基于高阶递归网络的单幅图像去雨滴模型

doi:10.12305/j.issn.1001-506X.2026.01.02

摘要/Abstract

摘要：

目前单幅图像去雨滴模型提取大尺度雨滴特征的能力较差，导致精度不高，无法很好地应用在复杂多变的实际场景中。为此，提出一种基于高阶递归网络的单幅图像去雨滴模型。首先，利用结合注意力机制的分组卷积构建一种双尺度注意力残差模块，更好地提取大尺度雨滴的有效特征。其次，设计一种高阶递归特征传递机制，有效强化了这些特征从局部到整体的传递作用。最后，提出一种双尺度残差门控循环单元，建立了对递归计算中逐阶段特征的反馈过程，进一步提高了模型的性能。实验结果表明，提出的高阶递归网络在公开的基准数据集上取得了当前最优的性能表现，较好解决了当前算法精度不足的问题。

关键词: 高阶递归, 深度学习, 单幅图像去雨滴, 双尺度残差, 分组卷积, 门控循环单元

Abstract:

At present, the ability of extracting large-scale raindrop features from single image raindrop models is poor, resulting in low accuracy and inability to be well applied in complex and varied practical scenes. Therefore, a single image raindrop removal model based on high-order recursive network is proposed. First, a dual-scale attention residual module by using group convolution combined with attention mechanism is constructed. It better extracts effective features of large-scale raindrops. Second, a high-order recursive feature transfer mechanism is designed to effectively enhance the transfer process of these features from local to global. Last, a dual-scale residual gated recurrent unit is proposed, which establishes feedback process on the features of each stage in the recursive calculation, further improving the model’s performance. The experimental results show that the proposed high-order recursive network has achieved the best performance on publicly available benchmark datasets, effectively solving the problem of insufficient accuracy in current algorithms.

Key words: high-order recursive, deep learning, single image raindrop removal, dual-scale residual, group convolution, gated recurrent unit

中图分类号:

TP 391.4

包玉刚, 贾皓翔, 赵旦峰. 基于高阶递归网络的单幅图像去雨滴模型[J]. 系统工程与电子技术, 2026, 48(1): 12-21.

Yugang BAO, Haoxiang JIA, Danfeng ZHAO. Single image raindrop removal model based on high-order recursive network[J]. Systems Engineering and Electronics, 2026, 48(1): 12-21.

