Hazy images present a challenging ill-posed problem, suffering from information loss and color distortion. Current deep learning-based dehazing methods enhance performance by increasing network depth but incur substantial parameter overhead. Meanwhile, standard convolutional layers concentrate on low-frequency details, often overlooking high-frequency information, which hinders the effective utilization of prior information presented in blurred images. In this paper, we propose TCL-Net, a lightweight dehazing network which emphasizes on frequency-domain features. Our network first includes a sophisticated layer for extracting high-frequency and low-frequency information, specifically designed using Fast Vision Transformers for the original blurred images. Concurrently, we have designed a frequency-domain information fusion module that integrates high-frequency and low-frequency information with the characteristics of convolutional networks for subsequent convolutional layers. Furthermore, to better leverage spatial information of the original image, we introduce a multi-angle attention module. With the aforementioned design, our network achieves superior performance with a total parameter size of only 0.48 MB, representing an order of magnitude reduction in parameters compared to other state-of-the-art lightweight networks.