Comparison of Random Forest, k-Nearest Neighbor, and Support Vector Machine Classifiers for Land Cover Classification Using Sentinel-2 Imagery

Sensors, Journal Year: 2017, Volume and Issue: 18(1), P. 18 - 18

Published: Dec. 22, 2017

In previous classification studies, three non-parametric classifiers, Random Forest (RF), k-Nearest Neighbor (kNN), and Support Vector Machine (SVM), were reported as the foremost classifiers at producing high accuracies. However, only a few studies have compared performances of these with different training sample sizes for same remote sensing images, particularly Sentinel-2 Multispectral Imager (MSI). this study, we examined RF, kNN, SVM land use/cover using image data. An area 30 × km² within Red River Delta Vietnam six types was classified 14 sizes, including balanced imbalanced, from 50 to over 1250 pixels/class. All results showed overall accuracy (OA) ranging 90% 95%. Among sub-datasets, produced highest OA least sensitivity followed consecutively by RF kNN. relation size, all similar (over 93.85%) when size large enough, i.e., greater than 750 pixels/class or representing an approximately 0.25% total study area. The achieved both imbalanced datasets.

Language: Английский

---

Review of studies on tree species classification from remotely sensed data

Remote Sensing of Environment, Journal Year: 2016, Volume and Issue: 186, P. 64 - 87

Published: Aug. 20, 2016

Language: Английский

---

First Experience with Sentinel-2 Data for Crop and Tree Species Classifications in Central Europe

Remote Sensing, Journal Year: 2016, Volume and Issue: 8(3), P. 166 - 166

Published: Feb. 23, 2016

The study presents the preliminary results of two classification exercises assessing capabilities pre-operational (August 2015) Sentinel-2 (S2) data for mapping crop types and tree species. In first case study, an S2 image was used to map six summer species in Lower Austria as well winter crops/bare soil. Crop type maps are needed account crop-specific water use agricultural statistics. information is also useful parametrize growth models yield estimation, retrieval vegetation biophysical variables using radiative transfer models. second aimed seven different deciduous coniferous Germany. Detailed about distribution important forest management assess potential impacts climate change. our assessment, were produced at 10 m spatial resolution by combining ten spectral channels with 20 pixel size. A supervised Random Forest classifier (RF) deployed trained appropriate ground truth. both studies, confirmed its expected produce reliable land cover maps. Cross-validated overall accuracies ranged between 65% (tree species) 76% (crop types). high value red-edge shortwave infrared (SWIR) bands mapping. Also, blue band sites. S2-bands near amongst least channels. object based analysis (OBIA) classical pixel-based achieved comparable results, mainly cropland. As only single date acquisitions available this full could not be assessed. future, twin satellites will offer global coverage every five days therefore permit concurrently exploit unprecedented temporal resolution.

Language: Английский

---

Flood hazard risk assessment model based on random forest

Journal of Hydrology, Journal Year: 2015, Volume and Issue: 527, P. 1130 - 1141

Published: June 14, 2015

Language: Английский

---

Remote Sensing Technologies for Enhancing Forest Inventories: A Review

Canadian Journal of Remote Sensing, Journal Year: 2016, Volume and Issue: 42(5), P. 619 - 641

Published: July 27, 2016

Forest inventory and management requirements are changing rapidly in the context of an increasingly complex set economic, environmental, social policy objectives. Advanced remote sensing technologies provide data to assist addressing these escalating information needs support subsequent development parameterization models for even broader range needs. This special issue contains papers that use a variety derive forest or inventory-related information. Herein, we review potential 4 advanced technologies, which posit as having greatest influence inventories designed characterize resource strategic, tactical, operational planning: airborne laser scanning (ALS), terrestrial (TLS), digital aerial photogrammetry (DAP), high spatial resolution (HSR)/very (VHSR) satellite optical imagery. ALS, particular, has proven be transformative technology, offering required detail accuracy across large areas diverse types. The coupling DAP with ALS will likely have impact on practices next decade, providing capacity suite attributes, well monitoring growth over time.