图/表 11

图1

图2

图3

图4

表1

表2

表3

表4

图5

图6

图7

参考文献 33

1	陈峥, 毕晓君. 基于轻量级神经网络的单幅图像去雨滴模型[J]. 哈尔滨工程大学学报, 2023, 44 (2): 292- 299.
	CHEN Z, BI X J. A single image raindrop removal model based on lightweight neural network[J]. Journal of Harbin Engineering University, 2023, 44 (2): 292- 299.
2	WANG J F, CHEN Y, DONG Z K, et al. Improved YOLOv5 network for real-time multi-scale traffic sign detection[J]. Neural Computing and Applications, 2023, 35 (10): 7853- 7865. doi: 10.1007/s00521-022-08077-5
3	RUAN W J, CHEN J, WU Y, et al. Multi-correlation filters with triangle-structure constraints for object tracking[J]. IEEE Trans. on Multimedia, 2018, 21 (5): 1122- 1134.
4	CHEN L C, PAPANDREOU G, KOKKINOS I, et al. Deeplab: semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected crfs[J]. IEEE Trans. on Pattern Analysis and Machine Intelligence, 2017, 40 (4): 834- 848.
5	LUO Y, XU Y, JI H. Removing rain from a single image via discriminative sparse coding[C]//Proc. of the IEEE International Conference on Computer Vision, 2015: 3397–3405.
6	YOU S, TAN R, KAWAKAMI R, et al. Adherent raindrop modeling, detection and removal in video[J]. IEEE Trans. on Pattern Analysis and Machine Intelligence, 2016, 38 (9): 1721- 1733. doi: 10.1109/TPAMI.2015.2491937
7	EIGEN D, KRISHNAN D, FERGUS R. Restoring an image taken through a window covered with dirt or rain[C]//Proc. of the IEEE International Conference on Computer Vision, 2013: 633–640.
8	QIAN R, TAN R T, YANG W H, et al. Attentive generative adversarial network for raindrop removal from a single image[C]//Proc. of the IEEE Conference on Computer Vision and Pattern Recognition, 2018: 2482–2491.
9	VALANARASU J M J, YASARLA R, PATEL V M. Transweather: transformer-based restoration of images degraded by adverse weather conditions[C]//Proc. of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022: 2353–2363.
10	TANG H, XU D, SEBE N, et al. Attention-guided generative adversarial networks for unsupervised image-to-image translation[C]//Proc. of the International Joint Conference on Neural Networks, 2019.
11	包玉刚, 贾皓翔, 赵旦峰. 基于判别性特征增强网络的单幅图像去雨模型[EB/OL]. [2025-03-18]. http://kns.cnki.net/kcms/detail/23.1390.U.20241105.1516.004.html.
	BAO Y G, JIA H X, ZHAO D F. A single image de-raining model based on discriminant feature enhancement network[EB/OL]. [2025-03-18] http://kns.cnki.net/kcms/detail/23.1390.U.20241105.1516.004.html.
12	ALLETTO S, CARLIN C, RIGAZIO L, et al. Adherent raindrop removal with self-supervised attention maps and spatio-temporal generative adversarial networks[C]//Proc. of the IEEE/CVF International Conference on Computer Vision Workshops, 2019: 2329–2338.
13	LIU X, SUGANUMA M, SUN Z, et al. Dual residual networks leveraging the potential of paired operations for image restoration[C]//Proc. of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2019: 7007–7016.
14	LI R T, TAN R T, CHEONG L F. All in one bad weather removal using architectural search[C]//Proc. of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020: 3172–3182.
15	ZHANG K H, LI D X, LUO W H, et al. Dual attention-in-attention model for joint rain streak and raindrop removal[J]. IEEE Trans. on Image Processing, 2021, 30, 7608- 7619. doi: 10.1109/TIP.2021.3108019
16	GUO Y, XIAO X Y, WANG X X, et al. A two-stage real image deraining method for gt-rain challenge cvpr 2023 workshop UG2+ Track 3[EB/OL]. [2024-08-05]. https: //arxiv.org/abs/2305.07979.
17	CAI L, FU Y L, HUO W L, et al. Multiscale attentive image de-raining networks via neural architecture search[J]. IEEE Trans. on Circuits and Systems for Video Technology, 2023, 33 (2): 618- 633. doi: 10.1109/TCSVT.2022.3207516
18	阳金霖, 李朝锋. 融合快速傅里叶卷积的域变换图像去雨滴方法[J]. 计算机工程, 2024, 50 (9): 296- 303. doi: 10.19678/j.issn.1000-3428.0068351
	YANG J L, LI C F. Domain transform image raindrop removal method by integrating fast Fourier convolution[J]. Computer Engineering, 2024, 50 (9): 296- 303. doi: 10.19678/j.issn.1000-3428.0068351
19	江炜. 基于机器视觉与深度学习的挡风玻璃雨滴检测方法研究[D]. 盐城: 盐城工学院, 2024.
	JIANG W. Research on raindrop detection on windshields based on machine vision and deep learning[D]. Yancheng: Yancheng Institute of Technology, 2024.
20	文渊博, 高涛, 安毅生, 等. 基于视觉提示学习的天气退化图像恢复[EB/OL]. [2024-08-05]. https://link.cnki.net/urlid/11.1826.tp.20240627.0925.002.
	WEN Y B, GAO T, AN Y S, et al. Weather degraded image restoration based on visual prompt learning[EB/OL]. [2024-08-05]. https://link.cnki.net/urlid/11.1826.tp.20240627.0925.002.
21	HU J, SHEN L, ALBANIE S, et al. Squeeze-and-excitation networks[C]//Proc. of the IEEE Conference on Computer Vision and Pattern Recognition, 2018: 7132–7141.
22	XIE G T, WANG J D, ZHANG T, et al. Interleaved structured sparse convolutional neural networks[C]//Proc. of the IEEE Conference on Computer Vision and Pattern Recognition, 2018: 8847–8856.
23	HE K M, ZHANG X Y, REN S Q, et al. Deep residual learning for image recognition[C]//Proc. of the IEEE Conference on Computer Vision and Pattern Recognition, 2016: 770–778.
24	HUA B S, TRAN M K, YEUNG S K. Pointwise convolutional neural networks[C]//Proc. of the IEEE Conference on Computer Vision and Pattern Recognition, 2018: 984–993.
25	TAN M, LE Q V. Mixconv: mixed depthwise convolutional kernels[EB/OL]. [2024-08-05].https://arxiv.org/abs/1907.09595.
26	HAN K, WANG Y H, TIAN Q, et al. Ghostnet: more features from cheap operations[C]//Proc. of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020: 1580–1589.
27	MA N, ZHANG X, ZHENG H T, et al. Shufflenet v2: practical guidelines for efficient CNN architecture design[C]//Proc. of the European Conference on Computer Vision, 2018: 116–131.
28	LI X, WU J L, LIN Z C, et al. Recurrent squeeze-and-excitation context aggregation net for single image deraining[C]//Proc. of the European Conference on Computer Vision, 2018: 254–269.
29	JIANG K, WANG Z Y, YI P, et al. Multi-scale progressive fusion network for single image deraining[C]//Proc. of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020: 8346–8355.
30	WANG Z, BOIVIK A C, SHEIKH H R, et al. Image quality assessment: from error visibility to structural similarity[J]. IEEE Trans. on Image Processing, 2004, 13 (4): 600- 612. doi: 10.1109/TIP.2003.819861
31	HUYNH T Q, GHANBARI M. Scope of validity of PSNR in image/video quality assessment[J]. Electronics Letters, 2008, 44 (13): 800- 801. doi: 10.1049/el:20080522
32	PASZKE A, GROSS S, MASSA F, et al. Pytorch: an imperative style, high-performance deep learning library[C]// Proc. of the Conference on Neural Information Processing Systems, 2020.
33	KINGMA D P, BA J. Adam: a method for stochastic optimization[EB/OL]. [2024-08-05].https://arxiv.org/abs/1412.6980.

