Towards a training data model for artificial intelligence in earth observation

Yue, Peng; Shangguan, Boyi; Hu, Lei; Jiang, Liangcun; Zhang, Chenxiao; Cao, Zhipeng; Pan, Yinyin

doi:10.1080/13658816.2022.2087223

Cited by 19 publications

(7 citation statements)

References 29 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Its bounding boxes are represented using either 2D or 3D geometries. Yue et al (2022) propose a Training data Markup Language (TDML) for producing Machine learning training data, which defines a UML model and encodings consistent with the OGC standards baseline. It supports the exchange and retrieval of the geospatial machine learning training data in the Web environment, which is consistent with the ubiquitous JSON/XML encoding on the Web.…”

Section: Training Data Specificationsmentioning

confidence: 99%

“…Annotation files in four kinds of formats are provided in FASDD, i.e., the JSON format defined by the TDML (Yue et al, 2022), the XML format adopted by the PASCAL VOC (Everingham et al, 2015) dataset, the JSON format adopted by the Microsoft COCO (Lin et al, 2014) dataset, and the text format adopted by models of YOLO (Redmon et al, 2016) series.…”

Section: Data Annotationmentioning

confidence: 99%

See 1 more Smart Citation

FASDD: An Open-access 100,000-level Flame and Smoke Detection Dataset for Deep Learning in Fire Detection

Wang

Jiang

Yue

et al. 2023

Preprint

View full text Add to dashboard Cite

Abstract. With the advancement of computer vision, artificial intelligence, and remote sensing technologies, deep learning algorithms are increasingly used in terrestrial, airborne, and spaceborne-based fire detection systems. The performance and generalization of these data-driven fire detection algorithms, however, are restricted by the limited number and source of fire detection datasets. A large-scale fire detection benchmark dataset covering complex and varied fire scenarios is urgently needed. This work constructs a 100,000-level Flame and Smoke Detection Dataset (FASDD) based on multi-source heterogeneous flame and smoke images. It holds rich variations in image size, resolution, illumination (day and night), scenario (indoor and outdoor), image range (far and near), viewing angle (top view and side view), platform (surveillance cameras, drones, and satellites), and data source (Internet, social media, and open-access fire datasets). To the best of our knowledge, FASDD is currently the most versatile and comprehensive dataset for fire detection. It provides a challenging benchmark to drive the continuous evolution of fire detection models. Additionally, we formulate a unified workflow for preprocessing, annotation, and quality control of fire samples. Out-of-the-box annotations are published in four different formats for training deep learning models. Extensive performance evaluations based on classical methods show that most of the object detection models trained on FASDD can achieve satisfactory fire detection results, and especially YOLOv5x achieves nearly 80 % mAP@0.5 accuracy on heterogeneous images spanning two domains of computer vision and remote sensing. And the application in wildfire location demonstrates that deep learning models trained on our dataset can be used in recognizing and monitoring forest fires. Deep learning models trained with FASDD can be simultaneously deployed on satellites, drones, and ground sensors, thus realizing collaborative fire observation and detection in a space-air-ground integrated network environment. The dataset is available from the Science Data Bank website at https://doi.org/10.57760/sciencedb.j00104.00103 (Wang et al., 2022a).

show abstract

Section: Training Data Specificationsmentioning

confidence: 99%

Section: Data Annotationmentioning

confidence: 99%

FASDD: An Open-access 100,000-level Flame and Smoke Detection Dataset for Deep Learning in Fire Detection

Wang

Jiang

Yue

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…61 Open access to research data and other digital research resources according to the FAIR principles, assuring the findability, accessibility, interoperability and reuse of data is essential. 62 For dermoscopy, the international imaging initiatives Human Against Machine (HAM) and International Skin Imaging Collaboration (ISIC) are good examples of collective open-source databases, providing many dermoscopy images. Acquiring a large data set of relatively new imaging modalities such as OCT is still challenging, as OCT is not widely used yet.…”

