基于自适应大核注意力的轻量级图像超分辨率网络

Abstract

Abstract:

To address the issue of large parameter quantities in high-performance image super-resolution networks， a lightweight model was proposed. First， three kinds of large kernel attention were integrated， namely dual-path large kernel attention， large kernel pixel attention and large kernel channel attention， aiming to expand the model perceptual field and establish the long-range dependence of pixels. Second， an adaptive attention fusion mechanism was introduced to enhance the characterization of features and improve the model performance. Experiments demonstrate that the model performs well on visual perception and quantitative tests. On the Urban100 dataset， the mean value of peak signal-to-noise ratio of ×4 reconstruction results was improved by 0.25 dB compared with the currently popular ARRFN algorithm. The reconstructed images have more realistic visual effects， clearer and more natural texture information， which fully demonstrates the proposed algorithm effectiveness.

Key words: computer version, super-resolution, large kernel attention, lightweight, adaptive feature fusion

CLC Number:

TP391.41

De-qiang CHENG,Gui LIU,Qi-qi KOU,Jian-ying ZHANG,He JIANG. Lightweight image super⁃resolution network based on adaptive large kernel attention fusion[J].Journal of Jilin University(Engineering and Technology Edition), 2025, 55(3): 1015-1027.

Figures/Tables 16

Fig.1

Table 1

Table 2

Fig.2

Table 3

Table 4

Table 5

Fig.3

Fig.4

Fig.5

Fig.6

Table 6

Index comparison under benchmark dataset when scaling factor is 2，3 and 4"

