Extracting Water Surfaces of the Dike-Pond System from High Spatial Resolution Images Using Deep Learning Methods DOI Creative Commons

Jinhao Zhou,

K. S. Fu, Shen Liang

et al.

Remote Sensing, Journal Year: 2024, Volume and Issue: 17(1), P. 111 - 111

Published: Dec. 31, 2024

A type of aquaculture pond called a dike-pond system is distributed in the low-lying river delta China’s eastern coast. Along with swift growth coastal economy, water surfaces (WDPS) play major role attributed to yielding more profits than dike agriculture. This study aims explore performance deep learning methods for extracting WDPS from high spatial resolution remote sensing images. We developed three fully convolutional network (FCN) models: SegNet, UNet, and UNet++, which are compared two traditional same testing regions Guangdong–Hong Kong–Macao Greater Bay Area. The extraction results five evaluated parts. first part general comparison that shows biggest advantage FCN models over P-score, an average lead 13%, but R-score not ideal. Our analysis reveals low problem due omission outer ring rather quantity WDPS. also analyzed reasons behind it provided potential solutions. second error, demonstrates have few connected, jagged, or perforated WDPS, beneficial assessing fishery production, pattern changes, ecological value, other applications extracted by visually close ground truth, one most significant improvements methods. third special scenarios, including various shape types, intricate configurations, multiple conditions. irregular shapes juxtaposed land types increases difficulty extraction, still achieve P-scores above 0.95 while conditions causes sharp drop indicators all methods, requires further improvement solve it. integrated performances provide recommendations their use. offers valuable insights enhancing leveraging practical applications.

Language: Английский

greenDBCE-Net: A Novel Deep Learning Framework for Annual Mapping of Coastal Aquaculture Ponds in China with Sentinel-2 Data DOI Creative Commons
Li Yin, Liaoying Zhao, Huaguo Zhang

et al.

Remote Sensing, Journal Year: 2025, Volume and Issue: 17(3), P. 362 - 362

Published: Jan. 22, 2025

Despite the promising advancements of deep learning techniques in coastal aquaculture pond extraction, their capacity for large-scale mapping tasks remains relatively limited. To address this challenge, study developed a novel framework, Dual-Branch Enhanced Network (DBCE-Net), annual ponds at national scale using Sentinel-2 imagery. The DBCE-Net framework effectively mitigates contextual information loss inherent traditional methods and reduces classification errors by processing both down-sampled images block original resolution. architecture comprises local feature extraction global along with fusion decoding. pivotal Multi-scale Dynamic Feature Fusion (DFF) module synthesizes features while incorporating complementary information, demonstrating strong robustness smaller training areas, compared to previous that required larger number samples distributed across different regions. By applying imagery from 2017 2023, we mapped spatiotemporal distribution all counties China, achieving an overall accuracy approximately 93%. results demonstrate substantial changes area China total declining 8970.25 km2 8261.17 km2, representing notable decrease 7.90%. most pronounced reduction was observed Shanghai, 38.92%, followed Zhejiang (31.57%) Jiangsu (19.07%). These reductions are primarily attributed policies converting into natural wetlands. In contrast, Liaoning Province slightly increased 5.75%. This demonstrates good generalizability is further expand its application areas worldwide, providing important scientific value practical significance industry.

Language: Английский

Citations

0
GCL_FCS30: a global coastline dataset with 30-m resolution and a fine classification system from 2010 to 2020 DOI Creative Commons
Jian Zuo, Li Zhang, Jingfeng Xiao

et al.

Scientific Data, Journal Year: 2025, Volume and Issue: 12(1)

Published: Jan. 22, 2025

The coastline reflects coastal environmental processes and dynamic changes, serving as a fundamental parameter for coast. Although several global datasets have been developed, they mainly focus on morphology, the typology of coastlines are still lacking. We produced Global CoastLine Dataset (GCL_FCS30) with detailed classification system. extraction employed combined algorithm incorporating Modified Normalized Difference Water Index an adaptive threshold segmentation method. was performed hybrid transect classifier that integrates random forest stable training samples derived from multi-source geophysical data. GCL_FCS30 offers significant advantages in capturing artificial coastlines, reflecting strong alignment location validation found to achieve overall accuracy Kappa coefficient over 85% 0.75. Each category accurately covered majority area represented third-party data exhibited high degree spatial relevance. Therefore, is first dataset covering latitudes continuous smooth line vector format.

Language: Английский

Citations

0
Autonomous Extraction Technology for Aquaculture Ponds in Complex Geological Environments Based on Multispectral Feature Fusion of Medium-Resolution Remote Sensing Imagery DOI Creative Commons

Zongxia Liang,

Yingzi Hou, Jianfeng Zhu

et al.

Remote Sensing, Journal Year: 2024, Volume and Issue: 16(22), P. 4130 - 4130

Published: Nov. 5, 2024

Coastal aquaculture plays a crucial role in global food security and the economic development of coastal regions, but it also causes environmental degradation ecosystems. Therefore, automation, accurate extraction, monitoring areas are for scientific management ecological zones. This study proposes novel deep learning- attention-based median adaptive fusion U-Net (MAFU-Net) procedure aimed at precisely extracting individually separable ponds (ISAPs) from medium-resolution remote sensing imagery. Initially, this analyzes spectral differences between interfering objects such as saltwater fields four typical along coast Liaoning Province, China. It innovatively introduces difference index field zones (DIAS) integrates new band into imagery to increase expressiveness features. A augmented module (MEA-FM), which adaptively selects channel receptive various scales, information channels, captures multiscale spatial achieve improved extraction accuracy, is subsequently designed. Experimental comparative results reveal that proposed MAFU-Net method achieves an F1 score 90.67% intersection over union (IoU) 83.93% on CHN-LN4-ISAPS-9 dataset, outperforming advanced methods U-Net, DeepLabV3+, SegNet, PSPNet, SKNet, UPS-Net, SegFormer. study’s provide data support areas, provides effective semantic segmentation tasks based images.

Language: Английский

Citations

2
Analysis of Surface Water Area Changes and Driving Factors in the Tumen River Basin (China and North Korea) Using Google Earth Engine (2015–2023) DOI Open Access
Di Wu, Donghe Quan, Ri Jin

et al.

Water, Journal Year: 2024, Volume and Issue: 16(15), P. 2185 - 2185

Published: Aug. 1, 2024

Understanding the dynamics of water bodies is crucial for managing resources and protecting ecosystems, especially in regions prone to climatic extremes. The Tumen River Basin, a transboundary area Northeast Asia, has seen significant body changes influenced by natural anthropogenic factors. Using Landsat 8 Sentinel-1 data on Google Earth Engine, we systematically analyzed spatiotemporal variations drivers this basin from 2015 2023. extraction process demonstrated high accuracy, with overall precision rates 95.75% 98.25% Sentinel-1. Despite observed annual fluctuations, exhibited an increasing trend, notably peaking 2016 due extraordinary flood event. Emerging Hot Spot Analysis revealed upstream areas as declining cold spots downstream hot spots, artificial showing growth trend. Utilizing Random Forest Regression, key factors such precipitation, potential evaporation, population density, bare land, wetlands were identified, accounting approximately 81.9–85.3% area. During anomalous period June September 2016, Geographically Weighted Regression (GWR) model underscored predominant influence density at sub-basin scale. These findings provide critical insights strategic resource management environmental conservation Basin.

Language: Английский

Citations

0
A Study on the Extraction of Satellite Image Information for Two Types of Coastal Fishery Facility Fish Cages and Rafts Influenced by Clouds and Vessels DOI Creative Commons
Ao Chen,

Jialu Yu,

Junbo Zhang

et al.

Journal of Marine Science and Engineering, Journal Year: 2024, Volume and Issue: 12(12), P. 2280 - 2280

Published: Dec. 11, 2024

Research on the extraction of satellite information for areas coastal fish cages and rafts is important to quickly grasp pattern structure fishery aquaculture industry. This study proposes a multi-feature rule-based object-oriented image classification (MROIC) model, integrating spatial-spectral enhancement techniques with object-based analysis methods. The MROIC model enhances spectral by constructing ratio bands alongside principal component analysis, subsequently employing rule sets, edge detection algorithms, comprehensive algorithmic merging techniques. It applicable tasks in complex environments, including influence clouds vessels. cage raft facilities extracted via southwest coast Xiapu County, Fujian Province, as an example. results showed that attained average total accuracy 90.43% Kappa coefficient 0.80. Extracting area fisheries under vessels can provide better lower omission error. proposed this demonstrates high strong applicability, offering technical support government planning facility aiding risk assessment management efficiency insurance.

Language: Английский

Citations

0
Extracting Water Surfaces of the Dike-Pond System from High Spatial Resolution Images Using Deep Learning Methods DOI Creative Commons

Jinhao Zhou,

K. S. Fu, Shen Liang

et al.

Remote Sensing, Journal Year: 2024, Volume and Issue: 17(1), P. 111 - 111

Published: Dec. 31, 2024

A type of aquaculture pond called a dike-pond system is distributed in the low-lying river delta China’s eastern coast. Along with swift growth coastal economy, water surfaces (WDPS) play major role attributed to yielding more profits than dike agriculture. This study aims explore performance deep learning methods for extracting WDPS from high spatial resolution remote sensing images. We developed three fully convolutional network (FCN) models: SegNet, UNet, and UNet++, which are compared two traditional same testing regions Guangdong–Hong Kong–Macao Greater Bay Area. The extraction results five evaluated parts. first part general comparison that shows biggest advantage FCN models over P-score, an average lead 13%, but R-score not ideal. Our analysis reveals low problem due omission outer ring rather quantity WDPS. also analyzed reasons behind it provided potential solutions. second error, demonstrates have few connected, jagged, or perforated WDPS, beneficial assessing fishery production, pattern changes, ecological value, other applications extracted by visually close ground truth, one most significant improvements methods. third special scenarios, including various shape types, intricate configurations, multiple conditions. irregular shapes juxtaposed land types increases difficulty extraction, still achieve P-scores above 0.95 while conditions causes sharp drop indicators all methods, requires further improvement solve it. integrated performances provide recommendations their use. offers valuable insights enhancing leveraging practical applications.

Language: Английский

Citations

0