融合多尺度残差与注意力机制的图像去雾算法

Image Dehazing Algorithm Integrating Multiscale Residual and Attention Mechanism

摘要

针对现有深度学习去雾算法因编码器与解码器间跨层特征失配，导致细节丢失与对比度下降的问题，提出一种融合多尺度残差与注意力机制的图像去雾算法。首先，设计多尺度残差感知下采样模块，通过多分支并行卷积融合残差连接与高效通道注意力机制，以增强细节特征提取能力；其次，引入自适应细粒度通道注意力机制，动态调整全局与局部信息的特征权重，从而有效抑制对比度下降。在 RESIDE−6K、SOTS 和 HSTS 数据集上的实验结果表明：本文算法在主观视觉质量及峰值信噪比 (PSNR)、结构相似性指数 (SSIM) 等客观指标上均优于 6 种主流对比算法，展现出较强的泛化能力。其中，在 RESIDE−6K 测试集上，PSNR 与 SSIM 分别达到 28.51 dB，0.965 1；在 SOTS 测试集上分别达到 27.82 dB，0.965 5；在 HSTS 测试集上分别达到 29.30 dB，0.960 1。本研究通过多尺度残差与注意力机制的有效协同，实现了细节保留与对比度恢复的双重优化，为有雾图像质量增强提供了有效的技术方案。

Abstract

To address the loss of detail and reduced contrast in foggy images caused by cross-layer feature mismatch between encoders and decoders in existing deep learning algorithms, an image defogging algorithm that integrates multi-scale residuals with attention mechanisms was proposed. Firstly, a multi-scale residual-aware subsampling module was designed at the encoder side. which enhanced detailed feature extraction capability through multi-branch parallel convolutions combined with residual connections and efficient channel attention mechanisms. Secondly, an adaptive fine-grained channel attention mechanism was introduced to dynamically adjust feature weights between global and local information, thereby suppressing contrast degradation. Experimental results on the RESIDE–6K, SOTS, and HSTS datasets demonstrate that the proposed algorithm outperforms six mainstream comparative algorithms in both subjective visual effects and objective metrics such as peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM), while exhibiting strong generalization capability. Specifically, PSNR and SSIM values of 28.51 dB and 0.965 1 are achieved on the RESIDE−6K test set, 27.82 dB and 0.965 5 on the SOTS test set, and 29.30 dB and 0.960 1 on the HSTS test set. Through the effective integration of multi-scale residual and attention mechanisms, dual optimization of detail preservation and contrast recovery is realized, providing an effective technical solution for enhancing the quality of hazy images.

关键词

图像去雾 / 深度学习 / 注意力机制 / 残差连接 / 多尺度特征融合 / 自适应权重

Key words

image dehazing / deep learning / attention mechanism / residual connection / multi-scale feature fusion / adaptive weight

参考文献

[1] KEHAR V, CHOPRA V, SINGH B K, et al. Efficient single image dehazing model using metaheuristics—based brightness channel prior[J]. Mathematical Problems in Engineering, 2021, 2021(1): 5584464.

[2] 岳永恒, 雷文朋. 基于改进的 CycleGAN 和 YOLOv8 联合雾天道路环境感知算法[J]. 华南理工大学学报 (自然科学版), 2025, 53(2): 48-57.

[3] YUE Y H, LEI W P. Foggy road environment perception algorithm based on an improved CycleGAN and YOLOv8[J]. Journal of South China University of Technology (Natural Science Edition), 2025, 53(2): 48-57.

[4] REN W Q, ZHANG J G, XU X Y, et al. Deep video dehazing with semantic segmentation[J]. IEEE Transactions on Image Processing, 2019, 28(4): 1895-1908.

[5] 董静薇, 赵春丽, 海博. 融合同态滤波和小波变换的图像去雾算法研究[J]. 哈尔滨理工大学学报 , 2019, 24(1): 66-70.

[6] DONG J W, ZHAO C L, HAI B. Research on image de—fog algorithm based on fusion homomorphic filtering and wavelet transform[J]. Journal of Harbin University of Science and Technology, 2019, 24(1): 66-70.

[7] SEOW M J, ASARI V K. Ratio rule and homomorphic filter for enhancement of digital colour image[J]. Neurocomputing, 2006, 69(7/8/9): 954-958.

[8] HE K M, SUN J, TANG X O. Single image haze removal using dark channel prior[C]// 2009 IEEE Conference on Computer Vision and Pattern Recognition. Miami, FL, USA: PIEEE, 2009: 1956-1963.

[9] ZHU Q S, MAI J M, SHAO L. A fast single image haze removal algorithm using color attenuation prior[J]. IEEE Transactions on Image Processing, 2015, 24(11): 3522-3533.

