AbstractIn this work, we present a dataset of banana leaf imagery, both with and without diseases. The dataset consists of 11,767 images, categorized as follows: 3,339 healthy images, 3,496 images of leaves affected by Black Sigatoka and 4,932 images of leaves affected by Fusarium Wilt Race 1. This data was collected to support machine learning diagnostics for disease detection. The data collection process involved farmers, researchers, agricultural experts and plant pathologists from the northern and southern highland regions of Tanzania. To ensure unbiased representation, farms were randomly selected from the Rungwe, Mbeya, Arumeru, and Arusha districts, based on the presence of banana crops and the targeted diseases. The dataset offers a comprehensive collection of images captured from November 2022 to January 2023, using a high-resolution smartphone camera across a wide geographical area. Researchers and developers can use this dataset to build machine learning solutions that automatically detect diseases in images, potentially enabling agricultural stakeholders, including farmers, to diagnose Fusarium Wilt Race 1 and Black Sigatoka early and take timely action.
Background & SummaryBananas are a staple food and cash crop for nearly 70 million farmers in Africa’s humid and sub-humid tropics1. The crop is, however, highly susceptible to diseases, particularly to Black Sigatoka and Fusarium Wilt Race 12. These diseases can cause yield losses ranging from 30% to 100%, depending on the banana cultivar and the severity of the infection3. Therefore, effective disease management strategies are essential to protect banana crops and ensure food security in Africa. The current method of identifying and confirming banana diseases relies on visual inspection of leaves, stems and other affected plant parts. This approach has limitations such as delayed diagnosis beyond the point of recovery which can result in substantial disease management and control expenses and potentially significant or even complete crop losses.Several researchers have demonstrated the potential of using machine learning tools for diagnosis of diseases in different crops such as tomatoes, cassavas, and bananas4,5,6. This approach has proven to assist agricultural stakeholders, including smallholder farmers and extension officers, in automatically identifying crop diseases as early as possible for proper and effective interventions. However, training machine learning models requires a large amount of data, and developing countries like Tanzania lack enough data to facilitate research activities in this field. To address this issue, the dataset of banana leaf images was published in the Harvard repository in 20227. The dataset comprised 16,092 images, including 5,628 healthy leaves, 5,767 affected by Black Sigatoka and 4,697 showing symptoms of Fusarium Wilt Race 1, all gathered between February and July 2021. Although the dataset was extensive and included a broad range of instances, it did not fully represent data from all farming seasons. Recognizing this limitation, this article contributes to the agricultural sector by updating the dataset to include missing farming season data, a crucial component for accurate disease detection and analysis.MethodsCollection of field dataTable 1 outlines the timeline for the imagery data acquisition process, which took place from November 22, 2022, to January 13, 2023, in Tanzania’s Northern and Southern Highlands regions. Imagery data of banana leaves were collected using a random sampling technique to select farms across four districts: Rungwe, Mbeya, Arumeru, and Arusha. A total of 51 farms participated in the study, with 14 farms in Rungwe, 11 in Mbeya, 12 in Arumeru and 14 in Arusha. These farms were chosen by agricultural extension officers and farmers’ representatives, focusing on areas with a high incidence of two diseases that significantly affect banana production.Table 1 Timeline of the data acquisition process.Full size tableImages were captured using a Samsung SM-A715F/DS smartphone camera equipped with the Open Data Kit tool, with a sample of these images presented in Fig. 1. The data collection process was carried out by farmers and researchers, under the supervision of experts from the International Institute of Tropical Agriculture (IITA) and the Nelson Mandela African Institution of Science and Technology (NM-AIST). To ensure accuracy and reliability, the collected data were thoroughly quality-checked by agricultural experts and plant pathologists.Fig. 1Sample of collected images (a) Healthy (b) Black Sigatoka (c) Fusarium Wilt Race 1.Full size imageData preprocessingThe banana leaf images were preprocessed by cleaning, cropping, renaming and annotating. The images were manually cropped to remove extraneous background elements and emphasize the key features. VisiPics and Duplicate Photo Finder were used to check for duplicate images8. The selection of this software was based on its open-source nature and user-friendliness. VisiPics removed strictly identical images, while Duplicate Photo Finder utilized the “same picture” filter for image comparison. Figure 2 illustrates the use of VisiPics in detecting and deleting identical duplicates. It shows the process where VisiPics scans the image library, compares the visual content to identify exact matches and provides the user with an interface to review and remove these duplicates.Fig. 2Method used for detecting and removing identical duplicate images.Full size imageThe cleaned images were renamed to reflect the class names (HEALTHY, FUSARIUM WILT RACE 1, BLACK SIGATOKA), and the corresponding image number, i.e., HEALTHY_105.jpg. Bulk Rename Utility software was used to rename images for all classes as shown in Fig. 3. Also, images were labelled for classification and segmentation, among other computer vision tasks.Fig. 3Method used to rename images.Full size imageData RecordsData file descriptionThis article presents the published dataset of banana leaf images which has been deposited in the Zenodo repository and published on February 23, 20239. The dataset of 11,767 images in Joint Photographic