Enhancement of spatially partially coherent diffraction patterns using neural blind deconvolution

X-ray ptychography technology, a variant of coherent diffraction imaging, enables high-resolution imaging of large samples at the nanoscale by virtue of probe overlapping scanning, which is widely used in the fields of medicine, biology, and materials science. However, the spatially partially coherent illumination severely limits its imaging quality, especially when using laboratory sources. In this regard, we propose a generalized data enhancement technique that, for the first time, to our knowledge, enhances spatially partially coherent diffraction patterns by an unsupervised neural blind deconvolution network, thus compensating for the impact of decoherence effects on imaging quality. The method can flexibly adapt to the enhancement needs of different coherence conditions and does not require any prior knowledge. It also does not require fully coherent diffraction patterns as labeled data. At the same time, the enhanced diffraction patterns can be applied to arbitrary ptychography reconstruction algorithms. Experimental results demonstrate that the enhanced diffraction patterns improve the reconstruction quality of the single-mode algorithm. Moreover, applying these enhanced patterns to a spatially partially coherent reconstruction algorithm further improves the spatial resolution of the recovered images. This work will be useful for conducting X-ray ptychography experiments under low-coherent illumination conditions, such as those using laboratory sources.