Introducing Eurosat: A Novel Dataset and Deep Learning Benchmark for Land Use and Land Cover Classification

Helber, Patrick; Bischke, Benjamin; Dengel, Andreas; Borth, Damian

doi:10.1109/igarss.2018.8519248

Cited by 139 publications

(81 citation statements)

References 8 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…With the release of the geo-referenced EuroSAT we aim to make the large amount of Sentinel-2 satellite imagery accessible for machine learning approaches. There effectiveness was successfully demonstrated in [32], [33], [34].…”

Section: B Dataset Creationmentioning

confidence: 93%

“…We chose satellite images associated with the cities covered in the European Urban Atlas. The covered cities are distributed over the 34 In order to improve the chance of getting valuable image patches, we selected satellite images with a low cloud level. Besides the possibility to generate a cloud mask, ESA provides a cloud level value for each satellite image allowing to quickly select images with a low percentage of clouds covering the land scene.…”

Section: A Satellite Image Acquisitionmentioning

confidence: 99%

See 1 more Smart Citation

EuroSAT: A Novel Dataset and Deep Learning Benchmark for Land Use and Land Cover Classification

Helber

Bischke

Dengel

et al. 2019

IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing

Self Cite

906

478

View full text Add to dashboard Cite

In this paper, we address the challenge of land use and land cover classification using Sentinel-2 satellite images. The Sentinel-2 satellite images are openly and freely accessible provided in the Earth observation program Copernicus. We present a novel dataset based on Sentinel-2 satellite images covering 13 spectral bands and consisting out of 10 classes with in total 27,000 labeled and geo-referenced images. We provide benchmarks for this novel dataset with its spectral bands using state-of-the-art deep Convolutional Neural Network (CNNs). With the proposed novel dataset, we achieved an overall classification accuracy of 98.57%. The resulting classification system opens a gate towards a number of Earth observation applications. We demonstrate how this classification system can be used for detecting land use and land cover changes and how it can assist in improving geographical maps. The geo-referenced dataset EuroSAT is made publicly available at https://github.com/phelber/eurosat.

show abstract

Section: B Dataset Creationmentioning

confidence: 93%

Section: A Satellite Image Acquisitionmentioning

confidence: 99%

EuroSAT: A Novel Dataset and Deep Learning Benchmark for Land Use and Land Cover Classification

Helber

Bischke

Dengel

et al. 2019

IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing

Self Cite

906

478

View full text Add to dashboard Cite

show abstract

“…• DL systems can have millions of parameters • RS data may not be labelled • Hyperspectral data is a very large data cube with many layers, while DL algorithms are typically trained from very small RGB images • Light detection and light detection and ranging (LiDAR) have insufficient literature as the data is not an image, but a point cloud • Gridded searches or random methods are required for optimization which can be very time consuming [10] Helber et al [12], have also emphasized the importance of having a high-quality dataset for training and classification. They state that one of the challenges to creating these training sets has been access to ground truth datasets that are reliably labelled.…”

Section: Training Datamentioning

confidence: 99%

Data Mining and Machine Learning in Earth Observation – An Application for Tracking Historical Algal Blooms

Farias¹,

Sun²

2020

8th International Conference on Signal, Image Processing and Pattern Recognition

View full text Add to dashboard Cite

The data produced from Earth Observation (EO) satellites has recently become so abundant that manual processing is sometimes no longer an option for analysis. The main challenges for studying this data are its size, its complex nature, a high barrier to entry, and the availability of datasets used for training data. Because of this, there has been a prominent trend in techniques used to automate this process and host the processing in massive online cloud servers. These processes include data mining (DM) and machine learning (ML). The techniques that will be discussed include: clustering, regression, neural networks, and convolutional neural networks (CNN). This paper will show how some of these techniques are currently being used in the field of earth observation as well as discuss some of the challenges that are currently being faced. Google Earth Engine (GEE) has been chosen as the tool for this study. GEE is currently able to display 40 years of historical satellite imagery, including publicly available datasets such as Landsat, and Sentinel data from Copernicus. Using EO data from Landsat and GEE as a processing tool, it is possible to classify and discover historical algal blooms over the period of ten years in the Baltic Sea surrounding the Swedish island of Gotland. This paper will show how these technical advancements including the use of a cloud platform enable the processing and analysis of this data in minutes.

show abstract

“…Within the scope of this work, we consider the problem of missing modalities on the exemplary use case of building footprints segmentation in a multi-modal setup. Our segmentation network, depicted in (2) in Fig 1, is trained on two modalities relying on optical (RGB) and depth information. Our goal is to apply this model even in the absence of one modality by using a second CNN that is able to create a synthetic representation for the missing modality.…”

Section: Fusing Complementary Information Across Modalities Can Leadmentioning

confidence: 99%

“…With the success of deep learning, convolutional neural networks (CNNs) have shown to be a powerful approach to extract such higher level features and outperform many traditional computer vision methods [1]. Applications of CNN's in this context are manifold and have been used to solve problems ranging from regression and classification [2] over to object detection to semantic & instance segmentation [3].…”

Section: Introductionmentioning

confidence: 99%

Overcoming Missing and Incomplete Modalities with Generative Adversarial Networks for Building Footprint Segmentation

Bischke

Helber

Koenig³

et al. 2018

2018 International Conference on Content-Based Multimedia Indexing (CBMI)

Self Cite

View full text Add to dashboard Cite

The integration of information acquired with different modalities, spatial resolution and spectral bands has shown to improve predictive accuracies. Data fusion is therefore one of the key challenges in remote sensing. Most prior work focusing on multi-modal fusion, assumes that modalities are always available during inference. This assumption limits the applications of multi-modal models since in practice the data collection process is likely to generate data with missing, incomplete or corrupted modalities. In this paper, we show that Generative Adversarial Networks can be effectively used to overcome the problems that arise when modalities are missing or incomplete. Focusing on semantic segmentation of building footprints with missing modalities, our approach achieves an improvement of about 2% on the Intersection over Union (IoU) against the same network that relies only on the available modality.

show abstract

Introducing Eurosat: A Novel Dataset and Deep Learning Benchmark for Land Use and Land Cover Classification

Cited by 139 publications

References 8 publications

EuroSAT: A Novel Dataset and Deep Learning Benchmark for Land Use and Land Cover Classification

EuroSAT: A Novel Dataset and Deep Learning Benchmark for Land Use and Land Cover Classification

Data Mining and Machine Learning in Earth Observation – An Application for Tracking Historical Algal Blooms

Overcoming Missing and Incomplete Modalities with Generative Adversarial Networks for Building Footprint Segmentation

Contact Info

Product

Resources

About