>Business >Leveraging machine learning to quicken up ecological research

Leveraging machine learning to quicken up ecological research

The Serengeti is one of the last remaining bastions of the planet that is host to an intact community of big mammals. These animals roam across massive swathes of land, some migrating thousands of miles across several countries upon seasonal rainfall. As human encroachment surrounding the area becomes more intensive, these species are forced to modify their behaviours in order to survive. Expanding agriculture, poaching, and climate abnormalities add up to alterations in animal behaviour and population dynamics, but these alterations have happened at spatial and temporal scales which are tough to survey leveraging conventional research strategies. There is a great rush to comprehend how these animal communities operate as human pressures grow, both in order to comprehend the dynamics of these final pristine ecosystems, and to formulate efficient administration plans to conserve and safeguard and protect the integrity of this one-of-a-kind biodiversity hotbed. 

With this in mind, DeepMind has collaborated with ecologists and conservationists to produce machine learning strategies to help research the behavioural dynamics of an entire African animal community in the Serengeti National Park and Grumeti Reserve in Tanzania. The Serengeti-Mara ecosystem is internationally unparalleled with regards to its biodiversity, hosting a projected seventy large mammal species and 500 bird species, owing partially to its unique geology and varied habitat types. Almost 10 years ago, the Serengeti Lion Research program setup hundreds of motion-sensitive, cameras in the core of the safeguarded region. The cameras have their triggers in passing wildlife, taking animal pictures consistently, across massive spatial scales, enabling scientists to research animal behaviour, distribution, and demography with amazing spatial and temporal resolution. 

Over the previous nine years, the team has gathered and recorded millions of photos. Up till now, volunteers from all over the world have assisted to identify and count the species in the photos by hand leveraging the Zooniverse web-based platform, which hosts many similar projects targeted at citizen-scientists. This has had the outcome of a rich dataset, Snapshot Serengeti, featuring labels and counts for approximately 50 differing species. Presently, the annotation procedure is labour intensive and time intensive. It takes up to twelve months from the time a camera us triggered up till labels are gathered from volunteers. This restriction has not just hindered scientist’s capability to execute basic research, but has made it difficult for conservationists to react adaptively to hurdles and perturbations which are disruptions to the ecosystem. To assist researchers in unlocking this information with improved efficiency, we’ve leveraged the Snapshot Serengeti dataset to train machine learning models to automatically go about detecting, identifying and counting animals. 

Leveraging machine learning in the conservation efforts is not really a new thing. For instance, scientists have prior leveraged tourist imagery and YouTube videos to trace animals, and audio recordings to detect species on the basis of their calls. Camera trap information can be tough to work with –  animals may seem to be out of focus, and can be at several differing distances and positions with regard to the camera. With specialist input from leading ecologist and conservationist Dr. Meredith Palmer, the project quickly began to take shape, and currently they have a model that can execute on par with, or better than human annotators for a majority of the species in the region. Critically, this strategy shortens the information processing pipeline by approximately 9 months, which has massive potential to assist scientists in the domain. 

Obviously, field research is a challenge, are full of unexpected hazards like failing power lines and restricted or no internet access. The software is being prepped for deployment on the field and they are searching for ways to safely execute their pre-trained model with modest hardware necessities and minimal internet access. They’ve fostered a close relationship with their collaborators in the domain to ensure that their technology is leveraged with responsibility. Once in place, scientists in the Serengeti will be capable to make direct utilization of this utility, assisting in providing them with up-to-date species data to better assist their conservation efforts. 

Add Comment