Author Affiliations
1School of Physics and Electronic Engineering, Jiangsu University, Zhenjiang , Jiangsu 212013, China2Department of Biomedical Engineering, Southern University of Science and Technology, Shenzhen , Guangdong 518055, Chinashow less
Fig. 1. Examples of each class in training dataset (left) and testing dataset (right) (a) Red blood cell; (b) polystyrene bead; (c) small lymphocyte; (d) noise map
Fig. 2. Structure of CNN implemented for 4-class phase recognition
Fig. 3. Performance of LeNet-5 model in training process. (a) Loss function value; (b) accuracy
Fig. 4. Phase distributions of red blood cell before and after change. (a) Before change; (b) after change
Fig. 5. Performance of new CNN model in training process. (a) Loss function value; (b) accuracy
Fig. 6. Interferograms of various samples. (a) Red blood cell; (b) polystyrene bead; (c) small lymphocyte; (d) noise map
Fig. 7. Performance of LeNet-5 model in training process. (a) Loss function value; (b) accuracy
Fig. 8. ResNet-17 network architecture. (a) Structure of residual block; (b) total architecture of network
Fig. 9. Performance of LeNet-5 model in training process. (a) Loss function value; (b) accuracy
Fig. 10. Polystyrene bead interferograms with different intensity ratios (reference wave∶object wave). (a) 1∶1; (b) 2∶1; (c) 3∶1; (d) 4∶1
Fig. 11. Polystyrene bead interferograms with different fringe spatial frequencies. (a) 1.05 rad/pixel; (b) 1.57 rad/pixel; (c) 2.09 rad/pixel; (d) 3 rad/pixel
Fig. 12. Performance of LeNet-5 model in training process. (a) Loss function value; (b) accuracy
Confusion matrix | Predicted class |
---|
Red blood cell | Polystyrene bead | Small lymphocyte | No phase object |
---|
Actual class | Red blood cell | 28 | | | 22 | Polystyrene bead | | 48 | 2 | | Small lymphocyte | | 15 | 35 | | No phase object | 6 | | | 44 |
|
Table 1. Confusion matrix of classification results of LeNet-5 model on interferogram testing dataset
Class | Accuracy /% | Recall /% | F1-Score | Overall accuracy /% |
---|
Red blood cell | 82.35 | 56 | 0.667 | 77.5 | Polystyrene bead | 76.19 | 96 | 0.850 | Small lymphocyte | 94.59 | 70 | 0.805 | No phase object | 66.67 | 88 | 0.759 |
|
Table 2. Performance of LeNet-5 model on each class of interferogram testing dataset
Class | Accuracy /% | Recall /% | F1-Score | Overall accuracy /% |
---|
Red blood cell | 100 | 100 | 1.000 | 98 | Polystyrene bead | 100 | 92 | 0.958 | Small lymphocyte | 92.59 | 100 | 0.962 | No phase object | 100 | 100 | 1.000 |
|
Table 3. Performance of ResNet-17 model on each class of interferogram testing dataset
Datasets with different intensity ratios(R∶O) | Different fringe spatial frequency datasets /(rad⋅pixel-1) | Correctly identify number of samples/number of all samples | Accuracy /% |
---|
1∶1 | 1.05 | 863/867 | 453/867 | 99.54 | 52.25 | 2∶1 | 1.57 | 849/867 | 863/867 | 97.92 | 99.54 | 3∶1 | 2.09 | 849/867 | 361/867 | 97.92 | 41.64 | 4∶1 | 3 | 840/867 | 654/867 | 96.89 | 75.43 |
|
Table 4. Performance of ResNet-17 model on interferogram datasets with different fringe spatial frequencies
Class | Accuracy /% | Recall /% | F1-Score | Overall accuracy /% |
---|
Red blood cell | 100 | 100 | 1.000 | 97.25 | Polystyrene bead | 90.09 | 100 | 0.948 | Small lymphocyte | 100 | 89 | 0.942 | No phase object | 100 | 100 | 1.000 |
|
Table 5. Performance of ResNet-17 model on each class of interferogram testing dataset with multiple spatial frequencies