Neural network segmentation methods for fatigue crack images obtained with X-ray tomography

Synchrotron X-ray tomography allows to observe fatigue crack propagation during in situ tests. Accurately segmenting the 3D shape of the cracks from the tomography image is essential for quantitative analysis. Fatigue cracks have small openings which result in low contrast images making crack segmentation difficult. Phase contrast available at synchrotron sources improves crack detection but it also increases the complexity of the image and human intervention is generally used to help traditional segmentation methods. In this work, an image segmentation method based on a convolutional neural network is developed to replace the user interpretation of images. Combined with a 'Hessian matrix' filter, this method can successfully extract 3D shapes of internal fatigue cracks in metals.

Automated summary

In this study, an image segmentation method based on convolutional neural network is developed to successfully extract the 3D shapes of internal fatigue cracks in metals, combined with a 'Hessian matrix' filter.