Fig. 1. Characterization of MoS₂ sample
(a), (d): MoS₂ optical images; (b), (e): AFM images; (c), (f): Thicknesses obtained by AFM analysis

Fig. 2. Partial labels of MoS₂ nanosheets marked by different thickness ranges

Fig. 3. Processed optical images and corresponding label images

Fig. 4. Diagram of Network structure

Fig. 5. Structure of residual convolution network

Fig. 6. Structure of pyramid pooling model

Fig. 7. Schematic diagram of low-level feature map reuse

Fig. 8. Loss curve and accuracy curve during 2D-Net network training

Fig. 9. Confusion matrix of test results

Fig. 10. The prediction results of the four network structures

Fig. 11. Output results of feature map visualization
(a): MoS₂ optical images and F2 feature maps; (b): MoS₂ optical images and F5 feature maps

Fig. 12. Transfer learning for exfoliated grapheme
(a): Predicted results; (b): Confusion matrix of testing