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

By Patrick Helber, Benjamin Bischke, Andreas R. Dengel, Damian Borth. 2017

2,804 citations recorded by the source corpus.

In this paper, we present a patch-based land use and land cover classification approach using Sentinel-2 satellite images. The Sentinel-2 satellite images are openly and freely accessible, and are provided in the earth observation program Copernicus. We present a novel dataset, based on these images that covers 13 spectral bands and is comprised of ten classes with a total of 27 000 labeled and geo-referenced images. Benchmarks are provided for this novel dataset with its spectral bands using state-of-the-art deep convolutional neural networks. An overall classification accuracy of 98.57% was achieved with the proposed novel dataset. The resulting classification system opens a gate toward 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.

Subfields: Atmospheric Science, Computer Vision and Pattern Recognition, Ecology, Media Technology, Evaluation & Benchmarks

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Atmospheric Science

In Computer Vision and Pattern Recognition

In Ecology

In Media Technology

In Evaluation & Benchmarks

Citations in the corpus

Learning Transferable Visual Models From Natural Language Supervision (2021, methodology): “Two datasets, Flowers102 and EuroSAT underperform one-shot models.”
On Robustness and Transferability of Convolutional Neural Networks (2020, background): “The specialized group contains remote sensing datasets EuroSAT [25] and Resisc45 [8], and medical image datasets Patch Camelyon [67] and Diabetic Retinopathy [34].”
A Large-scale Study of Representation Learning with the Visual Task Adaptation Benchmark (2019, methodology): “EuroSAT (Helber et al., 2019) The task consists in classifying Sentinel-2 satellite images into 10 different types of land use (Residential, Industrial, River, Highway, etc).”
Florence: A New Foundation Model for Computer Vision (2021, background): “The datasets in the benchmark include: CropDisease (Mohanty et al., 2016) (plant leaf images, 38 disease states over 14 plant species), EuroSAT (Helber et al., 2019) (RGB satellite images, 10 categories), ISIC 2018 (Codella et al., 2019; Tschandl et al., 2018) (dermoscopic images of skin lesions, 7…”
Learning to Prompt for Vision-Language Models (2021, background): “Tuning the sentence structure could bring further improvements, e.g., putting “a type of ﬂower” after the class token for Flowers102, keeping only “texture” in the context for DTD, and adding “centered” before “satellite photo” for EuroSAT.”
ELEVATER: A Benchmark and Toolkit for Evaluating Language-Augmented Visual Models (2022, background): “Visual recognition has become ubiquitous in our society [60], with applications in geolocalization [51], action recognition [57], street number transcription [44], satellite remote sensing [22], medical imaging [62], self-driving cars [19], etc.”
Visual Prompt Tuning (2022, neutral): “800/1000 200 10,000 Caltech101 [43] 102 6,084 DTD [11] 47 1,880 Flowers102 [55] 102 6,149 Pets [57] 37 3,669 SVHN [54] 10 26,032 Sun397 [76] 397 21,750 Patch Camelyon [69] Specialized 2 800/1000 200 32,768 EuroSAT [29] 10 5,400 Resisc45 [10] 45 6,300 Retinopathy [36] 5 42,670 Clevr/count [34]”
CLIP-Adapter: Better Vision-Language Models with Feature Adapters (2021, neutral): “Following CLIP and CoOp, we select 11 image classification datasets to validate CLIP-Adapter’s effectiveness, namely Ima-geNet [9], StanfordCars [31], UCF101 [55], Cal-tech101 [13], Flowers102 [47], SUN397 [67], DTD [7], EuroSAT [21], FGVCAircraft [45], OxfordPets [48], and Food101 [3].”
Conditional Prompt Learning for Vision-Language Models (2022, background): “Specifically, the benchmark includes ImageNet [6] and Caltech101 [11] for classification on generic objects; OxfordPets [38], StanfordCars [28], Flowers102 [36], Food101 [2] and FGVCAircraft [35] for fine-grained classification; SUN397 [55] for scene recognition; UCF101 [46] for action recognition; ”
Tip-Adapter: Training-free CLIP-Adapter for Better Vision-Language Modeling (2021, methodology): “We conduct experiments for Tip-Adapter on 11 image classification datasets: ImageNet [9], StandfordCars [36], UCF101 [56], Caltech101 [15], Flowers102 [47], SUN397 [66], DTD [8], EuroSAT [25], FGVCAircraft [44], OxfordPets [50], and Food101 [3].”
How Well Do Self-Supervised Models Transfer? (2020, background): “Thirdly: While core academic vision research is happy to focus on ImageNet as a benchmark, the wider community of computer vision practitioners work with diverse data types from medical [54] to agricultural [40], to earth-observation [24] data and beyond.”
Mind the Gap: Understanding the Modality Gap in Multi-modal Contrastive Representation Learning (2022, background): “We study whether changing the gap will affect CLIP (ViT-B/16)’s performances on various downstream tasks, including coarse-grained classification (CIFAR10 and CIFAR100), fine-grained classification (EuroSAT [22]), and optical character recognition (SVHN, HatefulMemes [28]).”
Editing Models with Task Arithmetic (2022, positive): “We ﬁne-tune ﬁve CLIP models on MNIST and ﬁve models EuroSAT, varying only the random seed.”
Remote Sensing Image Scene Classification Meets Deep Learning: Challenges, Methods, Benchmarks, and Opportunities (2020, neutral): “4& 6 [126], RSSCN7 [125], Brazilian Coffee Scene [85], RSC11 [128], SIRI-WHU [127], RSCI-CB [129], AID [79], NWPU-RESISC45 [80], OPTIMAL-31 [74], EuroSAT [130],”