Experts Group (JPEG) format was categorized into three classes: Healthy (3339 images), Black Sigatoka (3496 images) and Fusarium Wilt Race 1 (4932 images) as illustrated in Fig. 4.Fig. 4Distribution of the dataset.Full size imageThe data repository structure is presented in Fig. 5. In the repository, the root folder consists of nine separate zipped folders named to indicate the class name and folder number; three folders of Fusarium Wilt Race 1 (FUSARIUM WILT-1.zip, FUSARIUM WILT-2.zip, FUSARIUM WILT-3.zip), three folders of Black Sigatoka (BLACK SIGATOKA-1.zip, BLACK SIGATOKA-2.zip, BLACK SIGATOKA-3.zip) and three folders of healthy (HEALTHY-1.zip, HEALTHY-2.zip, HEALTHY-3.zip). The total size of the dataset is 34.3 GB, distributed as follows: BLACK SIGATOKA-1.zip (3.7 GB), BLACK SIGATOKA-2.zip (3.3 GB), BLACK SIGATOKA-3.zip (5.1 GB), FUSARIUM WILT-1.zip (3.1 GB), FUSARIUM WILT-2.zip (5.0 GB), FUSARIUM WILT-3.zip (7.2 GB), HEALTHY-1.zip (2.0 GB), HEALTHY-2.zip (2.3 GB) and HEALTHY-3.zip (2.6 GB).Fig. 5Data structure of the repository.Full size imageTechnical ValidationThe dataset contains clear and in focus images with sufficient resolution of 3480 × 3496 pixels in jpeg format. The study used Python scripts to perform the processing steps within a Conda virtual environment. This environment was configured with Python 3.9 and included the GoogleDriveDownloader 0.4 library for downloading datasets, along with Pandas 1.3.5 for data manipulation and analysis. Leveraging the freely available software, LabelMe was used to annotate images specifically for instance and semantic segmentation tasks. For instance, segmentation with Mask R-CNN, individual objects were identified by manually drawing irregular polygon masks around the Regions of Interest (ROIs) within the banana plant images. For semantic segmentation with U-Net, regions were segmented based on their semantic meaning. Each outlined ROI was then assigned a specific class label, providing crucial information for training both deep learning models.The dataset focuses on banana imagery leaves, both healthy and those infected with two specific diseases, i.e., Fusarium Wilt Race 1 and Black Sigatoka, which significantly impact productivity in Tanzania. All classes are equally distributed to avoid data imbalance which may led to biasness. Images in this dataset were taken using a high-quality smartphone camera to avoid bias during model development. Future datasets can explore the possibility of incorporating other banana diseases apart from the two presented in this article.
Code availability
The data was preprocessed using Bulk Rename Utility: https://www.bulkrenameutility.co.uk for renaming the data, VisiPics: https://visipics.en.softonic.com for removing duplicates and LabelMe: https://github.com/labelmeai/labelme for data annotation.
ReferencesFAO. Acting together against banana diseases in Africa from food and agriculture organizanion of United Nation. http://www.fao.org/agriculture/crops/news-eventsbulletins/detail/en/item/36259/icode/en/?no_cache=1 (2021).Sanga, S. L. et al. Mobile-based deep learning models for banana disease detection. Engineering, Technology & Applied Science Research 10(3), 5674–5677, https://doi.org/10.48084/etasr.3452 (2020).Article
MATH
Google Scholar
ProMusa Vézina, A., & Van den Bergh, I. Black leaf streak. https://www.promusa.org/Black+leaf+streak (2020).Mkonyi, L. et al. Early identification of Tuta absoluta in tomato plants using deep learning. Scientific African https://doi.org/10.1016/j.sciaf.2020.e00590 (2020).Nabenda, J. N. et al. A dataset of necrotized cassava root cross-section images. Data in Brief https://doi.org/10.1016/j.dib.2020.106170 (2020).Sanga, S. et al. Mobile-based deep learning models for banana diseases detection. The International Conference on Learning Representations (ICLR) https://arxiv.org/abs/2004.03718 (2020).Mduma, N., & Leo, J. Dataset of banana leaves and stem images for object detection, classification and segmentation: A case of Tanzania. Data in Brief https://doi.org/10.1016/j.dib.2023.109322 (2023).Softonic. Software to detect and remove duplicate pictures. https://visipics.en.softonic.com (accessed January 02, 2024).Mduma, N. & Elinisa, C. Banana imagery dataset – Tanzania. Zenodo https://doi.org/10.5281/zenodo.7670326 (2023).Download referencesAcknowledgementsThe authors would like to thank and acknowledge: Study sponsors – the Swedish International Development Cooperation Agency (SIDA) and the International Development Research Centre (IDRC). Scholarship Programme - Artificial Intelligence for Development (AI4D) Africa. Scholarship Fund Manager - Africa Center for Technology Studies (ACTS).Author informationAuthors and AffiliationsNelson Mandela African Institution of Science and Technology, P. O. Box 447, Tengeru, Arusha, TanzaniaNeema Mduma & Christian ElinisaAuthorsNeema MdumaView author publicationsYou can also search for this author inPubMed Google ScholarChristian ElinisaView author publicationsYou can also search for this author inPubMed Google ScholarContributionsNeema Mduma: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Resources, Writing - Original Draft, Writing - Review & Editing, Visualization, Supervision, Project administration, Funding acquisition. Christian Elinisa: Conceptualization, Writing - Review & Editing, Methodology, Data Curation.Corresponding authorCorrespondence to
Neema Mduma.Ethics declarations
Competing interests
The authors declare no competing interests.
Additional informationPublisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
Reprints and permissionsAbout this articleCite this articleMduma, N., Elinisa, C. Banana Leaves Imagery Dataset.
Sci Data 12, 477 (2025). https://doi.org/10.1038/s41597-025-04456-4Download citationReceived: 19 February 2024Accepted: 13 January 2025Published: 21 March 2025DOI: https://doi.org/10.1038/s41597-025-04456-4Share this articleAnyone you share the following link with will be able to read this content:Get shareable linkSorry, a shareable link is not currently available for this article.Copy to clipboard
Provided by the Springer Nature SharedIt content-sharing initiative