Senior Design Projects
ECS193 A/B Winter & Spring 2021
Using Computer vision to identify the phenological condition of herbarium specimens
| ********** | |
| Reed J Kenny | |
| Potter Lab UC Davis |
Project's details
| Using Computer vision to identify the phenological condition of herbarium specimens | |
| Natural history collections form the basis for nearly all of our historic taxonomic and biogeographic knowledge. The digital imaging of specimens has proven to be revolutionary, with platforms like Symbiota and iDigBio integrating natural history databases from around the world. At this moment there are over 31 million individual images of natural history collections available through the iDigBio portal. While these platforms have made collections more accessible, the images are still assessed by human eye, limiting the volume of data that can be extracted in a reasonable timeframe. Machine learning technology has enormous potential to revolutionize the utility of digitized natural history collections. Phenological changes, such as shifts in flowering time have already been documented using herbarium specimens, but the data is time consuming to generate and studies are limited in their scope. Flowers are blooming earlier and many pollinator species are also shifting their flight times. It is not known whether the phenology of mutualist partners will change at the same rate. A mismatch in phenology across trophic levels could result in ecosystem level disruption. |
|
| Students working on this project will create an open source training algorithm that is flexible enough to be applied across many plant taxa and requires minimal technical skills to implement. The initial aim will be to train one of the available open source computer vision algorithms on an existing data set of Streptanthus herbarium specimens which have been scored for phenology. Next steps would be to expand the methodology to other species, analyze the sensitivity of the algorithm to different numbers of training images, and develop either a user friendly web interface, or a methodology to embed the algorithm in to existing collections databases such as the California Consortium of Herbaria. | |
| - A trained algorithm to identify the phenological state of herbarium specimens - A web interface that is accessible to non-specialists - A method of incorporating the algorithm into existing collections databases |
|
| Familiarity with computer vision and web design | |
| ********** | |
| 30-60 min weekly or more | |
| Open source project | |
| Attachment | N/A |
| No | |
| Team members | N/A |
| N/A | |
| N/A |