《计算机应用》唯一官方网站 ›› 2023, Vol. 43 ›› Issue (8): 2572-2580.DOI: 10.11772/j.issn.1001-9081.2022070985
• 多媒体计算与计算机仿真 • 上一篇
收稿日期:
2022-07-08
修回日期:
2022-11-16
接受日期:
2022-11-21
发布日期:
2023-01-15
出版日期:
2023-08-10
通讯作者:
许立君
作者简介:
陈侃松(1972—),男,湖北沙市人,教授,博士生导师,博士,主要研究方向:人工智能、数字孪生、工业互联网基金资助:
Kansong CHEN, Yuan ZHENG, Lijun XU(), Zhouyu WANG, Zhe ZHANG, Fujuan YAO
Received:
2022-07-08
Revised:
2022-11-16
Accepted:
2022-11-21
Online:
2023-01-15
Published:
2023-08-10
Contact:
Lijun XU
About author:
CHEN Kansong, born in 1972, Ph. D., professor. His research interests include artificial intelligence, digital twin, industrial internet.Supported by:
摘要:
针对目前大规模网络不适合在手机、平板电脑等资源匮乏的移动设备上使用,以及池化层会导致特征图的稀疏性最终影响神经网络识别精度的问题,提出了一个轻量级人脸识别神经网络ShuffaceNet,设计了一个非线性平滑Log-Mean-Exp函数ThetaMEX,并提出了一种端到端可训练的ThetaMEX全局池化层(TGPL),从而在保证算法精度的前提下,减少网络参数、提高运算速度,进而达到有效地将该网络部署在资源匮乏的移动设备上的目的。ShuffaceNet约有3 600个参数,模型大小仅为3.5 MB。在LFW(Labled Faces in the Wild)、AgeDB-30 (Age Database-30)、CFP (Celebrities in Frontal Profile)人脸数据集上的识别测试的结果表明,ShuffaceNet的精度分别达到了99.32%、93.17%、94.51%。与MobileNetV1、SqueezeNet、Xception相比,所提网络的大小分别缩减了73.1%、82.1%、78.5%,在AgeDB-30数据集上的精度分别提高了5.0%、6.3%、6.7%。可见,基于ThetaMEX全局池化的所提网络能够提高模型精度。
中图分类号:
陈侃松, 郑园, 许立君, 王周宇, 张哲, 姚福娟. 基于ThetaMEX全局池化的人脸识别神经网络——ShuffaceNet[J]. 计算机应用, 2023, 43(8): 2572-2580.
Kansong CHEN, Yuan ZHENG, Lijun XU, Zhouyu WANG, Zhe ZHANG, Fujuan YAO. ShuffaceNet: face recognition neural network based on ThetaMEX global pooling[J]. Journal of Computer Applications, 2023, 43(8): 2572-2580.
输入 | 操作 | 通道数 | 步长 | 重复次数 |
---|---|---|---|---|
112×112×3 | Depthwise Conv3×3 | 32 | 1 | |
56×56×32 | Max Pool | 1 | 1 | |
28×28×32 | Shufface Stage1 | 64 | 2 | 1 |
64 | 1 | 2 | ||
14×14×144 | Shufface Stage2 | 240 | 2 | 1 |
240 | 1 | 4 | ||
7×7×288 | Shufface Stage3 | 480 | 2 | 1 |
480 | 1 | 4 | ||
5×5×960 | Shufface Stage4(5×5) | 960 | 2 | 1 |
960 | 1 | 2 | ||
1×1×960 | ThetaMEX Pool | |||
Fully Connect | 1 000 |
表1 ShuffaceNet网络结构
Tab. 1 ShuffaceNet network structure
输入 | 操作 | 通道数 | 步长 | 重复次数 |
---|---|---|---|---|
112×112×3 | Depthwise Conv3×3 | 32 | 1 | |
56×56×32 | Max Pool | 1 | 1 | |
28×28×32 | Shufface Stage1 | 64 | 2 | 1 |
64 | 1 | 2 | ||
14×14×144 | Shufface Stage2 | 240 | 2 | 1 |
240 | 1 | 4 | ||
7×7×288 | Shufface Stage3 | 480 | 2 | 1 |
480 | 1 | 4 | ||
5×5×960 | Shufface Stage4(5×5) | 960 | 2 | 1 |
960 | 1 | 2 | ||
1×1×960 | ThetaMEX Pool | |||
Fully Connect | 1 000 |
网络 | 精度/% | 耗时/ms | 参数量/106 | |||
---|---|---|---|---|---|---|
LFW | AgeDB-30 | cfp_fp | cfp_ff | |||
MobileNetV1 | 98.65 | 88.75 | 89.18 | 92.97 | 60 | 3.20 |
MobileNetV2 | 98.79 | 88.84 | 89.48 | 93.17 | 50 | 2.10 |
ShuffleNetV1 | 98.73 | 89.28 | 89.31 | 96.20 | 28 | 0.83 |
SqueezeNet | 98.53 | 87.61 | 87.78 | 94.67 | 30 | 4.80 |
Xception | 98.47 | 87.32 | 88.91 | 91.24 | 54 | 4.00 |
MobileFaceNet | 98.33 | 87.83 | 82.71 | 98.05 | 25 | 0.99 |
LightCNN-9 | 98.81 | 89.87 | 89.63 | 96.30 | 35 | 5.56 |
ShuffaceNet | 99.32 | 93.17 | 94.51 | 98.58 | 22 | 0.86 |
表2 不同数据集上不同网络结构的比较
Tab. 2 Comparison of different network structures on different datasets
网络 | 精度/% | 耗时/ms | 参数量/106 | |||
---|---|---|---|---|---|---|
LFW | AgeDB-30 | cfp_fp | cfp_ff | |||
MobileNetV1 | 98.65 | 88.75 | 89.18 | 92.97 | 60 | 3.20 |
MobileNetV2 | 98.79 | 88.84 | 89.48 | 93.17 | 50 | 2.10 |
ShuffleNetV1 | 98.73 | 89.28 | 89.31 | 96.20 | 28 | 0.83 |
SqueezeNet | 98.53 | 87.61 | 87.78 | 94.67 | 30 | 4.80 |
Xception | 98.47 | 87.32 | 88.91 | 91.24 | 54 | 4.00 |
MobileFaceNet | 98.33 | 87.83 | 82.71 | 98.05 | 25 | 0.99 |
LightCNN-9 | 98.81 | 89.87 | 89.63 | 96.30 | 35 | 5.56 |
ShuffaceNet | 99.32 | 93.17 | 94.51 | 98.58 | 22 | 0.86 |
网络 | 精度/% | 耗时 /ms | 参数量/106 | |||
---|---|---|---|---|---|---|
LFW | AgeDB-30 | cfp_fp | cfp_ff | |||
MobileFaceNet | 98.33 | 87.83 | 82.71 | 98.05 | 28 | 0.99 |
Theta⁃MobileFaceNet | 98.51 | 88.21 | 83.53 | 98.40 | 28 | 1.00 |
ShuffleNetV1 | 98.73 | 89.28 | 83.53 | 98.40 | 25 | 0.83 |
Theta⁃Shufflefacenet | 97.79 | 88.01 | 83.10 | 98.24 | 25 | 1.00 |
ShuffaceNet | 99.32 | 93.17 | 94.51 | 98.58 | 23 | 0.86 |
Theta⁃ShuffaceNet | 98.96 | 94.52 | 95.31 | 99.10 | 22 | 0.86 |
表3 不同模型的精度与参数比较
Tab. 3 Comparison of accuracy and parameters of different models
网络 | 精度/% | 耗时 /ms | 参数量/106 | |||
---|---|---|---|---|---|---|
LFW | AgeDB-30 | cfp_fp | cfp_ff | |||
MobileFaceNet | 98.33 | 87.83 | 82.71 | 98.05 | 28 | 0.99 |
Theta⁃MobileFaceNet | 98.51 | 88.21 | 83.53 | 98.40 | 28 | 1.00 |
ShuffleNetV1 | 98.73 | 89.28 | 83.53 | 98.40 | 25 | 0.83 |
Theta⁃Shufflefacenet | 97.79 | 88.01 | 83.10 | 98.24 | 25 | 1.00 |
ShuffaceNet | 99.32 | 93.17 | 94.51 | 98.58 | 23 | 0.86 |
Theta⁃ShuffaceNet | 98.96 | 94.52 | 95.31 | 99.10 | 22 | 0.86 |
1 | 武文娟,李勇. Emfacenet:一种轻量级人脸识别的卷积神经网络[J]. 小型微型计算机系统, 2023, 44(3): 560-564. |
WU W J, LI Y. Emfacenet: a lightweight convolutional neural network for face recognition[J]. Journal of Chinese Computer Systems, 2023, 44(3): 560-564. | |
2 | LI X L, DING L K, WANG L, et al. FPGA accelerates deep residual learning for image recognition[C]// Proceedings of the IEEE 2nd Information Technology, Networking, Electronic and Automation Control Conference. Piscataway: IEEE, 2017: 837-840. 10.1109/itnec.2017.8284852 |
