Umeå universitets logga

Zoonotic infectious diseases and cross-species pathogen spillover are driven by biodiversity and spatio-temporal ecological interactions. Generating high-resolution data to approximate the presence and absence of animals and their mobility, density, and interactions can enhance research efforts and improve public and veterinary health prediction capacities. Bioacoustics technology provides solutions for tracking vocalizing animals in space and time, contributing to infectious disease ecology research. We report on our adaptation and testing of a bioacoustics Internet-of-Things system for passive spatio-temporal monitoring of avian species, situating its application within the context of zoonoses. We tested alternatives for the physical sensor device, microphones, and battery and chose Raspberry Pi Zero 2 W as our base computer. We further embedded and configured machine learning (ML) classification algorithms building on the BirdNET-Pi for avian classification. We developed a software, that supports TensorFlow-based classification algorithms and facilitates easy and open uploads of alternative ML algorithms for non-avian taxa (faunanet) for future applications of the prototype for multi-species bioacoustics monitoring. We evaluated the performance of the physical system and validated its classification accuracy within the premises of a zoo. Across four field tests, we collected 700 h of audio data on avian vocalizations and identified 57 distinct species. The performance of the prototype depends on the parametrization of the classification algorithms and the positioning of the physical sensor. Overall, the developed prototype demonstrated that the system can be further piloted for studies on zoonotic infectious diseases.