Language: Английский

---

UAV Remote Sensing for Urban Vegetation Mapping Using Random Forest and Texture Analysis

Remote Sensing, Journal Year: 2015, Volume and Issue: 7(1), P. 1074 - 1094

Published: Jan. 19, 2015

Unmanned aerial vehicle (UAV) remote sensing has great potential for vegetation mapping in complex urban landscapes due to the ultra-high resolution imagery acquired at low altitudes. Because of payload capacity restrictions, off-the-shelf digital cameras are widely used on medium and small sized UAVs. The limitation spectral can be reduced by incorporating texture features robust classifiers. Random Forest been satellite applications, but its usage UAV image classification not well documented. objectives this paper were propose a hybrid method using analysis accurately differentiate land covers vegetated areas, analyze how accuracy changes with window size. Six least correlated second-order measures calculated nine different sizes added original Red-Green-Blue (RGB) images as ancillary data. A classifier consisting 200 decision trees was spectral-textural feature space. Results indicated following: (1) outperformed traditional Maximum Likelihood showed similar performance object-based classification; (2) inclusion improved significantly; (3) followed an inverted U relationship results demonstrate that provides efficient ideal platform mapping. proposed shows good differentiating drawbacks adopting same time.

Language: Английский

---

Random forest meteorological normalisation models for Swiss PM<sub>10</sub> trend analysis

Atmospheric chemistry and physics, Journal Year: 2018, Volume and Issue: 18(9), P. 6223 - 6239

Published: May 3, 2018

Abstract. Meteorological normalisation is a technique which accounts for changes in meteorology over time an air quality series. Controlling such helps support robust trend analysis because there more certainty that the observed trends are due to emissions or chemistry, not meteorology. Predictive random forest models (RF; decision tree machine learning technique) were grown 31 monitoring sites Switzerland using surface meteorological, synoptic scale, boundary layer height, and variables explain daily PM10 concentrations. The RF used calculate meteorologically normalised formally tested evaluated Theil–Sen estimator. Between 1997 2016, significantly decreasing ranged between −0.09 −1.16 µg m−3 yr−1 with urban traffic experiencing greatest mean decrease concentrations at −0.77 yr−1. Similar magnitudes have been reported earlier periods indicates continuing similar rates as past. ability be interpreted was leveraged partial dependence plots relevant physical chemical processes influencing Notably, two regimes suggested by cause elevated Switzerland: one related poor dispersion conditions second resulting from high of secondary PM generation deep, photochemically active layers. meteorological process found robust, user friendly simple implement, readily interpretable suggests could useful many exploratory data situations.

Language: Английский

---

How much does multi-temporal Sentinel-2 data improve crop type classification?

International Journal of Applied Earth Observation and Geoinformation, Journal Year: 2018, Volume and Issue: 72, P. 122 - 130

Published: July 25, 2018

Language: Английский

---

Urban Flood Mapping Based on Unmanned Aerial Vehicle Remote Sensing and Random Forest Classifier—A Case of Yuyao, China

Water, Journal Year: 2015, Volume and Issue: 7(4), P. 1437 - 1455

Published: March 31, 2015

Flooding is a severe natural hazard, which poses great threat to human life and property, especially in densely-populated urban areas. As one of the fastest developing fields remote sensing applications, an unmanned aerial vehicle (UAV) can provide high-resolution data with potential for fast accurate detection inundated areas under complex landscapes. In this research, optical imagery was acquired by mini-UAV monitor serious waterlogging Yuyao, China. Texture features derived from gray-level co-occurrence matrix were included increase separability different ground objects. A Random Forest classifier, consisting 200 decision trees, used extract flooded spectral-textural feature space. Confusion assess accuracy proposed method. Results indicated following: (1) showed good performance flood mapping overall 87.3% Kappa coefficient 0.746; (2) inclusion texture improved classification significantly; (3) outperformed maximum likelihood artificial neural network, similar support vector machine. The results demonstrate that UAV ideal platform monitoring method shows capability extraction

Language: Английский

---

Geographic object-based image analysis (GEOBIA): emerging trends and future opportunities

GIScience & Remote Sensing, Journal Year: 2018, Volume and Issue: 55(2), P. 159 - 182

Published: Jan. 10, 2018

Over the last two decades (since ca. 2000), Geographic Object-Based Image Analysis (GEOBIA) has emerged as a new paradigm to analyzing high-spatial resolution remote-sensing imagery. During this time, research interests have demonstrated shift from development of GEOBIA theoretical foundations advanced geo-object-based models and their implementation in wide variety real-world applications. We suggest that such rapid evolution warrants need for systematic review defines recent developments field. Therefore, main objective paper is elucidate emerging trends discuss potential opportunities future development. The were found multiple subfields GEOBIA, including data sources, image segmentation, object-based feature extraction, modeling frameworks. It our view understanding state-of-the-art will further facilitate support study geographic entities phenomena at scales with effective incorporation semantics, informing high-quality project design, improving model performance results.

Language: Английский

---