Partnering with the Curtin Innovation Centre, a data science student intern from Curtin University joined ConsMin for a 12-week period in early 2020. The challenge was to utilise artificial intelligence (machine learning) to identify the presence of the Northern Quoll within camera trap pictures.
Dubbed Quoll!Quoll (Quoll Not Quoll), a program was developed to successfully review large datasets of photos to identify specific species of animals and categorise them. This program effectively saves hours of time for environmental professionals by reducing, if not eliminating, their need to manually scroll through hundreds of images looking for animals of interest.
The final model was trained on approximately 3000 photos and can accurately determine whether a photo contains a picture of a quoll, cat, dingo, bird or rock rat and sort them based on these classifications. It can also assign the miscellaneous class of 'animal' so that photos containing only a small or indistinguishable part of an animal are not missed.
Quoll!Quoll can revolutionise the way in which environmental personnel review data. Automating even a portion of the process makes a huge impact in the time required on this task as well as having greater confidence in identification. This program can also be further refined (for individual species), expanded (for additional species) or extended (to identify individuals not just species). Such opportunities can continually impact the way in which environmental data is collected and analysed.
Effectively processing data can ensure conservation measures are targeted appropriately. With management measures constantly evolving, responses to threatened or endangered species needing to be swift, and ongoing research constantly occurring, an innovation such as this will simplify decision making for industry.
The Quoll!Quoll project took home the AMEC Environmental Award in December 2020. It was also presented with a Merit Award in the Social Impact Category at the Lateral INCITE Awards in 2020.