Totally defined nanocatalysis: Detection of polyelement nanoparticles by deep learning

High-entropy alloys (HEAs), which are near-equimolar alloys of four or more metal elements, have long been used to achieve the desired properties of catalytic materials. However, a novel alloying approach that includes multiple principal elements at high concentrations to generate HEAs as novel catalytic materials has been reported. The fabrication of well-defined ultrastable supported HEAs, which provide superior performance and stability of catalysts owing to their augmented entropy and lower Gibbs free energy, remains a critical challenge. Supported HEA catalysts are sophisticated because of the variety of their morphologies and large sizes at the nanoscale. To address these challenges, PtPdInGaP@TiO₂, comprising five different metals, is prepared via ultrasonic-assisted coincident electro-oxidation–reduction precipitation (U-SEO-P). The electronic structure and catalytic performance of HEA nanoparticles (NPs) are studied using hard scanning transmission electron microscopy (STEM), which is the first direct observation of the electronic structure of HEA NPs. This research takes an important step forward in fully describing individual HEA NPs. Combining STEM with deep learning with convolutional neural network (CNN) of selected individual HEA NPs reveals significant aspects of shape and size for widespread and commercially important PtPdInGaP@TiO₂ NPs. The proposed method facilitates the detection and segmentation of HEA NPs, which has the potential for the development of high-performance catalysts for the reduction of organic compounds. © 2025 The Author(s). InfoMat published by UESTC and John Wiley & Sons Australia, Ltd.