Improving Coating Degradation Detection with Generative Networks
Author Information
Author(s): Dominik Stursa, Pavel Rozsival, Petr Dolezel
Primary Institution: Faculty of Electrical Engineering and Informatics, University of Pardubice, Pardubice, Czechia
Hypothesis
Can generative adversarial networks (GANs) effectively augment datasets for better detection of coating degradation?
Conclusion
The study found that using synthetic data generated by GANs significantly improves the performance of models detecting coating degradation.
Supporting Evidence
- GANs were used to generate synthetic data that mimics real-world conditions.
- Models trained with augmented datasets showed improved accuracy and F1-scores.
- The optimal ratio of synthetic to original data for model training was found to be between 0.2 and 0.5.
Takeaway
This study shows that using computer-generated images can help machines better recognize damage on coated surfaces, making it easier to spot problems.
Methodology
The study used deep convolutional generative adversarial networks (DCGAN) to create synthetic input-target pairs for training semantic segmentation models.
Limitations
The method does not allow for direct control over the distribution of different degradation levels within the synthetic dataset.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website