<abstract xmlns="http://www.w3.org/1999/xhtml"><p>Aiming at the weakness of CNN that is not sensitive to the changes of relative position and angle, a method of digital handwritten recognition based on deep capsule network is researched. The capsule network represents multiple attributes of an entity through a group of capsules composed of neurons, which effectively preserves the information about the position and posture of the entity. Dynamic routing algorithm makes the information interaction between capsules more clearly, and can determine the pose of the entity more accurately. While solving the shortcomings of convolutional neural networks, it also integrates the advantages of CNN and considers the relative position of it’s lack, so that the recognition effect is improved. The design implements a deep capsule network, reduces the amount of trainable parameters by changing the size of the convolution kernel, expands on the original network structure, adds a convolution after the convolution layer, and a process of dynamic routing on the main dynamic routing is added, and the number of iterations is changed for experimentation, which makes the accuracy of network recognition higher on the MNIST data set.</p></abstract>

Aiming at the weakness of CNN that is not sensitive to the changes of relative position and angle, a method of digital handwritten recognition based on deep capsule network is researched. The capsule network represents multiple attributes of an entity through a group of capsules composed of neurons, which effectively preserves the information about the position and posture of the entity. Dynamic routing algorithm makes the information interaction between capsules more clearly, and can determine the pose of the entity more accurately. While solving the shortcomings of convolutional neural networks, it also integrates the advantages of CNN and considers the relative position of it’s lack, so that the recognition effect is improved. The design implements a deep capsule network, reduces the amount of trainable parameters by changing the size of the convolution kernel, expands on the original network structure, adds a convolution after the convolution layer, and a process of dynamic routing on the main dynamic routing is added, and the number of iterations is changed for experimentation, which makes the accuracy of network recognition higher on the MNIST data set.

Deep Capsule Network Handwritten Digit Recognition

International Journal of Advanced Network, Monitoring and Controls

This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

{"article-title":"Deep Capsule Network Handwritten Digit Recognition"}

Aiming at the weakness of CNN that is not sensitive to the changes of relative position and angle, a method of digital handwritten recognition based on deep...

	Capsule network	Deep capsule network
Convolution layer	Conv1: 25699	Conv1: 51299 Conv2: 25655
Primary Capsule	9*9	5*5
Digit Capsule	One time dynamic routingThree iterations	Twice dynamic routingThe main route has three iterations, and the secondary route has three iterations
FC

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

Deep Capsule Network Handwritten Digit Recognition

Publicado en línea: 11 ene 2021

Páginas: 1 - 8

DOI: https://doi.org/10.21307/ijanmc-2020-031

Palabras clave
Component, Deeplearning, Nerve Capsule, Deep Capsule Network, Handwritten Digit Recognition

© 2020 Yuxing Tan et al., published by Sciendo

This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Figure 1.

Figure 2.

Figure 3.

Figure 4.

Figure 5.

Figure 6.

Figure 7.

Figure 8.

Figure 9.

Figure 10.

Figure 11.

Figure 12.

Figure 13.

Figure 14.

Figure 15.

Figure 16.

STRUCTURE COMPARISON OF CAPSULE NETWORK AND DEEP CAPSULE NETWORK

EXPERIMENTAL ENVIRONMENT

Deep Capsule Network Handwritten Digit Recognition

Publicado en línea: 11 ene 2021

Páginas: 1 - 8

DOI: https://doi.org/10.21307/ijanmc-2020-031

Palabras claveComponent, Deeplearning, Nerve Capsule, Deep Capsule Network, Handwritten Digit Recognition

© 2020 Yuxing Tan et al., published by Sciendo

This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Figure 1.

Figure 2.

Figure 3.

Figure 4.

Figure 5.

Figure 6.

Figure 7.

Figure 8.

Figure 9.

Figure 10.

Figure 11.

Figure 12.

Figure 13.

Figure 14.

Figure 15.

Figure 16.

STRUCTURE COMPARISON OF CAPSULE NETWORK AND DEEP CAPSULE NETWORK

EXPERIMENTAL ENVIRONMENT

Palabras clave
Component, Deeplearning, Nerve Capsule, Deep Capsule Network, Handwritten Digit Recognition