Diffraction Image Screening of Radiation Facilities Based on Privacy Protection Mechanism

Kang Xu; Yongxin Zhu; Bo Wu; Xiaoying Zheng; Lingyao Chen

doi:10.3788/LOP220950

Laser & Optoelectronics Progress, Volume. 60, Issue 10, 1010020(2023)

Diffraction Image Screening of Radiation Facilities Based on Privacy Protection Mechanism

Kang Xu^1,2, Yongxin Zhu^2、*, Bo Wu², Xiaoying Zheng², and Lingyao Chen³

Author Affiliations

¹School of Integrated Circuits, University of Chinese Academy of Sciences, Beijing 100049, China

²Shanghai Advanced Research Institute, Chinese Academy of Sciences, Shanghai 201210, China

³Shanghai Information Technology Research Center, Shanghai 201210, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(14)

Fig. 1. Federated learning model

Download full size

$Crystal diffraction images. (a) Including a lot of Bragg spots; (b) including a few Bragg spots$

Fig. 2. Crystal diffraction images. (a) Including a lot of Bragg spots; (b) including a few Bragg spots

Download full size

Fig. 3. FedKL model

Download full size

Fig. 4. Process of mixed training method

Download full size

Fig. 5. Client data distribution

Download full size

Fig. 6. Comparison of accuracy of different models under FedAvg and FedKL

Download full size

Fig. 7. Comparison of running time and size of different models under FedAvg and FedKL

Download full size

Fig. 8. Comparison of precision of different models under FedAvg and FedKL

Download full size

Fig. 9. Comparison of recall of different models under FedAvg and FedKL

Download full size

Fig. 10. Comparison of F1 score of different models under FedAvg and FedKL

Download full size

Fig. 11. Time comparison between synchronized training and mixed training under FedAvg (e.g. DenseNet121)

Download full size

Table 1. L498 data distribution
View table
Table 1. L498 data distribution
Label Number of images Proportion /%
Hit 148 7.4
Miss 1355 67.76
Maybe 498 24.9

Table 2. Test set data distribution
View table
Table 2. Test set data distribution
Label Number of images Proportion /%
Hit 66 16.5
Miss 190 47.5
Maybe 144 36.0

Table 3. Classification results on the test set (e.g. DenseNet121)
View table
Table 3. Classification results on the test set (e.g. DenseNet121)
Model Predict
Hit or Maybe Miss
FedAvg-DenseNet121 Label Hit or Maybe 30.0% 70.0%
Miss 5.3% 94.7%
FedKL-DenseNet121 Label Hit or Maybe 84.8% 15.2%
Miss 12.6% 87.4%

Tools

Get Citation

Copy Citation Text

Kang Xu, Yongxin Zhu, Bo Wu, Xiaoying Zheng, Lingyao Chen. Diffraction Image Screening of Radiation Facilities Based on Privacy Protection Mechanism[J]. Laser & Optoelectronics Progress, 2023, 60(10): 1010020

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites