LFRNet：面向无透镜成像的多层级特征蒸馏人脸识别方法

王烨茹; 董相辰; 肖文凯; 崔庆春; 刘亚旭; 徐凡; 张国威; 秦飞巍; 丁佳俊

doi:10.12382/bgxb.2025.0250

您当前的位置：

首页 >

文章列表页 >

LFRNet：面向无透镜成像的多层级特征蒸馏人脸识别方法

更新时间：2026-05-06

- LFRNet：面向无透镜成像的多层级特征蒸馏人脸识别方法
- Face Recognition Based on Embedded Lensless Imaging
- 兵工学报 2026年47卷第4期页码：250250
- 作者机构：
  
  杭州电子科技大学网络空间安全学院，浙江杭州 310018
  杭州智元研究院有限公司，浙江杭州 310000
  杭州电子科技大学计算机学院，浙江杭州 310018
  杭州电子科技大学上虞科学与工程研究院，浙江杭州 312300
- 作者简介：
  
  通信作者邮箱：wangyeru@hdu.edu.cn
- 基金信息：
- DOI：10.12382/bgxb.2025.0250
  中图分类号： TP391
- 收稿：2025-04-03，
  
  网络首发：2026-02-10，
  
  纸质出版：2026-04
- 稿件说明：
移动端阅览
王烨茹, 董相辰, 肖文凯, 等. LFRNet：面向无透镜成像的多层级特征蒸馏人脸识别方法[J]. 兵工学报, 2026,47(4):250250.

WANG Yeru, DONG Xiangchen, XIAO Wenkai, et al. Face Recognition Based on Embedded Lensless Imaging[J]. Acta Armamentarii, 2026, 47(4): 250250.
王烨茹, 董相辰, 肖文凯, 等. LFRNet：面向无透镜成像的多层级特征蒸馏人脸识别方法[J]. 兵工学报, 2026,47(4):250250. DOI： 10.12382/bgxb.2025.0250.

WANG Yeru, DONG Xiangchen, XIAO Wenkai, et al. Face Recognition Based on Embedded Lensless Imaging[J]. Acta Armamentarii, 2026, 47(4): 250250. DOI： 10.12382/bgxb.2025.0250.

摘要

近年来，无透镜人脸识别作为生物特征认证在军事场景中的关键技术之一，因其低成本、小型化和隐私保护优势，在战场环境、生物安全防护及军事安防等领域具有广泛的应用潜力。然而，现有基于深度学习的方法普遍存在训练成本高、识别准确率较低等问题，限制了其在复杂军事场景中的实际应用。为解决上述问题，提出一种基于嵌入的无透镜成像人脸识别方法，通过优化网络结构提高模型性能。该方法采用教师-学生分支网络架构，通过教师分支网络监督学生分支网络的训练，降低学生分支网络的训练难度，显著减少训练成本。为进一步提升识别精度，在不同深度的卷积层间加入特征差异损失函数，用于约束教师分支和学生分支之间的中级与高级特征差异。实验结果显示，与VGGFace方法相比，该方法在无透镜人脸图像上的识别准确率提升了5.45%，同时加快了学生分支网络的学习速度，有效解决了无透镜人脸识别技术中的核心问题，为推动相关技术的应用提供了新的解决方案。

Abstract

In recent years

lensless face recognition has emerged as a key biometric authentication technology in military applications. It holds significant potential for applications in battlefield environments

biosecurity defense and military security due to its low cost

miniaturization and enhanced privacy protection. However

the existing deep learning-based methods often suffer from high training costs and low recognition accuracy

which limits their practical deployment in complex military scenarios. To address these issues

an embedding-based lensless face recognition net (LFRNet) method is proposed to enhance model performance through an optimized network structure. The method employs a teacherstudent network architecture

where the teacher branch network supervises the training of the student branch network

effectively reducing the training complexity of the student branch network and significantly lowering the training costs. To further improve the recognition accuracy

the feature discrepancy loss functions are incorporated between the different depth convolutional layers to constrain the intermediate and high-level feature differences between the teacher and student branches. Experimental results demonstrate that the proposed method achieves a 5.45% improvement in recognition accuracy on lensless face images compared to the VGGFace method

while also accelerating the learning process of the student branch network. The proposed method effectively addresses core issues in lensless face recognition

offering a novel solution to advance the application of this technology.

关键词

Keywords

references

BOOMINATHAN V, ADAMS J K, ASIF M S, et al. Lensless imaging:a computational renais-sance [J]. IEEE Signal Processing Magazine, 2016, 33 (5): 23-35.

