Research on improving TransUnet network for feature classification

Jiangjie Yin; Ci Wang; LiPing Liu; Haolong Yang; Xuehong Sun; Yanni Wang

doi:10.1117/12.3004083

11 October 2023 Research on improving TransUnet network for feature classification

Jiangjie Yin, Ci Wang, LiPing Liu, Haolong Yang, Xuehong Sun, Yanni Wang

Proceedings Volume 12800, Sixth International Conference on Computer Information Science and Application Technology (CISAT 2023); 128003D (2023) https://doi.org/10.1117/12.3004083
Event: 6th International Conference on Computer Information Science and Application Technology (CISAT 2023), 2023, Hangzhou, China

Abstract

To address the low accuracy of TransUent network in small-area feature recognition, two feature classification networks based on improved TransUent network are proposed: DS-TransUnet network and DS-A-TransUnet network.DS-TransUnet network introduces a residual module in the bottom layer of the Convolutional Neural Network (CNN) module of the hybrid encoder of TransUent network to deepen the network and uses convolutional operation instead of maximum pooling in the hybrid encoder to reduce the loss of feature information. The DS-TransUnet network introduces a residual module at the bottom of the Convolutional Neural Networks (CNN) module of the TransUent network to deepen the network, and uses convolutional operations instead of maximum pooling in the hybrid encoder to reduce the loss of feature information, and uses depth-separable convolution instead of regular convolution to reduce the residual blocks and the large number of parameters caused by convolutional operations; the DS-A-TransUnet network On the basis of DS-TransUnet network, a void space pyramidal pooling module is introduced, and the classification accuracy is further improved by using void convolution with different expansion rates in parallel to obtain different size of perceptual fields and fully extract multi-scale features. The experimental results show that the recognition accuracy of DS-TransUnet network is 3.06% and 2.85% higher than that of TransUent network for arable land and water bodies, respectively, and the overall recognition accuracy reaches 89.14%; the recognition accuracy of DS-A-TransUnet network is 3.17% and 4.79% higher than that of TransUent network for arable land and water bodies, respectively, and the overall recognition accuracy reaches 90.53%. The overall recognition accuracy reached 90.53%.

(2023) Published by SPIE. Downloading of the abstract is permitted for personal use only.

Citation Download Citation

Jiangjie Yin, Ci Wang, LiPing Liu, Haolong Yang, Xuehong Sun, and Yanni Wang "Research on improving TransUnet network for feature classification", Proc. SPIE 12800, Sixth International Conference on Computer Information Science and Application Technology (CISAT 2023), 128003D (11 October 2023); https://doi.org/10.1117/12.3004083

ACCESS THE FULL ARTICLE

INSTITUTIONAL
Select your institution to access the SPIE Digital Library.

SELECT YOUR INSTITUTION

PERSONAL
Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.

PERSONAL SIGN IN

No SPIE Account? Create one

PURCHASE THIS CONTENT

SUBSCRIBE TO DIGITAL LIBRARY

50 downloads per 1-year subscription

Members: $195

Non-members: $335 ADD TO CART

25 downloads per 1 - year subscription

Members: $145

Non-members: $250 ADD TO CART

PURCHASE SINGLE ARTICLE

Includes PDF, HTML & Video, when available

Members: $17.00

Non-members: $21.00 ADD TO CART

PROCEEDINGS
10 PAGES

DOWNLOAD PAPER SAVE TO MY LIBRARY

GET CITATION

RIGHTS & PERMISSIONS

Get copyright permission Get copyright permission on Copyright Marketplace

KEYWORDS

Image segmentation

Convolution

Education and training

Image classification

Remote sensing

Vegetation

Buildings

Show All Keywords

Keywords/Phrases

Search In:

Publication Years