Fig. 1. Schematic of SEM

Fig. 2. Schematic of DNN-KE method

Fig. 3. Preparation of sample set. (a) 36 character samples; (b) SEM test

Fig. 4. Parameter adjustment analysis of activation function and DNN layer size. (a) Loss function value; (b) training time

Fig. 5. Parameter adjustment analysis of DNN layer number and layer size. (a) Loss function values; (b) training time

Fig. 6. Convergence of DNN training

Fig. 7. Optical simulation comparison. (a) SEM; (b) DNN-KE