Full-field transmission X-ray microscopy (TXM)–X-ray absorption near-edge structural (XANES) (TXM-XANES) is an imaging method that combines TXM and XANES. By measuring the TXM images of multiple energy points before and after the K-edge of the element of interest, the distribution of elemental chemical states in the sample can be determined. Conventional TXM-XANES data requires the acquisition of images and background images at each energy point, which results in a large data volume and extended acquisition time. At the nanoscale, the instability of the mechanical structure and the movement of the sample may impact the TXM-XANES data analysis.

This study aims to use machine learning methods to achieve background-image sequence prediction modeling using only two spectral background images to reduce the data volume and shorten the acquisition time.

Machine learning, polynomial regression, and linear interpolation were used to generate background image sequences. A prediction model of the complex linear relationship between image grayscale values, pixel points, energy, and other related features based on the known data was established. Subsequently, the entire background image sequence could be predicted using only two spectral background images. Finally, 2D energy distribution maps obtained by conventional TXM-XANES method and this improved TXM-XANES method for standard powder samples and lithium battery cathode material sample were compared and analyzed in details.

The proposed method achieves complete background-image sequence prediction modeling using only two spectral background images. The comparison results show that the proposed method requires a lower data volume and shorter acquisition time than the conventional TXM-XANES methods, which can significantly improve the experimental efficiency of TXM-XANES.

This study addresses the issues of prolonged data gathering time and poor experimental efficiency in TXM-XANES by developing a machine learning model that builds complex linear relationships between pixel values and related features, such as location and color, using machine learning. Using the two TXM-XANES background images for full-sequence background prediction achieves rapid prediction of the entire background image sequence.