3D X-ray µCT is a powerful non-destructive tool for studying materials. Image processing is a necessary step for extracting quantitative information from the images. Image segmentation, a commonly used image processing workflow, is to classify the image voxels into different categories based on their grayscale values. A proper image segmentation provides the basis for further analysis. However, this process is prone to operator bias. This is because most of the existing methods for segmentation rely on thresholding in some way. Different operators may choose different thresholding limits, leading to biased results. 

Machine learning algorithms enable computers to learn from and make predictions based on data. Employing machine learning in image processing potentially has many advantages. Trained machine learning models can perform tasks without human interference, reducing operator bias. However, multiple challenges exist, including generating high-quality training data, building the most efficient model architect for a specific task, optimizing the results when using transfer learning etc. The X-ray tomography and applications group works closely with partners in the mining and steelmaking industry to reduce the carbon footprint. The collaboration provides more opportunities for the group but also brings new challenges: the large amount of data and the complex materials. Fast and accurate material characterization using X-ray µCT and machine learning will greatly help the industry to realize a sustainable future. 

In this project, the student will work to develop machine learning-based image processing workflows. The student may explore methods for generating training data, test different machine learning algorithms, develop bespoke models. 

Computer science, image processing, some experience in programming (e.g., python)