BaleUAVision: Hay Bales UAV Captured Dataset

BaleUAVision dataset comprises a comprehensive collection of UAV-captured images of agricultural fields with hay bales. It includes high-resolution RGB imagery (in both raw and annotated -COCO, CSV, JSON, YOLO, Segmentation Masks- formats), catering to a wide range of applications from precision agriculture to machine learning in computer vision and autonomous navigation. More specifically, it encompasses detailed UAV-captured data from agricultural fields, characterized by varied flight parameters to optimize image capture for machine learning applications. This dataset is distinctive due to its diverse altitude range (50-100m), multiple speed settings (3.7-5m/s), and different overlap ratios ensuring comprehensive field coverage. The total area covered by the dataset is 938,715 square meters, with a Ground Sampling Distance (GSD) ranging from 1.53 to 3.06 cm/pixel, facilitating fine-grained analysis. The data includes 2,599 high-resolution RGB images, each meticulously annotated for semantic segmentation, and is coupled with orthophotos to support simulation tasks such as autonomous hay bale collection scenarios. This dataset is a valuable asset for advancements in precision agriculture, offering extensive resources for developing and testing computer vision and path-planning algorithms. Dataset Details Images: High-resolution RGB images of 16 Hay bale fields Number of images: 2,599 Formats: Raw RGB images and Annotated images in {COCO, CSV, JSON, YOLO, Segmentation Masks} formats Annotations: Semantic segmentation with polygons Dataset Task Type Usage: Segmentation and Classification/Detection Tasks Annotation Software Used: Label Studio Captured Fields: The dataset includes imagery from 16 fields, with 14 located in the Xanthi region and 2 in the Drama region, both situated in the northern part of Greece Orthophotos: Orthomosaic views for each subset of the dataset, generated through an image stitching process, offering a macro-perspective of the fields Size: ~45.5GB Resolution: 4056x3040 (RGB) Flight Parameters: Various altitudes, speeds and overlaps Geo-location: Yes, each image is geo-referenced Total Area Covered: 938,715 square meters (m²) in total Additional Information: The number of hay bales has been manually counted for each field from the orthophoto representations, providing a reliable reference for users aiming to develop or evaluate algorithms for automated hay bale counting Files Structure ├── BaleUAVision ├── Annotated ├── Hay bales 1 ├── Hay-bales-1-YOLO # folder which contains **YOLO** formated .txt files ├── images # folder which contains images with prefixes ├── Masks # folder that contains image **Segmentation Masks** using the python script "segmentation_masks.py" ├── classes # .txt file which contains the name of the class ├── Hay-bales-1-COCO # .json file which is for **COCO** format ├── Hay-bales-1-CSV # classic .csv file for **CSV** format ├── Hay-bales-1-JSON # .json file for **JSON** format └── notes ├── Hay bales 2 ├── Hay-bales-2-YOLO ├── images ├── Masks ├── classes ├── Hay-bales-2-COCO ├── Hay-bales-2-CSV ├── Hay-bales-2-JSON └── notes ... └── Hay bales 16 ├── Hay-bales-16-YOLO ├── images ├── Masks ├── classes ├── Hay-bales-16-COCO ├── Hay-bales-16-CSV ├── Hay-bales-16-JSON └── notes ├── Orthophotos ├── Hay bales 1 orthophoto # .tiff images for classic orthomosaic/panorama representation ├── Hay bales 2 orthophoto ... └── Hay bales 16 orthophoto ├── Raw Data ├── Hay bales 1 # contains 205 .jpg images ├── Hay bales 2 # contains 423 .jpg images ... └── Hay bales 16 # contains 119 .jpg images └── Dataset Description.csv # contains details and metadata for each Hay bale sub-set