Figure 1.
Network structure of deep capsule
TABLE I
STRUCTURE COMPARISON OF CAPSULE NETWORK AND DEEP CAPSULE NETWORK
| Capsule network | Deep capsule network | |
| Convolution layer | Conv1: 256*9*9 | Conv1: 512*9*9 Conv2: 256*5*5 |
| Primary Capsule | 9*9 | 5*5 |
| Digit Capsule | One time dynamic routing Three iterations | Twice dynamic routing The main route has three iterations, and the secondary route has three iterations |
| FC | ||
Figure 2.
Decoder network
Figure 3.
Dynamic routing algorithm
TABLE II
EXPERIMENTAL ENVIRONMENT
| Operating system | Windows10(RAM16.0GB) |
| CPU | Intel(R)Core(TM)i7-9750H |
| GPU | NVIDIA GeForce GTX 1660 Ti |
| Dataset | MNIST |
| Other | pytorch1.5.0+cu101 python 3.7.7 |
Figure 4.
Test precision chart of capsule network under 50 epochs
Figure 5.
Test accuracy chart of deep capsule network under 50 epochs
Figure 6.
Test precision chart of capsule network under 30 epochs
Figure 7.
Test accuracy chart of deep capsule network under 30 epochs
Figure 8.
Comparison between the accuracy of capsule network and deep capsule network
Figure 9.
Impact of changing the number of routing iterations on the deep capsule network
Figure 10.
Influence of different iteration times of two routes on deep capsule network
Figure 11.
Influence of the same number of two routing iterations on deep capsule network
Figure 12.
Schematic diagram of some pictures in MNIST database
Figure 13.
Schematic diagram of reconstructed image
Figure 14.
Comparison of input and output images of the network
Figure 15.
Partial reconstruction results of the improved network
Figure 16.
Improved partial error reconstruction results