NeoDrone（Near-earth Observe via Drone，无人机近地观测数据集）

NeoDrone数据集面向低空无人机视觉感知技术研发，适用于智慧城市管理、应急救援响应、电力能源巡检、交通流量监控、精准农业监测、环境保护执法等领域。为高校科研团队、人工智能企业及低空经济相关机构提供15万+高质量图像、150万+精确标注实例和7,176对严格时空对齐的可见光-红外图像对。解决四大核心问题：1）通过15项结构化元数据（采集高度、天气、季节、方位角、GPS坐标等）弥补环境上下文缺失，支持场景自适应训练；2）30类细粒度目标覆盖（含特种工程车辆、各类无人机等稀有目标）提升小样本学习能力；3）八大典型场景（城市、乡村、水域、荒漠等）增强复杂环境泛化性，解决域偏移问题；4）多模态融合实现全天候鲁棒感知，克服单一可见光模态在夜间、雾雨等恶劣条件下的性能衰减。数据采用通用开放格式，可直接集成至YOLO、MMDetection等主流框架，显著提升低空智能系统在真实环境中的感知精度与工程部署效率，有力支撑低空经济安全、高效发展。

The NeoDrone dataset is designed for the research and development of low-altitude UAV visual perception technologies, and is applicable to scenarios such as smart city management, emergency rescue response, power and energy inspection, traffic flow monitoring, precision agriculture monitoring, and environmental law enforcement. It provides over 150,000 high-quality images, more than 1.5 million accurately annotated instances, and 7,176 pairs of strictly spatially-temporally aligned visible-light and infrared image pairs for university research teams, AI enterprises, and institutions related to the low-altitude economy.

The dataset addresses four core technical challenges:
1) Compensating for the lack of environmental context via 15 structured metadata items (including acquisition altitude, weather, season, azimuth angle, GPS coordinates, etc.) to support scene-adaptive training;
2) Covering 30 categories of fine-grained targets (including rare targets such as special engineering vehicles and various UAVs) to improve few-shot learning capabilities;
3) Including eight typical scenarios (urban, rural, water area, desert, etc.) to enhance generalization ability in complex environments and solve the domain shift problem;
4) Achieving all-weather robust perception through multimodal fusion, thereby overcoming the performance degradation of a single visible-light modality under harsh conditions such as nighttime, fog, and rain.

Adopting universal open formats, the dataset can be directly integrated into mainstream frameworks like YOLO and MMDetection. It significantly improves the perception accuracy and engineering deployment efficiency of low-altitude intelligent systems in real-world environments, and strongly supports the safe and efficient development of the low-altitude economy.