Create your own conference schedule! Click here for full instructions

The Virtual Conference is located at

Abstract Detail


Spagnuolo, Edward [1], Wilf, Peter [1], Serre, Thomas [2].

Decoding family-level features for modern and fossil leaves from computer-vision heat maps.

Computers can illustrate diagnostic regions in heat maps, providing novel insights for humans through direct visual feedback. Angiosperm leaves pose a classic identification problem due to the complexity of leaf architecture. We show that analyzing leaf heat maps is a novel way to generate human-friendly botanical information, with value for identifying extant and fossil leaves. We emphasize the family level because it is the traditional starting point for plant identification, especially for fossil angiosperm leaves, which nearly all represent extinct species and genera from extant families. We developed a scoring system based on traditional leaf architecture and scored the hotspots' locations on previously published computer-vision heat maps of cleared leaves (Wilf, Serre et al. 2016, PNAS). Heat maps, illustrating diagnostic regions for family identification based on a sparse coding algorithm, of 3115 cleared leaves of 930 genera in 14 angiosperm families were analyzed. The top-5 and top-1 16x16 pixel hotspot patches of highest diagnostic value were scored for their occurrences at 21 locations of the base, midsection, apex, margin, venation, and teeth. The resulting data were visualized through box plots and analyzed using cluster and principal component analyses. We then visually identified similar features in fossil leaves to informally demonstrate the potential for applications to the fossil record. Hotspots were concentrated on secondary veins for many families (Salicaceae, Apocynaceae, Myrtaceae, Anacardiaceae), at tooth apices in families with most or all toothed leaves (Betulaceae, Rosaceae), and on the little-studied leaf margins of all untoothed families (Rubiaceae, Annonaceae, Ericaceae). Multivariate analyses supported high family distinctiveness, driven by scores for margins, tooth apices, and secondary veins. The results echo some traditional observations, including signals from Myrtaceae intramarginal veins and tooth architecture in Rosaceae, Malvaceae, and Betulaceae. They also show that most diagnostic features of leaves have probably never been noticed. Potential analogs of the hotspot regions in fossils were detected on Fagaceae primary veins, Salicaceae and Anacardiaceae secondary veins, and Myrtaceae higher-order veins. The interpretation of computer vision heat maps highlights paths forward for botanists to discover new, family-diagnostic botanical characters with value for the fossil record, including families with minimally defined leaf architecture. The heat maps also show that most leaf architecture features have yet to be discovered and described. Filtering and observing the strongest signals from heat maps, as done here, also enhances intuitive, “gestalt” visual training, even without the development of formal characters.

Log in to add this item to your schedule

1 - Pennsylvania State University, Dept. of Geosciences, University Park, PA, 16802, USA
2 - Department of Cognitive, Linguistic and Psychological Sciences, Carney Institute for Brain Science, Brown University, Providence, RI, 02912, USA

Computer Vision
Fossil Leaves
Heat Maps
Leaf Architecture
Leaf Margins
Leaf Identification
Leaf Teeth
Cleared Leaves.

Presentation Type: Oral Paper
Session: PL2, Paleobotany: Cookson Student Presentations - Session II
Location: /
Date: Monday, July 19th, 2021
Time: 1:30 PM(EDT)
Number: PL2005
Abstract ID:77
Candidate for Awards:Isabel Cookson Award,Maynard F. Moseley Award

Copyright © 2000-2021, Botanical Society of America. All rights reserved