Extensive species identification

Biodiversity research at the amateur level


As the amount of accessible biodiversity data, such as observation data, is growing, there is a need for better tools to help identify the contents of this data. More and more biodiversity data have multimedia associated with them due to the ubiquitous presence of smartphones. Recent advances in machine-learning technology in the form of deep learning open up the possibility of using these large pools of data to create advanced image recognition tools for automatically identifying species from these media.


"The model assists both citizen scientists and validators with the identification of species."

Within the work area of biodiversity research, an important goal is to support citizen science initiatives concerned with biodiversity research that could benefit from automatic species identification.

Such efforts are expected to be increasingly important given the rapid changes that occur in biodiversity worldwide due to climate change, for example. Another goal is to further the knowledge of the public about the natural world by helping them to access and understand its large diversity.

Recognizing species

Recognizing 16,148 species

One of the first results of this collaboration is the application of deep learning to the identification of species from pictures taken in the field. This type of picture poses unique problems due to the complex appearance of the species and the background in the picture. For the project, several models were developed based on a subset of Observation.org's Dutch and Belgian observations database. Several million observations covering 16,148 species and subspecies were used.

Sites and apps

Sites and apps

The resulting image recognition models are implemented in the workflow of Waarneming.nl, the site of the Dutch branch of Observation.org, assisting both citizen scientists and validators with the identification of species. Additionally, the models have been deployed as the backbone of a series of apps, further assisting citizen scientists with identifications in the field. These apps were explicitly designed to work offline - without a Wi-Fi or data connection - thus extending their usability.


TDWG 2018 presentation (PDF)