Semi-automated detection of tagged animals from camera trap images using artificial intelligence

The use of technology in ecology and conservation offers unprecedented opportunities to survey and monitor wildlife remotely, for example by using camera traps. However, such solutions typically cause challenges stemming from the big datasets gathered, such as millions of camera trap images. Artificial intelligence is a proven, powerful tool to automate camera trap image analyses, but this is so far largely been restricted to species identification from images. Here, we develop and test an artificial intelligence algorithm that allows discrimination of individual animals carrying a tag (in this case a patagial yellow tag on vultures) from a large array of camera trap images. Such a tool could assist scientists and practitioners using similar patagial tags on vultures, condors and other large birds worldwide. We show that the overall performance of such an algorithm is relatively good, with 88.9% of all testing images (i.e. those not used for training or validation) correctly classified using a cut-off discrimination of 0.4. Specifically, performance was high for correctly classifying images with a tag (95.2% of all positive images correctly classified), but less so for images without a tag (87.0% of all negative images). The correct classification of images with a tag was, however, significantly higher when the tag code was at least partly readable compared with the other cases. Overall, this study underscores the potential of artificial intelligence for assisting scientists and practitioners in analysing big datasets from camera traps.