3 | ZHOU E J, CAO Z M, YIN Q. Naive-deep face recognition: touching the limit of LFW benchmark or not?[EB/OL]. (2015-01-20) [2022-10-26].. |
4 | WEN Y D, ZHANG K P, LI Z F, et al. A discriminative feature learning approach for deep face recognition[C]// Proceedings of 2016 European Conference on Computer Vision, LNCS 9911. Cham: Springer, 2016: 499-515. |
5 | LIU W Y, WEN Y D, YU Z D, et al. SphereFace: deep hypersphere embedding for face recognition[C]// Proceedings of the 2017 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2017: 6738-6746. 10.1109/cvpr.2017.713 |
6 | DENG J K, GUO J, XUE N N, et al. ArcFace: additive angular margin loss for deep face recognition[C]// Proceedings of the 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2019: 4685-4694. 10.1109/cvpr.2019.00482 |
7 | XIAO B, LI X Y, LI C G, et al. A novel pooling block for improving lightweight deep neural networks[J]. Pattern Recognition Letters, 2020, 135: 307-312. 10.1016/j.patrec.2020.05.012 |
8 | SANDLER M, HOWARD A, ZHU M L, et al. MobileNetV2: inverted residuals and linear bottlenecks[C]// Proceedings of the 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2018: 4510-4520. 10.1109/cvpr.2018.00474 |
9 | ZHANG X Y, ZHOU X Y, LIN M X, et al. ShuffleNet: an extremely efficient convolutional neural network for mobile devices[C]// Proceedings of the 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2018: 6848-6856. 10.1109/cvpr.2018.00716 |
10 | CHOLLET F. Xception: deep learning with depthwise separable convolutions[C]// Proceedings of the 2017 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2017: 1800-1807. 10.1109/cvpr.2017.195 |
11 | HE K M, ZHANG X Y, REN S Q, et al. Deep residual learning for image recognition[C]// Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2016: 770-778. 10.1109/cvpr.2016.90 |
12 | IANDOLA F N, HAN S, MOSKEWICZ M W, et al. SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <0.5 MB model size[EB/OL]. (2016-11-04) [2022-10-26].. |
13 | 赵锋,张鹏,张冉. 基于Ghostnet轻量级人脸识别算法研究[J]. 电子测量技术, 2022, 45(16):130-136. |
ZHAO F, ZHANG P, ZHANG R. Research on Ghostnet-based lightweight face recognition algorithm[J]. Electronic Measurement Technology, 2022, 45(16): 130-136. | |
14 | KATZ G, BARRETT C, DILL D L, et al. Reluplex: an efficient SMT solver for verifying deep neural networks[C]// Proceedings of the 2017 International Conference on Computer Aided Verification, LNCS 10426. Cham: Springer, 2017: 97-117. |
15 | COHEN N, SHARIR O, SHASHUA A. Deep SimNets[C]// Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2016: 4782-4791. 10.1109/cvpr.2016.517 |