[10] LIANG J Y, CAO J Z, SUN G L, et al. SwinIR: image restoration using swin transformer[C]// 2021 IEEE/CVF International Conference on Computer Vision Workshops. Montreal, BC, Canada: IEEE, 2021: 1833-1844.

[11] LIU Z, LIN Y T, CAO Y, et al. Swin transformer: hierarchical vision transformer using shifted windows[C]// 2021 IEEE/CVF International Conference on Computer Vision. Montreal, QC, Canada: IEEE, 2021: 9992-10002.

[12] 王浩宇. 面向智能交通监控的无人机视觉感知关键技术研究[D]. 武汉: 华中科技大学, 2024.

[13] WANG H Y. Key Technologies of UAV Visual Perception for Intelligent Traffic Surveillance[D]. Wuhan: Huazhong University of Science and Technology, 2024.

[14] CAI B L, XU X M, JIA K, et al. DehazeNet: an end—to—end system for single image haze removal[J]. IEEE Transactions on Image Processing, 2016, 25(11): 5187-5198.

[15] LI B Y, PENG X L, WANG Z Y, et al. AOD−Net: all—in—one dehazing network[C]// 2017 IEEE International Conference on Computer Vision. Venice, Italy: IEEE, 2017: 4780-4788.

[16] CHEN D D, HE M M, FAN Q N, et al. Gated context aggregation network for image dehazing and deraining[C]// 2019 IEEE Winter Conference on Applications of Computer Vision. Waikoloa Village, HI, USA: IEEE, 2019: 1375-1383.

[17] QIN X, WANG Z L, BAI Y C, et al. FFA−Net: feature fusion attention network for single image dehazing[J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2020, 34(7): 11908-11915.

[18] YANG H H, YANG C H, JAMES TSAI Y C. Y−Net: multi—scale feature aggregation network with wavelet structure similarity loss function for single image dehazing[C]// ICASSP 2020—2020 IEEE International Conference on Acoustics, Speech and Signal Processing. Barcelona, Spain: IEEE, 2020: 2628-2632.

[19] SUN H, WEN Y, FENG H J, et al. Unsupervised bidirectional contrastive reconstruction and adaptive fine—grained channel attention networks for image dehazing[J]. Neural Networks, 2024, 176: 106314.

[20] 董倩瑜, 杨秋翔, 赵印. 基于 U−Net 的多尺度特征融合去雾网络[J]. 激光与光电子学进展, 2025, 62(6): 450-460.

[21] DONG Q Y, YANG Q X, ZHAO Y. Multi—scale feature fusion dehazing network based on U−Net[J]. Laser & Optoelectronics Progress, 2025, 62(6): 450-460.

[22] ZHOU H, WANG Y L, ZHANG Q, et al. A dul—stage residual diffusion model with perceptual decoding for remote sensing image dehazing[J]. IEEE Transactions on Geoscience and Remote Sensing, 2025, 63: 109312.

[23] WANG Q L, WU B G, ZHU P F, et al. ECA−Net: efficient channel attention for deep convolutional neural networks[C]// 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Seattle, WA, USA: IEEE, 2020: 11531-11539.

[24] 翟凤文, 朱玉彤, 金静. 双分支卷积结合细节增强的图像去雾[J]. 西北工业大学学报, 2025, 43(1): 109-118.

[25] ZHAI F W, ZHU Y T, JIN J. Image dehazing based on double branch convolution and detail enhancement[J]. Journal of Northwestern Polytechnical University, 2025, 43(1): 109-118.

[26] LU L P, XIONG Q, XU B R, et al. MixDehazeNet: mix structure block for image dehazing network[C]// 2024 International Joint Conference on Neural Networks. Yokohama, Japan: IEEE, 2024: 1-10.

[27] WANG J B, LU K, XUE J, et al. Single image dehazing based on the physical model and MSRCR algorithm[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2017, 28(9): 2190-2199.

[28] WANG Y Z, YAN X F, WANG F, et al. UCL−Dehaze: toward real—world image dehazing via unsupervised contrastive learning[J]. IEEE Transactions on Image Processing, 2024, 33: 1361-1374.

[29] 金海波, 马琳琳, 田桂源. 自适应 Transformer 网络下的单幅图像去雾方法[J]. 计算机工程与应用 , 2024, 60(3): 237-245.

[30] JIN H B, MA L L, TIAN G Y. Single image defogging method under adaptive transformer network[J]. Computer Engineering and Applications, 2024, 60(3): 237-245.

[31] LI B Y, REN W Q, FU D P, et al. Benchmarking single image dehazing and beyond[J]. IEEE Transactions on Image Processing, 2019, 28(1): 492-505.

基金资助

国家自然科学基金项目(51574004)