Dynamic Convolution Self-Attention Network for Land-Cover Classification in VHR Remote-Sensing Images
The current deep convolutional neural networks for very-high-resolution (VHR) remote-sensing image land-cover classification often suffer from two challenges. First, the feature maps extracted by network encoders based on vanilla convolution usually contain a lot of redundant information, which easi...
Main Authors: | , , , , , |
---|---|
Format: | Article |
Language: | English |
Published: |
MDPI AG
2022-10-01
|
Series: | Remote Sensing |
Subjects: | |
Online Access: | https://www.mdpi.com/2072-4292/14/19/4941 |
_version_ | 1797477110429777920 |
---|---|
author | Xuan Wang Yue Zhang Tao Lei Yingbo Wang Yujie Zhai Asoke K. Nandi |
author_facet | Xuan Wang Yue Zhang Tao Lei Yingbo Wang Yujie Zhai Asoke K. Nandi |
author_sort | Xuan Wang |
collection | DOAJ |
description | The current deep convolutional neural networks for very-high-resolution (VHR) remote-sensing image land-cover classification often suffer from two challenges. First, the feature maps extracted by network encoders based on vanilla convolution usually contain a lot of redundant information, which easily causes misclassification of land cover. Moreover, these encoders usually require a large number of parameters and high computational costs. Second, as remote-sensing images are complex and contain many objects with large-scale variances, it is difficult to use the popular feature fusion modules to improve the representation ability of networks. To address the above issues, we propose a dynamic convolution self-attention network (DCSA-Net) for VHR remote-sensing image land-cover classification. The proposed network has two advantages. On one hand, we designed a lightweight dynamic convolution module (LDCM) by using dynamic convolution and a self-attention mechanism. This module can extract more useful image features than vanilla convolution, avoiding the negative effect of useless feature maps on land-cover classification. On the other hand, we designed a context information aggregation module (CIAM) with a ladder structure to enlarge the receptive field. This module can aggregate multi-scale contexture information from feature maps with different resolutions using a dense connection. Experiment results show that the proposed DCSA-Net is superior to state-of-the-art networks due to higher accuracy of land-cover classification, fewer parameters, and lower computational cost. The source code is made public available. |
first_indexed | 2024-03-09T21:13:03Z |
format | Article |
id | doaj.art-78977d7002d041148ec382316fb66411 |
institution | Directory Open Access Journal |
issn | 2072-4292 |
language | English |
last_indexed | 2024-03-09T21:13:03Z |
publishDate | 2022-10-01 |
publisher | MDPI AG |
record_format | Article |
series | Remote Sensing |
spelling | doaj.art-78977d7002d041148ec382316fb664112023-11-23T21:41:14ZengMDPI AGRemote Sensing2072-42922022-10-011419494110.3390/rs14194941Dynamic Convolution Self-Attention Network for Land-Cover Classification in VHR Remote-Sensing ImagesXuan Wang0Yue Zhang1Tao Lei2Yingbo Wang3Yujie Zhai4Asoke K. Nandi5Department of Electrical and Computer Engineering, University of Wisconsin–Madison, Madison, WI 53706, USAShaanxi Joint Laboratory of Artificial Intelligence, Shaanxi University of Science and Technology, Xi’an 710021, ChinaShaanxi Joint Laboratory of Artificial Intelligence, Shaanxi University of Science and Technology, Xi’an 710021, ChinaShaanxi Joint Laboratory of Artificial Intelligence, Shaanxi University of Science and Technology, Xi’an 710021, ChinaShaanxi Joint Laboratory of Artificial Intelligence, Shaanxi University of Science and Technology, Xi’an 710021, ChinaDepartment of Electronic and Electrical Engineering, Brunel University, London UB8 3PH, UKThe current deep convolutional neural networks for very-high-resolution (VHR) remote-sensing image land-cover classification often suffer from two challenges. First, the feature maps extracted by network encoders based on vanilla convolution usually contain a lot of redundant information, which easily causes misclassification of land cover. Moreover, these encoders usually require a large number of parameters and high computational costs. Second, as remote-sensing images are complex and contain many objects with large-scale variances, it is difficult to use the popular feature fusion modules to improve the representation ability of networks. To address the above issues, we propose a dynamic convolution self-attention network (DCSA-Net) for VHR remote-sensing image land-cover classification. The proposed network has two advantages. On one hand, we designed a lightweight dynamic convolution module (LDCM) by using dynamic convolution and a self-attention mechanism. This module can extract more useful image features than vanilla convolution, avoiding the negative effect of useless feature maps on land-cover classification. On the other hand, we designed a context information aggregation module (CIAM) with a ladder structure to enlarge the receptive field. This module can aggregate multi-scale contexture information from feature maps with different resolutions using a dense connection. Experiment results show that the proposed DCSA-Net is superior to state-of-the-art networks due to higher accuracy of land-cover classification, fewer parameters, and lower computational cost. The source code is made public available.https://www.mdpi.com/2072-4292/14/19/4941land-cover classificationfeature fusionself-attentionlightweight |
spellingShingle | Xuan Wang Yue Zhang Tao Lei Yingbo Wang Yujie Zhai Asoke K. Nandi Dynamic Convolution Self-Attention Network for Land-Cover Classification in VHR Remote-Sensing Images Remote Sensing land-cover classification feature fusion self-attention lightweight |
title | Dynamic Convolution Self-Attention Network for Land-Cover Classification in VHR Remote-Sensing Images |
title_full | Dynamic Convolution Self-Attention Network for Land-Cover Classification in VHR Remote-Sensing Images |
title_fullStr | Dynamic Convolution Self-Attention Network for Land-Cover Classification in VHR Remote-Sensing Images |
title_full_unstemmed | Dynamic Convolution Self-Attention Network for Land-Cover Classification in VHR Remote-Sensing Images |
title_short | Dynamic Convolution Self-Attention Network for Land-Cover Classification in VHR Remote-Sensing Images |
title_sort | dynamic convolution self attention network for land cover classification in vhr remote sensing images |
topic | land-cover classification feature fusion self-attention lightweight |
url | https://www.mdpi.com/2072-4292/14/19/4941 |
work_keys_str_mv | AT xuanwang dynamicconvolutionselfattentionnetworkforlandcoverclassificationinvhrremotesensingimages AT yuezhang dynamicconvolutionselfattentionnetworkforlandcoverclassificationinvhrremotesensingimages AT taolei dynamicconvolutionselfattentionnetworkforlandcoverclassificationinvhrremotesensingimages AT yingbowang dynamicconvolutionselfattentionnetworkforlandcoverclassificationinvhrremotesensingimages AT yujiezhai dynamicconvolutionselfattentionnetworkforlandcoverclassificationinvhrremotesensingimages AT asokeknandi dynamicconvolutionselfattentionnetworkforlandcoverclassificationinvhrremotesensingimages |