Fig. 1. Neuron model

Fig. 2. Convolution operation diagram

Fig. 3. Pooling operation diagram

Fig. 4. Cross image taken far and near the center of curvature radius of the measured mirror

Fig. 5. Cross image of contaminated mirror

Fig. 6. Image preprocessing process

Fig. 7. Three ways of annotation

Fig. 8. Structural diagram of convolutional neural network

Fig. 9. Change of loss value under different marking methods

Fig. 10. Five-fold cross validation to evaluate model performance

Fig. 11. Loss of the five-fold cross validation test set

Fig. 12. Key point prediction and center point calculation of four different types

Fig. 13. Principle of SUSAN algorithm