LGMOD-UAV: A Lightweight Deployment System for Ground Micro-Object Detection by UAV

Unmanned Aerial Vehicle (UAV) are crucial for rapid identification and tracking of unknown targets, reducing risks to personnel and ensuring safe patrols. However, lightweight, real-time models, and challenges like poor environmental conditions and small target sizes hinder effectiveness. To address this, we propose a lightweight deployment system for ground micro-object detection by UAV(LGMOD-UAV). Using MobileNetV3 as the backbone, replacing YOLOv8n, we reduce computational load and parameters. The BiFPN module and Inner-IoU loss function improve multi-scale feature capture and performance. On the Atlas 200I DK A2 platform, operator optimization further boosts performance, achieving 93.7% mAP, 2.1 GFLOPs, 2.8 MB model size, and 38.2 FPS.