16 | ZHANG B X, ZHAO Q, FENG W Q, et al. AlphaMEX: a smarter global pooling method for convolutional neural networks[J]. Neurocomputing, 2018, 321: 36-48. 10.1016/j.neucom.2018.07.079 |
17 | CHEN S, LIU Y, GAO X, et al. MobileFaceNets: efficient CNNs for accurate real-time face verification on mobile devices[C]// Proceedings of the 2018 Chinese Conference on Biometric Recognition, LNCS 10996. Cham: Springer, 2018: 428-438. |
18 | LUO W J, LI Y J, URTASUN R, et al. Understanding the effective receptive field in deep convolutional neural networks[C]// Proceedings of the 30th International Conference on Neural Information Processing Systems. Red Hook, NY: Curran Associates Inc., 2016: 4905-4913. |
19 | CAO Q, SHEN L, XIE W D, et al. VGGFace2: a dataset for recognising faces across pose and age[C]// Proceedings of the 13th IEEE International Conference on Automatic Face and Gesture Recognition. Piscataway: IEEE, 2018:67-74. 10.1109/fg.2018.00020 |
20 | MOSCHOGLOU S, PAPAIOANNOU A, SAGONAS C, et al. AgeDB: the first manually collected, in-the-wild age database[C]// Proceedings of the 2017 IEEE Conference on Computer Vision and Pattern Recognition Workshops. Piscataway: IEEE, 2017: 1997-2005. 10.1109/cvprw.2017.250 |
[1] | 马胜位, 黄瑞章, 任丽娜, 林川. 基于多层语义融合的结构化深度文本聚类模型[J]. 《计算机应用》唯一官方网站, 2023, 43(8): 2364-2369. |
[2] | 李校林, 杨松佳. 基于深度学习的多用户毫米波中继网络混合波束赋形[J]. 《计算机应用》唯一官方网站, 2023, 43(8): 2511-2516. |
[3] | 李淦, 牛洺第, 陈路, 杨静, 闫涛, 陈斌. 融合视觉特征增强机制的机器人弱光环境抓取检测[J]. 《计算机应用》唯一官方网站, 2023, 43(8): 2564-2571. |
[4] | 衡红军, 杨鼎诚. 知识增强的方面词交互图神经网络[J]. 《计算机应用》唯一官方网站, 2023, 43(8): 2412-2419. |
[5] | 张琨, 杨丰玉, 钟发, 曾广东, 周世健. 基于混合代码表示的源代码脆弱性检测[J]. 《计算机应用》唯一官方网站, 2023, 43(8): 2517-2526. |
[6] | 何嘉明, 杨巨成, 吴超, 闫潇宁, 许能华. 基于多模态图卷积神经网络的行人重识别方法[J]. 《计算机应用》唯一官方网站, 2023, 43(7): 2182-2189. |
[7] | 夏子芳, 于亚新, 王子腾, 乔佳琪. 融合协同知识图谱与反事实推理的可解释推荐机制[J]. 《计算机应用》唯一官方网站, 2023, 43(7): 2001-2009. |
[8] | 秦源源, 张鸿. 基于注意力特征金字塔网络的肺结节检测算法[J]. 《计算机应用》唯一官方网站, 2023, 43(7): 2311-2318. |
[9] | 杨盼, 张敏情, 葛虞, 狄富强, 张英男. 基于风格迁移过程的彩色图像信息隐藏算法[J]. 《计算机应用》唯一官方网站, 2023, 43(6): 1730-1735. |
[10] | 张慧斌, 冯丽萍, 郝耀军, 王一宁. 基于注意力机制和迁移学习的古壁画朝代识别[J]. 《计算机应用》唯一官方网站, 2023, 43(6): 1826-1832. |
[11] | 王辉, 李建红. 基于Transformer的三维模型小样本识别方法[J]. 《计算机应用》唯一官方网站, 2023, 43(6): 1750-1758. |
[12] | 郑智雄, 刘建华, 孙水华, 徐戈, 林鸿辉. 融合多窗口局部信息的方面级情感分析模型[J]. 《计算机应用》唯一官方网站, 2023, 43(6): 1796-1802. |
[13] | 许睿, 梁爽, 万航, 文益民, 沈世铭, 李建. 基于烛台图模式匹配的PM2.5扩散特征的提取[J]. 《计算机应用》唯一官方网站, 2023, 43(5): 1394-1400. |
[14] | 隋佳宏, 毛莺池, 于慧敏, 王子成, 平萍. 基于图注意力网络的全局图像描述生成方法[J]. 《计算机应用》唯一官方网站, 2023, 43(5): 1409-1415. |
[15] | 蒋瑞林, 覃仁超. 基于深度可分离卷积的多神经网络恶意代码检测模型[J]. 《计算机应用》唯一官方网站, 2023, 43(5): 1527-1533. |
阅读次数 | ||||||
全文 |
|
|||||
摘要 |
|
|||||