EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

CleanCam: A labelled underwater image dataset for camera-viewport fouling assessment in aquaculture monitoring

Khoa Nguyen, Anh Hoang, Nhat-Anh Tran, Nguyet-Anh Tran, Minh-Hoang Pham, Khoi Phan, Van-Dinh Nguyen, Dzung V Dinh, Cuong Do

Data in Brief · 2026-08 · 68 · 113137

DOI: 10.1016/j.dib.2026.113137 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Data in Brief, 2026-08. Verfasst von Khoa Nguyen und 8 weiteren Autor:innen. Beteiligte Einrichtung: College of Engineering and Computer Science, VinUniversity, Vinhomes Ocean Park, Gia Lam District, Hanoi, 100000, Vietnam. Die Arbeit nennt 20 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Trường Đại học VinUni.

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

CleanCam is a labelled underwater image dataset for assessing fouling on camera viewports in aquaculture monitoring. Real images were sampled every 10 s from fixed underwater videos recorded by two camera nodes in golden pompano (Trachinotus blochii) rearing tanks at the Research Institute for Aquaculture No. 1 in Hai Phong, Viet Nam. The real subset contains 18,972 RGB JPEG images collected across 20 effective collection days and annotated on a five-level ordinal scale from L1 clean to L5 heavy obstruction. The labels describe material attached to the camera viewport, rather than generic image quality. Temporal context was used during annotation to distinguish persistent viewport deposits from turbidity, suspended particles, lighting variation, and other water-column effects. The release also includes 3,600 split-consistent synthetic images for Levels 3 to 5, giving 22,572 images in total. The Zenodo package provides image folders, image-level metadata, synthetic provenance records, deterministic capture-disjoint train, validation, and test splits, a data dictionary, release manifests, and documentation. The accompanying public code reproduces the manuscript figures and tables and supports release validation, dataset characterization, synthetic-subset analysis, and reference classification workflows. CleanCam can be reused for ordinal image classification, cleaning/no-cleaning threshold studies, synthetic augmentation experiments, leakage-aware benchmarking, and image-stream quality control in long-term aquatic monitoring systems.

Schlagwörter

Stichwörter der Autor:innen

AquacultureBiofoulingCamera viewportImage quality controlOrdinal labelsReproducibilitySynthetic augmentationUnderwater imaging

Autor:innen und Institutionen

  • College of Engineering and Computer Science, VinUniversity, Vinhomes Ocean Park, Gia Lam District, Hanoi, 100000, Vietnam.
  • Smart Green Transformation Center (GREEN-X), VinUniversity, Vinhomes Ocean Park, Gia Lam District, Hanoi, 100000, Vietnam.

Förderung

  • Trường Đại học VinUni

Zitieren

APA

Nguyen, K., Hoang, A., Tran, N. A., Tran, N. A., Pham, M. H., Phan, K., Nguyen, V. D., Dinh, D. V., & Do, C. (2026). CleanCam: A labelled underwater image dataset for camera-viewport fouling assessment in aquaculture monitoring. Data in Brief, 68, 113137. https://doi.org/10.1016/j.dib.2026.113137

Vancouver

Nguyen K, Hoang A, Tran NA, Tran NA, Pham MH, Phan K, et al. CleanCam: A labelled underwater image dataset for camera-viewport fouling assessment in aquaculture monitoring. Data in Brief. 2026;68:113137. doi:10.1016/j.dib.2026.113137

BibTeX
@article{Nguyen2026,
  author = {Nguyen, Khoa and Hoang, Anh and Tran, Nhat-Anh and Tran, Nguyet-Anh and Pham, Minh-Hoang and Phan, Khoi and Nguyen, Van-Dinh and Dinh, Dzung V and Do, Cuong},
  title = {CleanCam: A labelled underwater image dataset for camera-viewport fouling assessment in aquaculture monitoring},
  journal = {Data in Brief},
  year = {2026},
  volume = {68},
  pages = {113137},
  doi = {10.1016/j.dib.2026.113137}
}
Kostenlos starten →