The Classification Method Study of Crops Remote Sensing with Deep Learning, Machine Learning, and Google Earth Engine

Remote Sensing, Journal Year: 2022, Volume and Issue: 14(12), P. 2758 - 2758

Published: June 8, 2022

The extraction and classification of crops is the core issue agricultural remote sensing. precise crop types great significance to monitoring evaluation planting area, growth, yield. Based on Google Earth Engine Colab cloud platform, this study takes typical oasis area Xiangride Town, Qinghai Province, as an example. It compares traditional machine learning (random forest, RF), object-oriented (object-oriented, OO), deep neural networks (DNN), which proposes a random forest combined with network (RF+DNN) framework. In study, spatial characteristics band information, vegetation index, polarization main in were constructed using Sentinel-1 Sentinel-2 data. temporal phenology growth state analyzed curve curvature method, data screened time space. By comparing analyzing accuracy four methods, advantages RF+DNN model its application value illustrated. results showed that for during period good development, better result could be obtained whose accuracy, training, predict spent than DNN alone. overall Kappa coefficient 0.98 0.97, respectively. also higher (OA = 0.87, 0.82), object oriented 0.78, 0.70) 0.93, 0.90). scalable simple method proposed paper gives full play platform operation, can effectively improve accuracy. Timely accurate at different scales pattern change, yield estimation, safety warning.

Language: Английский

---

Twenty-meter annual paddy rice area map for mainland Southeast Asia using Sentinel-1 synthetic-aperture-radar data

Earth system science data, Journal Year: 2023, Volume and Issue: 15(4), P. 1501 - 1520

Published: April 4, 2023

Abstract. Over 90 % of the world's rice is produced in Asia–Pacific region. Synthetic-aperture radar (SAR) enables all-day and all-weather observations distribution tropical subtropical regions. The complexity cultivation patterns regions makes it difficult to construct a representative data-relevant crop model, increasing difficulty extracting distributions from SAR data. To address this problem, area mapping method for large regional or areas based on time-series Sentinel-1 data proposed study. Based analysis backscattering characteristics mainland Southeast Asia, combination spatiotemporal statistical features with good generalization ability was selected then input into U-Net semantic segmentation combined WorldCover reduce false alarms, finally 20 m resolution map five countries Asia 2019 obtained. achieved an accuracy 92.20 validation sample set, agreement obtained when comparing our other maps at national provincial levels. maximum coefficient determination R2 0.93 level 0.97 level. These results demonstrate advantages complex cropping reliability generated maps. annual paddy available https://doi.org/10.5281/zenodo.7315076 (Sun et al., 2022b).

Language: Английский

---

Automatic Rice Early-Season Mapping Based on Simple Non-Iterative Clustering and Multi-Source Remote Sensing Images

Remote Sensing, Journal Year: 2024, Volume and Issue: 16(2), P. 277 - 277

Published: Jan. 10, 2024

Timely and accurate rice spatial distribution maps play a vital role in food security social stability. Early-season mapping is of great significance for yield estimation, crop insurance, national policymaking. Taking Tongjiang City Heilongjiang Province with strong heterogeneity as study area, hierarchical K-Means binary automatic classification method based on phenological feature optimization (PFO-HKMAR) proposed, using Google Earth Engine platform Sentinel-1/2, Landsat 7/8 data. First, SAR backscattering intensity time series reconstructed used to construct optimize polarization characteristics. A new index named VH-sum built, which defined the summation VH specific periods temporal changes characteristics different land cover types. Then comes selection, optimization, reconstruction optical Finally, PFO-HKMAR established Simple Non-Iterative Clustering. can achieve early-season one month before harvest, overall accuracy, Kappa, F1 score reaching 0.9114, 0.8240 0.9120, respectively (F1 greater than 0.9). Compared two datasets Northeast China ARM-SARFS, scores are improved by 0.0507–0.1957, 0.1029–0.3945, 0.0611–0.1791, respectively. The results show that be promoted enable mapping, provide valuable timely information stakeholders decision makers.

