Harmonizing Habitat: P-Yolov5 Enhanced Computer Vision For Mitigating Human-Wildlife Conflicts In Rural Areas

Abstract

In rural areas, the incursion of forest animals onto roads and into villages poses a substantial safety risk to residents. To address this issue, our research introduces an innovative variant of the You Only Look Once (YOLO) model, designated as P-YOLOv5. This model utilizes a MobileNetV3 backbone and a Feature Pyramid Network (PAnet) neck to enhance real-time object detection and image recognition. P-YOLOv5 excels in balancing speed and accuracy, making it particularly suitable for applications like autonomous driving, surveillance, and robotics. The MobileNetV3 backbone provides an efficient framework for feature extraction, while the PAnet neck enhances the model's capacity to capture contextual and spatial information across various scales. Experimental results showcase exceptional object detection performance, revealing high precision and recall rates on a meticulously pre-processed dataset. The Precision-Recall Curve further emphasizes the model's ability to strike a balance between accuracy and false positive rates, highlighting its practical applicability. Noteworthy is P-YOLOv5's achievement of a remarkable 0.93 mAP@0.5 for all classes, underscoring its robust capabilities for real-world object detection tasks in situations where forest animals pose a threat to human safety.