Decoding family-level features for modern and fossil leaves from computer-vision heat maps
Open Access
- Author:
- Spagnuolo, Edward Joseph
- Area of Honors:
- Geobiology
- Degree:
- Bachelor of Science
- Document Type:
- Thesis
- Thesis Supervisors:
- Peter Daniel Wilf, Thesis Supervisor
Peter Daniel Wilf, Thesis Honors Advisor
Mark E Patzkowsky, Faculty Reader - Keywords:
- cleared leaves
computer vision
fossil identification
fossil leaves
heat maps
leaf architecture
leaf identification
leaf margin
leaf teeth
leaf venation
paleobotany - Abstract:
- Angiosperm leaves present a classic identification problem due to their morphological complexity. Computer-vision heat maps illustrate diagnostic regions for identification, providing novel insights through visual feedback. I investigate the potential of analyzing leaf heat maps to reveal novel, human-friendly botanical information with applications for extant- and fossil-leaf identification. I developed a manual scoring system for hotspot locations on published computer-vision heat maps of cleared leaves that showed diagnostic regions for family identification. Heat maps of 3114 cleared leaves of 930 genera in 14 angiosperm families were analyzed. The top-5 and top-1 hotspot regions of highest diagnostic value were scored for 21 leaf locations. The resulting data were analyzed using cluster and principal component analyses and visualized using box plots. I manually identified similar features in fossil leaves to informally demonstrate potential fossil applications. The method successfully mapped machine feedback using standard botanical language, and distinctive patterns emerged for each family. Hotspots were concentrated on secondary veins (Salicaceae, Myrtaceae, Anacardiaceae, Rubiaceae, Celastraceae), tooth apices (Betulaceae, Rosaceae), and on the little-studied leaf margins of untoothed leaves (Rubiaceae, Annonaceae, Ericaceae, Apocynaceae, Fabaceae). Results from multivariate analyses were driven by similar leaf features. The results echo many traditional observations, while also showing that most diagnostic leaf features remain undescribed. Heat maps that initially appear to be noise can be translated, and the knowledge obtained can be used offline, highlighting paths forward for botanists and paleobotanists to discover new, family-diagnostic botanical characters.