Identifying Forest Burned Area Using a Deep Learning Model Based on Post-Fire Optical and SAR Remote Sensing Images

Identifying wildfire burned areas using satellite images is significant for effectively monitoring the status of forests. The full utilization of multi-source satellite images that provide complementary information is beneficial for accurate monitoring of Forest-Burned Area (FBA), which, however, is...

Full description

Saved in:
Bibliographic Details
Main Authors: Xiaofei Xi, Man Kang, Long Dai, Yan Jing, Peng Han, Congqiang Hou
Format: Article
Language:English
Published: IEEE 2024-01-01
Series:IEEE Access
Subjects:
Online Access:https://ieeexplore.ieee.org/document/10792922/
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1850251074119663616
author Xiaofei Xi
Man Kang
Long Dai
Yan Jing
Peng Han
Congqiang Hou
author_facet Xiaofei Xi
Man Kang
Long Dai
Yan Jing
Peng Han
Congqiang Hou
author_sort Xiaofei Xi
collection DOAJ
description Identifying wildfire burned areas using satellite images is significant for effectively monitoring the status of forests. The full utilization of multi-source satellite images that provide complementary information is beneficial for accurate monitoring of Forest-Burned Area (FBA), which, however, is ignored by many current studies. In this paper, inspired by the Residual-based U-Net (RU-Net), an innovative deep learning-based model, DARU-Net, for FBA Identification (FBAI) using multi-source satellite images is presented based on a dual-path mechanism and an attention module. The proposed DARU-Net employs a dual-path mechanism to mine complementary information from Sentinel-1 Synthetic Aperture Radar (SAR) image and Sentinel-2 optical image. Besides, a channel-spatial attention residual (CSAR) module is embedded into the network, aiming at helping the network to focus on useful information. The experimental results on benchmark FBAI datasets demonstrate the good performance of DARU-Net in identifying wildfire burned areas, with an overall accuracy of 93.14% and a F-score of 83.01%, outperformed some widely-used U-Net-based detection models. The DARU-Net is more capable of accurately identifying FBAs by preserving geometrical details due to the use of dual-path integration module. Besides, it is found that, the CSAR module is helpful to promote both the precision and the training efficiency of the proposed model.
format Article
id doaj-art-aa2a6b95bb0d47a581d8470b6a579856
institution OA Journals
issn 2169-3536
language English
publishDate 2024-01-01
publisher IEEE
record_format Article
series IEEE Access
spelling doaj-art-aa2a6b95bb0d47a581d8470b6a5798562025-08-20T01:58:00ZengIEEEIEEE Access2169-35362024-01-011218810218811310.1109/ACCESS.2024.351520510792922Identifying Forest Burned Area Using a Deep Learning Model Based on Post-Fire Optical and SAR Remote Sensing ImagesXiaofei Xi0Man Kang1Long Dai2Yan Jing3Peng Han4Congqiang Hou5https://orcid.org/0009-0008-3499-4054Beijing Skysight Technology Company Ltd., Beijing, ChinaBeijing Skysight Technology Company Ltd., Beijing, ChinaBeijing Skysight Technology Company Ltd., Beijing, ChinaState Grid (Xi’an) Environmental Protection Technology Center Company Ltd., Xi’an, ChinaBeijing Skysight Technology Company Ltd., Beijing, ChinaBeijing Skysight Technology Company Ltd., Beijing, ChinaIdentifying wildfire burned areas using satellite images is significant for effectively monitoring the status of forests. The full utilization of multi-source satellite images that provide complementary information is beneficial for accurate monitoring of Forest-Burned Area (FBA), which, however, is ignored by many current studies. In this paper, inspired by the Residual-based U-Net (RU-Net), an innovative deep learning-based model, DARU-Net, for FBA Identification (FBAI) using multi-source satellite images is presented based on a dual-path mechanism and an attention module. The proposed DARU-Net employs a dual-path mechanism to mine complementary information from Sentinel-1 Synthetic Aperture Radar (SAR) image and Sentinel-2 optical image. Besides, a channel-spatial attention residual (CSAR) module is embedded into the network, aiming at helping the network to focus on useful information. The experimental results on benchmark FBAI datasets demonstrate the good performance of DARU-Net in identifying wildfire burned areas, with an overall accuracy of 93.14% and a F-score of 83.01%, outperformed some widely-used U-Net-based detection models. The DARU-Net is more capable of accurately identifying FBAs by preserving geometrical details due to the use of dual-path integration module. Besides, it is found that, the CSAR module is helpful to promote both the precision and the training efficiency of the proposed model.https://ieeexplore.ieee.org/document/10792922/Forest monitoringsatellite imageattention mechanismconvolutional neural network
spellingShingle Xiaofei Xi
Man Kang
Long Dai
Yan Jing
Peng Han
Congqiang Hou
Identifying Forest Burned Area Using a Deep Learning Model Based on Post-Fire Optical and SAR Remote Sensing Images
IEEE Access
Forest monitoring
satellite image
attention mechanism
convolutional neural network
title Identifying Forest Burned Area Using a Deep Learning Model Based on Post-Fire Optical and SAR Remote Sensing Images
title_full Identifying Forest Burned Area Using a Deep Learning Model Based on Post-Fire Optical and SAR Remote Sensing Images
title_fullStr Identifying Forest Burned Area Using a Deep Learning Model Based on Post-Fire Optical and SAR Remote Sensing Images
title_full_unstemmed Identifying Forest Burned Area Using a Deep Learning Model Based on Post-Fire Optical and SAR Remote Sensing Images
title_short Identifying Forest Burned Area Using a Deep Learning Model Based on Post-Fire Optical and SAR Remote Sensing Images
title_sort identifying forest burned area using a deep learning model based on post fire optical and sar remote sensing images
topic Forest monitoring
satellite image
attention mechanism
convolutional neural network
url https://ieeexplore.ieee.org/document/10792922/
work_keys_str_mv AT xiaofeixi identifyingforestburnedareausingadeeplearningmodelbasedonpostfireopticalandsarremotesensingimages
AT mankang identifyingforestburnedareausingadeeplearningmodelbasedonpostfireopticalandsarremotesensingimages
AT longdai identifyingforestburnedareausingadeeplearningmodelbasedonpostfireopticalandsarremotesensingimages
AT yanjing identifyingforestburnedareausingadeeplearningmodelbasedonpostfireopticalandsarremotesensingimages
AT penghan identifyingforestburnedareausingadeeplearningmodelbasedonpostfireopticalandsarremotesensingimages
AT congqianghou identifyingforestburnedareausingadeeplearningmodelbasedonpostfireopticalandsarremotesensingimages