| Abstract Detail
Conference Wide Hodel, Richard [1], Soltis, Pamela [2], Ickert-Bond, Steffi [3], Fochesatto , Stefano [1], weaver, William [1]. Using deep learning with digitized herbarium specimen image data. The rapid increase in digitized herbarium specimens available in natural history collections has enabled us to study many aspects of plant biology, such as morphological comparisons, shifts in phenology, and changes in species distributions. Currently, tens of millions of herbarium specimen images are available via online databases, with more image data added every day. In recent years, machine learning methods have shown promise in efficiently extracting data from herbarium specimens. Deep learning analyses, such as convolutional neural networks, leverage computer vision to automate analyses of herbarium sheet images. Researchers can use deep learning methods to classify digitized herbarium specimens by species or other user-specified categories. Even relatively simple neural networks can contain hundreds of billions of parameters and require the use of graphical processing units (GPUs), which can process multiple computations simultaneously. Accessing and interfacing with GPUs can represent a barrier to entry for some users. This workshop will help participants to vault past the initial technical challenges of using deep learning on herbarium sheets to conduct analyses for object detection, image segmentation, and taxonomic classification. Participants will work with curated sets of images, or they can use their own datasets--although this will require some advance preparation by the participant to organize their images. The workshop will be conducted in Jupyter notebooks, and some knowledge of Python will be helpful, but not required. A laptop computer and a Google account will be required.
1 - National Museum of Natural History, Botany, MRC 166, Smithsonian Institution, Washington, DC, USA
2 - University Of Florida, Florida Museum Of Natural History, Gainesville, FL, 32611.0, United States
3 - University of Alaska Fairbanks
Keywords:
none specified
Presentation Type: Workshop
Number: W04001
Abstract ID:8
Candidate for Awards:None |
