Clustering-Based Detection of Debye–Scherrer Rings

Abstract

Calibration of the X-ray powder diffraction (XRPD) experimental setup is a crucial step before data reduction and analysis, and requires correctly extracting individual Debye–Scherrer rings from the 2D XRPD image. This problem is approached using a clustering-based machine learning framework, thus interpreting each ring as a cluster. This allows automatic identification of Debye–Scherrer rings without human intervention and irrespective of detector type and orientation. Various existing clustering techniques are applied to XRPD images generated from both orthogonal and nonorthogonal detectors, and the results are visually presented for images with varying inter-ring distances, diffuse scatter, and ring graininess. The accuracy of predicted clusters is quantitatively evaluated using an annotated gold standard and multiple cluster analysis criteria. These results demonstrate the superiority of density-based clustering for the detection of Debye–Scherrer rings. Moreover, the given algorithms impose no prior restrictions on detector parameters such as sample-to-detector distance, alignment of the center of diffraction pattern, or detector type and tilt, as opposed to existing automatic detection approaches.