Characterizing soot in TEM images using a convolutional neural network

Soot is an important material with impacts that depend on particle morphology. Transmission electron microscopy (TEM) represents one of the most direct routes to qualitatively assess characteristics. However, producing quantitative information requires robust image processing tools, which complicated by low contrast and complex aggregated morphologies characteristic soot. The current work presents a new convolutional neural network explicitly trained characterize soot, using pre-classified images particles from natural gas engine; laboratory flare; marine engine. results are compared against other existing classifiers before considering effect have automated primary size methods. Estimates overall uncertainties between fully approaches aggregate characterization range 25% in dp,100 85% DTEM. A consistent correlation observed projected-area equivalent diameter across all techniques.