Deep learning-assisted TEM image analysis for morphological characterization of molybdenum disulfide catalysts

The morphology of molybdenum disulfide (MoS₂) plays a key role in its catalytic performance, as active sites are associated with the edges of its layered structure. Transmission electron microscopy (TEM) is widely used to characterize MoS₂ catalysts, but manual image analysis is time-consuming and operator-dependent. Here, a deep-learning-assisted workflow combines YOLO instance segmentation of individual slabs, automatic scale calibration, particle grouping and interactive correction to extract morphological parameters. YOLO26s was selected as a practical compromise between segmentation quality and model complexity following comparison with YOLO26n and YOLO26m. Agreement was evaluated using expert-reviewed slab lengths from one held-out image and independent manual morphological measurements by two experts on five additional images. For 720 paired slab lengths, the mean difference was +0.09 nm and the root-mean-square error was 0.26 nm; these values describe agreement within a single image and with a reference obtained by reviewing the model masks. The workflow supports quantitative analysis of the investigated MoS₂ images, while broader validation across specimens and imaging conditions remains necessary.