To the naked eye, it’s a pitch-black photograph with a few shrubs and the petals of a white flower glinting in the moonlight. But a closer look by artificial intelligence reveals the forehead of a giant forest elephant.
Artificial intelligence has emerged as a new weapon in the battle for wildlife conservation in Lopé National Park in Gabon. Through machine learning, an algorithm is now able to identify over 25 species for researchers in the Congo Basin, which otherwise would have been missed by humans.
Researchers from Stirling University, who have a long-running partnership with the Agence National des Parcs Nationaux du Gabon (ANPN), supported by funding from the European Union’s ECOFAC 6 programme, set up camera traps in the rainforest to monitor wildlife.
They teamed up with artificial intelligence company Appsilon to use image recognition technology, known as Mbaza AI, to analyse large numbers of photographs taken in the rainforest. The 200 cameras, spread out across the Lopé National Park, are triggered by motion sensors and each take hundreds of pictures per day.
The head of the biodiversity monitoring programme, Stirling’s Dr Robin Whytock, invited Appsilon to apply its technology to automate and improve the analysis of photos, which otherwise must be performed manually by rangers and researchers. A database of over 1.5m photos trained the algorithm to classify as many as 3000 images per hour with an accuracy rate of 96 per cent.
Dr Whytock said: “Central African forest mammals such as elephants, gorillas and pangolins are threatened by unsustainable trade, land-use change and the global climate crisis. Appsilon’s work on the Mbaza AI app, will allow conservationists working on the ground in Central Africa to rapidly analyse wildlife data collected using automated camera traps, reducing the lag between data collection and analysis from months or years to just days.
“This will enable conservationists to rapidly identify and respond to threats to biodiversity and improve conservation management at a time of global biodiversity crisis.”
Mbaza AI, which does not require any specialist hardware or even an internet connection, allows rangers to estimate populations of animals that are difficult to monitor, like forest elephants. It also allows them to track animals and their movements, as well as gain a better understanding of their behaviour through facial recognition, which is especially important for primates with complex social hierarchies.
Dr Lee White, Gabon’s minister for the environment and a British-born conservationist, added that the software has potential for anti-poaching purposes too. Over 200 animal and plant species are considered threatened in Gabon, including the African Forest Elephant, which across Africa has seen 70 per cent of its population wiped out in the last 15 years, leaving Gabon as a crucial sanctuary for the remaining population.
The Appsilon algorithm has also been used in Gabon to track the spread of a skin disease known as Yaws, which is spread by bacteria to great apes as well as humans. There is currently no vaccine for this disease, which mainly affects children.
Dr White said: “It is our ambition to use ‘Mbaza’, the name for a traditional guard house, to roll wildlife monitoring out across all of Gabon’s protected areas and forestry concessions.”