Dataset from drone and system camera for bilberry counting

General description This dataset comprises two sets of 156 images of bilberry (Vaccinium myrtillus) shrubs collected in Ilomantsi, Finland. One image set was captured using a Canon EOS 6D system camera, and the other using a DJI Mavic 2 Pro drone. Images from both platforms were acquired from the same reference plots during the 2023 field season. Images were collected across multiple forest sites to capture a broad range of bilberry growing conditions, including variation in plant height, canopy density, and ground vegetation. The dataset was used as a test set in the study evaluating the performance of a YOLOv8 object detection model and the Segment Anything Model (SAM) combined with a Random Forest classifier for bilberry detection and counting. The data was collected with a 0.5-m-by-0.5-m square berry frame, such that the outlines of the frame remain visible in each image. Images were subsequently cropped around the frame to ensure that only bilberries within the frame were considered. System camera images were captured from an oblique angle optimized for berry visibility. Drone images were acquired from a forward-looking camera angle at approximately 1m altitude. This configuration was selected to minimize disturbance effects, including potential berry displacement caused by drone rotor wash. The dataset has been utilized in the publication, and this publication should be cited for technical details when using the data. The publication also describes the details of how the data was used: Hyyppä, Susanna, Josef Taher, Harri Kaartinen, Teemu Hakala, Kirsi Karila, Leena Matikainen, Marjut Turtiainen, Antero Kukko, and Juha Hyyppä. 2026. "Quantifying Bilberry Counts and Densities: A Comparative Assessment of Segmentation and Object Detection Models from Drone and Camera Imagery" Forests 17, no. 2: 253. https://doi.org/10.3390/f17020253 Data can be used for comparing various machine learning and deep learning solutions for berry counting from images. Such data is expected to be valuable for automating field reference collection for wild berry yield maps. Dataset structure The dataset is provided as a ZIP archive containing: drone/: Folder containing 156 images captured using a drone system_camera/: Folder containing 156 images captured with a system camera. These images correspond to the same test plots as the drone images. image_annotations_and_field_measurements.xlsx: Excel file containing image-level annotations and field measurements. The file includes 156 rows, each corresponding to a paired system camera and drone image from the same test plot. Excel file description The spreadsheet includes the following columns: canon_filename: Filename of the system camera image drone_filename: Filename of the corresponding drone image. Each row links images acquired from the same test plot canon_annotation_count: Number of annotated bilberries in the system camera image drone_annotation_count: Number of annotated bilberries in the drone image actual_count: Number of bilberries counted manually in the field Additional information Acquisition date: The dataset was collected on July 24-27, 2023. Acknowledgements We would like to acknowledge the following people for participating in the study: Kirsi Karila, Leena Matikainen, Marjut Turtiainen, Antero Kukko and Emmi-Lotta Vesala. Cite Any scientific publication using this dataset should cite the following paper: Hyyppä, Susanna, Josef Taher, Harri Kaartinen, Teemu Hakala, Kirsi Karila, Leena Matikainen, Marjut Turtiainen, Antero Kukko, and Juha Hyyppä. 2026. "Quantifying Bilberry Counts and Densities: A Comparative Assessment of Segmentation and Object Detection Models from Drone and Camera Imagery" Forests 17, no. 2: 253. https://doi.org/10.3390/f170202534