Improved Road Extraction Models through Semi-Supervised Learning with ACCT

Improving the performance and reducing the training cost of road extraction models in the absence of samples is important for updating road maps. Despite the success of recent road extraction models on standard datasets, they often fail to perform when applied to new datasets or real-world scenarios...

Full description

Saved in:
Bibliographic Details
Main Authors: Hao Yu, Shihong Du, Zhenshan Tan, Xiuyuan Zhang, Zhijiang Li
Format: Article
Language:English
Published: MDPI AG 2024-09-01
Series:ISPRS International Journal of Geo-Information
Subjects:
Online Access:https://www.mdpi.com/2220-9964/13/10/347
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1850205483950931968
author Hao Yu
Shihong Du
Zhenshan Tan
Xiuyuan Zhang
Zhijiang Li
author_facet Hao Yu
Shihong Du
Zhenshan Tan
Xiuyuan Zhang
Zhijiang Li
author_sort Hao Yu
collection DOAJ
description Improving the performance and reducing the training cost of road extraction models in the absence of samples is important for updating road maps. Despite the success of recent road extraction models on standard datasets, they often fail to perform when applied to new datasets or real-world scenarios where labeled samples are not available. In this paper, our focus diverges from the typical quest to pinpoint the optimal road extraction model or evaluate generalization prowess across models. Instead, we propose a method called Asymmetric Consistent Co-Training (ACCT) to train existing road extraction models faster and make them perform better in new scenarios lacking samples. ACCT uses two models with different structures and a supervision module to enhance accuracy through mutual learning. Labeled and unlabeled images are processed by both models to generate road maps from different perspectives. The supervision module ensures consistency between predictions by computing losses based on labeling status. ACCT iteratively adjusts parameters using unlabeled data, improving generalization. Empirical evaluations show that ACCT improves IoU by 2.79% to 10.26% using only 1/8 of the labeled data compared to fully supervised methods. It also reduces parameters by over 49% compared to state-of-the-art semi-supervised methods while maintaining similar accuracy. These results highlight the potential of leveraging large amounts of unlabeled data to enhance road extraction models as data acquisition technology advances.
format Article
id doaj-art-b6be41b8c65044beaeafc59d83ee30d8
institution OA Journals
issn 2220-9964
language English
publishDate 2024-09-01
publisher MDPI AG
record_format Article
series ISPRS International Journal of Geo-Information
spelling doaj-art-b6be41b8c65044beaeafc59d83ee30d82025-08-20T02:11:04ZengMDPI AGISPRS International Journal of Geo-Information2220-99642024-09-01131034710.3390/ijgi13100347Improved Road Extraction Models through Semi-Supervised Learning with ACCTHao Yu0Shihong Du1Zhenshan Tan2Xiuyuan Zhang3Zhijiang Li4School of Earth and Space Sciences, Peking University, Beijing 100871, ChinaCollege of Urban and Environmental Sciences, Peking University, Beijing 100871, ChinaSchool of Computer Science, Nanjing University of Information Science and Technology, Nanjing 210044, ChinaCollege of Urban and Environmental Sciences, Peking University, Beijing 100871, ChinaSchool of Information Management, Wuhan University, Wuhan 430072, ChinaImproving the performance and reducing the training cost of road extraction models in the absence of samples is important for updating road maps. Despite the success of recent road extraction models on standard datasets, they often fail to perform when applied to new datasets or real-world scenarios where labeled samples are not available. In this paper, our focus diverges from the typical quest to pinpoint the optimal road extraction model or evaluate generalization prowess across models. Instead, we propose a method called Asymmetric Consistent Co-Training (ACCT) to train existing road extraction models faster and make them perform better in new scenarios lacking samples. ACCT uses two models with different structures and a supervision module to enhance accuracy through mutual learning. Labeled and unlabeled images are processed by both models to generate road maps from different perspectives. The supervision module ensures consistency between predictions by computing losses based on labeling status. ACCT iteratively adjusts parameters using unlabeled data, improving generalization. Empirical evaluations show that ACCT improves IoU by 2.79% to 10.26% using only 1/8 of the labeled data compared to fully supervised methods. It also reduces parameters by over 49% compared to state-of-the-art semi-supervised methods while maintaining similar accuracy. These results highlight the potential of leveraging large amounts of unlabeled data to enhance road extraction models as data acquisition technology advances.https://www.mdpi.com/2220-9964/13/10/347road extractionsemi-supervised learningfeature learningtraining cost
spellingShingle Hao Yu
Shihong Du
Zhenshan Tan
Xiuyuan Zhang
Zhijiang Li
Improved Road Extraction Models through Semi-Supervised Learning with ACCT
ISPRS International Journal of Geo-Information
road extraction
semi-supervised learning
feature learning
training cost
title Improved Road Extraction Models through Semi-Supervised Learning with ACCT
title_full Improved Road Extraction Models through Semi-Supervised Learning with ACCT
title_fullStr Improved Road Extraction Models through Semi-Supervised Learning with ACCT
title_full_unstemmed Improved Road Extraction Models through Semi-Supervised Learning with ACCT
title_short Improved Road Extraction Models through Semi-Supervised Learning with ACCT
title_sort improved road extraction models through semi supervised learning with acct
topic road extraction
semi-supervised learning
feature learning
training cost
url https://www.mdpi.com/2220-9964/13/10/347
work_keys_str_mv AT haoyu improvedroadextractionmodelsthroughsemisupervisedlearningwithacct
AT shihongdu improvedroadextractionmodelsthroughsemisupervisedlearningwithacct
AT zhenshantan improvedroadextractionmodelsthroughsemisupervisedlearningwithacct
AT xiuyuanzhang improvedroadextractionmodelsthroughsemisupervisedlearningwithacct
AT zhijiangli improvedroadextractionmodelsthroughsemisupervisedlearningwithacct