Drone-based dataset of annotated sunflower images from Bangladesh

Abstract

Accurate and automated detection of sunflower plants, along with assessments of their growth stages and health conditions, is crucial for enabling precision agriculture and improving crop management. In this work, we present a drone-based dataset of annotated sunflower images, derived from high-resolution videos captured at two distinct locations in Bangladesh. The original dataset comprises 1649 images extracted from drone footage of the BARI Surjomukhi-3 variety under various orientations, health conditions, and weather scenarios. After meticulous annotation using the Roboflow platform and augmentation with seven distinct techniques, the dataset expanded to 4286 images in Pascal VOC format. Detailed metadata—including geospatial coordinates, timestamped acquisition conditions, and camera settings—accompanies the dataset to support reproducibility and model generalization. By offering a comprehensive suite of annotated and augmented images, this dataset provides a valuable resource for developing and refining computer vision models geared toward sunflower detection, maturity evaluation, and yield prediction, ultimately advancing sustainable farming practices and decision-making tools in agricultural research.