Section: Ch a L L E Nge S A N D Fu T U R E Dir Ec Tionsmentioning

confidence: 99%

“…In order to train a DL model, enormous amounts of correctly labelled data should be fed to the model 61 . Open access to research data and other digital research resources according to the FAIR principles, assuring the findability, accessibility, interoperability and reuse of data is essential 62 . For dermoscopy, the international imaging initiatives Human Against Machine (HAM) and International Skin Imaging Collaboration (ISIC) are good examples of collective open‐source databases, providing many dermoscopy images.…”

Section: Challenges and Future Directionsmentioning

confidence: 99%

The application of artificial intelligence in the detection of basal cell carcinoma: A systematic review

Widaatalla

Wolswijk

Adan

et al. 2023

Acad Dermatol Venereol

View full text Add to dashboard Cite

Basal cell carcinoma (BCC) is one of the most common forms of cancer, 1-3 with a rising incidence worldwide. To date, histopathological examination of a punch biopsy is the gold standard to distinguish BCC from alternative diagnoses and to determine the BCC subtype. 1,4 However, a punch biopsy is an invasive procedure, with risks of pain and bleeding during the procedure and the additional chance of infection and/or scarring. 5 Moreover, awaiting histopathological assessment may be stressful for many patients. Considering the rising incidence of BCC, which causes a major burden on healthcare systems, finding alternatives to an invasive biopsy is desired.The last century has seen many advances in medical imaging from the first X-ray to more sophisticated imaging techniques, such as computed tomography (CT) and magnetic resonance imaging (MRI). 6 This progress, concurrent with advances in computational power and machine learning techniques, often termed artificial intelligence (AI) in the literature, eased the transformation of medical images into quantitative minable data that could be used to build diagnostic, predictive and prognostic clinical decision support systems (cDSS). 7,8 Quantitative imaging analysis techniques that are being extensively investigated to develop cDSS include handcrafted radiomics and deep learning. Handcrafted radiomic features

show abstract

“…In the European Union (EU) the Sentinel for Common Agricultural Policy program (Sen4CAP) [2] focuses on developing tools and analytics to support the verification of direct payments to farmers with underlying environmental conditionalities such as the adoption of environmentally-friendly [50] and crop diversification [51] practices based on real-time monitoring by the European Space Agency's (ESA) Sentinel high-resolution satellite constellation [1] to complement on site verification. Recently, the volume and diversity of space-borne Earth Observation (EO) data [63] and post-processing tools [18,61,70] has increased exponentially. This wealth of resources, in combination with important developments in machine learning for computer vision [20,28,53], provides an important opportunity for the development of tools for the automated monitoring of crop development.…”

Section: Introductionmentioning

confidence: 99%

ViTs for SITS: Vision Transformers for Satellite Image Time Series

Tarasiou¹,

Chavez²,

Zafeiriou³

2023

Preprint

View full text Add to dashboard Cite

In this paper we introduce the Temporo-Spatial Vision Transformer (TSViT), a fully-attentional model for general Satellite Image Time Series (SITS) processing based on the Vision Transformer (ViT). TSViT splits a SITS record into non-overlapping patches in space and time which are tokenized and subsequently processed by a factorized temporo-spatial encoder. We argue, that in contrast to natural images, a temporal-then-spatial factorization is more intuitive for SITS processing and present experimental evidence for this claim. Additionally, we enhance the model's discriminative power by introducing two novel mechanisms for acquisition-time-specific temporal positional encodings and multiple learnable class tokens. The effect of all novel design choices is evaluated through an extensive ablation study. Our proposed architecture achieves state-of-the-art performance, surpassing previous approaches by a significant margin in three publicly available SITS semantic segmentation and classification datasets. All model, training and evaluation codes are made publicly available in https://github.com/michaeltrs/DeepSatModels.

show abstract

Towards a training data model for artificial intelligence in earth observation

Cited by 19 publications

References 29 publications

FASDD: An Open-access 100,000-level Flame and Smoke Detection Dataset for Deep Learning in Fire Detection

FASDD: An Open-access 100,000-level Flame and Smoke Detection Dataset for Deep Learning in Fire Detection

The application of artificial intelligence in the detection of basal cell carcinoma: A systematic review

ViTs for SITS: Vision Transformers for Satellite Image Time Series

Contact Info

Product

Resources

About