Deep Learning Approach for Forest Change Detection and Prediction Using Satellite Images DOI

Bodavula Sneha,

M. Padmaja,

Yarasani Kundana

и другие.

Опубликована: Ноя. 14, 2024

Язык: Английский

Automatized Sentinel-2 mosaicking for large area forest mapping DOI Creative Commons
Timo P. Pitkänen, András Balázs, Sakari Tuominen

и другие.

International Journal of Applied Earth Observation and Geoinformation, Год журнала: 2024, Номер 127, С. 103659 - 103659

Опубликована: Янв. 22, 2024

Creating maps of forest inventory variables is commonly taking advantage satellite images, which are mosaicked together for gaining larger coverage. Recently, mosaicking has increasingly shifted towards user friendly cloud-based online environments such as Google Earth Engine (GEE), equipped with huge image repositories and extensive processing capabilities. This enables the easy transferability workflows into new sets diversifies range methodological options mosaicking. The quality control output mosaic, ensuring that reflectance values representative to targeted land cover, however primarily based on certain assumptions or pre-set rules may not always produce an optimal result. Our study focuses assessing comparing performance three different algorithms predicting variables, set field data main site type, fertility class, volume biomass growing stock. One compared mosaics derives from manual selection, thus enabling rigorous visual control, two others resting GEE-assisted automatized methods include applying a percentile-based statistic over all input selecting best pixels using predefined indicators. results indicate generally providing relatively equal levels. Compared them, quality-based mosaic slightly lower accuracy particularly when continuous (i.e., stock) it suffers minor defects. For total stock, example, RMS errors 56.22 % manual, 56.33 percentile-based, 59.47 mosaics, respectively. These perspective large area mapping, automatically generated provide approximately similar manually controlled workflow at fraction workload.

Язык: Английский

Процитировано

5

Artificial Intelligence Models to Prevent Forest Fires DOI
Wasswa Shafik

Advances in computational intelligence and robotics book series, Год журнала: 2024, Номер unknown, С. 78 - 106

Опубликована: Май 6, 2024

The main goal is to appropriately utilize advanced algorithms analyze environmental data, improve early disease detection and intervention tactics, reduce the harmful effects of forest fires on human beings. Analyze challenges faced by traditional methods in addressing constantly evolving nature wildfires need for more adaptable proactive approaches, highlight advantages AI. Discusses constituents incorporated into AI model, comprising meteorological satellite imagery, historical fire records. It analyzes selection specifically tailored prevention, considering parameters. during creation implementation models prevention viability integrating artificial intelligence existing management infrastructure emergency response systems. showcases current research, progress, use AI-driven solutions address posed provides a concise overview chapter's findings.

Язык: Английский

Процитировано

5

ICSF: Integrating Inter-Modal and Cross-Modal Learning Framework for Self-Supervised Heterogeneous Change Detection DOI
Erlei Zhang, He Zong, Xinyu Li

и другие.

IEEE Transactions on Geoscience and Remote Sensing, Год журнала: 2024, Номер 63, С. 1 - 16

Опубликована: Дек. 19, 2024

Язык: Английский

Процитировано

5

Transformative Trends in AI for Environmental Monitoring DOI Open Access

R. Leena Sri,

Divya Vetriveeran, Rakoth Kandan Sambandam

и другие.

Опубликована: Фев. 16, 2025

Язык: Английский

Процитировано

0

Decreased and Fragmented Greenspaces in and around Rural Residential Areas of Eastern China in the Process of Urbanization DOI

W Li,

Jun Wang,

Yuan Luo

и другие.

Remote Sensing Applications Society and Environment, Год журнала: 2025, Номер unknown, С. 101518 - 101518

Опубликована: Март 1, 2025

Язык: Английский

Процитировано

0

Using Multisource High-Resolution Remote Sensing Data (2 m) with a Habitat–Tide–Semantic Segmentation Approach for Mangrove Mapping DOI Creative Commons
Ziyu Sun, Weiguo Jiang, Ziyan Ling

и другие.

Remote Sensing, Год журнала: 2023, Номер 15(22), С. 5271 - 5271

Опубликована: Ноя. 7, 2023

Mangrove wetlands are hotspots of global biodiversity and blue carbon reserves in coastal wetlands, with unique ecological functions significant socioeconomic value. Annual fine-scale monitoring mangroves is crucial for evaluating national conservation programs implementing sustainable mangrove management strategies. However, annual mapping over large areas using remote sensing remains a challenge due to spectral similarities vegetation, tidal periodic fluctuations, the need consistent dependable samples across different years. In previous research, there has been lack strategies that simultaneously consider spatial, temporal, methodological aspects extraction. Therefore, based on an approach considers habitat, tides, semantic segmentation approach, we propose method suitable long time-series data. This optimized hybrid model integrates considerations. The uses five sensors (GF-1, GF-2, GF-6, ZY-301, ZY-302) combine deep learning U-Net models habitat information algorithms during low-tide periods. produces map spatial resolution 2 m. We applied this algorithm three typical regions Beibu Gulf Guangxi Province. results showed following: (1) scored above 0.9 terms its F1-score all study at time training, average accuracy 92.54% (2) overall (OA) extraction distribution was 93.29%. When comparing validation years, OA exceeded 89.84% Kappa coefficient 0.74. (3) reliable extracting sparse slow-growing young narrow belts along roadsides. some where flooding occurs, existing dataset underestimates certain extent. provides foundation implementation ecosystems, support species diversity conservation, recovery, development goals related development.

Язык: Английский

Процитировано

9

Application of an improved U-Net with image-to-image translation and transfer learning in peach orchard segmentation DOI Creative Commons

Jiayu Cheng,

Yihang Zhu, Yiying Zhao

и другие.

International Journal of Applied Earth Observation and Geoinformation, Год журнала: 2024, Номер 130, С. 103871 - 103871

Опубликована: Май 2, 2024

Peach cultivation holds a significant economic importance, and obtaining the spatial distribution of peach orchards is helpful for yield prediction precision agriculture. In this study, we introduce new U-Net semantic segmentation model, utilizing ResNet50 as backbone network, augmented with an Efficient Multi-Scale Attention (EMA) mechanism module LayerScale adaptive scaling parameter. To address style differences between images from Unmanned Aerial Vehicle (UAV), Google Earth, Sentinel-2 satellite, incorporate Cycle-Consistent Generative Adversarial Networks (CycleGAN). This synthesis ensures that UAV conform to comparable found in Earth images, while feature details high resolution are transferred through transfer learning. The results demonstrate using network model yields higher accuracy compared VGG16 model. Specifically, Mean Intersection over Union (MIoU) values by 0.49 % 0.95 %, respectively. MIoU UAV, increased 0.87 1.71 1.74 respectively, introduction EMA. Additionally, parameters, 0.31 0.33 1.44 further enhancing After applying CycleGAN learning, 1.02 0.15 1.57 resulting 97.39 92.08 84.54 %. comparative analysis DeepLabV3+, PSPNet, HRNet models demonstrates superior mapping performance proposed method. Moreover, method exhibits good generalization speed across six test sites research area. Overall, approach efficiency orchard mapping, accommodating various resolutions, potential addressing diverse requirements applications.

Язык: Английский

Процитировано

3

Potential of Earth Observation to Assess the Impact of Climate Change and Extreme Weather Events in Temperate Forests—A Review DOI Creative Commons
Marco Wegler,

Claudia Kuenzer

Remote Sensing, Год журнала: 2024, Номер 16(12), С. 2224 - 2224

Опубликована: Июнь 19, 2024

Temperate forests are particularly exposed to climate change and the associated increase in weather extremes. Droughts, storms, late frosts, floods, heavy snowfalls, or changing climatic conditions such as rising temperatures more erratic precipitation having an increasing impact on forests. There is urgent need better assess impacts of extreme events (EWEs) temperate Remote sensing can be used map at multiple spatial, temporal, spectral resolutions low cost. Different approaches forest assessment offer promising methods for a broad analysis EWEs. In this review, we examine potential Earth observation assessing EWEs by reviewing 126 scientific papers published between 1 January 2014 31 2024. This study provides comprehensive overview sensors utilized, spatial temporal resolution studies, their distribution, thematic focus various abiotic drivers resulting responses. The indicates that multispectral, non-high-resolution timeseries were employed most frequently. A predominant proportion studies droughts. all instances EWEs, dieback prevailing response, whereas trends, phenology shifts account largest share response categories. detailed in-depth differentiation implies area-wide have so far barely distinguished effects different species level.

Язык: Английский

Процитировано

3

Spatial classification model of port facilities and energy reserve prediction based on deep learning for port management―A case study of Ningbo DOI
Huixiang Huang,

Yan Qiaoling,

Yang Yang

и другие.

Ocean & Coastal Management, Год журнала: 2024, Номер 258, С. 107413 - 107413

Опубликована: Окт. 11, 2024

Язык: Английский

Процитировано

3

Cloud–Aerosol Classification Based on the U-Net Model and Automatic Denoising CALIOP Data DOI Creative Commons

Xingzhao Zhou,

Bin Chen,

Qia Ye

и другие.

Remote Sensing, Год журнала: 2024, Номер 16(5), С. 904 - 904

Опубликована: Март 4, 2024

Precise cloud and aerosol identification hold paramount importance for a thorough comprehension of atmospheric processes, enhancement meteorological forecasts, mitigation climate change. This study devised an automatic denoising cloud–aerosol classification deep learning algorithm, successfully achieving in vertical profiles utilizing CALIPSO L1 data. The algorithm primarily consists two components: classification. task integrates module that comprehensively assesses various methods, such as Gaussian filtering bilateral filtering, automatically selecting the optimal approach. results indicated is more suitable data, yielding SNR, RMSE, SSIM values 4.229, 0.031, 0.995, respectively. involves constructing U-Net model, incorporating self-attention mechanisms, residual connections, pyramid-pooling modules to enhance model’s expressiveness applicability. In comparison with machine models, model exhibited best performance, accuracy 0.95. Moreover, it demonstrated outstanding generalization capabilities, evaluated using harmonic mean F1 value, which accounts both precision recall. It achieved 0.90 0.97 samples from lidar during spring 2019. endeavored predict low-quality data VFM revealing significant differences consistency 0.23 clouds 0.28 aerosols. Utilizing confidence 532 nm attenuated backscatter coefficient validate medium- predictions cases 8 February 2019, was found align closely observational high confidence. Statistical comparisons predicted geographical distribution revealed specific patterns regional characteristics aerosols, showcasing proficiency identifying aerosols within layers.

Язык: Английский

Процитировано

2