Source-linked AI summary
An open access repository of images on plant health to enable the development of mobile disease diagnostics
David. P. Hughes, Marcel Salathe
TL;DR
PlantVillage addresses crop-disease yield losses by releasing an openly accessible, expert-curated image dataset for mobile diagnostics and computer-vision research. The dataset contains over 50,000 plant images, and the platform provides information on over 150 crops and 1,800 diseases.
Problem
Crop diseases reduce potential yields, while many growers lack access to expert diagnostic help, motivating mobile disease-diagnostic approaches.
Method
The paper releases expertly curated images of healthy and diseased plant leaves through the PlantVillage platform and describes the dataset and platform.
Results
54,309 images span 14 crop species, while PlantVillage provides information on over 150 crops and over 1,800 diseases.
Takeaways & Limitations
The openly available dataset supports ongoing crowdsourcing and computer-vision efforts toward mobile crop-disease diagnostics.
Takeaways & Limitations
The collection intentionally includes varied conditions such as strong sun or cloud, but its end-user setting is a grower with those conditions.
Abstract
from arXiv · showhide
Human society needs to increase food production by an estimated 70% by 2050 to feed an expected population size that is predicted to be over 9 billion people. Currently, infectious diseases reduce the potential yield by an average of 40% with many farmers in the developing world experiencing yield losses as high as 100%. The widespread distribution of smartphones among crop growers around the world with an expected 5 billion smartphones by 2020 offers the potential of turning the smartphone into a valuable tool for diverse communities growing food. One potential application is the development of mobile disease diagnostics through machine learning and crowdsourcing. Here we announce the release of over 50,000 expertly curated images on healthy and infected leaves of crops plants through the existing online platform PlantVillage. We describe both the data and the platform. These data are the beginning of an on-going, crowdsourcing effort to enable computer vision approaches to help solve the problem of yield losses in crop plants due to infectious diseases.
4 Digital Epidemiology Lab, School of Life Sciences, School of Computer and Communication Sciences, EPFL, Switzerland
The effort aims to use crowdsourcing and computer vision to address crop yield losses caused by infectious diseases.
- The data begin an ongoing crowdsourcing effort to enable computer vision approaches for addressing infectious-disease-related yield losses.
BACKGROUND & SUMMARY
Infectious diseases and pests continue to threaten food production, especially for smallholder farmers, while diagnostic expertise remains difficult to scale. PlantVillage responds by combining an online crop-health platform with an openly released, expert-curated image dataset for future machine-assisted diagnosis.
- Food-security motivation: Smallholder farmers may routinely lose 80-100% of a crop to pests and diseases, intensifying yield gaps in poorer regions.
- PlantVillage platform: PlantVillage is an online crop-health and crop-disease platform modeled after community-driven programming forums.
- PlantVillage platform: PlantVillage traffic grew 250% year over year and reached its 2 millionth visitor in fall 2015.
- PlantVillage platform: The platform provides open-access information covering over 150 crops and over 1,800 diseases, written by plant pathology experts for growers.
- Machine-assisted diagnosis: Machine-assisted diagnosis is proposed as a way to support or provide visual disease diagnosis where human expertise is unavailable or difficult to scale.
- Dataset contribution: The paper releases tens of thousands of expert-labeled images of healthy and diseased plants, with over 50,000 openly stored on PlantVillage and continued growth planned.
METHODS
The dataset was assembled from field-trial leaves and research-station collections, photographed under varied outdoor conditions and processed into standardized images.
- Image sources: Images were collected from experimental research stations associated with U.S. Land Grant Universities, with additional sources planned.
- Image collection: Technicians collected leaves from field trials of crops infected with one disease and photographed them against grey or black paper backgrounds in full light.
- Image collection: Researchers intentionally included varied lighting and conditions to reflect the range encountered by growers using smartphones.
- Image collection: Each leaf typically received 4-7 automatic-mode photographs while being rotated around 360 degrees.
- Image collection: Large leaves were photographed in sections to retain high-resolution, close-proximity views.
- Image processing: Images were cropped to remove much of the background and oriented with leaf tips pointing upward.
DATA RECORDS
The PlantVillage dataset contains 54,309 images spanning 14 crop species and multiple disease classes, with healthy-leaf images available for 12 species. Records are organized by crop and disease status and made available through the PlantVillage website.
- Dataset scope: 54,309 images span 14 crop species, including apple, blueberry, cherry, corn, grape, orange, peach, bell pepper, potato, raspberry, soybean, squash, strawberry, and tomato.The supplied crop list is distributed across the dataset description passages.
- Disease coverage: The collection includes 17 fungal, 4 bacterial, 2 mold, 2 viral, and 1 mite-caused disease classes.These counts describe the disease categories represented in the records.
- Disease coverage: Healthy leaves not visibly affected by disease are included for 12 crop species.Healthy images provide a non-diseased class alongside infected leaves.
- Access: The data records are available through the PlantVillage website, with an image-URL-to-classification mapping deposited separately.The supplied text names the website and describes the accompanying mapping file.
- Organization: Table 1 summarizes each crop’s current disease status, while parenthesized values indicate the number of images in each class.The table is dated April 4, 2016 in the supplied caption.
TECHNICAL VALIDATION
Disease identities were established through field-based assessment by expert plant pathologists using standard phenotyping approaches. Experts also diagnosed diseases in experimentally infected crops and sentinel plots before images were curated into PlantVillage.
- Expert assessment: Expert plant pathologists determined disease states directly in the field with technicians.The experts provided the diagnoses used for the dataset labels.
- Expert assessment: Diagnoses followed standard phenotyping approaches used by plant pathologists.The supplied passages identify standard phenotyping as the basis for disease-state determination.
- Validation settings: In many cases, experts infected crops directly using standard experimental approaches, making diagnosis straightforward.This procedure was used for some disease cases in the field trials.
- Validation settings: For diseases occurring in sentinel plots, experts diagnosed their presence in regional experimental research stations.Sentinel plots were maintained to identify disease presence in a region.
- Curation: Only expertly identified leaves were retained in the PlantVillage database.The database was curated using the experts’ diagnoses.