基于差异特征反投影融合的图像超分辨重建

doi:10.12305/j.issn.1001-506X.2026.06.03

系统工程与电子技术 ›› 2026, Vol. 48 ›› Issue (6): 1809-1818.doi: 10.12305/j.issn.1001-506X.2026.06.03

基于差异特征反投影融合的图像超分辨重建

黄琼丹(), 刘露露(), 韩洁婧(), 王佳鹏(), 康仕林()

西安邮电大学通信与信息工程学院，陕西西安 710121

收稿日期:2025-04-24 修回日期:2025-08-14 接受日期:2025-08-18 出版日期:2026-06-25 发布日期:2025-10-29
通讯作者: 黄琼丹 E-mail:limitless010@163.com;liululu0222@163.com;18729432603@163.com;dmyprincess2022@163.com;15686479558@163.com
作者简介:刘露露（1999—），女，硕士研究生，主要研究方向为信号处理、图像处理
韩洁婧（2001—），女，硕士研究生，主要研究方向为信号处理、视觉场景理解
王佳鹏（2001—），男，硕士研究生，主要研究方向为三维视觉配准、深度学习
康仕林（2002—），男，硕士研究生，主要研究方向为信号处理、人工智能
基金资助:
国家重点研发计划（2022YFB4601700）

Image super-resolution reconstruction based on differential feature back-projection fusion

Qiongdan HUANG(), Lulu LIU(), Jiejing HAN(), Jiapeng WANG(), Shilin KANG()

School of Communication and Information Engineering，Xi’an University of Posts and Telecommunications，Xi’an 710121，China

Received:2025-04-24 Revised:2025-08-14 Accepted:2025-08-18 Online:2026-06-25 Published:2025-10-29
Contact: Qiongdan HUANG E-mail:limitless010@163.com;liululu0222@163.com;18729432603@163.com;dmyprincess2022@163.com;15686479558@163.com

摘要/Abstract

摘要：

基于卷积神经网络的超分辨算法，难以充分学习低分辨率到高分辨率图像的复杂映射，导致重建图像精度低、细节不保真，提出一种基于差异特征反投影融合（difference feature back-projection fusion，DFBPF）的图像超分辨重建算法。该算法通过DFBPF模块，在迭代上下投影过程中融合原始特征与分支间差异特征，利用误差反馈引导重建优化。同时，通过通道交互型全局上下文模块，增强网络的全局理解能力。在各测试集上，峰值信噪比/结构相似度取得了更好的结果，缩放因子为×2时依次为38.18 dB/0.961 2、33.89 dB/0.920 5、32.30 dB/0.901 5、32.80 dB/0.934 2、39.14 dB/0.978 6。实验结果表明，所提算法重建的图像边缘更清晰、细节更丰富。

关键词: 卷积神经网络, 图像超分辨, 注意力机制, 特征融合

Abstract:

The super-resolution algorithm based on convolutional neural networks is difficult to fully learn the complex mapping from low resolution to high-resolution images, resulting in low reconstruction accuracy and inaccurate details. Therefore, a super-resolution reconstruction algorithm based on difference feature back projection fusion （DFBPF） is proposed. This algorithm utilizes a DFBPF module to integrate original features and branch differential features during iterative up-and-down projection processes, leveraging error feedback to guide reconstruction optimization. Additionally, a channel-interactive global context module is employed to enhance the network’s global understanding capability. On various test datasets, the algorithm achieves better results in improved peak signal-to-noise ratio and structural similarity index metrics （SSIM）. For a scaling factor of 2, the results are 38.18 dB/0.961 2, 33.89 dB/0.920 5, 32.30 dB/0.901 5, 32.80 dB/0.934 2 and 39.14 dB/0.978 6 respectively. Experimental results demonstrate that the proposed algorithm produces reconstructed images with sharper edges and richer details.

Key words: convolutional neural network （CNN）, image super-resolution, attention mechanism, feature fusion

中图分类号:

TN 911.73

黄琼丹, 刘露露, 韩洁婧, 王佳鹏, 康仕林. 基于差异特征反投影融合的图像超分辨重建[J]. 系统工程与电子技术, 2026, 48(6): 1809-1818.