Language: Английский

---

Mapping upland crop–rice cropping systems for targeted sustainable intensification in South China

The Crop Journal, Journal Year: 2024, Volume and Issue: 12(2), P. 614 - 629

Published: March 24, 2024

Upland crop-rice cropping systems (UCR) facilitate sustainable agricultural intensification. Accurate UCR cultivation mapping is needed to ensure food security, water management, and rural revitalization. However, datasets describing are limited in spatial coverage crop types. Mapping more challenging than identification most existing approaches rely heavily on accurate phenology calendars representative training samples, which limits its applications over large regions. We describe a novel algorithm (RRSS) for automatic of upland crop–rice using Sentinel-1 Synthetic Aperture Radar (SAR) Sentinel-2 Multispectral Instrument (MSI) data. One indicator, the VV backscatter range, was proposed discriminate another two indicators were designed by coupling greenness pigment indices further tobacco or oilseed UCR. The RRSS applied South China characterized complex smallholder rice diverse topographic conditions. This study developed 10-m maps major bowl China, Xiang-Gan-Min (XGM) region. performance validated based 5197 ground-truth reference sites, with an overall accuracy 91.92 %. There 7348 km2 areas UCR, roughly one-half them located plains. represented mainly oilseed-UCR tobacco-UCR, contributed respectively 69 % 15 area. patterns accounted only one-tenth production, can be tripled intensification from single cropping. Application fragmented subtropical regions suggested spatiotemporal robustness algorithm, could generate application at national global scales.

Language: Английский

---

Hybridization of process-based models, remote sensing, and machine learning for enhanced spatial predictions of wheat yield and quality

Computers and Electronics in Agriculture, Journal Year: 2025, Volume and Issue: 234, P. 110317 - 110317

Published: March 23, 2025

Language: Английский

---

Integration of Sentinel-1 and Sentinel-2 Data for Ground Truth Sample Migration for Multi-Temporal Land Cover Mapping

Remote Sensing, Journal Year: 2024, Volume and Issue: 16(9), P. 1566 - 1566

Published: April 28, 2024

Reliable and up-to-date training reference samples are imperative for land cover (LC) classification. However, such datasets not always available in practice. The sample migration method has shown remarkable success addressing this challenge recent years. This work investigated the application of Sentinel-1 (S1) Sentinel-2 (S2) data migration. In addition, impact various spectral bands polarizations on accuracy migrated was also assessed. Subsequently, combined S1 S2 images were classified using Support Vector Machines (SVM) Random Forest (RF) classifiers to produce annual LC maps from 2017 2021. results showed a higher (98.25%) migrations both comparison (87.68%) (96.82%) independently. Among classes, highest found water, built-up, bare land, grassland, cropland, wetland. Inquiries efficiency different polarization used that 4 8 VV water class more important, while wetland class, 5, 6, 7, 8, 8A together with superior performance. RF classifier provided better performance than SVM (higher overall, producer, user accuracy). Overall, our findings suggested shared use can be as suitable means producing high-quality samples.

Language: Английский

---

Early Mapping Method for Different Planting Types of Rice Based on Planet and Sentinel-2 Satellite Images

