Aluminum Plate Defect Image Segmentation Using Improved Generative Adversarial Networks for Eddy Current Detection

Fig. 8. Segmentation results of different methods. (a) Original image; (b) truth image; (c) Otsu method; (d) FCN-8s model; (e) FCN-32s model; (f) U-Net model; (g) proposed method

Download full size

Fig. 9. Segmentation results of eddy current testing images under different signal-to-noise ratios. (a) Original image; (b) truth image; (c) Otsu method; (d) FCN-8s model; (e) FCN-32s model; (f) U-Net model; (g) proposed method

Download full size

Table 1. Comparison of segmentation results using different methods
View table
Table 1. Comparison of segmentation results using different methods
Method Precision Recall F1
Otsu 0.9006 0.6452 0.7518
FCN-8s 0.8640 0.7910 0.826
FCN-32s 0.7840 0.8434 0.813
U-Net 0.7609 0.8502 0.803
Proposed method 0.8799 0.9775 0.926

Table 2. Comparison of segmentation methods for eddy current testing image under different signal-to-noise ratios

View table

Table 2. Comparison of segmentation methods for eddy current testing image under different signal-to-noise ratios

Method	Signal-to-noise ratio	Precision	Recall	F1
Otsu	50 dB	0.6648	0.8259	0.7366
	60 dB	0.8588	0.7936	0.8249
	70 dB	0.8222	0.8142	0.8182
FCN-8s	50 dB	0.7120	0.7835	0.7460
	60 dB	0.8155	0.8968	0.8539
	70 dB	0.9369	0.8011	0.8637
FCN-32s	50 dB	0.7773	0.8962	0.8325
	60 dB	0.7348	0.9050	0.8111
	70 dB	0.9155	0.7464	0.8223
U-Net	50 dB	0.6994	0.7952	0.7442
	60 dB	0.8685	0.8166	0.8418
	70 dB	0.9236	0.7882	0.8505
Proposed method	50 dB	0.8725	0.9694	0.9184
	60 dB	0.8696	0.9894	0.9256
	70 dB	0.8710	0.9794	0.9220

Tools

Get Citation

Copy Citation Text

Qi Zhang, Bo Ye, Siqi Luo, Honggui Cao. Aluminum Plate Defect Image Segmentation Using Improved Generative Adversarial Networks for Eddy Current Detection[J]. Laser & Optoelectronics Progress, 2021, 58(8): 0815002

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites