Rain streaks degrade the quality of optical images and seriously affect the effectiveness of subsequent vision-based algorithms. Although the applications of a convolutional neural network (CNN) and self-attention mechanism (SA) in single image deraining have shown great success, there are still unresolved issues regarding the deraining performance and the large computational load. The work in this paper fully coordinates and utilizes the advantages between CNN and SA and proposes a hybrid content-aware deraining network (CAD) to reduce complexity and generate high-quality results. Specifically, we construct the CADBlock, including the content-aware convolution and attention mixer module (CAMM) and the multi-scale double-gated feed-forward module (MDFM). In CAMM, the attention mechanism is used for intricate windows to generate abundant features and simple convolution is used for plain windows to reduce computational costs. In MDFM, multi-scale spatial features are double-gated fused to preserve local detail features and enhance image restoration capabilities. Furthermore, a four-token contextual attention module (FTCA) is introduced to explore the content information among neighbor keys to improve the representation ability. Both qualitative and quantitative validations on synthetic and real-world rain images demonstrate that the proposed CAD can achieve a competitive deraining performance.

The problem of image deraining has drawn widespread attention as the blurry images caused by rain streaks can severely impair the practical value of images. Although exiting deep learning-based methods have achieved significant success, single-image deraining still faces great challenges due to the diverse orientations, multi-scale distributions and intense texture entanglement of rain streaks. Specifically, existing state-space-based deraining methods still suffer from relatively uniform local enhancement across different feature hierarchies, representation bias caused by fixed single-direction scanning in two-dimensional (2D) images, and insufficient selection of cross-level information during feature fusion. In this paper, a novel deraining network named HLDA-Net is proposed to address these issues, which unifies hierarchical local enhancement (HLE) and direction-aware dual-branch state-space module (DDS) within a Convolutional Neural Network (CNN) encoder–decoder framework. Level-adaptive detail refinement is embedded in HLE for better background texture restoration, while complementary dual-branch spatial modeling is introduced in DDS to mitigate the directional bias of single-path scanning and strengthen long-distance dependency modeling. In addition, gated skip fusion (GSF) is integrated into the encoder–decoder restoration pipeline to strengthen cross-level feature aggregation and optimize stability. Extensive experiments on synthetic and real-world benchmarks demonstrate the effectiveness of HLDA-Net for single-image deraining, with the peak signal-to-noise ratio (PSNR) and Structural Similarity Index Measure (SSIM) on the Rain100H dataset reaching 31.034 and 0.914, respectively.