Wildlife Monitoring on the Edge: A Performance Evaluation of Embedded Neural Networks on Microcontrollers for Animal...

Area of Science:

• Animal behavior monitoring
• Biologging technology
• Edge computing

Background:

• Monitoring wild animal behavior is crucial but challenging due to the cost and difficulty of implanting devices.
• Existing animal monitoring devices require low power consumption and robustness for extended battery life.
• The need for custom smart devices capable of detecting multiple behaviors under these constraints is significant.

Purpose of the Study:

• To design and implement a custom edge-computing solution for monitoring animal behavior.
• To develop a device capable of detecting multiple horse gaits using an embedded artificial neural network (ANN).
• To optimize the balance between energy consumption and computing performance for animal behavior monitoring devices.

Main Methods:

• An edge-computing approach embedding an artificial neural network (ANN) within a microcontroller.
• Utilizing data from an Inertial Measurement Unit (IMU) sensor to capture animal movement.
• Implementing and deploying multiple ANNs on various microcontroller architectures for performance comparison.

Main Results:

• The embedded ANNs achieved up to 97.96% ± 1.42% accuracy in detecting three different horse gaits.
• The system demonstrated high energy efficiency, reaching 450 Mops/s/watt.
• On-device computation significantly reduced data transmission, minimizing power consumption.

Conclusions:

• The proposed edge-computing solution effectively monitors animal behavior with high accuracy and low power consumption.
• Embedding ANNs in microcontrollers is a viable strategy for developing robust and energy-efficient biologging devices.
• This approach offers a promising solution for long-term, non-invasive monitoring of animals in their natural habitats.