Acta Optica Sinica, Volume. 45, Issue 10, 1011005(2025)
Single-Photon Imaging Algorithms Based on Multi-Scale Fusion Networks
Figures & Tables(11)
Fig. 2. Schematic diagram of multi-scale fusion neural network structure based on attention guidance
Fig. 4. Comparison of reconstruction results on Art scene using different algorithms. (a1) Reconstructed images under SBR of 2∶50; (a2) close-up views of reconstructed images under SBR of 2∶50; (b1) reconstructed images under SBR of 2∶100; (b2) close-up views of reconstructed images under SBR of 2∶100
Fig. 5. Comparison of reconstruction results on Reindeer scene using different algorithms. (a1) Reconstructed images under SBR of 2∶50; (a2) close-up views of reconstructed images under SBR of 2∶50; (b1) reconstructed images under SBR of 2∶100; (b2) close-up views of reconstructed images under SBR of 2∶100
Fig. 6. Comparison of robustness of five deep learning algorithms. (a) RMSE of different algorithms under five SBRs; (b) accuracy of different algorithms under five SBRs for
Fig. 7. Reconstruction results of different algorithms in four real-world scenarios
Fig. 8. Qualitative comparison of ablation experiments on different modules under SBR of 2∶100. (a1) Ground truth; (b1) reconstructed image using single-scale network; (c1) reconstructed image with ablated ADDF module; (d1) reconstructed image with ablated CBAM; (e1) reconstructed image using complete AMSF-Net; (a2)‒(e2) close-up views corresponding to Figs. 8(a1)‒(e1)
Table 1. Quantitative comparison of imaging performances of different algorithms under various signal-to-background ratios
View tableTable 1. Quantitative comparison of imaging performances of different algorithms under various signal-to-background ratios
Condition Method RMSE /m SSIM A /% Number of signal photons is 2, number of background photons is 10, SBR is 0.2 Harmany et al.[ 31 ]3.742 0.005 42.22 49.87 50.81 SPADnet[ 32 ]0.061 0.869 88.97 95.68 96.84 Non-local[ 18 ]0.049 0.926 91.36 98.37 99.18 EeNet[ 19 ]0.089 0.816 85.21 93.38 95.06 NLSA-Encoder[ 20 ]0.091 0.766 23.51 51.66 73.56 AMSF-Net 0.038 0.942 94.30 98.64 99.05 Number of signal photons is 2, number of background photons is 50, SBR is 0.04 Harmany et al.[ 31 ]4.609 0.002 26.69 31.37 32.01 SPADnet[ 32 ]0.064 0.868 88.46 95.56 96.77 Non-local[ 18 ]0.050 0.904 89.31 97.54 98.34 EeNet[ 19 ]0.131 0.807 83.40 92.04 93.94 NLSA-Encoder[ 20 ]0.127 0.642 18.36 37.57 55.71 AMSF-Net 0.044 0.927 92.51 98.02 98.63 Number of signal photons is 2, number of background photons is 100, SBR is 0.02 Harmany et al.[ 31 ]4.905 0.001 18.49 21.95 22.66 SPADnet[ 32 ]0.097 0.857 87.86 94.99 96.30 Non-local[ 18 ]0.062 0.886 87.51 96.87 97.89 EeNet[ 19 ]0.264 0.715 75.16 84.74 87.47 NLSA-Encoder[ 20 ]0.191 0.519 10.30 20.82 31.61 AMSF-Net 0.049 0.913 90.99 97.49 98.26
Table 2. Quantitative comparison of ablation experiment results for different modules
View tableTable 2. Quantitative comparison of ablation experiment results for different modules
SBR RMSE /m Single w/o ADDF w/o CBAM AMSF-Net 2∶10 0.046 0.042 0.041 0.038 2∶50 0.045 0.046 0.044 0.044 2∶100 0.059 0.062 0.053 0.049
Table 3. Quantitative comparison of ablation experiment results for loss functions
View tableTable 3. Quantitative comparison of ablation experiment results for loss functions
SBR RMSE /m OR+TV w/o TV AMSF-Net 2∶10 0.046 0.040 0.038 2∶50 0.051 0.046 0.044 2∶100 0.068 0.051 0.049
Tools
Get Citation
Copy Citation Text
Pengfei Zhou, Yuyang Zhao, Chenghao Jiang, Tianpeng Xie, Yan Jiang, Zhonghe Liu, Jingguo Zhu. Single-Photon Imaging Algorithms Based on Multi-Scale Fusion Networks[J]. Acta Optica Sinica, 2025, 45(10): 1011005
Paper Information
Category: Imaging Systems
Received: Jan. 24, 2025
Accepted: Apr. 2, 2025
Published Online: May. 19, 2025
The Author Email: Jingguo Zhu (zhujingguo@ime.ac.cn)
CSTR:32393.14.AOS250542
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20