Evaluating DETR, RetinaNet, and RTMDet Models for Object Detection in Rural Drone Imagery

This study compares real-time object detection models-RetinaNet, DETR, and RTMDet-focusing on their effectiveness in analyzing aerial images of rural landscapes. The research examines each model's performance in detecting objects within rural environments. Rural areas present unique challenges for object detection due to scale variations, environmental obstacles, and potential obstructions that obscure objects of interest, all of which require robust, adaptive detection capabilities. In particular, the study addresses real-time processing demands critical for applications like agricultural monitoring, wildlife tracking, and rural infrastructure assessment. Transfer learning with pre-trained models is employed to tailor these models to the complex visual characteristics of rural aerial imagery, allowing for efficient adaptation and enhanced accuracy. This approach leverages prior knowledge to overcome data scarcity. Through testing and evaluation, this study aims to provide insights into the strengths and limitations of RetinaNet, DETR, and RTMDet in supporting practical applications where rapid, reliable object detection is essential. The findings contribute perspectives on optimizing object detection systems for aerial surveillance in rural settings, advancing research and practical deployments.