<p>Image restoration under adverse weather conditions represents a significant and ongoing research challenge in computer vision. In this paper, we propose an end-to-end image restoration model for image degradation under multiple weather conditions—Multi-scale Aggregated and Residual Image Restoration Network (MARNet). MARNet employs a dense residual structure with multi-scale fusion and consists of three core modules. First, in the encoder module, we utilize a dense connection with multi-scale fusion, incorporating a TransXNet structure that combines convolution and Transformers to ensure the integration of global and local features, thereby enhancing information acquisition efficiency. Second, in the feature extraction module, we use inner and outer layers of densely connected residual blocks for deep feature extraction, leveraging both overall residual connections of the blocks and residual connections within the transformer blocks. Finally, we introduce a pixelshuffle and transpose convolution integration as an upsampling module. This design not only allows for flexible image recovery but also achieves fine-grained restoration. Through extensive training and fine-tuning on multiple datasets, MARNet effectively accomplishes dehazing, deraining, and desnowing for multi-weather image restoration, yielding improved results in testing across various datasets.</p>