UAV based Tomato Dataset
Description
This dataset contains UAV-acquired tomato field imagery collected for tomato counting and yield prediction research under real agricultural conditions. The images were captured using an unmanned aerial vehicle (UAV) equipped with high-resolution RGB and/or multispectral sensors over multiple tomato cultivation plots during different crop growth stages, including flowering, fruit setting, ripening, and harvest stages. The dataset includes raw aerial images, orthomosaic patches, and manually annotated tomato instances generated using Roboflow annotation tools. The annotations are provided in object detection format suitable for deep learning applications such as YOLO, COCO, or Pascal VOC. Each image is associated with metadata including flight altitude, overlap settings, environmental conditions, GPS coordinates, and plot-level yield measurements. The primary purpose of the dataset is to support research in: -- Tomato counting using computer vision -- UAV-based precision agriculture -- Yield prediction modeling -- Small-object detection in aerial imagery -- Agricultural AI and deep learning applications The dataset was collected under varying illumination conditions and field environments to improve model robustness and generalization. Ground-truth yield measurements were recorded from corresponding plots to enable supervised learning and benchmarking for yield estimation tasks. This dataset can be used by researchers in: -- Precision agriculture -- Remote sensing -- Computer vision -- Machine learning -- Crop phenotyping The dataset is organized into structured directories containing raw images, annotations, metadata files, and experimental documentation to facilitate reproducibility and reuse in future agricultural AI research.
Files
Steps to reproduce
Steps to Reproduce the UAV Tomato Dataset 1. Study Area Preparation Select a tomato field with defined plots and sufficient plant density. Record GPS coordinates, mark plot boundaries, and assign unique plot IDs for metadata consistency. 2. UAV Platform and Sensor Configuration Use a UAV with high-resolution RGB or multispectral camera. Multirotor drone operated at low altitude for high spatial resolution. Recommended settings: - Flight altitude: 20–40 m - Front overlap: 75–85% - Side overlap: 65–75% - Camera angle: 90° (nadir) - Speed: 3–5 m/s 3. Flight Mission Planning Plan automated flights using DJI GS Pro, Pix4Dcapture, or DroneDeploy. Ensure full coverage and consistent overlap. Flights should be conducted under: - Stable lighting (10:00 AM–12:00 PM) - Wind <5 m/s 4. Ground Control Points (GCPs) Place 4–6 visible GCPs across the field. Record coordinates using GPS to improve orthomosaic accuracy. 5. UAV Image Acquisition Capture images at key growth stages: flowering, fruit set, ripening, and harvest. Store raw images with structured naming. 6. Image Annotation Upload images/patches to annotation tools (e.g., Roboflow). Label each tomato using bounding boxes. Verify annotations to reduce errors (missing, duplicate, incorrect boxes). Export in YOLO format. 7. Ground Truth Yield Collection Collect plot-wise harvest data aligned with image IDs: Fruit count - Total weight (kg) - Average fruit weight - Disease observations 8. Metadata Compilation Prepare metadata including: image ID, flight ID, plot ID, GPS location, altitude, weather, growth stage, and yield data. 9. Data Quality Control Perform: - Removal of blurred images - Annotation verification - Bounding box consistency checks - Metadata validation - Orthomosaic inspection 10. Applications Dataset supports: object detection, tomato counting, yield prediction, crop monitoring, UAV analysis, precision agriculture, and small-object detection research. 11. Citation Morshed, M. S. J., Adnan, M. N. & Islam, M. R. (2025). Comparative Analysis of Classical Feature Detection Methods for UAV-Based Tomato Detection. Journal of Computer Science, 21(9), 2153–2170. 12. License For research and educational use only. Check repository for commercial use terms. 13. Contact Muhammad Sarwar Jahan Morshed, PhD Student Jashore University of Science and Technology, Bangladesh Email: p220102.cse@student.just.edu.bd 14. Acknowledgements Thanks to field staff (Md. Aslam Sheikh), UAV operator (Abdullah Morshed), and annotation contributors (Kamrul Islam Ridoy, Muhammad Ashaheed, Mohammed Mahathir). 15. Keywords UAV, tomato counting, yield prediction, precision agriculture, object detection, deep learning, computer vision, remote sensing, agricultural AI.