DDDNet: A lightweight and robust deep learning model for accurate segmentation and analysis of TEM images

Abstract

The primary aim of this study was to develop an optimal, lightweight model for the segmentation of transmission electron microscopy (TEM) images. Our model is designed with a minimal parameter count, superior performance metrics, and robust adaptability to variations in substrates, nanoparticle sizes, and nanomaterial diversity within TEM images. In achieving this, we benchmarked our model against four deep learning models using subsets from the Bright-Field TEM(BF-TEM) and Au-TEM datasets. Our model demonstrated exceptional segmentation performance, requiring only 0.34 M parameters and 39.33 G floating-point operations. It also provided the most accurate estimates of average nanoparticle sizes, closely matching true labeled values. These results confirm the model’s proficiency and precision in TEM image processing and introduce a powerful tool for nanoscale image analysis. Our work sets a new standard for lightweight and efficient TEM segmentation models, paving the way for future advancements in nanotechnology research.