Agronomy, Journal Year: 2024, Volume and Issue: 14(1), P. 137 - 137

Published: Jan. 4, 2024

In Northeast China, transplanted rice cultivation has been adopted to extend the growing season and boost yields, responding limitations of cumulative temperature zone high food demand. However, direct-seeded offers advantages in water conservation labour efficiency. The precise timely monitoring distribution different planting types is key ensuring security promoting sustainable regional development. This study explores feasibility mapping various using only early-stage satellite data from season. We focused on Daxing Farm Fujin City, Jiamusi Heilongjiang Province, for cropland plot extraction Planet imagery. Utilizing Sentinel-2 imagery, we analysed differences rice’s modified normalized difference index (MNDWI) during specific phenological periods. A multitemporal Gaussian mixture model (GMM) was developed, integrated with maximum expectation algorithm, produce binarized classification outcomes. These results were employed detect surface changes map corresponding types. probability within arable plots quantified, yielding a plot-level rice-cultivation-type product. achieved an overall accuracy 91.46% classifying types, Kappa coefficient 0.89. area based land parcels showed higher R2 by 0.1109 compared pixel-based lower RMSE 0.468, indicating more accurate aligned real statistics surveys, thus validating our study’s method. approach, not requiring labelled samples or many predefined parameters, new method rapid feasible mapping, especially suitable areas China. It fills gap supporting management fields policies planting-method changes.

Language: Английский

---

Novel Harmonic-Based Scheme for Mapping Rice-Crop Intensity at a Large Scale Using Time-Series Sentinel-1 and ERA5-Land Datasets

IEEE Transactions on Geoscience and Remote Sensing, Journal Year: 2024, Volume and Issue: 62, P. 1 - 23

Published: Jan. 1, 2024

Rice-crop intensity is the annual number of rice growth cycles in a field. Monitoring on large scale vital evaluating grain production and its ecological impact. Synthetic Aperture Radar (SAR) has an all-weather imaging capability. However, existing SAR-based rice-crop mapping methods mostly focus small regions due to diversity backscatter patterns, inefficiency time-series feature extraction, unavailability phenological information scale. In this study, harmonic-based method proposed identify essential periodicities. It also suppresses short-term disturbance Sentinel-1 SAR data without setting filtering windows or assuming profile shapes. The detects troughs, eliminating requirement for point-by-point traversal mathematical operations. Annual temperature profiles are derived from ERA5-Land troughs related under various agro-climatic conditions. Then, single (135,537 km ² ), double (19,036 triple (259 ) intensities covering entire Southern China 2020 mapped 10m resolution, relying region-specific prior information. achieves overall accuracy 82.26%, can potentially support continental global task.

Language: Английский

---

Mapping Paddy Rice Planting Area in Dongting Lake Area Combining Time Series Sentinel-1 and Sentinel-2 Images

Remote Sensing, Journal Year: 2023, Volume and Issue: 15(11), P. 2794 - 2794

Published: May 27, 2023

Accurate and timely acquisition of cropping intensity spatial distribution paddy rice is not only an important basis for monitoring growth predicting yields, but also ensuring food security optimizing the agricultural production management system cropland. However, due to monsoon climate in southern China, it cloudy rainy throughout year, which makes difficult obtain accurate information on cultivation based optical time series images. Conventional image synthesis prone omission or redundancy spectral temporal features that are potentially rice-growth identification, making determine optimal high-precision mapping rice. To address these issues, we develop a method granulate effective use interval classification by extracting phenological signatures cost-effective highly results. Two steps involved this method: (1) analyzing various (spectra, polarization, seasonal regularity) identify three key periods lifespan rice; (2) identifying with highest class separation between rice, non-paddy crops, wetlands under different stages; (3) removing redundant retain feature combinations. Subsequently, obtained sets used as input data random forest classifier. The results showed overall accuracy identified was 95.44% F1 scores above 93% both single- double-cropping Meanwhile, correlation coefficient our mapped area compared official statistics at county district levels 0.86. In addition, found combining Sentinel-1 Sentinel-2 images recognition better than using alone, improved 5.82% 2.39%, confirms synergistic can effectively overcome problem missing caused clouds rain. Our study demonstrates potential distinguishing mixed rice-cropping systems subtropical regions fragmented rice-field environment, provides rational layout improvement systems.

Language: Английский

---

A comprehensive review of rice mapping from satellite data: Algorithms, product characteristics and consistency assessment

Science of Remote Sensing, Journal Year: 2024, Volume and Issue: 10, P. 100172 - 100172

Published: Oct. 25, 2024

Language: Английский

---