Drone Imagery Dataset of Aedes Mosquito Breeding Habitats

This dataset provides manually annotated drone imagery of potential Aedes mosquito breeding habitats for use in computer vision and public health research. The data were collected using unmanned aerial vehicles (UAVs) - DJI Mavic 2 Pro to capture high-resolution RGB images of outdoor environments containing common water-holding containers that are known to serve as breeding sites for Aedes mosquitoes.

The dataset consists of 2,821 images in PNG format, each with a resolution of 512 × 512 pixels. All images were annotated manually with bounding boxes to identify five categories of breeding habitats: water tanks, drums, tires, coconut shells, and other open containers. Manual annotation was performed using the Roboflow platform to ensure accurate and consistent ground-truth labels suitable for object detection tasks.

To facilitate immediate use in machine learning pipelines, the dataset is provided with a predefined train, validation, and test split of 70%, 15%, and 15%, respectively. No image preprocessing or data augmentation has been applied, allowing users full flexibility to perform custom preprocessing and augmentation according to their specific research needs.

The dataset is released in three widely used annotation formats: COCO (JSON), YOLOv8 (TXT), and Pascal VOC (XML), enabling compatibility with a broad range of object detection frameworks. This multi-format release supports reproducibility and ease of adoption across different research communities.

This dataset is intended for academic research and development of automated methods for mosquito breeding habitat detection, environmental monitoring, and vector-borne disease surveillance using aerial imagery. It may also be used for benchmarking object detection algorithms on real-world, imbalanced datasets derived from UAV observations.