缩放因子	模型	参数量/M	浮点运算量/G	Set5^［24］ PSNR/SSIM	Set14^［25］ PSNR/SSIM	B100^［26］ PSNR/SSIM	Urban100^［27］ PSNR/SSIM
x₂	SRCNN^［10］	0.05	52.7	36.66/0.9542	32.42/0.9063	31.36/0.8879	29.50/0.8946
	FSRCNN^［28］	0.01	6.0	37.00/0.9558	32.63/0.9088	31.53/0.8920	29.88/0.9020
	VDSR^［29］	0.66	612.6	37.53/0.9587	33.03/0.9124	31.90/0.8960	30.76/0.9140
	DRCN^［30］	1.77	17974.0	37.63/0.9588	33.04/0.9118	31.85/0.8942	30.75/0.9133
	CARN^［31］	1.59	222.8	37.76/0.9590	33.52/0.9166	32.09/0.8978	31.92/0.9256
	EDSR-baseline^［12］	1.37	316.2	37.99/0.9604	33.57/0.9175	32.16/0.8994	31.98/0.9272
	IMDN^［32］	0.69	158.8	38.00/0.9605	33.63/0.9177	32.19/0.8996	32.17/0.9283
	LAMRN^［35］	1.39	320.1	38.09/0.9608	33.78^/0.9195^	32.22/0.9003	32.31/0.9299
	FMEN^［34］	0.75	172.0	38.10^/0.9609^	33.75^/0.9192^	32.26^/0.9007^	32.41/0.9311
	A²F-L^［33］	1.36	306.1	38.09/0.9607	33.78^*/0.9192^	32.23/0.9002	32.46^/0.9313^
	ARRFN^［36］	0.98	-	38.01/0.9606	33.66/0.9179	32.20/0.8999	32.27/0.9295
	ALSR（本文）	1.51	189.3	38.14^/0.9616^	33.74/0.9187	32.24^/0.9004^	32.48^/0.9317^
x₃	SRCNN^［10］	0.05	52.7	32.75/0.9090	29.28/0.8209	28.41/0.7863	26.24/0.7989
	FSRCNN^［28］	0.01	5.0	33.16/0.9140	29.43/0.8242	28.53/0.7910	26.43/0.8080
	VDSR^［29］	0.66	612.6	33.66/0.9213	29.77/0.8314	28.82/0.7976	27.14/0.8279
	DRCN^［30］	1.77	17974.0	33.82/0.9226	29.76/0.8311	28.80/0.7963	27.15/0.8276
	CARN^［31］	1.59	118.8	34.29/0.9255	30.29/0.8407	29.06/0.8034	28.06/0.8493
	EDSR-baseline^［12］	1.55	160.4	34.37/0.9270	30.28/0.8417	29.09/0.8052	28.15/0.8527
	IMDN^［32］	0.70	71.5	34.36/0.9270	30.32/0.8417	29.09/0.8046	28.17/0.8519
	LAMRN^［35］	1.41	145.3	34.55^/0.9285^	30.41^*/0.8431	29.17^/0.8067^	28.43^*/0.8571
	FMEN^［34］	0.76	77.2	34.45/0.9275	30.40/0.8435	29.17^*/0.8063	28.33/0.8562
	A²F-L^［33］	1.37	136.3	34.54/0.9283	30.41^/0.8436^	29.14/0.8062	28.40/0.8574^*
	ARRFN^［36］	0.99	-	34.38/0.9272	30.36/0.8422	29.09/0.8050	28.22/0.8533
	ALSR（本文）	1.69	122.5	34.58^/0.9293^	30.47^/0.8451^	29.18^/0.8069^	28.44^/0.8576^
x₄	SRCNN^［10］	0.05	52.7	30.48/0.8628	27.49/0.7503	26.90/0.7101	24.52/0.7221
	FSRCNN^［28］	0.01	4.6	30.71/0.8657	27.59/0.7535	26.98/0.7150	24.62/0.7280
	VDSR^［29］	0.66	612.6	31.35/0.8838	28.01/0.7674	27.29/0.7251	25.18/0.7524
	DRCN^［30］	1.77	17974.0	31.53/0.8854	28.02/0.7670	27.23/0.7233	25.14 /0.7510
	CARN^［31］	1.59	90.9	32.13/0.8937	28.60/0.7806	27.58/0.7349	26.07/0.7837
	EDSR-baseline^［12］	1.51	114.2	32.09/0.8938	28.58/0.7813	27.57/0.7357	26.04/0.7849
	IMDN^［32］	0.70	40.9	32.21/0.8948	28.58/0.7811	27.56/0.7353	26.04/0.7838
	LAMRN^［35］	1.41	85.0	32.37^/0.8971^	28.70^/0.7839^	27.61/0.7384^*	26.32^*/0.7925
	FMEN^［34］	0.77	44.2	32.24/0.8955	28.70^/0.7839^	27.63/0.7379	26.28/0.7908
	A²F-L^［33］	1.37	77.2	32.32/0.8964	28.67/0.7839	27.62^*/0.7379	26.32^/0.7931^
	ARRFN^［36］	1.00	-	32.22/0.8952	28.60/0.7817	27.57/0.7355	26.09/0.7858
	ALSR（本文）	1.66	69.9	32.40^/0.8976^	28.71^/0.7843^	27.65^/0.7386^	26.34^/0.7937^

