14 Free Datasets for Agricultural Crop Detection
14 Free Datasets for Agricultural Crop Detection
In the field of machine learning, having relevant, clean data to train better models is necessary. For some common uses, there are plenty of free, labeled datasets available, but for more specialized applications, where, for example, the images must be captured by drones so the models can be properly trained for the target data, there are far fewer datasets available. This is because of the cost of creating such datasets. Although there are other limitations, in the case of drone-based image datasets, the target’s approximate altitude is a key point. If models are trained on images taken at lower or higher heights than those in the target dataset, they won’t produce the expected results. On the other hand, good results also depend on the similarity between training and target images in terms of brightness, contrast, and other image features.
Roboflow has a great, useful database of public datasets and, more importantly, enables us to augment our own or others’ public datasets (by forking or cloning the datasets), changing image sizes, orientations, zoom, brightness, blur, color, noise, and many important image features, so that, in the end, not only do we have a richer dataset, but we can also train our models on images that are more likely to be seen in the end-user’s data. Unfortunately Roboflow’s datasets usually don’t include meta-data like the average height of drone’s flight, but they are still a good help for starting on crop-detection projects.
At Saiwa, we wanted to start training our models for new crop-counting tenants, so we were seeking free, drone-based datasets for agricultural crop detection. This article is based on the results of this search, including crop datasets, some of which also include weeds related to the specific crops. Below we included some of these datasets.
1. Plant Detection and Counting Dataset
This dataset contains 189 labeled, drone-based pictures of three classes: maize, sugar beet, and sunflower. It’s CC BY 4.0 license lets all uses with attribution. [1]
The images of this dataset are captured at 20-50 meters above ground targeting 2-5 mm/pixel GSD.
2. Soybean Detection Dataset
The dataset owns 12,907 images with soybean labels, augmented by rotating and/or changing the brightness of the train set images, from the original dataset of 5,389 images. The train set includes 11,290 images, validation set has 1,078 and 539 images used for test set. The usage is also assumed to remain under the CC BY 4.0 license conditions. [2]
All images are resized (fit in black edges) to 1280×1280 during preprocessing.
3. Tobacco Detection Dataset
This Dataset contains 11,976 images, 11,035 images in training set, 623 for validation and 318 for test set, augmented by horizontally and/or vertically flipping and/or 90° Rotation in three directions, Clockwise, Counter-Clockwise, Upside Down from an original dataset, contained 4692 images. Object classes are five, including tobacco, diente_leon, kikuyo, lengua_vaca and others for tobacco and it’s weeds. The license is CC BY 4.0 too. [3]
4. Potato Detection Dataset
This dataset includes 1,680 images: 1,341 images for training set, 169 images for validation and 170 for test set. It also has 5 classes: potato(papa), diente-de-leon, kikuyo, lengua-de-vaca and others. It helds CC BY 4.0 license too. [4]
5. Corn Detection Dataset
This dataset originally included 455 labeled images, which is augmented to 1091 images, by rotating, cropping or blurring per training images, gives us totally 954 training images, 91 validation images and 46 test images from corn fields. It also has the CC BY 4.0 license. [5]
6. Potato Segmentation Dataset
This segmentation dataset includes 1,441 labeled images: 1,068 images in the training set, 230 images in the validation set, and 143 images in the test set, from potato fields. During preprocessing, auto-orientation was applied and images were resized by stretching to 640x640. The dataset holds CC BY 4.0 license too. [6]
7. Sugar-beet, Chardon Detection Dataset
This dataset originally included 724 images from sugar-beet and chardons, which is augmented to 1,954 images, by rotating, cropping and blurring train set, gives us 1,845 train images and 109 validation images. The dataset is auto-oriented and resized by stretching to 416x416. This dataset also holds CC BY 4.0 for license. [7]
8. Tobacco Detection Dataset
This dataset contains 662 labeled images with 19,768 labels from tobacco fields, 463 images for training set, 133 images for validation and 66 images for test set. The dataset is auto-oriented and resized to 640x640 by stretching. The detection model is trained with 50% confidence and overlap threshold, giving 84.6% precision, and 84.9% recall on test set, and it’s public to use. The license is CC BY 4.0. [8]
9. Sugar-beet Detection Dataset
This dataset has 1,684 images which are gained by augmentation from 624 original images, by rotating, blurring and zooming on train dataset, which gives us 1,590 train images and 94 images for validation from sugar-beet fields holding CC BY 4.0 license. The dataset is auto-oriented and resized by stretching to 416x416. [9]
10. Corn Detection Dataset
This corn detection dateset originally included 1,914 auto-oriented 416x416 images, augmented to 5,168 images by rotating, cropping and/or blurring on training set, which gives us 4,881 train images and 287 images on validation set. The license is also CC BY 4.0. [10]
11. Potatoes Weeds Dataset
This is a dataset containing 400 images of potato and it’s weeds, including 320 images for training, 40 for validation, and 40 images for the test set. The license is CC BY 4.0 and classes are potato (papa), diente_leon, lengua_vaca, kikuyo, and others. [11]
12. Cabbage and Radish Detection Dataset
The Dataset includes 323 images with three classes: cabbage, radish_1 and radish_2 which contains: 195 images for train the model, 64 validation images and 64 test images. The license is also CC BY 4.0. This dataset is cleaned after the test and images with no annotations or wrong annotations are removed. [12]
13. Rice Seedling Detection Dataset
This Dataset contains 280 images which is gained by augmentation using bounding boxes exposure, from -10% to +10%, for rice crop fields. 270 images is used for train set and 10 images for validation set. The license is CC BY 4.0.