ASIF M S, AYREMLOU A, SANKARANARAYANAN A. Flatcam:thin, lensless cameras using coded aperture and computation [J]. IEEE Transactions Computational Imaging, 2017, 3 (3): 384-397.

SHIMANO T, NAKAMURA Y, TAJIMA K, et al. Lensless lightfield imaging with Fresnel zone aperture: quasi-coherent coding [J].Applied Optics, 2018, 57 (11): 2841-2850.

WU J C, CAO L C, BARBASTATHIS G. DNN-FZA camera:a deep learning approach toward broadband FZA lensless imaging [J]. Optics Letters, 2021, 46 (1): 130-133.

ANTIPA N, KUO G, HECKEL R, et al. DiffuserCam: lensless single-exposure 3D imaging [J]. Optica, 2018, 5:1-9.

PAN X X, NAKAMURA T, CHEN X, et al. Lensless inference camera:incoherent object recognition through a thin mask with LBP map generation [J]. Optics Express, 2021, 29 (7): 9758-9771.

PAN X X, CHEN X, NAKAMURA T, et al. Incoherent reconstruction-free object recognition with mask-based lensless optics and the transformer [J].Optics Express, 2021, 29 (23): 37962-37978.

ZHANG Y E, WU Z Y, LIN P Y, et al. Text detection and recognition based on a lensless imaging system [J]. Applied Optics, 2022, 61 (14): 4177-4186.

TAN J, NIU L, ADAMS J K, et al. Face detection and verification using lensless cameras [J]. IEEE Transactions on Computational Imaging, 2018, 5 (2): 180-194.

GERA D, BALASUBRAMANIAN S. Facial expression recognition with flatcam lensless imaging [C]∥Proceedings of International Conference on Data Science and Applications. Singapore:Springer, 2022, 2:661-677.

WU M H, LEE Y T C, TIEN C H. Lensless facial recognition with encrypted optics and a neural network computation [J]. Applied Optics, 2022, 61 (26): 7595-7601.

NGUYEN CANH T, NAGAHARA H. Deep compressive sensing for visual privacy protection in flatcam imaging[C]∥Proceedings of the IEEE/CVF International Conference on Computer Vision Workshops. Seoul, Korea:IEEE, 2019.

HENRY C, ASIF M S, LI Z. Privacy-preserving face recognition and verification with lensless camera [J]. IEEE Transactions on Biometrics, Behavior, and Identity Science, 2025, 7 (3): 354-367.

马子垚.基于深度学习的低分辨率人脸识别研究[D].哈尔滨：哈尔滨工业大学，2021.

MA Z Y.Research on low-resolution face recognition based on deep learning [D]. Harbin: Harbin Institute of Technology, 2021. (in Chinese)

王晨.基于知识蒸馏的轻量级低分辨率人脸识别方法[D].赣州：赣南师范大学，2023.

WANG C.Lightweight low-resolution face recognition method based on knowledge distillation [D]. Ganzhou: Gannan Normal University, 2023. (in Chinese)

WANG M, DENG W H. Deep face recognition with clustering based domain adaptation [J]. Neurocomputing, 2020, 393:1-14.

LIU W Y, WEN Y D, YU Z D, et al. Sphereface: deep hypersphere embedding for face recognition:arXiv:1704.08063 [R].Ithaca, NY, US:Cornell University, 2017:1704.08063.

DENG J K, GUO J, XUE N N, et al. Arcface: additive angular margin loss for deep face recognition[C]∥Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Long Beach, CA, US:IEEE, 2019:4690-4699.

HUANG Y G, WANG Y H, TAI Y, et al. Curricularface:adaptive curricu-lum learning loss for deep face recognition: arXiv:2004.00288 [R].Ithaca, NY, US: Cornell University, 2020:2004.00288.

HUANG Z H, YANG S Z, ZHOU M C, et al. Feature map distillation of thin nets for low-resolution object recognition [J]. IEEE Transactions on Image Processing, 2022, 31:1364-1379.

浏览量

107

下载量

CNKI被引量

文章被引用时，请邮件提醒。

提交

工具集

关联资源

基于迭代式师生训练的协同区域覆盖算法

面向战场低空目标的轻量化识别方法

面向无人装备的智能边缘计算软技术分析

面向多物理特征地形的特种车辆能耗预测协同优化算法

融合不确定性感知的非结构场景可通行地图构建方法研究