Most existing deraining networks can effectively remove rain streaks, but they often ignore the natural light and shadow details in the image. This can cause artifacts or blurry edges around objects in the restored image. To solve this problem, a two-stage linear iterative deraining and restoration network is proposed. The network has two stages: a rain streak enhancement module and a deraining restoration module. These stages use multiple linear iterations to handle both deraining and restoration tasks together. The rain streak enhancement module uses residual attention to identify local features, adjust details, distinguish rain streaks from the background and refine the image through multiple iterations. The deraining restoration module combines multi-scale parallel depth-wise separable convolutions and hybrid attention mechanism to understand features at different scales and restore image details accurately. Moreover, an enhanced edge loss is introduced to improve and assist in restoring edge details. The experimental results show that this method achieves good performance on the Rain100L, Rain100H and real-world datasets. It not only effectively removes rain but also restores image lighting details and edge details with high accuracy. It also performs well in the simulated restoration of natural damage in Thangka and murals, which demonstrates the proposed network model’s strong generalization ability.