14. Cotton Detection Dataset
This dataset contains 71 labeled auto-oriented images with 640x640 size, for cotton detection, including 64 training images, 4 validation images and 3 test images. The license is CC BY 4.0. [14]
Conclusion
Nowadays, the use of machine learning applications such as image processing is becoming more popular in the field of agriculture. Good supervision of our fields and crops leads to better, healthier products. In this article, we took a deep dive into publicly available, drone-based, crop-detection datasets. Thanks to the Ninja and Roboflow teams, we gathered datasets with annotations for several crops, including corn, rice, tobacco, potato, cotton, cabbage and radish, sugar beet, sunflowers, maize, and soybean, useful for training detection or segmentation models. We hope this effort will be useful to other researchers and companies that, like us, have concerns about global health, for similar uses. We will return with further improvements to our research, more datasets, and comments on datasets’ usefulness after training our models.
[1] E. David, Plant detection and counting from high-resolution RGB images acquired from UAVs: comparison between deep-learning and handcrafted methods with application to maize, sugar beet, and sunflower crops, ver. 1. Zenodo, Jun. 2021. doi: 10.1101/2021.04.27.441631.
[2] Weed Data Soybean 2023, “Data_Soybean_2023 Dataset” [Dataset], Roboflow Universe, Roboflow, Nov. 2024. [Online]. Available: https://universe.roboflow.com/weed-data-soybean-2023/data_soybean_2023. [Accessed: Oct. 25, 2025].
[3] Nahian, “crop_count Dataset,” Roboflow Universe, Sep. 2025. [Online]. Available: https://universe.roboflow.com/nahian-ljdvh/crop_count-tjjve. [Accessed: Oct. 25, 2025].
[4] Universidad, “Dataset_Unificado_Grupo1YLuis_SinModificar Dataset,” Roboflow Universe, Sep. 2024. [Online]. Available: https://universe.roboflow.com/universidad-rltbk/dataset_unificado_grupo1yluis_sinmodificar. [Accessed: Oct. 25, 2025].
[5] T. Orlando, “train_yolo_bbox Dataset,” Roboflow Universe, Oct. 2024. [Online]. Available: https://universe.roboflow.com/thiago-orlando/train_yolo_bbox. [Accessed: Oct. 25, 2025].
[6] Qgis, “Potato Dataset,” Roboflow Universe, Jun. 2025. [Online]. Available: https://universe.roboflow.com/qgis-cdke3/potato-2l2sf. [Accessed: Oct. 25, 2025].
[7] New Workspace, “22_01_25_center_beet+chardon Dataset,” Roboflow Universe, Jan. 2022. [Online]. Available: https://universe.roboflow.com/new-workspace-evkln/22_01_25_center_beet-chardon. [Accessed: Oct. 25, 2025].
[8] J. Liu, “Detection-tobacco Dataset,” Roboflow Universe, May 2025. [Online]. Available: https://universe.roboflow.com/liu-jinhong/detection-tobacco. [Accessed: Oct. 25, 2025].
[9] New Workspace, “22_02_09_faster_beet_last Dataset,” Roboflow Universe, Feb. 2022. [Online]. Available: https://universe.roboflow.com/new-workspace-evkln/22_02_09_faster_beet_last. [Accessed: Oct. 25, 2025].
[10] New Workspace, “21_12_13__WITH_DOUBLE_WITH_FLOU_WITHOUT_HERBS Dataset,” Roboflow Universe. Roboflow, Dec. 2021. [Online]. Available: https://universe.roboflow.com/new-workspace-evkln/21_12_13__with_double_with_flou_without_herbs. [Accessed: Oct. 25, 2025].
[11] transformer, “potatoes-weeds-dataset Dataset,” Roboflow Universe, Jun. 2024. [Online]. Available: https://universe.roboflow.com/transformer-dkslr/potatoes-weeds-dataset. [Accessed: Oct. 25, 2025].
[12] sensor, “field_crop_20221006 Dataset,” Roboflow Universe, Nov. 2022. [Online]. Available: https://universe.roboflow.com/sensor-hq923/field_crop_20221006. [Accessed: Oct. 25, 2025].
[13] Jungwoong, “rice-seedling-detection Dataset,” Roboflow Universe, Sep. 2024. [Online]. Available: https://universe.roboflow.com/jungwoong/rice-seedling-detection. [Accessed: Oct. 25, 2025].
[14] West Texas AM University, “crop Dataset,” Roboflow Universe, Oct. 2024. [Online]. Available: https://universe.roboflow.com/west-texas-am-university/crop-r3m1m. [Accessed: Oct. 25, 2025].
