Create your own conference schedule! Click here for full instructions

Abstract Detail



Biodiversity Informatics & Herbarium Digitization

Weaver, William [1], Ng, Julienne [2], Laport, Robert [3].

LeafMachine: Using Machine Learning to Automate Phenotypic Trait Extraction from Herbarium Vouchers.

Herbaria across the United States are digitizing millions of specimens, drastically increasing the accessibility of meticulously preserved vouchers spanning hundreds of years. These specimens can contain a wealth of information about the species' ecology, such as flower and fruit phenology, leaf size and shape, signatures of herbivory, and changes in response to climate change. However, extracting such phenotypic data from digitized images can be time-consuming and require significant manual user input. Here, we present LeafMachine, a machine learning application for the autonomous analysis of digitized herbarium images. We take advantage of recent advancements in computer vision and machine learning to autonomously identify and measure leaves from digitized herbarium specimens. This involves using convolutional neural networks for image segmentation, as well as machine learning and contextual algorithms to locate and interpret distance scales in images to convert pixel-distance into metric distance. Machine learning and contextual algorithms are used to locate and interpret distance scales in images to convert pixel-distance into metric distance. Training data for our algorithms consists of 2,684 randomly sampled specimen vouchers from 147 herbaria across the United States to enhance generalizability. Processing several herbaria collections showed that LeafMachine can extract leaf morphometric data from images that vary widely in quality, resolution, and layout. Our application was developed using Matlab (v. 2019a) and is also available as a standalone Windows application called LeafMachine, available at https://github.com/Gene-Weaver/LeafMachine. Our novel application of machine learning has the potential to vastly increase available trait information and inform ecologically-relevant hypotheses related to community dynamics, adaptation, and global climate change. 


Log in to add this item to your schedule

Related Links:
Download LeafMachine
LeafMachine Website


1 - University of Colorado Boulder, 1900 Pleasant St, Boulder, CO, 80302, USA
2 - University Of Colorado Boulder, Department Of Ecology & Evolutionary Biology, Campus Box 334, Boulder, CO, 80309, United States
3 - Rhodes College, Department Of Biology, 2000 North Parkway, Memphis, TN, 38112, United States

Keywords:
machine learning
convolutional neural networks
support vector machines
herbarium
herbarium specimen images
digitized herbarium data
morphology.

Presentation Type: Oral Paper
Session: BIHD1, Biodiversity Informatics & Herbarium Digitization
Location: Tucson I/Starr Pass
Date: Monday, July 29th, 2019
Time: 2:30 PM
Number: BIHD1005
Abstract ID:259
Candidate for Awards:None


Copyright © 2000-2019, Botanical Society of America. All rights reserved