Fast Segmentation Of Synchrotron Tomography Data Using Convolutional Neural Networks

Kurzfassung

The continuously increasing brilliance of synchrotron sources as well as the use of fast imaging detectors gives material scientists access to a large amount of three- or four-dimensional data. Such microstructural data are important to investigate the relationships between microstructure and properties of materials. However, human-based segmentation of tomographic images can be a tedious time-consuming task and often represents a bottleneck in the research process. Deep learning algorithms and, particularly, convolutional neural networks are state-of-the-art techniques for pattern recognition in digital images, and their use in materials science is explored. However, their application needs to be adapted to the specific needs of this field. In our work, a convolutional neural network was trained to segment the microstructural components of an Al-Si cast alloy imaged using synchrotron X-ray tomography. A pixel-wise weighted error function is implemented to account for microstructural features which are difficult to identify in the topographies but play a relevant role for the correct description of the 3D architecture of the investigated alloy. The results show that the total working time for the segmentation with the trained convolutional neural network was reduced to <1% of the time needed for human-based segmentation.