模块结构	Test_A		Test_B
模块结构	PSNR/dB	SSIM	PSNR/dB	SSIM
标准卷积（3,3）	25.11	0.883	22.89	0.785
标准卷积（3,5）	25.33	0.883	22.93	0.783
单尺度组卷积（1,1）	26.51	0.895	23.46	0.793
单尺度组卷积（3,3）	31.11	0.931	25.89	0.820
单尺度组卷积（5,5）	31.16	0.932	25.90	0.822
双尺度组卷积（1,3）	30.70	0.929	25.76	0.819
双尺度组卷积（1,5）	31.25	0.933	26.03	0.823
双尺度组卷积（3,5）	31.41	0.934	26.17	0.824

超参数组合	Test_A		Test_B
超参数组合	PSNR/dB	SSIM	PSNR/dB	SSIM
R₁=1，R₂=1	25.09	0.815	22.39	0.726
R₁=1，R₂=2	21.37	0.642	19.80	0.594
R₁=1，R₂=3	17.94	0.488	17.42	0.477
R₁=2，R₂=1	30.61	0.928	25.92	0.820
R₁=2，R₂=2	31.41	0.934	26.17	0.824
R₁=2，R₂=3	30.70	0.932	25.75	0.822
R₁=3，R₂=1	30.28	0.924	25.73	0.816
R₁=3，R₂=2	30.96	0.928	26.04	0.820
R₁=3，R₂=3	30.78	0.928	25.93	0.821

模型结构	Test_A		Test_B
模型结构	PSNR/dB	SSIM	PSNR/dB	SSIM
无门控循环单元	30.74	0.927	25.72	0.813
局部递归循环单元	31.09	0.932	25.93	0.822
全局递归循环单元	30.86	0.930	25.79	0.819
双尺度残差循环单元	31.41	0.934	26.17	0.824

