, adversely affecting downstream vision tasks. Existing convolutional neural network (CNN) and transformer hybrids often suffer from frequency aliasing and inadequate decoupling of spectral components. To address this, we propose the multi-scale wavelet transformer network (MWT-Net), which performs orthogonal Haar wavelet decomposition in the feature domain to separate high- and low-frequency components. The high-frequency branch employs a lightweight CNN for detail recovery, while the low-frequency branch utilizes a swin transformer for global structure modeling. Additionally, we introduce an unmanned aerial vehicle (UAV)-based remote sensing dehazing dataset with multi-level synthetic haze. Experiments demonstrate that MWT-Net achieves state-of-the-art performance on both synthetic and real-world benchmarks, with superior detail preservation and structural consistency.</p>