Image dehazing serves as a fundamental low-level computer vision task that aims to recover high-quality images from haze-corrupted observations. This technology plays a critical role in safety-critical applications such as autonomous driving, intelligent transportation systems, and urban security monitoring. The performance of image dehazing directly impacts the perception accuracy and overall safety of vision-based systems. However, owing to the complexity of degradation factors in real-world hazy images and the difficulty in capturing paired hazy-clear data, existing methods face significant challenges in real-world image dehazing. In this paper, we focus on three key aspects to improve adaptability in real-world scenarios. Specifically, (1) to simulate the complex haze degradation process, we design a haze degradation model incorporating multiple scattering effects and diverse degradation factors. (2) To improve frequency-domain distortions, we introduce a Spectral-Spatial Fusion Module into the network, which adaptively processes both spatial and frequency information. (3) To effectively incorporate prior knowledge, we incorporate a Prior-Guided Feedforward Network into the network architecture. Comprehensive experiments show that the proposed method achieves better dehazing performance than other state-of-the-art real-world image dehazing methods. This method can improve the perception accuracy of vision-based systems and enhance their reliability in safety-critical tasks.