Qiongdan HUANG, Lulu LIU, Jiejing HAN, Jiapeng WANG, Shilin KANG. Image super-resolution reconstruction based on differential feature back-projection fusion[J]. Systems Engineering and Electronics, 2026, 48(6): 1809-1818.

图/表 17

图1

图2

图3

图4

图5

图6

图7

表1

表2

表3

表4

表5

表6

图8

图9

图10

图11

参考文献 35

1	SU H, LI Y, XU Y F, et al. A review of deep-learning-based super-resolution: from methods to applications[J]. Pattern Recognition, 2025, 157 (1): 110935.
2	KARWOWSKA K, WIERZBICKI D. MCWESRGAN: improving enhanced super-resolution generative adversarial network for satellite images[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2023, 16, 9459- 9479.
3	ALBARAKATI H M, UR REHMAN S, KHAN M A, et al. A unified super-resolution framework of remote-sensing satellite images classification based on information fusion of novel deep convolutional neural network architectures[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2024, 17, 14421- 14436.
4	WANG C, SUN W J. Semantic guided large scale factor remote sensing image super-resolution with generative diffusion prior[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2025, 220, 125- 138.
5	WANG J R, LU Y T, WANG S Z, et al. Two-stage spatial-frequency joint learning for large-factor remote sensing image super-resolution[J]. IEEE Trans. on Geoscience and Remote Sensing, 2024, 62, 5606813.
6	朱福珍, 邹丹妮, 王志芳, 等. 双参数Beta过程联合字典遥感图像超分辨[J]. 系统工程与电子技术, 2019, 41 (9): 1955- 1960.
	ZHU F J, ZOU D N, WANG Z F, et al. Remote sensing image super-resolution based on double parameters Beta process joint dictionary[J]. Systems Engineering and Electronics, 2019, 41 (9): 1955- 1960.
7	BECK L M, SHOCHER A, ROSSMAN U, et al. Improving correlation based super-resolution microscopy images through image fusion by self-supervised deep learning[J]. Optics Express, 2024, 32 (16): 28195- 28205. doi: 10.1364/OE.521577
8	QU Y L, ZHAO S Q, HUANG Y Y, et al. Self-inspired learning for denoising live-cell super-resolution microscopy[J]. Nature Methods, 2024, 21 (10): 1895- 1908. doi: 10.1038/s41592-024-02400-9
9	DONG C, LOY C C, HE K, et al. Learning a deep convolutional network for image super-resolution[C]//Proc. of 13th European Conference on Computer Vision, 2014: 184−199.
10	DONG C, LOY C C, TANG X. Accelerating the super-resolution convolutional neural network[C]//Proc. of the 14th European Conference on Computer Vision, 2016: 391−407.
11	KIM J, LEE J K, LEE K M. Accurate image super-resolution using very deep convolutional networks[C]//Proc. of the IEEE Conference on Computer Vision and Pattern Recognition, 2016: 1646−1654.
12	ZHANG Y L, TIAN Y P, KONG Y, et al. Residual dense network for image super-resolution[C]//Proc. of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2018: 2472−2481.
13	HUANG Y F, LI J, HU Y T, et al. Deep convolution modulation for image super-resolution[J]. IEEE Trans. on Circuits and Systems for Video Technology, 2023, 34 (5): 3647- 3662.
14	LAI W S, HUANG J B, AHUJA N, et al. Deep Laplacian pyramid networks for fast and accurate super-resolution[C]//Proc. of the IEEE Conference on Computer Vision and Pattern Recognition, 2017: 5835−5843.
15	LIM B, SON S H, KIM H W, et al. Enhanced deep residual networks for single image super-resolution[C]//Proc. of the IEEE Conference on Computer Vision and Pattern Recognition Workshops, 2017: 136−144.
16	TIAN C W, ZHANG Y N, ZUO W M, et al. A heterogeneous group CNN for image super-resolution[J]. IEEE Trans. on Neural Networks and Learning Systems, 2024, 35 (5): 6507- 6519. doi: 10.1109/TNNLS.2022.3210433
17	ZHANG Y L, LI K P, LI K, et al. Image super-resolution using very deep residual channel attention networks[C]//Proc. of the 15th European Conference on Computer Vision, 2018: 286−301.
18	陈一鸣, 周登文. 基于自适应级联的注意力网络的超分辨率重建[J]. 自动化学报, 2022, 48 (8): 1950- 1960.
	CHEN Y M, ZHOU D W. Adaptive attention network for image super-resolution[J]. Acta Automatica Sinica., 2022, 48 (8): 1950- 1960.
19	DAI T, CAI J Y, ZHANG Y B, et al. Second-order attention network for single image super-resolution[C]//Proc. of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2019: 11065−11074.
20	CHEN X Y, WANG X T, ZHOU J T, et al. Activating more pixels in image super-resolution transformer[C]//Proc. of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023: 22367−22377.
21	WANG Y, LI Y S, WANG G, et al. Multi-scale attention network for single image super-resolution[C]//Proc. of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024: 5950−5960.
22	CAO Y, XU J R, LIN S, et al. Gcnet: non-local networks meet squeeze-excitation networks and beyond[C]//Proc. of the IEEE/CVF International Conference on Computer Vision Workshops, 2019: 1971−1980.
23	IRANI M, PELEG S. Improving resolution by image registration[J]. CVGIP: Graphical Models and Image Processing, 1991, 53 (3): 231- 239. doi: 10.1016/1049-9652(91)90045-L
24	HARIS M, SHAKHNAROVICH G, UKITA N. Deep back-projection networks for super-resolution[C]//Proc. of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2018: 1664−1673.
25	LI Z, YANG J L, LIU Z, et al. Feedback network for image super-resolution[C]//Proc. of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2019: 3862−3871.
26	MEI Y Q, FAN Y C, ZHOU Y Q, et al. Image super-resolution with cross-scale non-local attention and exhaustive self-exemplars mining[C]//Proc. of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020: 5689−5698.
27	WOO S, PARK J, LEE J Y, et al. CBAM: Convolutional block attention module[C]//Proc. of the 15th European Conference on Computer Vision, 2018: 3−19.
28	WANG Q L, WU B G, P. ZHU P F, et al. ECA-Net: efficient channel attention for deep convolutional neural networks[C]//Proc. of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020: 11531−11539.
29	AGUSTSSON E, TIMOFTE R. Ntire 2017 challenge on single image super-resolution: dataset and study[C]//Proc. of the 30th IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, 2017: 1122−1131.
30	WANG X L, GIRSHICK R, GUPTA A, et al. Non-local neural networks[C]//Proc. of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2018: 7794−7803.
31	HU J, SHEN L, SUN G. Squeeze-and-excitation networks[C]//Proc. of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2018: 7132−7141.
32	AHN N, KANG B, SOHN K A. Fast, accurate, and lightweight super-resolution with cascading residual network[C]//Proc. of the European conference on computer vision, 2018: 252-268.
33	LI J C, FANG F M, MEI K F, et al. Multi-scale residual network for image super-resolution[C]//Proc. of the 15th European Conference on Computer Vision, 2018: 527−542.
34	FANG F F, LI J C, ZENG T Y. Soft-edge assisted network for single image super-resolution[J]. IEEE Trans. on Image Processing, 2020, 29 (1): 4656- 4668.
35	WANG Z X, GAO G W, LI J C, et al. Lightweight feature de-redundancy and self-calibration network for efficient image super-resolution[J]. ACM Transaction on Multimedia Computing Communications and Applications, 2023, 19 (3): 110.

