RealFusion: A reliable deep learning-based spatiotemporal fusion framework for generating seamless fine-resolution imagery

Remote Sensing of Environment, Journal Year: 2025, Volume and Issue: 321, P. 114689 - 114689

Published: March 5, 2025

Language: Английский

---

Refining landsat-based annual NDVImax estimation using shape model fitting and phenological metrics

Ecological Informatics, Journal Year: 2025, Volume and Issue: unknown, P. 103107 - 103107

Published: March 1, 2025

Language: Английский

---

Multilevel Feature Cross-Fusion-Based High-Resolution Remote Sensing Wetland Landscape Classification and Landscape Pattern Evolution Analysis

Remote Sensing, Journal Year: 2025, Volume and Issue: 17(10), P. 1740 - 1740

Published: May 16, 2025

Analyzing wetland landscape pattern evolution is crucial for managing resources. High-resolution remote sensing serves as a primary method monitoring patterns. However, the complex types and spatial structures of wetlands pose challenges, including interclass similarity intraclass heterogeneity, leading to low separability landscapes difficulties in identifying fragmented small objects. To address these issues, this study proposes multilevel feature cross-fusion classification network (MFCFNet), which combines global modeling capability Swin Transformer with local detail-capturing ability convolutional neural networks (CNNs), facilitating discerning consistency differences. alleviate semantic confusion caused by different-level features gaps during fusion, we introduce deep–shallow (DSFCF) module between encoder decoder. We incorporate global–local attention block (GLAB) aggregate contextual information detail. The constructed Shengjin Lake Wetland Gaofen Image Dataset (SLWGID) utilized evaluate performance MFCFNet, achieving evaluation metric results OA, mIoU, F1 score 93.23%, 78.12%, 87.05%, respectively. MFCFNet used classify from 2013 2023. A analysis conducted, focusing on transitions, area changes, characteristic variations. demonstrates effectiveness dynamic patterns, providing valuable insights conservation.

Language: Английский

---

A fast adaptive spatio-temporal fusion method to enhanced Fit-FC

PLoS ONE, Journal Year: 2024, Volume and Issue: 19(7), P. e0301077 - e0301077

Published: July 31, 2024

Space-time fusion is an economical and efficient way to solve "space-time contradiction". Among all kinds of space-time methods, Fit-FC method based on weight Function widely used. However, this the linear model depict phase change, but change in real scene complicated, difficult accurately capture resulting spectral distortion image. In addition, pixel-by-pixel scanning with moving Windows leads inefficiency issues, limiting its use large-scale long-term tasks. To overcome these limitations, paper developed a simple fast adaptive remote sensing image Spatio-Temporal Fit-FC, called Adapt Lasso-Fit-FC (AL-FF). Firstly, sparse characteristics time between images are explored, estimation regression constructed, which overcomes fuzzy problem caused by failure complex nonlinear transition weighted method, making algorithm better at capturing details. Secondly, window selection established manually setting parameters different data sets, improve convenience robustness application make simpler more efficient. Finally, improved AL-FF compared other algorithms verify performance improvement. Compared current advanced has stronger detail ability can generate accurate results. computational efficiency significantly improved, increased than 20 times mainstream method.

Language: Английский

---

Monitoring and analysis of the Lake Poyang wetland drought process in 2022 based on spatiotemporal information fusion model

Journal of Lake Sciences, Journal Year: 2024, Volume and Issue: 36(5), P. 1525 - 1536

Published: Jan. 1, 2024

Language: Английский

---

Spatiotemporal evolution and driving mechanism of Dongting Lake based on 2005–2020 multi-source remote sensing data

Ecological Informatics, Journal Year: 2024, Volume and Issue: unknown, P. 102822 - 102822

Published: Sept. 1, 2024

Language: Английский

---

Unmixing-Based Spatiotemporal Image Fusion Based on the Self-Trained Random Forest Regression and Residual Compensation

IEEE Transactions on Geoscience and Remote Sensing, Journal Year: 2023, Volume and Issue: 61, P. 1 - 19

Published: Jan. 1, 2023

Spatiotemporal satellite image fusion (STIF) has been widely applied in land surface monitoring to generate high spatial and temporal reflectance images from sensors. This paper proposed a new unmixing-based spatiotemporal method that is composed of self-trained random forest machine learning regression (R), low resolution (LR) endmember estimation (E), (HR) reconstruction residual compensation (C), is, RERC. RERC uses train predict the relationship between spectra corresponding class fractions. process flexible without any ancillary training dataset, does not possess limitations linear spectral unmixing, which requires number endmembers be no more than bands. The running time about ~1% mixture model. In addition, adopts approach refine fused make full use information LR image. was assessed prediction MODIS with Landsat using two benchmark datasets, fusing different numbers bands by known (seven used) very-high-resolution PlanetScope (four bands). MODIS-Landsat imagery large areas at national scale for Republic Ireland France. code available https://www.researchgate.net/proiile/Xiao_Li52.

