Ultrasound image super-resolution reconstruction based on semi-supervised CycleGAN

Area of Science:

• Non-destructive testing
• Artificial intelligence in engineering
• Signal processing

Background:

• Diffraction artifacts in ultrasonic testing complicate accurate defect characterization.
• Quantitative analysis of defects is crucial for structural integrity assessment.

Purpose of the Study:

• To develop a deep learning model for accurate defect morphology characterization in ultrasonic testing images.
• To eliminate diffraction artifacts and enhance defect reconstruction precision.

Main Methods:

• Proposed a label-enhanced semi-supervised CycleGAN network (LESS-CycleGAN) for cross-domain image transformation.
• Introduced a novel authenticity loss function for high-precision defect reconstruction.
• Utilized simulated and actual ultrasonic phased array images for validation.

Main Results:

• LESS-CycleGAN effectively eliminates diffraction artifacts from ultrasonic testing images.
• Achieved subwavelength-scale defect reconstruction with high precision.
• Demonstrated convenience, effectiveness, and robustness in experimental validation.

Conclusions:

• The LESS-CycleGAN model offers a robust and effective solution for artifact reduction and defect characterization in ultrasonic testing.
• The method shows significant potential for improving quantitative defect analysis in non-destructive evaluation.