Enabling artificial intelligence capabilities in Savu software

Savu is a Python package, developed in the Data Analysis Group, used to process and reconstruct parallel-beam tomography data. Data processing pipelines in Savu can be built using plugins stacked together to form a sequential chain of processing events, e.g., reconstruction, filtering, data correction etc. The plugins are written to work with data from different imaging modalities and beamlines.

Tomography data processing is experiment and sample dependent, making it problematic for inexperienced users and time consuming for beamline staff. With the choice of plugins in Savu broadening and the amount of data collected increasing, particularly with Diamond II upgrade in mind, it is important for the community to look towards automation. Now is a suitable time to capitalise on the amount of Machine Learning/Artificial Intelligence research in progress both on campus and further afield, to investigate the potential to automatically determine which plugins are suitable for which data.

One of such tasks can be automatically identifying noise levels and type of artifacts (outliers) in the data, to create bespoke processing for a particular sample and experimental setup. This would have a significant impact on the tomography user experience, and ensure we are using state-of-the-art processing techniques to remain at the forefront of scientific research.

In the past there have been successful attempts by a Year in Industry (YiI) student to automate data analysis when using deep learning apparatus and convolutional neural networks for semantic image segmentation of tomographic data. Therefore, this work will be partially a continuation of the already existing project.

The selected candidate will be communicating and working closely with another YiI student, who will be working on a similar topic.