Design and Construction of an Automated Multipurpose Avian Deterrent System

Abstract

This study focused on the development and integration of a bimodal bird scaring system, comprising multiple subsystems: bird detection, power and control, audio and laser, and a robotic arm. Visual data was collected using cameras, while avian vocalisations were sourced from the Federal University of Agriculture, Abeokuta (FUNAAB). An ornithologist analysed the collected data, differentiating birds from other objects and identifying distress and predator calls. The bird detection subsystem was crafted with YOLOv5 software, the audio subsystem with Audacity, and the control subsystem used microcontroller tools and UART protocols. Power modelling was conducted with Proteus, and manipulator modelling was done in SOLIDWORKS. Integration of subsystems involved serial communication protocols for synchronised operation. The efficacy of the Bimodal Bird Scaring system was evaluated by tracking bird invasions pre- and post-implementation, alongside assessing response accuracy. The bird detection model exhibited a precision range of 0.84208 to 0.84978 and a maximum recall of 0.82708, reflecting improved detection of real birds. Moreover, the mean Average Precision (mAP) rose from 0.70455 to 0.82104, confirming the model's effectiveness. Performance metrics showed a high true positive rate (0.89) and a low false positive rate (0.01). The aggregate deterrence system efficiently processed 10 commands per second, highlighting operational competence. Daily repulsion rates of the system at varying frequencies (1 kHz to 14 kHz) indicated increasing effectiveness in deterrence—62.00%, 70.00%, 72.20%, and 75.54% over three experiments. The findings concluded that using multiple strategies in an integrated management plan significantly boosted the ability to deter birds from rice farms.