航空学报 > 2025, Vol. 46 Issue (3): 630531-630531   doi: 10.7527/S1000-6893.2024.30531

深空光电测量与智能感知技术专栏

弱光环境下基于深度学习的单光子计数成像去噪方法

赵志浩1, 杨照华1, 吴云2, 余远金3()   

  1. 1.北京航空航天大学 仪器科学与光电工程学院,北京 100191
    2.北京控制工程研究所,北京 100090
    3.北京理工大学 自动化学院,北京 100081
  • 收稿日期:2024-04-16 修回日期:2024-05-23 接受日期:2024-06-12 出版日期:2024-06-19 发布日期:2024-06-17
  • 通讯作者: 余远金 E-mail:yuanjin.yu@bit.edu.cn
  • 基金资助:
    国家自然科学基金(62222304);北京控制工程研究所空间光电测量与感知实验室开放基金(LabSOMP-2022-04)

Single-photon counting imaging denoising method based on deep learning in low-light environment

Zhihao ZHAO1, Zhaohua YANG1, Yun WU2, Yuanjin YU3()   

  1. 1.School of Instrumentation and Optoelectronic Engineering,Beihang University,Beijing 100191,China
    2.Beijing Institute of Control Engineering,Beijing 100090,China
    3.School of Automation,Beijing Institute of Technology,Beijing 100081,China
  • Received:2024-04-16 Revised:2024-05-23 Accepted:2024-06-12 Online:2024-06-19 Published:2024-06-17
  • Contact: Yuanjin YU E-mail:yuanjin.yu@bit.edu.cn
  • Supported by:
    National Natural Science Foundation of China(62222304);Space Optoelectronic Measurement and Perception Laboratory, Beijing Institute of Control Engineering(LabSOMP-2022-04)

摘要:

单像素单光子计数成像灵敏度高的特点使其在弱光环境下探测极具优势,但其重构图像质量依然会随着光信号的减弱而退化,为此设计了基于深度学习去噪的单像素单光子计数成像方法以提高弱光条件下重构图像的信噪比。首先,建立单像素单光子计数成像系统;然后,利用压缩感知算法对图像进行重构;最后,分别使用三维块匹配算法和深度学习算法对重构图像进行去噪,并对2种方法去噪效果进行了比较分析。实验结果表明,三维块匹配算法和深度学习算法都能提高图像信噪比,基于深度学习图像去噪的单像素单光子计数成像方法获得的图像信噪比提升了12.97 dB,极大地提升了弱光环境下的图像信噪比,且优于三维块匹配算法。因此,所提方法为提升单像素单光子成像系统在弱光环境中重构图像质量提供了一种新的思路。

关键词: 单像素成像, 光子计数成像, 图像去噪, 深度学习, 弱光探测

Abstract:

The high sensitivity of single-pixel single-photon counting imaging makes it extremely advantageous in low-light detection, but the quality of the reconstructed images with this method will still degrade with the weakening of light flux. A single-pixel single-photon counting imaging method based on deep learning denoising is designed to improve the signal-to-noise ratio of reconstructed images in low-light environment. Firstly, a single-pixel single-photon counting imaging system is established. Then, the compressed sensing algorithm is used to reconstruct the image. Finally, the 3D block matching algorithm and the deep learning algorithm are used to denoise the reconstructed image, and the denoising effects of the two algorithms are compared. Ther results show that both of the deep learning algorithm and the 3D block matching algorithm can improve the signal-to-noise ratio of the image. The signal-to-noise ratio of the image obtained by the single-pixel single-photon counting imaging method based on deep learning image denoising is increased by 12.97 dB,which has a great increase of the image signal-to-noise ratio in the low-light environment, and has a higher signal-to-noise ratio than that obtained by the 3D block matching algorithm. Therefore, this method provides a new idea for improving the quality of the image reconstructed by the single-pixel single-photon imaging system in the low-light environment.

Key words: single-pixel imaging, photon counting imaging, image denoising, deep learning, low-light detection

中图分类号: