Author Affiliations
1College of Information Science and Engineering, Northeastern University, Shenyang, Liaoning 110004, China2School of Optical-Electrical and Computer Engineering, University of Shanghai for Science and Technology, Shanghai 200093, Chinashow less
Fig. 1. Design of optical experiment platform
Fig. 2. Overall network structure. (a) Conv block; (b) dense block; (c) DenseUnet; (d) DCNN
Fig. 3. ReLU nonlinear activation function
Fig. 4. Physical diagram of multimode optical fiber imaging system
Fig. 5. Digital images in imaging system performance test experiment
Fig. 6. Results of reconstruction and classification
Fig. 7. Classification accuracy of training set and validation set for reconstructed images under 0.1 m fiber
Fig. 8. Reconstruction accuracy under different fiber lengths. (a) 0.1 m; (b) 10 m
Fig. 9. Classification accuracy of different fiber lengths in training set and validation set. (a) 0.1 m; (b) 10 m
Fig. 10. Results of speckle reconstruction under different fiber lengths
Fig. 11. Comparison of reconstruction results of different network models. (a) Speckle images output from multimode fiber; (b) reconstruction results of DenseUnet; (c) reconstruction results of traditional Unet; (d) original image (ground truth)
Fiberlength /m | Model | Classification accuracy /% |
---|
Speckleimage | Reconstructedimage |
---|
0.1 | DCNN | 91.5±0.4 | 98.2±0.2 | 1 | DCNN | 90.2±0.3 | 97.6±0.2 | 10 | DCNN | 86.6±0.6 | 97.2±0.5 | 20 | DCNN | 76.4±1.1 | 97.0±0.4 |
|
Table 1. Classification accuracy of different fiber lengths using two input modes
Fiber length /m | Index | DenseUnet | Unet | DenseUnet- |
---|
| SSIM | 0.898 | 0.719 | 0.658 | 0.1 | PSNR | 41.943 | 34.022 | 11.978 | | Classification accuracy /% | 98.20 | 95.33 | 85.25 | | SSIM | 0.790 | 0.712 | 0.622 | 1 | PSNR | 36.440 | 28.108 | 11.402 | | Classification accuracy /% | 97.60 | 92.67 | 79.67 | | SSIM | 0.780 | 0.540 | 0.594 | 10 | PSNR | 36.072 | 12.125 | 11.344 | | Classification accuracy /% | 97.20 | 73.33 | 68.72 | | SSIM | 0.760 | 0.422 | 0.517 | 20 | PSNR | 35.943 | 11.917 | 13.002 | | Classification accuracy /% | 97 | 53 | 68.67 |
|
Table 2. Reconstruction quality of different models