Fieldguide is powered by machine-learning software designed for image recognition of natural history specimens and live images of species in the field. It is the only image-recognition software that is tailored to both the scientific community and citizen scientists. We estimate that over 300,000 species can be identified to species using image recognition neural nets along with locality and phenology filters. Over one million natural history images are uploaded to the internet each week, and of these 250,000 have associated scientific names. Thus, the amount of natural history images available for training neural networks is increasing exponentially, as are images that need identifications. Fieldguide is working with Caltech, Cornell Tech and Google to continually improve the accuracy and performance of its machine learning services through improved neural network processes. Fieldguide will help museums screen unidentified material and aid taxonomists by categorizing above the species level, identify reliable taxa to the species level, and provide a check on human identifications. We will also use Fieldguide results to ask questions about why species look alike, even when they can be on different continents and totally unrelated.
Fieldguide offers two products, the first is an app for iOS and Android devices that provides image recognition in the field and museum specimen images. The apps also include key features for creating a natural history community by taxa, region, or theme. Museum specimen images are uploaded to the national database (iDigBio) and Fieldguide servers. People can also use the Leps by Fieldguide to upload images of leps in the field. They are added to the national database as observations and not museum specimens. The signature project is Leps by Fieldguide , which targets moths and butterflies and secondarily all other arthropods. They will also be developing apps for other groups, such as fungi and plants.
We are also using the “Fieldguide Batch” tool, which provides image recognition for all specimen images for a Symbiota portal like SCAN/LepNet. It can be used for both identification of rough-sorted specimens and confirmation of human identification of species.