Table 6

Table 7

Fig.7

Fig.8

Table 8

References 39

1	Cheng D Q, Chen L L, Lv C, et al. Light-guided and cross-fusion U-Net for anti-illumination image super-resolution[J]. IEEE Transaction on Circuits and Systems for Video Technology, 2022, 32(12): 8436-8449.
2	Jebkins W K, Mather B C, Jr D C M. Nearest neighbor and generalized inverse distance interpolation for fourier domain image reconstruction[C]∥Proceedings of the IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP), Tampa, USA, 1985:1069-1072.
3	Blu T, Thevenaz P, Unser M. Linear interpolation revitalized[J]. IEEE Trans Image Process, 2004, 13(5): 710-719.
4	Lin C, Shen M, Chang H, et al. The efficient VLSI design of BICUBIC convolution interpolation for digital image processing[C]∥Proceedings of the International Symposium on Circuitsand Systems (ISCAS), Seattle, USA, 2008:480-483.
5	Dai S, Han M, Xu W,et al. Soft-cuts: a soft edge smoothness prior for color image super-resolution[J]. IEEE Transactions on Image Processing, 2009, 18(5): 969-981.
6	Sun J, Xu Z, Shun H Y. Image super-resolution using gradient profile prior[C]∥Proceedings of the IEEE Conference on Computer Vision and Pattern Reco Anchorage, Anchorage, AK, USA, 2008: 1-8.
7	Yan Q, Xu Y, Yang X, et al. Single image super resolution based on gradient profile sharpness[J]. IEEE Transactions on Image Processing, 2015,24(10):3187-3202 .
8	iang H J, Mujtaba A S A D, Liu J J, et al. Single image detail enhancement via metropolis theorem[J]. Multimedia Tools and Applications,2024, 83: 36329-36353.
9	程德强, 郭昕, 陈亮亮, 等. 多通道递归残差网络的图像超分辨率重建[J].中国图象图形学报, 2021, 26(3): 605-618.
	Cheng De-qiang, Guo Xin, Chen Liang-liang, et al. Image super resolution reconstruction by multi-channel recursive residual networks[J]. Journal of Image and Graphics, 2021, 26(3): 605-618.
10	Dong C, Loy C C, He K M, et al. Image super-resolution using deep convolutional networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2016, 38(2): 295-307.
11	Shi W, Caballero J, Huszár F,et al. Real-time single image and video super-resolution using an efficient sub-pixel convolutional neural network[C]∥IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, USA, 2016:1874-1883.
12	Lim B, Son S, Kim H,et al. Enhanced deep residual networks for single image super-resolution[C]∥IEEE Conference on Computer Vision and Pattern Recognition Workshops, Honolulu,USA, 2016:1132-1140.
13	郭继昌, 吴洁, 郭春乐, 等. 基于残差连接卷积神经网络的图像超分辨率重构[J]. 吉林大学学报: 工学版,2019, 49(5): 1726-1734.
	Guo Ji-chang, Wu Jie, Guo Chun-le, et al.Image super-resolution reconstruction based on residual-connected convolutional neural networks[J]. Journal of Jilin University(Engineering and Technology Edition,2019, 49(5): 1726-1734.
14	Zhang Y, Li K. Image super-resolution using very deep residual channel attention networks[C]∥Proceedings of the European Conference on Computer Vision, Munich Germany, 2018: 286-301.
15	Niu B, Wen W L, Ren W Q,et al. Single image super-resolution via a holistic attention network[C]∥ 2020 European Conference on Computer Vision (ECCV), Glasgow, USA, 2020:191-207.
16	Vaswan A, Shazeer N, Parmar, et al. Attention is all you need[J/OL].[2023-09-15].
17	Liang J Y, Cao J Z, Sun G L,et al. Image restoration using swin transformer[C]∥Proceedings of the IEEE/CVF International Conference on Computer Vision, Montreal, Canada, 2021: 1833-1844.
18	Lu Z S, Li J C, Liu H, et al. Transformer for single image super-resolution[C]∥Proc IEEE/CVF Conf Comput Vis Pattern Recognit Workshops, New Orleans, USA, 2022: 456-465.
19	Kou Q Q, Cheng D Q, Zhang H et al. Single Image super resolution via multi-attention fusion recurrent network[J]. IEEE Access,2023(11): 98653-98665.
