Unsupervised Classification of Polymers Using Low‐Loss Spectral Features in Scanning Transmission Electron Microscopy With Electron Energy‐Loss Spectroscopy

ABSTRACT This study presents a classification model for nanoscale polymer characterization, utilizing low‐loss spectral data obtained through scanning transmission electron microscopy combined with electron energy‐loss spectroscopy. To enhance interpretability, spectral features are extracted via mixed Gaussian model fitting, where each Gaussian peak corresponds to a specific chemical bond state, facilitating the identification of key descriptors. These features enable effective clustering of seven standard polymers, achieving clear statistical separation based on four principal features. Furthermore, the extracted features are employed to visualize the progression of electron irradiation damage within an adequate data space. Feature changes under varying irradiation conditions are significantly correlated with alterations in the summed spectra. This methodology holds promise for advancing qualitative analyses of polymer alloys without electron staining, allowing detailed investigations into local denaturation and the presence of reaction layers. With the expansion of standard datasets, this approach offers a robust framework for characterizing and understanding polymer behavior at the nanoscale.