Abstract. Invention of Coherent X-ray Diffractive Imaging (CXDI) technique allows to reconstruct inner structure of nanoparticles, such as proteins and viruses, with 1 Å resolution. In CXDI experiments, free electron laser radiation scatters at sample of object under study and a diffraction image is recorded. On a basis of many recorded diffraction images, original 3D structure is reconstructed. However, not all diffraction images can be used for reconstruction, many images are empty, others contain diffraction pattern of some contaminant or include contributions of several particles. Therefore, classification of recorded images by structure type becomes an important step of data analysis. This paper presents a comparison of several approaches for images classification by the structure type. The comparison was performed on different experimental datasets. New European X-ray Free-Electron Laser (XFEL) will start operating in 2017; it will allow collecting up to 27,000 diffraction images per second. The possibility of image classification in European XFEL experiments at the rate of data collection was also investigated with considered approaches.

 

Key words: coherent X-ray diffractive imaging, biological particles, correlation coefficients, support vector machine, k-means, spectral clustering, multilayer perceptron, convolutional neural network.