Abstract by Abigail Jenkins
An application of machine learning and computer vision models in the field of taxonomy
One of the major tasks of herbarium curation staff is the appropriate classification of new specimens that are collected in the field. Traditionally, specimens are classified by hand, a time-intensive process, often involving experts worldwide. Samples of interest are then preserved, mounted on herbarium sheets, and stored at various locations (such as the Smithsonian) throughout the world. Recently, the Smithsonian has undergone a digitization effort and are in the process of databasing and imaging their extensive herbarium collection. Due to the time intensive process of classification, we are developing machine learning and computer vision models to classify over 500,000 images of ferns into their proper species. An ideal algorithm will be able to locate a fern mounted on an herbarium sheet, segment the background of the image from the fern to isolate the subject, and then classify the fern into its proper species. Progress toward the development of such an algorithm will be presented.