Created
April 17, 2018 07:26
-
-
Save stefanv/649cd0bd8caf33fe503c62fca7e313b7 to your computer and use it in GitHub Desktop.
Bee Wing ID Project Description
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
[In collaboration with Lauren Ponisio.] | |
We are in the midst of Earth's sixth mass extinction due to | |
anthropogenic impacts such as climate change and habitat | |
destruction. Biodiversity provides essential services to people, such as | |
crop pollination and water filtration, which will also be lost as | |
species go extinct. Conservation biologists and ecologists have the | |
immense task of describing the processes that maintain biodiversity and | |
how we can harness those processes to restore species and the services | |
they provide. | |
One of the most fundamental steps in understanding how to maintain and | |
restore biodiversity is identifying what species are in a landscape, an | |
often expensive and time consuming task, particularly in understudied | |
groups like insects. This task will become even more difficult in the | |
future as the field of taxonomy continues to dwindle. | |
If tools were developed to quickly and cheaply identify species, this | |
would propel new advances in ecology that would hopefully translate to | |
slowing the loss of biodiversity. | |
We propose to build software to identify insects by their wing | |
venation. We will first focus on bees, an important group for their role | |
as crop pollinators but also particularly diverse and difficult to | |
identify to species (for example, a key to the genus Lasioglossum has | |
over 300 couplets which must be navigated to get to a species | |
identification). Bee wings can be distinguished based on relationships | |
between vein junctions, cell perimeters and areas, and veins. | |
While these features can be extracted through heuristic, designed | |
features, we will focus here on automatic discovery of underlying | |
patterns through the application of deep convolutional neural networks. | |
Previous work [http://idbee.ece.wisc.edu/]—using manually constructed | |
features—achieved classification rates of 70–80%, and we hope to see | |
significant improvements over these results. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment