Fig. 1. Connection mechanism of DenseNet

Fig. 2. Structure of DenseNet

Fig. 3. Structure of dense block

Fig. 4. Structures of bottleneck layer and transition layer. (a) Structure of bottleneck layer; (b) structure of transition layer

Fig. 5. Scattering medium

Fig. 6. Experimental device

Fig. 7. Diagram of polarization state control

Fig. 8. Structure of neural network

Fig. 9. Speckle pattern formed after scattering by ZnO with the same initial phase and different polarized light. (a) θ=20.2°; (b) θ=20.4°; (c) θ=20.6°; (d) θ=20.8°

Fig. 10. Speckle pattern of light scattered by ZnO with different initial phase and the same polarization state. (a) φ₀=0.3π; (b) φ₀=0.6π; (c) φ₀=0.9π; (d) φ₀=1.2π

Fig. 11. Simulation diagram of effect of learning rate on model training

Fig. 12. Recognition accuracy curve of test data during training. (a) Δθ=1.0°; (b) Δθ=0.5°; (c) Δθ=0.2°

Fig. 13. Recognition accuracy curve of test data after fixed learning rate. (a) Learning rate is 0.001; (b) learning rate is 0.00001