Deep learning-based water body segmentation in SAR imagery: enhancing accuracy with CNN-U-Net and EfficientNet
-
Suresh Devaraj
Abstract
Accurate water body segmentation in synthetic aperture radar (SARSAR) imagery is crucial for hydrological analysis, flood monitoring, and environmental management. This study investigates the effectiveness of deep learning (DL) techniques, particularly the convolutional neural network-based U-NetU-Net modelU-Net model with EfficientNet as the encoder, to enhance water body classification in Sentinel-1 SAR images. The proposed model demonstrates significant improvements over traditional methods, such as KMeansKMeans clustering, achieving an accuracy of 97.78%. Integrating contrast-stretched SAR imagery and speckle simulation augmentation enhances model robustness, addressing common issues such as speckle noisespeckle noise and boundary misclassification. The model reduces segmentation errors, thus improving the precision of water body delineation and providing a reliable method for SAR-based water mapping. Despite these improvements, challenges persist, including backscatter intensity variations, seasonal changes, and complex land-water interfaces. The research demonstrated that DL methods significantly improve SAR image classificationimage classification of water bodieswater bodies, serving as powerful tools for environmental monitoring and resource management.
Abstract
Accurate water body segmentation in synthetic aperture radar (SARSAR) imagery is crucial for hydrological analysis, flood monitoring, and environmental management. This study investigates the effectiveness of deep learning (DL) techniques, particularly the convolutional neural network-based U-NetU-Net modelU-Net model with EfficientNet as the encoder, to enhance water body classification in Sentinel-1 SAR images. The proposed model demonstrates significant improvements over traditional methods, such as KMeansKMeans clustering, achieving an accuracy of 97.78%. Integrating contrast-stretched SAR imagery and speckle simulation augmentation enhances model robustness, addressing common issues such as speckle noisespeckle noise and boundary misclassification. The model reduces segmentation errors, thus improving the precision of water body delineation and providing a reliable method for SAR-based water mapping. Despite these improvements, challenges persist, including backscatter intensity variations, seasonal changes, and complex land-water interfaces. The research demonstrated that DL methods significantly improve SAR image classificationimage classification of water bodieswater bodies, serving as powerful tools for environmental monitoring and resource management.
Kapitel in diesem Buch
- Frontmatter I
- Preface V
- Contents VII
- Integrating Sentinel-1 satellite data with machine learning for land use classification 1
- A systematic review of deep learning techniques in microwave remote sensing: challenges, applications, and future directions 17
- Fundamentals of active and passive microwave remote sensing: principles and applications 31
- Comprehensive overview of active and passive microwave remote sensing satellite sensors 55
- Essentials of RADAR remote sensing and AI integration 73
- Fusion of scatterometer and optical remote sensing: enhanced classification and change detection 91
- AI-powered urban infrastructure monitoring using RADAR-based remote sensing 103
- Fusion of the optical and microwave images for cloud removal 123
- Integrating AI in RADAR remote sensing: enhancing data processing, interpretation, and decision-making 141
- Revolutionizing precision agriculture: the synergy of RADAR, Internet of things (IoT), and satellite technology 155
- Integrating AI with RADAR remote sensing: applications in disaster mitigation, defense, and climate change 171
- Computational techniques in RADAR remote sensing from a machine and deep learning perspective 189
- Deep learning-based water body segmentation in SAR imagery: enhancing accuracy with CNN-U-Net and EfficientNet 205
- Artificial intelligence in RADAR remote sensing: advances, challenges, and future prospects 215
- Revolutionizing agricultural and environmental analytics with synthetic aperture radar (SAR): innovations, challenges, and future directions 229
- Editors’ biographies
- Index 247
Kapitel in diesem Buch
- Frontmatter I
- Preface V
- Contents VII
- Integrating Sentinel-1 satellite data with machine learning for land use classification 1
- A systematic review of deep learning techniques in microwave remote sensing: challenges, applications, and future directions 17
- Fundamentals of active and passive microwave remote sensing: principles and applications 31
- Comprehensive overview of active and passive microwave remote sensing satellite sensors 55
- Essentials of RADAR remote sensing and AI integration 73
- Fusion of scatterometer and optical remote sensing: enhanced classification and change detection 91
- AI-powered urban infrastructure monitoring using RADAR-based remote sensing 103
- Fusion of the optical and microwave images for cloud removal 123
- Integrating AI in RADAR remote sensing: enhancing data processing, interpretation, and decision-making 141
- Revolutionizing precision agriculture: the synergy of RADAR, Internet of things (IoT), and satellite technology 155
- Integrating AI with RADAR remote sensing: applications in disaster mitigation, defense, and climate change 171
- Computational techniques in RADAR remote sensing from a machine and deep learning perspective 189
- Deep learning-based water body segmentation in SAR imagery: enhancing accuracy with CNN-U-Net and EfficientNet 205
- Artificial intelligence in RADAR remote sensing: advances, challenges, and future prospects 215
- Revolutionizing agricultural and environmental analytics with synthetic aperture radar (SAR): innovations, challenges, and future directions 229
- Editors’ biographies
- Index 247
