Optimizing automated classification for zooplankton in coastal conditions: the impact of model selection, imaging instruments, and colour information

This dataset contains the code and data that were used in the article "Optimizing automated classification for zooplankton in coastal conditions: the impact of model selection, imaging instruments, and colour information" to be published at bioRxiv. This paper was based on annotated zooplankton images of the ISIIS-DPI (also referred to as mDPI) in the North Sea in June 2023, with CPICS on the southwest of The Netherlands, the VPR in the Belgian North Sea and the Pi-10 in the North Sea. The paper compares and analyses methods for automated classification on data from all these instruments with Convolutional Neural Networks. This dataset contains i) the code in which the CNNs were trained, the results were analysed and the figures from the manuscript were created ii) the annotated training sets that are not already published elsewhere, iii) the test set predictions of each trained CNN and iv) some example images of each training set. To use the code, download the folder zip-archive 'code', 'ml_data', the training sets, and to reproduce Fig. 1 of the manuscript, 'training_set_examples.zip'. Then, check the required libraries in requirements.txt and, if desired, modify the paths to data files in CNN_comparison_init.py. The figures were generated in the .ipynb-notebook, and CNN training was performed in the scripts 'cnn_train_CPICS.py', 'cnn_train_ISIIS.py', 'cnn_train_PI10.py', and 'cnn_train_VPR.py'. Future updated versions of the code may be posted to the related github-repository. The CPICS and Pi-10 training data are provided here. The training data of the ISIIS-DPI can be found at: https://doi.org/10.5281/zenodo.20762203. A more recent version of the Pi-10 training data is at: https://doi.org/10.5194/essd-2026-215. The VPR training data was downloaded from: https://doi.org/10.14284/607.