Improved YOLOv8n Models for Object Detection in Remote Sensing Images

Object detection in remote sensing images remains a critical challenge due to varying object sizes, complex backgrounds, and the need for real-time processing. This project proposes an enhanced object detection approach using the YOLO (You Only Look Once) framework, specifically leveraging YOLOv8n, YOLOv11n, and TBC-YOLOv8 models for detecting objects in remote sensing imagery. These models are fine-tuned with advanced techniques such as Coordinate Attention and Bidirectional Feature Pyramid Networks (BiFPN) to enhance the detection of small objects and improve localization accuracy.

The dataset used in this project, referred to as RSOD (Remote Sensing Object Detection), includes diverse classes like aircraft, oiltank, overpass, and playground, with images representing varying environmental conditions. The main goal is to address issues related to multi-scale object detection and complex background handling, which are common in remote sensing data.

Through a series of experiments, the models’ performance is evaluated based on precision, recall, mean Average Precision (mAP), and F1 score. Results demonstrate that the modified YOLO models outperform the baseline in detecting small objects and handling cluttered environments. The approach presented in this project significantly improves the accuracy and efficiency of remote sensing object detection, making it suitable for a wide range of applications in environmental monitoring, urban planning, and disaster management.

Keywords: Object detection, remote sensing, YOLOv8n, YOLOv11n, TBC-YOLOv8, small object detection, Bidirectional Feature Pyramid Networks, Coordinate Attention, model evaluation, satellite imagery.