The illegal wildlife trade is estimated to be a multi-billion dollar industry, with hundreds of species traded worldwide. A significant portion of the illegal wildlife trade uses online marketplaces to advertise and sell live animals and animal products. With transactions taking place on the Internet, manually searching through thousands of posts is extremely difficult and requires methods of automated filtering.
Compared to using computer vision to identify species from imagery, identifying imagery associated with the illegal wildlife trade is more challenging as it requires identifying the context in which the species is depicted. .
In a new article published in Biological Conservation, scientists based at the Helsinki Institute of Interdisciplinary Conservation Sciences at the University of Helsinki fill this gap, using machine learning to identify such image content in the digital space. Developed an automatic algorithm.
“This is the first machine vision model applied to infer the context of images to identify live animal sales. Accompanied by images of animals in captivity, which is different from non-capture images, such as photographs of animals taken by tourists in national parks, using a technique called feature visualization. We have demonstrated that the model can consider both the presence of the animal in the image and the environment surrounding the animal in the image, thus flagging posts that may be illegally selling animals. says Dr. Litwik Kulkarnithe lead author of this study.
As part of their research, scientists trained 24 different neural network models on newly created datasets under various experimental conditions. The best performing models achieved very high accuracy and were able to distinguish well between natural and captive contexts. The model was also tested with data retrieved from the source and worked well. Therefore, it has been shown to work well for identifying other content on the Internet.
“These methods are a game changer in our work seeking to enhance the automated identification of illegal wildlife trade content from digital sources. We are now expanding this research to include taxa beyond mammals. , we are developing a new model that can simultaneously identify image and text content,” said Associate Professor. Enrico Di Mininanother co-author leading the Helsinki Institute of Interdisciplinary Conservation Sciences.
The scientists plan to make their method openly available for use by the wider scientific and practitioner community.
Original: New AI Techniques to Tackling Illegal Wildlife Trade on the Internet
Than: University of Helsinki
