Several datasets representing cross-validation of convoluted neural network (CNN) classification of pollen images. In each data file, one variable represents the pollen species (or pollen type when pollen are not identified to species level, but to a group of morphologically similar species represented by a type), and one the model predicted species. The remaining variables are the model likelihoods of the individual pollen grain belonging to a particular species. The maximum of these likelihoods will identify the predicted species. Each row in the datasets represent one pollen grain. An R-script to reproduce results in the paper is included.