Many citizen science projects exist where conservation or government organizations ask the public to help with monitoring species and ecosystems. Citizen science is a great tool both from a data collection point of view where conservation organizations can effectively collect much more data with citizen scientists than they could with trained staff, but also from an educational point of view where participants of citizen science programs are much more motivated to help with conservation efforts. It creates a sense of ownership with the participants in the areas they are monitoring.
Traditionally those projects make use of specific smartphone apps to report sightings of species. This is a barrier to entry though because of the multitude of apps people already have on their phones, the varying quality of those apps and because people have to know the apps in the first place. Facebook on the other hand is used by a much larger public, and people constantly post pictures and sightings of animals on facebook already anyway.
Together with LAMAVE and Marine Wildlife Watch of the Philippines we are proposing to develop a machine learning algorithm that can interpret facebook posts with a specific hashtag, and extract animal sightings from them. This data is then fed into a database for marine wildlife monitoring currently being developed for the Department of Environment and Natural Resources (DENR), specifically it's Biodiversity Management Bureau.
The machine learning algorithm needs to identify sightings with the following attributes for which it can use both the picture and the post text:
- Picture with animal?
- Type of sighting (stranding, in water sighting, fisheries bycatch)
- Taxa (elasmobranch, cetacean, marine turtle, bony fish, sirena)
See also the Photo ID page
Marine Wildlife Watch of the Philippines already has collected a large dataset of citizen reports on facebook with pictures that they manually classified. This can be used as a training set. There are probably a few NGO and Government Organizations that would want to participate in the project to collect data.
- Create training dataset with data from MWWP from facebook
- Research the different machine learning methods that can be applied to this use case and draw conclusions on the methods most likely to succeed
- Implement the most likely method on the data provided by MWWP and evaluate it's accuracy on the existing training data.