About AI Analysis Tools

Using our in house machine learning models, the site is able to utilize dedicated GPUs to provide enhanced images for all datasets uploaded to NRDS. The images can be generated and downloaded or they can be compared to the original image on the site.

Super Resolution

This implements a transfer learning approach using a GAN model that is nearly identical to that in Haan et al. [ de Haan, K., Ballard, Z. S., Rivenson, Y., Wu, Y. & Ozcan, A. Resolution enhancement in scanning electron microscopy using deep learning. Sci. Reports 9, DOI: 10.1038/s41598-019-48444-2 (2019). ] but where the GAN is trained from a large number of non-scanning electron microscopy images originating from DIV2K, Flickr2K, and OST and then followed by training on 16-bit grayscale images originating from X-ray computed tomography (XCT) on surrogate tristructural isotropic particles.

Activity Detection

This implements a Yolov11 dislocation defect object tracking model that shows how defects evolve over time and produces histograms of the object behavior rate of change.

Dislocation Segmentation

This runs Model B of the Predictive Automation of Novel Defect Anomalies code available at: Idaho National Laboratory GitHub - PANDA.
For addition details see: Wu, M., Sharapov, J., Anderson, M. et al. Quantifying dislocation-type defects in post irradiation examination via transfer learning. Sci Rep 15, 15889 (2025). https://doi.org/10.1038/s41598-025-00238-5

Dislocation Lines

This model detects dislocation lines based on contrast in TEM and SEM images.

HiPerClust

This model works on 3-D point clouds from atom probe tomography and reproducibly identifies clusters of atoms in those datasets with high accuracy using a combination of machine learning and HDBSCAN.

For addition details see:
https://github.com/IdahoLabResearch/HiPerClust