Language: Английский

---

Object-Level Hybrid Spatiotemporal Fusion: Reaching a Better Tradeoff Among Spectral Accuracy, Spatial Accuracy, and Efficiency

IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, Journal Year: 2023, Volume and Issue: 16, P. 8007 - 8021

Published: Jan. 1, 2023

Spatiotemporal fusion (STF) is a cost-effective way to complement the spatiotemporal resolution of multi-source images, which has been employed in various applications requiring image sequences. In real-world applications, spectral accuracy, spatial accuracy and efficiency STF play critical role. Despite this, most methods focus on improving while challenges information loss low have received limited attention. Additionally, improvements are contradictory, existing cannot balance them well, limits their reliability applicability for tasks. To solve above issues, this study proposes an object-level hybrid method (OL-HSTFM), incorporates advantage strategy, three-step (Fit-FC), temporal adaptive reflectance model (STARFM). The performance OL-HSTFM was compared with two classic eight state-of-the-art at sites. experimental results indicate that outperforms other 10 overall excellent efficiency. Furthermore, new metric can assess both domains STF, provides more comprehensively intuitively measurement quality fused images commonly used metrics. program openly available https://github.com/Andy-cumt/Object-level-spatiotemporal-fusion-models .

Language: Английский

---

A task decoupled framework for enhancing the deep learning-based spatiotemporal fusion method

International Journal of Remote Sensing, Journal Year: 2023, Volume and Issue: 44(13), P. 4163 - 4189

Published: July 3, 2023

ABSTRACTSpatiotemporal fusion (STF) is a cost-effective way to reconstruct time-series images. In recent years, deep learning-based (DL-based) STF methods have received substantial attention. However, two limitations of DL-based still remain: (1) existing require simultaneous learning both the multi-source images correction model and model, which complicates training task. The high complexity poses challenge for network accurately learn underlying mathematical principles STF, thereby reducing method's reliability generalization ability; (2) tend generate blurry predictions. To address these limitations, this study proposes task decoupled (TD) framework that offers simple yet effective solution enhancing method. consists are trained using actual simulated image pairs, respectively, model. loss edge feature added in function ameliorate its detailed information preservation ability. proposed evaluated on three five different sites root-mean-square error (RMSE) Robert's (Edge) assess spectral spatial accuracy. experimental results indicate can significantly improve models' ability predict (average increase rate = 5.3% accuracy), preserve 16.2% retrieve land cover change, generalize new data. These findings demonstrate effectiveness addressing potential advancing applications.KEYWORDS: Spatiotemporal fusiontask-decoupled frameworkdeep learningloss AcknowledgementsThis work was supported part by Otto Poon Charitable Foundation Smart Cities Research Institute, Hong Kong Polytechnic University (Work Program: CD03) Urban Informatics Cities, (1-ZVN6). authors thank Dr. Tan providing source code EDCSTFN GAN-STFM, Ms. Cao MANet. would also like Editors all reviewers their helpful constructive comments paper.Disclosure statementNo conflict interest reported author(s).Data Availability statementThe data openly available https://github.com/Andy-cumt/Spatiotemporal-fusion-dataset-DeepLearning.Supplementary materialSupplemental article be accessed online at https://doi.org/10.1080/01431161.2023.2232548

Language: Английский

---

Faster inference from state space models via GPU computing

Ecological Informatics, Journal Year: 2024, Volume and Issue: 80, P. 102486 - 102486

Published: Jan. 24, 2024

Inexpensive Graphics Processing Units (GPUs) offer the potential to greatly speed up computation by employing their massively parallel architecture perform arithmetic operations more efficiently. Population dynamics models are important tools in ecology and conservation. Modern Bayesian approaches allow biologically realistic be constructed fitted multiple data sources an integrated modelling framework based on a class of statistical called state space models. However, model fitting is often slow, requiring hours weeks computation. We demonstrate benefits GPU computing using for population British grey seals, with particle Markov chain Monte Carlo algorithm. Speed-ups two orders magnitude were obtained estimations log-likelihood, compared traditional ‘CPU-only’ implementation, allowing accurate method inference used where this was previously too computationally expensive viable. has enormous potential, but one barrier further adoption steep learning curve, due GPUs' unique hardware architecture. provide detailed description software setup, our case study provides template other similar applications. also tutorial-style architectures, examples GPU-specific programming practices.

Language: Английский

---