Fig. 1. Learning and design processes of optical system design based on deep learning

Fig. 2. Architecture of deep learning network

Fig. 3. Path of light ray in main plane of optical array

Fig. 4. Training diagram under different learning rates

Fig. 5. Untrained optical system and trained optical system. (a) Untrained optical system; (b) trained optical system

Fig. 6. Comparison between output glass obtained by network model and glass library

Fig. 7. Grouping of deep learning training data and test data

Fig. 8. Comparison results of four groups of depth learning designed lenses and reference lenses under different apertures at F^# of 14.0, 8.0 and 7.0

Fig. 9. Comparison results of four groups of depth learning designed lenses and reference lenses under different apertures at F^# of 6.4, 5.0, 4.0 and 2.8

Fig. 10. Test results of network model generalization capability. (a) Spot radius thermal diagram of optical system at focal length of 1 mm; (b) spot radius thermal diagram of optical system at focal length of 100 mm; (c) spot radius thermal diagram of optical system at focal length of 1000 mm; (d) success proportion of optical system design at focal length of 1 mm; (e) success proportion of optical system design at focal length of 100 mm; (f) success proportion of optical system design at focal length of 1000 mm