Search

The hybrid input-output algorithm is a phase retrieval method that provides solution for the phase problem of coherent X-ray diffraction imaging of micro- and nano-objects from the diffraction pattern alone without using any focusing optics. In this paper, we have studied a tolerance of this algorithm to missing information at the center of the diffraction pattern, which is a frequent problem in problems of small-angle scattering. We considered the particular problem of the stability of the algorithm in the case of scattering from an ordered structure and provided a qualitative and quantitative description of the degradation of image reconstruction with an increase in the amount of missing information.

The paper presents the results of application of machine learning methods: principle component analysis and support vector machine for classification of diffraction images produced in experiments at free-electron lasers. High efficiency of this approach presented by application to simulated data of adenovirus capsid and bluetongue virus core. This dataset were simulated with taking into account the real conditions of the experiment on lasers free electrons such as noise and features of used detectors.