YOLO-Based Vehicle Detection in Drone Aerial Imagery Using Transfer Learning

Abstract

Vehicle detection in drone aerial imagery has become an important research area due to the growing use of Unmanned Aerial Vehicles (UAVs) in intelligent transportation systems, surveillance, disaster monitoring, border security, and urban traffic management. Traditional vehicle detection methods based on handcrafted features and conventional machine learning techniques often struggle to achieve high accuracy under complex aerial imaging conditions such as varying illumination, occlusion, small object size, dense traffic regions, and complex backgrounds. This study presents a YOLO-based deep learning framework integrated with transfer learning for efficient vehicle detection in aerial drone images. The proposed approach utilizes pretrained COCO dataset weights along with aerial datasets including VEDAI, DOTA, COWC, and VAID for model training and evaluation. Data preprocessing, annotation conversion into YOLO format, and optimized network parameter configuration are employed to improve detection performance and computational efficiency. The customized YOLO model effectively detects vehicles in real-time aerial scenes while operating with moderate GPU resources. Experimental results demonstrate improved detection accuracy, reduced training time, and robust performance under challenging environmental conditions. Evaluation metrics such as Precision, Recall, Intersection over Union (IoU), Average Precision (AP), and Mean Average Precision (mAP) validate the reliability and effectiveness of the proposed framework for UAV-based intelligent transportation and surveillance applications.