Designing an Automated Barking Drone to Detect and Repulse Cattle

The growth and expansion of the Australian livestock industry have given rise to several challenges associated with the age-old mustering method. Specifically, the direct and indirect costs of mustering livestock have escalated due to increased fatalities associated with quadbikes use, soaring prices of quality sheepdogs, and concerns related to livestock welfare. In response, this research proposes an innovative approach using quadcopters equipped with image recognition technology, specifically the YOLOv4-tiny Object Detection Model (ODM), to assist livestock farmers in conducting these activities more efficiently. The primary objective of this study is to investigate the overall efficiency of such a system, with a focus on the real-time application of an Object Detection Algorithm, Animal-Machine Interaction (AMI), and a machine-vision-based drone positioning architecture for cattle mustering in a real-world environment. There are three main stages to this study: the construction of a quadcopter, the training and evaluation of an object detection model, and the testing of employing such a system to assist farmers in mustering activities. The larger goal of this study is to contribute to modernising livestock management practices and address the challenges posed by traditional mustering methods. Our results showed that the onboard ODM achieved a real-time average precision of 85.35% in detecting cattle on the farm. Additionally, cattle responded predictably to the drone, moving away as it approached. When strategically positioned, the drone effectively guided the cattle toward the desired location. On the other hand, short-term habituation to the presence of barking drones was observed, where the cattle’s escape distance decreased from an initial 85m to 12m after three consecutive trials within the same day. It is important to note that this habituation effect was not long-lasting, as similar behaviour was observed in a preliminary test conducted three months before the experiments. In conclusion, this study demonstrates the feasibility of using a quadcopter to assist with herding activities. Through real-world trials, it has established a foundation for further development and refinement of such systems, paving the way for advancements in livestock management.