Abstract
As awareness of the negative impact of humans on wildlife grows, understanding how animals respond to short-term disturbances (e.g., human-induced disruptions) or long-term changes (e.g., habitat reduction, temperature increase) becomes crucial. Over the past few decades, ecologists have greatly benefited from advances in miniaturized electronics: biologging, i.e., the development and deployment of animal-borne tracking devices, has become a key tool essential to numerous studies. Various types of animal-borne tracking devices have been developed, enabling the tracking of an animal's location over time using GPS, monitoring its activity with accelerometers, or recording environmental information through sound or video collection. Biologging is a highly active research field, with new solutions regularly emerging, each bringing new improvements in terms of size, autonomy, or cost. This thesis is part of the REPOS project and aims to study the impact of human activity on the sleep cycles of animals and the resulting behavioral disturbances. To achieve this, it requires the use of animal-borne tracking devices directly carried by the animals.However, animal-borne tracking devices are limited in terms of acceptable size and maximum weight, with the battery typically representing a significant, if not major, portion of these limitations.Reducing the battery size obviously reduces the data acquisition duration, potentially rendering data collection pointless considering the logistical and financial efforts required to deploy tracking devices and the fact that capturing an animal biases its behavior for some time, further reducing the truly usefull temporal window of data collection.To address this problem, we first focus on the development of new energy-efficient animal-borne tracking devices. We will use the Audiologger as a case study to establish a methodology for designing energy-efficient biological tracking devices. Since observation campaigns involve multiple tracking devices, the amount of data to be analyzed is substantial and requires some form of data processing automation to assist biologists in inferring animal behavior. The literature reveals a growing interest in artificial intelligence approaches in studies related to animal ecology and conservation issues, while demonstrating several successful case studies at the same time. Using these works as a reference, we will explore various audio classification approaches. Furthermore, we will demonstrate that there is a highly complex relationship between data acquisition and data processing phases. We will show that the data processing step can be incorporated into the data acquisition phase to improve the autonomy of the tracking device.