An Acoustic-based system for early detection of elephants in Nyangores Forest

Abstract

Elephants greatly contribute to their ecosystem through seed dispersal, creation of water sources when it is dry through digging and even carbon sequestration. However, their population has been reducing due human-elephant conflict, poaching and habitat degradation. In Nyangores Forest, which is a part of Kenya’s greater Mau Region, frequent encounters between forest-adjacent communities and elephants often lead to injuries or fatalities on either side, exacerbating the human-elephant conflict. Elephants communicate through vocalization and one of the most common and frequently used are the low rumbles which are very low frequency sounds used by the elephants and can travel over long distances and are below human hearing level. The study aimed to develop an early detection system based on the elephant vocalizations to reduce conflict and enhance conservation. There have been many methods that have been used in the detection of elephants such as imaging with the use of camera traps which does not work well with harsh weather conditions which can affect the angles and the clarity and seismic monitoring using vibrations from their vocalizations which are felt on the ground which has limited range and high sensitivity to environmental noise. This was achieved using Rapid Application Development (RAD), integrating a machine learning model with an IoT device for sound classification and localization. The CNN model was trained and tested and attained an accuracy of 99%. The IoT device was developed using Raspberry Pi Pico which resulted in a working integrated system which can detect elephant vocalizations and send an alert demonstrating real-time functionality. The results confirmed the viability of acoustic localization for elephant detection. The system developed offers a low-cost, efficient and scalable alternative for elephant detection and monitoring. Its implementation has the potential to strengthen existing conservation efforts and infrastructure used by different agencies and bodies, improve ranger response times and integrate with deterrent mechanisms such as bee fences.