GreenFlyDB: An Image Dataset for Greenfly-Based Urban Waste Detection and Drone-Assisted Environmental Monitoring

GreenFlyDB is a publicly available image dataset designed for greenfly (aphid) detection in agricultural and urban environmental monitoring scenarios. The dataset is intended for object detection tasks, where greenflies are exploited as biological indicators of organic waste presence and environmental degradation.

The dataset contains a total of 2,405 RGB images collected from publicly accessible online sources under diverse real-world conditions, including variations in illumination, background complexity, camera viewpoints, and insect scale. These variations aim to reflect realistic operational conditions encountered in drone-assisted and ground-based monitoring systems.

Images are organized into two classes: Green Fly, containing visible aphids annotated with bounding boxes, and Not Green Fly, which includes visually similar insect species deliberately selected as hard negative samples to reduce false detections. All annotations were manually produced using bounding boxes and are provided in YOLO format, where each image is associated with a corresponding text file containing normalized bounding-box coordinates.

The dataset is split into fixed and mutually exclusive subsets for reproducible research: 70% training, 20% validation, and 10% testing. The same split is maintained across all experiments to ensure fair benchmarking of object detection models.

GreenFlyDB is suitable for training and evaluating real-time deep learning–based object detection models, particularly YOLO-based architectures, and supports research in precision agriculture, urban waste monitoring, bio-inspired environmental assessment, and autonomous drone-assisted inspection systems.