Benchmark of plankton images classification: emphasizing features extraction over classifier complexity

Panaïotis, Thelma; Amblard, Emma; Boniface-Chang, Guillaume; Dulac-Arnold, Gabriel; Woodward, Benjamin; Irisson, Jean-Olivier

doi:10.5194/essd-18-945-2026

Articles | Volume 18, issue 2

https://doi.org/10.5194/essd-18-945-2026

© Author(s) 2026. This work is distributed under
the Creative Commons Attribution 4.0 License.

https://doi.org/10.5194/essd-18-945-2026

© Author(s) 2026. This work is distributed under
the Creative Commons Attribution 4.0 License.

Articles | Volume 18, issue 2

Review article

|

05 Feb 2026

Review article |

| 05 Feb 2026

Benchmark of plankton images classification: emphasizing features extraction over classifier complexity

Thelma Panaïotis, Emma Amblard, Guillaume Boniface-Chang, Gabriel Dulac-Arnold, Benjamin Woodward, and Jean-Olivier Irisson

Supplement

https://doi.org/10.5194/essd-18-945-2026-supplement

Data sets

ISIISNet : plankton images captured with the ISIIS (In-situ Ichthyoplankton Imaging System) Thelma Panaïotis et al. https://doi.org/10.17882/101950

FlowCAMNet : plankton images captured with the FlowCAM Laetitia Jalabert et al. https://doi.org/10.17882/101961

UVP6Net : plankton images captured with the UVP6 Marc Picheral et al. https://doi.org/10.17882/101948

ZooCAMNet : plankton images captured with the ZooCAM Jean-Baptiste Romagnan et al. https://doi.org/10.17882/101928

ZooScanNet: plankton images captured with the ZooScan Amanda Elineau et al. https://doi.org/10.17882/55741

WHOI-Plankton. Annotated Plankton Images - Data Set for Developing and Evaluating Classification Methods H. M. Sosik et al. https://doi.org/10.1575/1912/7341

Model code and software

ThelmaPana/plankton_classif Thelma Panaïotis and Emma Amblard https://doi.org/10.5281/zenodo.17937437

Download

Article (2037 KB)
Full-text XML

Short summary

To address the lack of performance benchmark in plankton image classification, we evaluated machine learning methods on six large and realistic datasets. Testing both traditional and more recent convolutional neural networks (deep learning), we find that relatively small deep networks performed best, particularly for uncommon classes, because they extract richer features. Our results indicate that such compact models are sufficient for classifying small grayscale plankton images.