Fig 1.
Overall flow chart, the process structure.
Fig 2.
Traditional U-net network structure.
Fig 3.
Improved U-net network structure diagram.
Table 1.
Proportion of different scene categories in the data.
Fig 4.
Hybrid expansion convolution structure diagram.
Fig 5.
Sensory field changes before and after.
Fig 6.
Histogram changes before and after equalization.
Fig 7.
Improved loss, IOU and Dice of model training.
Fig 8.
IOU changes across different models.
Table 2.
Comparison of the metrics between the network model proposed in this paper and other network models.
Fig 9.
Comparison of segmentation results of each algorith.
Table 3.
Time comparison of several network models.
Table 4.
Comparison of maximum detail coefficient under different defects.
Fig 10.
Grayscale map of ROI area.
Fig 11.
Comparison of the histograms of Image 2.
Fig 12.
Comparison of the histograms of Image 1.
Fig 13.
Flow of defect detection.
Fig 14.
Image comparison before and after correction.
Fig 15.
Frequency curve obtained by projection.
Fig 16.
Steps for detecting anomalies.
Fig 17.
Analytical filter bank.
Fig 18.
Transformation trend of diferrent signal.
Table 5.
Comparison of maximum detail coefficient under different defects.
Fig 19.
Comparison of normal, deformed and damaged wavelengths.
Table 6.
The d-max range of different types of defects.
Table 7.
Experimental results of defect detection.
Fig 20.
Detail coefficients for various diseases.