The Development of Monitoring System for Elephant Intrusion Detection in Agricultural Areas to Reduce Human-elephant Conflict with Convolutional Neural network technology

Abstract

Human-elephant conflict occurs due to the migration of elephants from their habitat to human agricultural areas in search of food and water resource. This research has therefore introduced Convolutional Neural Network Technology to reduce Human-elephant conflicts, with a framework called YOLOv5 is utilized for real-time object detection from video footage using embedded neural brain devices to monitor and prevent wild elephant intrusions into agricultural areas. First, researchers collected elephant image datasets from monitoring areas, leveraging the advancements in deep learning frameworks to develop Yolo-based architecture models suitable for embedded devices, ensuring both speed and accuracy. In this work, the researchers adjusted the hyperparameters for the YOLO model variants: YOLOv5N, YOLOv5S and YOLOv5M. The computations were conducted with reduced complexity and the proposed models are well-suited for embedded devices. After testing, it was observed that the YOLOv5S model achieved an average accuracy of 95.68% mAP@.5 with a speed increase of up to 50% compared to the YOLOv5M model, which had a maximum accuracy of 95.46% mAP@.5. Then, before deploying the models on embedded devices, researchers augmented the elephant image dataset from the internet for the YOLOv5S model. This augmentation improved the accuracy to a new value of 98.02% mAP@.5. After deployment, it was found that deep learning could accurately detect instances of wild elephant intrusions into agricultural areas, even though it may sometimes be slightly slower than human observation.