YOLOv8-AF: A Novel Customized YOLOv8-Accurate Fast Deep Learning Model for Enhancing Border Security in Various Global Regions

The proliferation of ground-pit tunnels along international borders poses significant security threats, including illegal immigration, smuggling, human trafficking, and potential terrorist activities. To address these challenges, this paper presents an innovative surveillance system that integrates drones equipped with high-resolution cameras and a deep learning-based object detection model, specifically the YOLOv8-AF (You Only Look Once v8-Accurate Fast) variant. The proposed model has been designed to autonomously identify ground pits from aerial imagery, providing real-time alerts to security authorities. The YOLOv8-AF model, trained on a publicly available comprehensive Ground Pit Image Dataset (GPID), demonstrates superior performance with over $95 \%$ accuracy and a mean average precision (mAP) of 0.895, outperforming other models like Faster-RCNN, Fast-RCNN, and SSD. The proposed model ability to process 116.28 frames per second makes it highly efficient for real-time applications, offering a cost-effective and robust solution for enhancing border security in various global regions.