F_L	F_up1	F_up2	PSNR/dB
−	−	−	37.78
√	−	−	37.84
−	√	−	37.90
−	−	√	37.88
√	√	−	38.02
√	−	√	38.01
−	√	√	38.04
√	√	√	38.11

注意力模块	参数量/×10⁶	推理时间/S	PSNR/dB	SSIM
GC	4.72	0.91	29.18	0.806 4
ECA	4.61	0.73	29.10	0.804 2
NL	5.17	1.25	29.25	0.808 5
SE	4.62	0.76	29.12	0.803 9
CIGC	4.61	0.77	29.23	0.808 7

模块数	参数量×10⁶	Set5		Set14		BSD100		Urban100		Manga109
模块数	参数量×10⁶	PSNR/dB	SSIM	PSNR/dB	SSIM	PSNR/dB	SSIM	PSNR/dB	SSIM	PSNR/dB	SSIM
2	1.63	37.97	0.960 1	33.63	0.917 5	32.19	0.899 4	32.17	0.928 3	38.88	0.977 1
4	3.12	38.12	0.960 7	33.69	0.919 0	32.22	0.900 4	32.50	0.931 5	38.89	0.977 5
6	4.61	38.18	0.961 2	33.88	0.920 5	32.30	0.901 5	32.80	0.934 5	39.14	0.978 6
8	6.10	38.20	0.961 2	33.86	0.920 3	32.28	0.901 5	32.81	0.934 5	39.13	0.978 6

模型	Set5		Set14		BSD100		Urban100		Manga109
模型	PSNR/dB	SSIM	PSNR/dB	SSIM	PSNR/dB	SSIM	PSNR/dB	SSIM	PSNR/dB	SSIM
SRCNN^[9]	36.66	0.954 2	32.45	0.906 7	31.36	0.887 9	29.50	0.894 6	35.60	0.966 3
FSRCNN^[10]	37.00	0.955 8	32.63	0.908 8	31.53	0.892 0	29.88	0.902 0	36.67	0.971 0
ACAN^[18]	38.10	0.960 8	33.60	0.917 7	32.21	0.900 1	32.29	0.929 7	38.81	0.977 3
SRFBN^[25]	38.11	0.960 9	33.82	0.919 6	32.29	0.901 0	32.62	0.932 8	39.08	0.977 9
CARN^[32]	37.76	0.959 0	33.52	0.916 6	32.09	0.897 8	31.92	0.925 6	38.36	0.976 5
MSRN^[33]	38.08	0.960 5	33.74	0.917 0	32.23	0.901 3	32.22	0.932 6	38.82	0.986 8
SeaNet^[34]	38.08	0.960 9	33.86	0.919 8	32.27	0.900 8	32.68	0.933 2	38.76	0.944 7
FDSCSR^[35]	38.12	0.960 9	33.69	0.919 1	32.24	0.900 4	32.50	0.931 5	38.89	0.977 5
DFBPFSR	38.18	0.961 2	33.89	0.920 5	32.30	0.901 5	32.80	0.934 2	39.14	0.978 6

模型	Set5		Set14		BSD100		Urban100		Manga109
模型	PSNR/dB	SSIM	PSNR/dB	SSIM	PSNR/dB	SSIM	PSNR/dB	SSIM	PSNR/dB	SSIM
SRCNN^[9]	32.75	0.909 0	29.30	0.821 5	28.41	0.786 3	26.24	0.798 9	30.48	0.911 7
FSRCNN^[10]	33.18	0.914 0	29.37	0.824 0	28.53	0.791 0	26.43	0.808 0	31.10	0.921 0
ACAN^[18]	34.46	0.927 7	30.39	0.843 5	29.11	0.805 5	28.28	0.855 0	33.61	0.944 7
SRFBN^[25]	34.70	0.929 2	30.51	0.846 1	29.24	0.808 4	28.73	0.864 1	34.18	0.948 1
CARN^[32]	34.29	0.925 5	30.29	0.840 7	29.06	0.803 4	28.06	0.849 3	33.50	0.944 0
MSRN^[33]	34.38	0.926 2	30.34	0.839 5	29.08	0.804 1	28.08	0.855 4	33.44	0.942 7
SeaNet^[34]	34.55	0.928 2	30.42	0.844 4	29.17	0.807 1	28.50	0.859 4	33.73	0.946 3
FDSCSR^[35]	34.50	0.928 1	30.43	0.844 2	29.15	0.806 8	28.40	0.857 6	33.78	0.946 0
DFBPFSR	34.66	0.929 0	30.53	0.846 0	29.23	0.808 7	28.70	0.863 0	34.03	0.948 1

基于差异特征反投影融合的图像超分辨重建

Image super-resolution reconstruction based on differential feature back-projection fusion

RichHTML

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

图/表 17

参考文献 35

相关文章 15

编辑推荐

Metrics

本文评价

[1]	王鹏昊, 何迪, 尤明懿, 郭细平. 基于随机共振和多尺度特征融合调制识别方法[J]. 系统工程与电子技术, 2026, 48(6): 2096-2104.
[2]	罗雨泉, 何雨强, 李雅鑫, 梁松, 王俊. 基于CNN-BiLSTM-MHA时空融合框架的毫米波雷达人体姿态估计[J]. 系统工程与电子技术, 2026, 48(5): 1502-1514.
[3]	陈垣光, 刘晓凯, 肖思瑶, 张顺生, 崔红元, 陈晓莹, 刘莹. 低信噪比下面向时空特征融合的雷达目标智能检测方法[J]. 系统工程与电子技术, 2026, 48(5): 1523-1531.
[4]	孙景荣, 张华, 陈哲哲, 赵方正. 基于卷积神经网络的雾天道路能见度测量方法[J]. 系统工程与电子技术, 2026, 48(4): 1112-1124.
[5]	田弘宇, 葛松虎, 郭宇, 崔中普, 梁潇. 低干噪比下多特征融合的通信干扰识别方法[J]. 系统工程与电子技术, 2026, 48(3): 1061-1071.
[6]	王瑶, 胥辉旗, 曹司磊, 王磊. 基于深度学习的舰船关键部位检测算法[J]. 系统工程与电子技术, 2026, 48(2): 410-421.
[7]	刘恒燕, 方君, 凌青, 闫文君, 于柯远, 张立民. 基于1D-2D-GRU-ResNet的辐射源个体识别方法[J]. 系统工程与电子技术, 2026, 48(2): 727-735.
[8]	张振兴, 杨任农, 李永林, 左家亮, 胡利平, 陈双艳. 基于双重注意力生成对抗网络的文本到图像生成[J]. 系统工程与电子技术, 2026, 48(1): 34-43.
[9]	蒋明煜, 张顺生, 肖思瑶. 面向轻量级交叉注意力卷积网络的SAR目标识别[J]. 系统工程与电子技术, 2025, 47(9): 2853-2861.
[10]	陈文洁, 张浦, 史高翔, 刘林, 刘烜. 基于余弦校验关系的卷积神经网络LDPC码盲识别[J]. 系统工程与电子技术, 2025, 47(9): 3117-3125.
[11]	付卫红, 彭文洪, 刘乃安. 混合注意力优化的SAR图像小目标检测方法[J]. 系统工程与电子技术, 2025, 47(8): 2519-2526.
[12]	倪康, 贾文杰, 邹旻瑞, 郑志忠. 基于动态聚合网络的SAR目标检测[J]. 系统工程与电子技术, 2025, 47(8): 2527-2539.
[13]	周东平, 阮航, 沙明辉, 王崇宇, 崔念强, 鲁耀兵. 基于特征融合的Fast-CNN的复杂波形调制识别[J]. 系统工程与电子技术, 2025, 47(7): 2165-2175.
[14]	符小卫, 王辛夷, 乔哲. 基于APIQ算法的多无人机攻防对抗策略[J]. 系统工程与电子技术, 2025, 47(7): 2205-2215.
[15]	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.

模型	Set5		Set14		BSD100		Urban100		Manga109
模型	PSNR/dB	SSIM	PSNR/dB	SSIM	PSNR/dB	SSIM	PSNR/dB	SSIM	PSNR/dB	SSIM
SRCNN^[9]	30.48	0.862 6	27.50	0.751 3	26.90	0.710 1	24.52	0.722 1	27.58	0.855 5
FSRCNN^[10]	30.72	0.866 0	27.61	0.755 0	26.98	0.715 0	24.62	0.728 0	27.90	0.861 0
ACAN^[18]	32.24	0.895 5	28.62	0.782 4	27.59	0.736 6	26.17	0.789 1	30.53	0.908 6
SRFBN^[25]	32.47	0.898 3	28.80	0.786 8	27.70	0.740 9	26.60	0.801 5	31.15	0.916 0
CARN^[32]	32.13	0.893 7	28.60	0.780 6	27.58	0.734 9	26.07	0.783 7	30.47	0.908 4
MSRN^[33]	32.07	0.890 3	28.60	0.775 1	27.52	0.727 3	26.04	0.789 6	30.17	0.903 4
SeaNet^[34]	32.33	0.897 0	28.72	0.785 5	27.65	0.738 8	26.32	0.794 2	30.74	0.912 9
FDSCSR^[35]	32.36	0.897 0	28.67	0.784 0	27.63	0.738 4	26.33	0.793 5	30.69	0.911 3
DFBPFSR	32.44	0.898 2	28.80	0.785 9	27.70	0.741 1	26.51	0.798 1	30.93	0.915 5