20	Wang Z, Gao G, Li J,et al. Lightweight image super-resolution with multi-scale feature interaction network[C]∥IEEE International Conference on Multimedia and Expo, Shenzhen, China, 2021: No.9428136.
21	Lan R, Sun L, Liu Z,et al. MADNet:a fast and lightweight network for single-image super resolution[J]. IEEE Transactions on Cybernetics, 2021,51(3):1443-1453.
22	Guo M H, Lu C Z, Liu Z N,et al. Visual attention network[J/OL].[2023-09-18].
23	Jie H, Li S, Samuel A, et al. Squeeze-and-excita⁃tion networks[C]∥Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogni⁃tion, Wellington, New Zealand, 2018: 7132-7141.
24	Bevilacqua M, Roumy A, Guillemot C, et al. Low-complexity single image super-resolution based on nonnegative neighbour embedding[C]∥British Machine Vision Conference, Oxford,UK, 2012:1-10.
25	Zeyde R, Elad M, Protter M. On single image scale-up using sparse-representations[C]∥International Conference on Curves and Surfaces,Avignon, France, 2010: 711-730.
26	Kim J, Lee J K, Lee K M. Accurate image super-resolution using very deep convolutional networks[C]∥2016 IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, USA, 2016:1646-1654.
27	Huang J B, Singh A, Ahuja N. Single image super-resolution from transformed self-exemplars[C]∥IEEE Conference on Computer Vision and Pattern Recognition,Boston, USA, 2015:5197-5206.
28	Dong C, Loy C C, Tang X. Accelerating the super-resolution convolutional neural network[C]∥European Conference on Computer Vision,Amsterdam, Holland, 2016:391-407.
29	Kim J, Lee J K, Lee K M. Accurate image super-resolution using very deep convolutional networks[C]∥Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, USA, 2016: 1646-1654.
30	Kim J, Lee J K, Lee K M. Deeply-recursive convolutional network for image super-resolution[C]∥Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, USA,2016:1637-1645.
31	Ahh N, Kang B, Sohn K A. Fast, accurate, and lightweight super-resolution with cascading residual network[C]∥European Conference on Computer Vision(ECCV), Munich, Germany, 2018: 256-272.
32	Hui Z, Gao X, Yang Y, et al. Lightweight image super-resolution with information multi-distillation network[C]∥Proceedings of the 27th ACM International Conference on Multimedia, Xi'an, China, 2019: 2024-2032.
33	Wang X H, Wang Q, Zhao Y Z, et al. Lightweight single-image super-resolution network with attentive auxiliary feature learning[C]∥Asian Conference on Computer Vision(ACCV),Kyoto, Japan, 2020: 3023-3041.
34	Du Z C, Liu D, Liu J, et al. Fast and memory-efficient network towards efficient image super-resolution[C]∥Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition,New Orleans, USA, 2022 :853-862.
35	Yan Y, Xu X, Chen W, et al. Lightweight attended multi-scale residual network for single image super-resolution[J]. IEEE Access, 2021, 9: 52202-52212.
36	Qin J, Zhang R. Lightweight single image super-resolution with attentive residual refinement network[J].Neurocomputing, 2022, 500: 846-855.
37	Gao G W, Wang Z X, Li J C, et al. Lightweight bimodal network for single-image super-resolution via symmetric CNN and recursive transformer[C]∥International Joint Conference on Artificial Intelligence,Shenzhen, China, 2022: 913-919.
38	Choi H, Lee J, Yang J. N-Gram in swin transformers for efficient lightweight image super-resolution[C]∥Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Vancouver, Canadian, 2023: 2071-2081.
39	Jie H, Li S, Samuel A, et al. Squeeze-and-excitation networks[C]∥Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Wellington, New Zealand, 2018: 7132-7141.