模型	Test_A		Test_B
模型	PSNR/dB	SSIM	PSNR/dB	SSIM
ATT-GAN^[7]	24.91	0.884	24.92	0.809
TransWeather^[8]	28.83	0.911	—	—
AG-GAN^[10]	25.09	0.822	22.72	0.738
SA-GAN^[12]	25.56	0.847	22.77	0.753
DuRN-S-P^[13]	28.07	0.887	24.65	0.824
All-in-One^[14]	23.41	0.872	—	—
D-DAM^[15]	30.12	0.926	—	—
Two-stage^[16]	30.12	0.910	24.86	0.803
MANAS^[17]	28.88	0.907	23.66	0.787
DTN^[18]	28.28	0.905	24.96	0.809
QPC-EPG^[20]	29.48	0.913	—	—
本文模型	31.41	0.934	26.17	0.824

[1]	陈海青, 汪刘应, 刘顾, 王龙, 葛超群, 陈孟州. 合成孔径雷达图像去噪算法研究进展[J]. 系统工程与电子技术, 2026, 48(1): 94-105.
[2]	李开明, 张袁鹏, 罗迎, 代肖楠. 弹道导弹雷达目标识别研究进展[J]. 系统工程与电子技术, 2025, 47(9): 2870-2889.
[3]	季策, 马相宇, 牟晓宇, 赵家毅. TS-GRU-VTA：基于深度学习的车辆信道估计方案[J]. 系统工程与电子技术, 2025, 47(9): 3093-3098.
[4]	闫小伟, 凌冲, 石胜斌. 地表未爆子弹药快速检测系统设计与实现[J]. 系统工程与电子技术, 2025, 47(8): 2639-2645.
[5]	张琬滢, 高由兵, 李泽一, 李鹏飞, 张伟. 基于双路径特征融合网络的背景信号抑制算法[J]. 系统工程与电子技术, 2025, 47(7): 2406-2413.
[6]	张新征, 闫梦可, 朱晓林. 噪声伪标签容忍的半监督SAR目标识别[J]. 系统工程与电子技术, 2025, 47(6): 1796-1805.
[7]	王彩云, 贾一帆, 李晓飞, 王佳宁, 吴钇达. 基于改进鲸鱼算法和BiGRU的弹道目标HRRP识别方法[J]. 系统工程与电子技术, 2025, 47(6): 1824-1832.
[8]	姜智杰, 宋恒, 胡楠, 段兰茜, 曹平. 隧道环境毫米波雷达目标识别与分类算法[J]. 系统工程与电子技术, 2025, 47(5): 1453-1460.
[9]	聂千祁, 沙明辉, 朱应申, 王崇宇, 崔念强. 基于盲源分离结合奇异谱分析的雷达多分量信号识别方法[J]. 系统工程与电子技术, 2025, 47(4): 1168-1175.
[10]	袁磊, 杨艳娟, 郭毅, 戴鹏. 相关噪声下基于深度学习的LDPC码码率半盲识别算法[J]. 系统工程与电子技术, 2025, 47(4): 1335-1345.
[11]	沈夏闰, 李若楠, 张昊田. 基于CVAE-LSTM的服务器KPI异常检测[J]. 系统工程与电子技术, 2025, 47(3): 1019-1027.
[12]	刘晓琳, 郭梦娇, 李卓. Dueling DQN优化下的航班延误自适应图卷积循环网络预测方法[J]. 系统工程与电子技术, 2025, 47(2): 568-579.
[13]	付卫红, 张鑫钰, 刘乃安. 基于多尺度融合神经网络的同频同调制单通道盲源分离算法[J]. 系统工程与电子技术, 2025, 47(2): 641-649.
[14]	张文中, 侯长波, 赵鹏旗, 刘思成. 基于GCN的无人机目标航迹关联[J]. 系统工程与电子技术, 2025, 47(12): 4005-4011.
[15]	陆恩绩, 汪玲, 朱岱寅, 周晔. 基于轻量化三维时域卷积网络（3D-TCN）的气象雷达雷暴预测方法[J]. 系统工程与电子技术, 2025, 47(12): 4084-4092.