A Database for Automatic Identification of Herbarium Specimens in Piperaceae Family

This is the first version of a herbarium specimen dataset for the Piperaceae family. It contains 10,503 samples distributed across 236 species of Piperaceae collected in Brazil. These samples were selected from the speciesLink repository and are part of the collection of 38 herbaria. They were identified by a team of 11 specialists in Piperaceae.

Among the 10,503 images of Piperaceae, 7,678 (73.10%) belong to the genus Piper L., 2,806 (26.72%) to the genus Peperomia Ruiz & Pav., and 19 (0.18%) to the genus Manekia Trelease. The dataset includes 156 species of Piper L., 79 species of Peperomia Ruiz & Pav., and a single species of Manekia Trel.

A comprehensive set of experiments evaluated segmentation, feature extraction, and classification algorithms for the dataset performance as reference values. The source code for the experiments is available on GitHub.

The metadata includes information about the herbarium that physically stores the specimen and the Brazilian region where the plant was collected. The file original.zip contains the images in their original version without any processing. The files grayscale.zip and rgb.zip contain the images in three dimensions (256 x 256, 400 x 400, and 512 x 512 pixels), organized into three folders for each size::

jpeg: The resized exsiccatae images.

mask: The masks predicted by the U-Net model.

segmented: The images with the background removed.