Table 1 Summary of literature analysis.
References | Techniques | Observations |
|---|---|---|
Edge-Constrained Neural Fill Process (ECNFP) | ECNFP outperforms current methods by delivering better visual results, especially in complex textures and boundaries | |
Deep learning methodologies | Highly effective across different degradation conditions, with each model offering distinct strengths in handling noise, resolution loss, and complex restoration challenges. | |
Revitalizing Convolutional Network (RCN) | RCN outperforms conventional CNN-based methods, delivering superior quality, detail recovery, and restoration precision. | |
Swin Transformer and ResNet | This approach efficiently captures local and global features, leading to superior image quality in low-light conditions by balancing noise reduction, contrast enhancement, and detail preservation. | |
Vision transformers | Vision transformers deliver improved performance in image restoration by effectively handling complex dependencies and enhancing overall restoration quality. | |
U2-Former | U2-Former enhances feature alignment, reduces inconsistencies, and restores complex details, while CT-U-Net effectively addresses complex tasks like noise reduction and detail recovery, both boosting restoration accuracy and resilience. | |
A Vision Transformer (ViT)-Based Approach | The ViT-based approach demonstrates robustness against variations in lighting, texture, and crack morphology, outperforming traditional CNN methods and showing promise for real-time dam monitoring | |
U-Net architecture | comparably to existing models, effectively identifying and addressing cracks in tunnel linings caused by aging, topography changes, and stress |
