Deep learning in cropland field identification: A review

Computers and Electronics in Agriculture, Journal Year: 2024, Volume and Issue: 222, P. 109042 - 109042

Published: May 17, 2024

Language: Английский

---

Large-scale rice mapping under spatiotemporal heterogeneity using multi-temporal SAR images and explainable deep learning

ISPRS Journal of Photogrammetry and Remote Sensing, Journal Year: 2024, Volume and Issue: 220, P. 395 - 412

Published: Dec. 31, 2024

Language: Английский

---

Cultivated land segmentation in RGB remote sensing images: non-uniform regularization with kernel space and graph cut

Published: June 13, 2024

To improve the application efficiency of RGB remote sensing images in agricultural land resource surveys, a cultivated segmentation algorithm based on kernel space non-uniform regularization classification and improved graph cut was proposed. Firstly, extracting texture color features using Local Binary Pattern (LBP), Gabor filters, RGB, HSV space, respectively. Next, introducing method to map data from lowdimension high-dimension, construct space-based sparse representation model classify segment pixel level. Finally, an innovative is enhanced by incorporating Gaussian distribution redefine penalty term for homogeneous regions new gradient measure define boundaries. This approach effectively removes scatter restricts boundary. The average accuracy F1 score classifier proposed this paper are about 2% 3% higher than those recent regularized subspace classifiers, Compared with Graph algorithm, has mIoU improvement 9%. whole 95.43%, 88.56%. comparison accuracy, which proves that can adapt scene effective.

Language: Английский

---

R-Unet: A Deep Learning Model for Rice Extraction in Rio Grande do Sul, Brazil

Remote Sensing, Journal Year: 2023, Volume and Issue: 15(16), P. 4021 - 4021

Published: Aug. 14, 2023

Rice is one of the world’s three major food crops, second only to sugarcane and corn in output. Timely accurate rice extraction plays a vital role ensuring security. In this study, R-Unet for was proposed based on Sentinel-2 time-series Sentinel-1, including an attention-residual module multi-scale feature fusion (MFF) module. The deepened network depth encoder prevented information loss. MFF fused high-level low-level features at channel spatial scales. After training, validation, testing seven datasets, performed best test samples Dataset 07, which contained optical synthetic aperture radar (SAR) features. Precision, intersection, union (IOU), F1-score, Matthews correlation coefficient (MCC) were 0.948, 0.853, 0.921, 0.888, respectively, outperforming baseline models. Finally, comparative analysis between classic models completed 07. results showed that had effect, highest scores precision, IOU, MCC, F1-score increased by 5.2%, 14.6%, 11.8%, 9.3%, respectively. Therefore, study can combine open-source sentinel images extract timely accurately, providing important governments implement decisions agricultural management.

Language: Английский

---

A Novel Rice Mapping Method Based on Multi-Temporal Polarization Decomposition Components for Single-Season Rice

Published: Sept. 20, 2023

Currently, Synthetic Aperture Radar (SAR) time-series data are more widely used in rice mapping for their ability to work all-day and all-weather. However, most researches only use backscattering coefficients while neglecting polarimetric information SAR data. Polarimetric can characterize the scattering mechanism of ground objects thus greatly simplify task. To address this issue, study proposes a feature combination based on m/X decomposition method classic UNet model extract planting areas with Sentinel-1 dual-polarized The training accuracy validation set reach 98.7%, mIoU mPA achieve 91.74% 95.66%, respectively. These experimental results indicate that proposed is simple effective, be generalized large-scale mapping.

Language: Английский

---

Single-Season Rice Area Mapping by Combining Multi-Temporal Polarization Decomposition Components and the Two-Stage Segmentation Method

Agriculture, Journal Year: 2023, Volume and Issue: 14(1), P. 2 - 2

Published: Dec. 19, 2023

Recently, Synthetic Aperture Radar (SAR) data, especially Sentinel-1 have been increasingly used in rice mapping research. However, current studies usually use long time series data as the source to represent differences between and other ground objects, crops, which results complex models large computational complexity during classification. To address this problem, a novel method for single season is proposed, based on principle that scattering mechanism of paddies early flooding period strongly influenced by water bodies, causing volume be lower than crops. Thus, feature combination can effectively stably extract planting areas was constructed combining multi-temporal using dual-polarization SAR so simple semantic segmentation model could realize high-precision tasks. A two-stage structure introduced further improve result with Omni-dimensional Dynamic Convolution Residual Segmentation (ODCRS model) bone model. In experiment, Suihua City, Heilongjiang Province selected study site, VH/VV polarized satellite 2022 source. The accuracy ODCRS 88.70%, user 84.19% field survey data. Furthermore, experiments different years regions also proved effectiveness stability proposed method.

Language: Английский

---

A Novel Wetland Monitoring Method Based on an Improved Cascade Forest Model Using Temporal-Spatial-Polarimetric Data from Gf-3 Images: A Case Study of the Yellow River Delta

Published: Jan. 1, 2024

Download This Paper Open PDF in Browser Add to My Library Share: Permalink Using these links will ensure access this page indefinitely Copy URL DOI

Language: Английский

---