UBD Herbarium - Dataset of Intact Leaves for Reconstruction

This dataset consists of a total of 2933 augmented individual intact herbarium leaves with a uniform background collected from Universiti Brunei Darussalam Herbarium (UBDH). These leaves were automatically extracted using a deep learning pipeline consisting of semantic segmentation model, connected component analysis and a single-leaf classifier trained on binary images to automatically detect intact leaves. All leaves were collected from ten different plant families namely Anacardiaceae, Annonaceae, Dipterocarpaceae, Ebenaceae, Euphorbiaceae, Malvaceae, Phyllanthaceae, Polygalaceae, Rubiaceae and Sapotaceae.

Family Name Training Validation Testing
Anacardiaceae 209 29 61
Annonaceae 199 28 58
Dipterocarpaceae 202 28 59
Ebenaceae 201 30 60
Euphorbiaceae 207 29 61
Malvaceae 207 29 61
Phyllanthaceae 210 30 60
Polygalaceae 200 28 58
Rubiaceae 205 29 60
Sapotaceae 200 28 58
Total 2040 288 596

Reconstruction of damaged herbarium leaves using deep learning techniques for improving classification accuracy, Ecological Informatics
Available online 2 February 2021, 101243, https://doi.org/10.1016/j.ecoinf.2021.101243