Author Affiliations
1Institute of Information Science, Beijing Jiaotong University, Beijing 100044, China2Beijing Key Laboratory of Advanced Information Science and Network Technology, Beijing 100044, China3Photonics Research Institute, Department of Electronic and Information Engineering, The Hong Kong Polytechnic University, Hong Kong 999077, Chinashow less
Fig. 1. Basic structure of GAN model
Fig. 2. Structure of SCGAN
Fig. 3. Experimental device diagram of BOTDA sensing system
Fig. 4. BGS data. (a) Experimental BGS (last 640 m); (b) simulated BGS; (c) cropped experiment BGS
Fig. 5. Test data noise reduction results of different networks. (a) (d) DnCNN; (b) (e) ADNet; (c) (f) BRDNet
Fig. 6. Temperature distribution obtained after denoising test data using BRDNet trained on SCGAN noise
Fig. 7. Comparison of noise reduction effects of three convolutional networks training by two noise on test data at different temperatures. (a) (d) DnCNN; (b) (e) ADNet; (c) (f) BRDNet
Fig. 8. Comparison of noise reduction effect of three convolutional networks training by two noise on test data with different averaging times. (a) (b) DnCNN; (c) (d) ADNet; (e) (f) BRDNet
Fig. 9. Noise analysis. (a) Noisy data; (b) noise generated by SCGAN; (c) Gaussian noise; (d) noise comparison along frequency direction; (e) noise comparison along the direction of sampling point
Fig. 10. Noise histograms. (a) Gaussian noise; (b) SCGAN noise
Fig. 11. Noise amplitude spectra. (a) Gaussian noise; (b) SCGAN generated noise; (c) noise obtained from collected data
Temperature /℃ | Raw SNR /dB | SNR in DnCNN /dB | | SNR in ADNet /dB | | SNR in BRDNet /dB |
---|
Gaussian | SCGAN | | Gaussian | SCGAN | | Gaussian | SCGAN |
---|
26 | 13.5075 | 27.7086 | 28.0158 | | 27.0201 | 28.6580 | | 28.2479 | 29.6317 | 40 | 13.6129 | 27.8218 | 28.2357 | | 27.2484 | 28.8876 | | 28.3616 | 29.1959 | 50 | 13.5443 | 27.6914 | 28.1440 | | 27.1203 | 28.7308 | | 28.2401 | 29.6505 | 60 | 13.6348 | 27.8036 | 28.1672 | | 27.1721 | 28.7854 | | 28.3438 | 29.5065 | 70 | 12.3321 | 26.9403 | 28.0238 | | 26.4081 | 28.2484 | | 27.5273 | 29.4539 |
|
Table 1. SNR comparison of experimental data at different temperatures by denoising networks trained on two types of noise
Averaging times | Raw SNR /dB | SNR in DnCNN /dB | | SNR in ADNet /dB | | SNR in BRDNet /dB |
---|
Gaussian | SCGAN | | Gaussian | SCGAN | | Gaussian | SCGAN |
---|
1 | -5.4766 | 3.9548 | 3.8272 | | 4.1264 | 3.3162 | | 4.4908 | 4.3208 | 5 | -4.5424 | 7.7427 | 11.4619 | | 8.5955 | 9.5528 | | 8.5702 | 9.7921 | 10 | -2.3383 | 11.1577 | 14.2687 | | 11.9138 | 12.7696 | | 11.5519 | 14.9751 | 25 | 1.0609 | 14.6377 | 17.4872 | | 15.6181 | 16.4871 | | 15.0304 | 17.5681 | 50 | 3.8668 | 17.4716 | 20.3283 | | 18.6267 | 19.4804 | | 17.9637 | 20.5342 | 100 | 6.6848 | 20.7458 | 23.2891 | | 21.9693 | 22.7074 | | 21.3077 | 23.5499 | 150 | 8.4353 | 21.6659 | 24.1926 | | 22.9131 | 23.6646 | | 22.2674 | 24.5940 | 200 | 9.6116 | 22.8649 | 25.2255 | | 24.0988 | 24.8258 | | 23.4732 | 25.6493 | 250 | 10.5648 | 23.7411 | 25.9811 | | 24.9672 | 25.7109 | | 24.3580 | 26.4586 | 500 | 13.5075 | 27.7086 | 28.0158 | | 27.0201 | 28.6580 | | 28.2479 | 28.7317 |
|
Table 2. SNR comparison of experimental data of different averaging times by denoising networks trained on two types of noise