In the digital era, significant challenges are posed to public safety and information credibility by low-resolution images. Existing super-resolution models are hindered by excessive parameters and high computational complexity, making them difficult to be deployed on edge devices, while deep network architectures often lead to the loss of high-frequency details. To address these issues, an efficient convolutional image super-resolution network integrated with a dual attention mechanism was proposed. A lightweight channel attention module and an enhanced spatial attention module were constructed to achieve cooperative optimization of dynamic channel-wise weight allocation and spatial key region enhancement. Combined with a partial convolution-based feed-forward network, the local feature extraction range was constrained, effectively suppressing interference from redundant receptive field expansion on texture reconstruction. The end-to-end architecture was designed with a single global residual connection and modular structure, which reduced parameters while enhancing cross-scale feature fusion capability. Experimental results demonstrate that the proposed model achieves 22 optimal results out of 30 core metrics across five benchmark datasets including Set5 and Set14, with an average PSNR(peak signal-to-noise ratio) improvement of 0.13 dB and an average parameter reduction of 37.5%. Local attribution map analysis further confirms that the model maintains a larger effective receptive field while key pixel information is utilized more precisely, significantly improving the reconstruction quality of high-frequency details. These findings adequately validate the superior efficiency-performance balance of the proposed model.
YAN Hang, WANG Jia-xing, SONG Zhen-feng.
Efficient Convolutional Image Super-resolution Network with Dual Attention Mechanism Fusion[J].
Science Technology and Engineering, 2026, 26(15): 6457-6467 DOI:10.12404/j.issn.1671-1815.2504162
FNPC(feedforward network based on partial convolution)相较于传统的前馈网络,通过部分卷积策略在通道维度上拆分特征,仅对部分通道进行空间卷积处理,以平衡计算量与性能并提高空间维度信息交换,如图5所示。首先通过1×1 卷积和 GELU 激活函数对特征图x∈RC×H×W 进行通道扩展与激活,然后将特征分成两个部分,此过程表示为
WangRuiqi. Review of image super-resolution reconstruction[J]. Computer Science and Applications, 2024, 14(2): 350-359.
[3]
ZhangY L, LiK, WangL, et al. Image super-resolution using very deep residual channel attention networks[C]// Proceedings of the European Conference on Computer Vision (ECCV). München: ECCV, 2018: 286-301.
[4]
DaiT, CaiJ, ZhangY, et al. Second-order attention network for single image super-resolution[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2019: 11065-11074.
[5]
AnwarS, BarnesN. Densely residual laplacian super-resolution[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2020, 44(3): 1192-1204.
[6]
DongC, LoyC C, HeK, et al. Image super-resolution using deep convolutional networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2015, 38(2): 295-307.
[7]
KimJ, LeeJ K, LeeK M. Accurate image super-resolution using very deep convolutional networks[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2016: 1646-1654.
[8]
LimB, SonS, KimH, et al. Enhanced deep residual networks for single image super-resolution[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops. New York: IEEE, 2017: 136-144.
[9]
ZhangY, TianY, KongY, et al. Residual dense network for image super-resolution[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2018: 2472-2481.
[10]
WuG, JiangJ, JiangK, et al. Fully 1×1 convolutional network for lightweight image super-resolution[J]. Machine Intelligence Research, 2024, 21(6): 1062-1076.
[11]
SunL, PanJ, TangJ. Shufflemixer: an efficient convnet for image super-resolution[J]. Advances in Neural Information Processing Systems, 2022, 35: 17314-17326.
[12]
FinderS E, AmoyalR, TreisterE, et al. Wavelet convolutions for large receptive fields[C]// European Conference on Computer Vision. Cham: Springer Nature Switzerland, 2024: 363-380.
ChenChunyi, WuXinyi, HuXiaojuan, et al. Image super-resolution reconstruction based on multi-scale attention fusion[J]. Chinese Optics, 2023, 16(5): 1034-1044.
[15]
NiuB, WenW, RenW, et al. Single image super-resolution via a holistic attention network[C]// Computer Vision-ECCV 2020: 16th European Conference. Glasgow: Springer International Publishing, 2020: 191-207.
[16]
ChenR, ZhangH, LiuJ. Multi-attention augmented network for single image super-resolution[J]. Pattern Recognition, 2022, 122: 108349.
[17]
WangY, LiY, WangG, et al. Multi-scale attention network for single image super-resolution[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2024: 5950-5960.
[18]
WooS, ParkJ, LeeJ Y, et al. CBAM: Convolutional block attention module[C]// Proceedings of the European Conference on Computer Vision (ECCV). Munich: ECCV, 2018: 3-19.
WangTingwei, ZhaoJianwei, ZhouZhenghua. Image super-resolution reconstruction method based on lightweight symmetric CNN-Transformer[J]. Pattern Recognition and Artificial Intelligence, 2024, 37(7): 626-637.
[21]
XiaB, HangY, TianY, et al. Efficient non-local contrastive attention for image super-resolution[C]// Proceedings of the AAAI Conference on Artificial Intelligence. New York: AAAI, 2022: 2759-2767.
[22]
RuanH, TanZ, ChenL, et al. Efficient sub-pixel convolutional neural network for terahertz image super-resolution[J]. Optics Letters, 2022, 47(12): 3115-3118.
[23]
ZhaoH, KongX, HeJ, et al. Efficient image super-resolution using pixel attention[C]// Computer Vision-ECCV 2020 Workshops. Glasgow: Springer International Publishing, 2020: 56-72.
LiPeiyu, ZhangYali, ZhangYibo, et al. Face image super-resolution reconstruction based on adaptive convolution and joint loss function[J]. Science Technology and Engineering, 2025, 25(6): 2442-2452.
LiuQirui, OuJikun, RenChao, et al. Image super-resolution reconstruction via adaptive weighted compressed sensing[J]. Science Technology and Engineering, 2023, 23(13): 5647-5654.
[30]
ChoiH, LeeJ, YangJ. N-gram in swin transformers for efficient lightweight image super-resolution[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2023: 2071-2081.
[31]
AhnN, KangB, SohnK A. Fast, accurate, and lightweight super-resolution with cascading residual network[C]// Proceedings of the European Conference on Computer Vision (ECCV). Munich: ECCV, 2018: 252-268.
[32]
ZhouL, CaiH, GuJ, et al. Efficient image super-resolution using vast-receptive-field attention[C]// European Conference on Computer Vision. Cham: Springer Nature Switzerland, 2022: 256-272.
[33]
DengW, YuanH, DengL, et al. Reparameterized residual feature network for lightweight image super-resolution[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2023: 1712-1721.
[34]
WanC, YuH, LiZ, et al. Swift parameter-free attention network for efficient super-resolution[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2024: 6246-6256.
[35]
DongC, LoyC C, TangX. Accelerating the super-resolution convolutional neural network[C]// Computer Vision-ECCV 2016: 14th European Conference. Amsterdam: Springer International Publishing, 2016: 391-407.
[36]
KongF, LiM, LiuS, et al. Residual local feature network for efficient super-resolution[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2022: 766-776.
[37]
LiZ, LiuY, ChenX, et al. Blueprint separable residual network for efficient image super-resolution[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2022: 833-843.
[38]
GuJ, DongC. Interpreting super-resolution networks with local attribution maps[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2021: 9199-9208.