fig7

Figure 7. (A) The outputs of each layer of a typical convolutional network architecture applied to the image of a Samoyed dog; each rectangular image is a feature map corresponding to the output for one of the learned features, detected at each image position^[64]. Copyright 2015 Springer Nature. (B) The schematic view of all-metal species quantification process. (C) Architecture of BT-Net, which includes the main encoder-decoder framework, the boundary attention module to extract the edge information of nanoparticles and the fusion module. (D) Visualized performance of electron-microscopy-based atom recognition statistics (EMARS) analysis procedures applied on a HAADF image of Pt/Al₂O₃ reforming catalyst^[65]. Copyright 2021, American Chemical Society. (E) Visualization of nanoparticle identification results with core-shell structure. (C and E) have been reproduced from ref.^[66] Copyright 2021, Elsevier.