Related Articles 15

[1]	Hua-song DONG,Yuan-feng LIAN. Lightweight detection algorithm for lossless transcoding and heavy compression of massive digital media videos [J]. Journal of Jilin University(Engineering and Technology Edition), 2025, 55(2): 741-747.
[2]	Hua CAI,Yan-yang ZHENG,Qiang FU,Sheng-yu WANG,Wei-gang WANG,Zhi-yong MA. Three-dimensional object detection algorithm based on multi-scale candidate fusion and optimization [J]. Journal of Jilin University(Engineering and Technology Edition), 2025, 55(2): 709-721.
[3]	Hua CAI,Ting-ting KOU,Yi-ning YANG,Zhi-yong MA,Wei-gang WANG,Jun-xi SUN. Three-dimensional vehicle multi-target tracking based on trajectory optimization [J]. Journal of Jilin University(Engineering and Technology Edition), 2024, 54(8): 2338-2347.
[4]	Xin-gang GUO,Ying-chen HE,Chao CHENG. Noise-resistant multistep image super resolution network [J]. Journal of Jilin University(Engineering and Technology Edition), 2024, 54(7): 2063-2071.
[5]	Zi-chao ZHANG,Jian CHEN. A mapping method using 3D orchard point cloud based on hawk-eye-inspired stereo vision and super resolution [J]. Journal of Jilin University(Engineering and Technology Edition), 2024, 54(5): 1469-1481.
[6]	Guang HUO,Da-wei LIN,Yuan-ning LIU,Xiao-dong ZHU,Meng YUAN,Di GAI. Lightweight iris segmentation model based on multiscale feature and attention mechanism [J]. Journal of Jilin University(Engineering and Technology Edition), 2023, 53(9): 2591-2600.
[7]	Ben-huai LI,Yan-wen LIU,Lu WANG,Xue-qian CHEN,Wen-jie ZUO. Crashworthiness optimization for frontal⁃end structure of rail vehicle constrained with absorbed energy and crushed displacements [J]. Journal of Jilin University(Engineering and Technology Edition), 2023, 53(4): 982-988.
[8]	Bin-xiang JIANG,Hong-kui XU,Dan HE. Drug Efficiency improvement of drug detection big data based on blockchain [J]. Journal of Jilin University(Engineering and Technology Edition), 2022, 52(7): 1666-1678.
[9]	Fei-yue DENG, LYUHao-yang,Xiao-hui GU,Ru-jiang HAO. Fault diagnosis of high⁃speed train axle bearing based on a lightweight neural network Shuffle⁃SENet [J]. Journal of Jilin University(Engineering and Technology Edition), 2022, 52(2): 474-482.
[10]	Yong CHEN,Chen-tao LU,Zhen WANG. Detection of foreign object intrusion in railway region of interest based on lightweight network [J]. Journal of Jilin University(Engineering and Technology Edition), 2022, 52(10): 2405-2418.
[11]	Yan-lei XU,Run HE,Yu-ting ZHAI,Bin ZHAO,Chen-xiao LI. Weed identification method based on deep transfer learning in field natural environment [J]. Journal of Jilin University(Engineering and Technology Edition), 2021, 51(6): 2304-2312.
[12]	Chao MA,Yun-kai GAO,Zhe LIU,Yue-xing DUAN,Lin-li TIAN. Optimization of multi⁃material and beam cross⁃sectional shape and dimension of skeleton⁃type body [J]. Journal of Jilin University(Engineering and Technology Edition), 2021, 51(5): 1583-1592.
[13]	Li-sheng JIN,Bai-cang GUO,Fang-rong WANG,Jian SHI. Dynamic multiple object detection algorithm for vehicle forward based on improved YOLOv3 [J]. Journal of Jilin University(Engineering and Technology Edition), 2021, 51(4): 1427-1436.
[14]	Zhe-ming YUAN,Hong-jie YUAN,Yu-xuan YAN,Qian LI,Shuang-qing LIU,Si-qiao TAN. Automatic recognition and classification of field insects based on lightweight deep learning model [J]. Journal of Jilin University(Engineering and Technology Edition), 2021, 51(3): 1131-1139.
[15]	Hui YE,Chang LIU,Kang-kang YAN. Application of fiber reinforced composite in auto⁃body panel [J]. Journal of Jilin University(Engineering and Technology Edition), 2020, 50(2): 417-425.

Metrics

Viewed

Full text

Abstract

Cited

Shared

Discussed

Comments

Recommended 0

No Suggested Reading articles found!

