YOLO-Based Harmful Wildlife Recognition and Repellent System Using Large AI Learning Dataset

AI-based harmful wildlife repellent devices and systems introduced in several studies are unable to smoothly detect and repel many and various types of harmful wildlifes due to a lack of learning data. To solve this problem, this study proposes an AI object recognition-based harmful wildlife repellent system that learns a YOLO-Based AI model using a 250,000-page wildlife activity dataset and applies it to a harmful wildlife repellent system. By training a new YOLO AI model using large AI training data, mAP (IoU=0.50～0.95) performance is improved by about 4% and prediction speed by about 18% compared to the existing YOLO model, and by applying this to a commercial MCU board, it is possible to construct a low-cost, high-performance harmful wildlife repellent system.