Because mesocarnivores (carnivores at intermediate trophic levels) are generally solitary, elusive, and nocturnal, they are very difficult to research. One technology that can remedy this is the camera trap, but the large quantity of video footage produced by cameras creates the challenge of long data-processing time for researchers. Scientists have combatted this by developing computer algorithms that sort through footage to produce species classifications, and some of these algorithms are designed to recognize individual animals based on unique features like coat patterns. Unfortunately, these algorithms work on few mesocarnivore species. This study expands and analyzes the individual identification capabilities of two image recognition algorithms, the eigenface and speeded up robust features (SURF), on images of the African civet (Civettictis civetta), a nocturnal and understudied mesocarnivore. Camera traps were placed across Gorongosa National Park (GNP) in Mozambique from June 2022 to August 2022 at carcasses and civet latrines to maximize the amount of video footage collected. Next, pattern features of the civets were extracted from the images and manually compared to identify individual civets and build a labeled dataset, which was used to test the two algorithms. While eigenface failed to properly distinguish individuals, SURF obtained a macro-weighted accuracy of 97.85%. Immediate applications of this study include calculating the African civet population size in GNP, a task that has been nearly impossible to complete until now, as well as understanding the species’ territory dynamics.

Catherine Keim, '23: linkedin.com/in/catherine-keim