LKA 3-5-1	LKA 5-7-1	LKA 7-9-1	参数量/K	浮点运算量/G	Set5 PSNR/SSIM	Set14 PSNR/SSIM	B100 PSNR/SSIM	Urban100 PSNR/SSIM
√	×	×	767.96	113.69	37.93/0.9587	33.43/0.9152	32.08/0.8975	31.74/0.9237
×	√	×	778.20	115.20	37.94/0.9592	33.46/0.9159	32.10/0.8978	31.74/0.9238
×	×	√	791.01	117.31	37.93/0.9584	33.44/0.9154	32.09/0.8977	31.73/0.9234
√	√	×	837.08	123.73	37.97/0.9598	33.50/0.9168	32.12/0.8983	31.80/0.9243
×	√	√	864.25	127.35	37.98/0.9601	33.51/0.9171	32.11/0.8981	31.78/0.9238
√	×	√	853.47	125.84	37.96/0.9594	33.50/0.9166	32.10/0.8976	31.77/0.9235
√	√	√	904.38	133.47	37.98/0.9604	33.52/0.9170	32.11/0.8984	31.80/0.9246

1×1Conv	7×7DPConv	1×1Conv	参数量/M	浮点运算量/G	Set5 PSNR/SSIM	Set14 PSNR/SSIM	B100 PSNR/SSIM	Urban100 PSNR/SSIM
× × × √	×	×	0.58	66.56	37.76/0.9590	33.22/0.9135	31.92/0.8960	31.12/0.9171
	×	√	0.60	68.98	37.78/0.9595	33.23/0.9136	31.93/0.8961	31.13/0.9173
	√	×	0.59	68.41	37.79/0.9596	33.24/0.9139	31.93/0.8962	31.13/0.9172
	×	√	0.61	71.37	37.77/0.9592	33.23/0.9136	31.93/0.8962	31.12/0.9172
√	√	√	0.63	73.22	37.81/0.9593	33.26/0.9142	31.94/0.8963	31.14/0.9175

DLKA	LKPA	AWLM	参数量/M	浮点运算量/G	Set5 PSNR/SSIM	Set14 PSNR/SSIM	B100 PSNR/SSIM	Urban100 PSNR/SSIM
× √ × √	×	×	0.74	109.99	37.81/0.9573	33.34/0.9142	32.03/0.8969	31.62/0.9213
	×	×	0.84	123.73	37.97/0.9598	33.50/0.9168	32.12/0.8983	31.80/0.9243
	√	×	0.77	113.37	37.93/0.9589	33.47/0.9161	32.11/0.8980	31.73/0.9234
	√	×	0.86	127.11	38.02/0.9608	33.50/0.9169	32.12/0.8983	31.98/0.9270
√	√	√	0.88	127.18	38.04/0.9610	33.52/0.9171	32.14/0.8987	32.01/0.9275

ALKA	LKCA	参数量/M	浮点运算量/G	Set5 PSNR/SSIM	Set14 PSNR/SSIM	B100 PSNR/SSIM	Urban100 PSNR/SSIM
× √ × √	×	0.74	109.99	37.81/0.9573	33.34/0.9142	32.03/0.8969	31.62/0.9213
	×	0.88	127.18	38.04/0.9609	33.52/0.9171	32.14/0.8987	32.01/0.9275
	√	0.92	116.73	37.96/0.9594	33.43/0.9152	32.06/0.8972	31.65/0.9217
	√	1.06	133.90	38.05/0.9610	33.56/0.9175	32.16/0.8991	32.04/0.9279

尺度因子	LBNet^［37］	ESRT^［18］	NGswin^［38］	ALSR（本文）
2	-	38.03/0.9600	38.05/0.9610	38.14/0.9616
3	34.47/0.9277	34.42/0.9268	34.52/0.9282	34.58/0.9293
4	32.29/0.8960	32.19/0.8947	32.33/0.8963	32.40/0.8976

Lightweight image super⁃resolution network based on adaptive large kernel attention fusion

RICH HTML

PDF (PC)