Plants Diseases Detection And Classification

## Overview

Plants Diseases Detection And Classification is a dataset for object detection tasks - it contains Plants annotations for 2,516 images.

## Getting Started

You can download this dataset for use within your own projects, or fork it into a workspace on Roboflow to create your own model.

  ## License

  This dataset is available under the [CC BY 4.0 license](https://creativecommons.org/licenses/CC BY 4.0).

Explore our Plant Disease Image Dataset, featuring a diverse collection of labeled images for developing and testing machine learning models in agriculture.

This dataset consists of five subsets with annotated images in COCO format, designed for object detection and tracking plant growth: 1. Cucumber_Train Dataset (for Faster R-CNN) - Includes training, validation, and test images of cucumbers from different angles. - Annotations: Bounding boxes in COCO format for object detection tasks.

1. Tomato Dataset
2. Contains images of tomato plants for 24 hours at hourly intervals from a fixed angle.

3. Annotations: Bounding boxes in COCO format.

4. Pepper Dataset

5. Contains images of pepper plants for 24 hours at hourly intervals from a fixed angle.

6. Annotations: Bounding boxes in COCO format.

7. Cannabis Dataset

8. Contains images of cannabis plants for 24 hours at hourly intervals from a fixed angle.

9. Annotations: Bounding boxes in COCO format.

10. Cucumber Dataset

11. Contains images of cucumber plants for 24 hours at hourly intervals from a fixed angle.

12. Annotations: Bounding boxes in COCO format.

This dataset supports training and evaluation of object detection models across diverse crops.

A collection of non-framed multi-spectral images of tomato plants infected with the Tuta Absoluta leafminer disease.

(1) In agriculture, plant diseases are primarily responsible for the reduction in production which causes economic losses. In plants, citrus is used as a major source of nutrients like vitamin C throughout the world. However, ‘Citrus’ diseases badly effect the production and quality of citrus fruits.

(2) The computer vision and image processing techniques have been widely used for detection and classification of diseases in plants.

(3) The dataset contains an image gallery of healthy and unhealthy citrus fruits and leaves that could be usable for the researchers to prevent plants from diseases using advanced computer vision techniques. The disease targeted in the data sets are the Blackspot, Canker, Scab, Greening, and Melanose.

(4) The dataset contains 759 images of healthy and unhealthy images for both Citrus fruits and leaves collectively. Each image contains 256 * 25 dimensions with 72 dpi resolution.

(5) All images were acquired from the Sargodha region, a tropical area of Pakistan under the supervision of Dr. Basharat ALi Saleem, Endeavour Executive Fellow Curtin University · Horticulture Research Laboratory Postharvest Australia · Bentley

(6) All images were annotated manually by the domain expert Dr. Basharat ALi Saleem to represent their every class such as : For Citrus fruits (Black Spot, Canker, Greening, Scab, and healthy with total number of 150 images ), For Citrus Leaves (Black Spot, Canker, Greening, Melanose, and healthy with total number of 609 image)

(6) Further details can be found in the associated publications with the dataset.

The Plant RNA-Image Repository is a compiled database of plant images and omics data. The dataset contains images of four distinct plant maladies, including powdery mildew, rust, leaf spot, and blight, as well as gene expression and metabolite data. Using a high-resolution camera in a controlled environment at the facility of the various Agriculture Universities of the Khyber Pakhtunkhwa, Pakistan. We captured 26940 images of plants, where each class has different number of samples for each disease type. Each image was labelled with the disease type corresponding to it. The images were preprocessed by resizing them to 224x224 pixels and standardizing the pixel values. In addition to collecting images of the same plants, we also collected gene expression and metabolite data. We extracted RNA from the plant leaves using a commercial reagent and sequenced it on an Illumina HiSeq 4000 platform. The average length of the 100 million pairedend readings obtained was 150 base pairs. The unprocessed reads were trimmed with Trimmomatic and aligned with STAR against the reference genome. We counted the number of reads that mapped to each gene using featureCounts, and then identified differentially expressed genes between healthy and diseased plants using the DESeq2 package in R. Using gas chromatography-mass spectrometry (GC-MS), we gathered additional metabolite information. Using a methanol-water extraction protocol, we extracted metabolites from the plant leaves and analyzed the extracts using GC-MS.

Medicinal Plant Detect

## Overview

Medicinal Plant Detect is a dataset for object detection tasks - it contains Image annotations for 413 images.

## Getting Started

You can download this dataset for use within your own projects, or fork it into a workspace on Roboflow to create your own model.

  ## License

  This dataset is available under the [CC BY 4.0 license](https://creativecommons.org/licenses/CC BY 4.0).

This paper presents a novel image dataset with high intrinsic ambiguity and a long-tailed distribution built from the database of Pl@ntNet citizen observatory. It consists of 306146 plant images covering 1081 species. We highlight two particular features of the dataset, inherent to the way the images are acquired and to the intrinsic diversity of plants morphology:

(i) the dataset has a strong class imbalance, i.e. a few species account for most of the images, and,

(ii) many species are visually similar, rendering identification difficult even for the expert eye.

These two characteristics make the present dataset well suited for the evaluation of set-valued classification methods and algorithms. Therefore, we recommend two set-valued evaluation metrics associated with the dataset (macro-average top-k accuracy and macro-average average-k accuracy) and we provide baseline results established by training deep neural networks using the cross-entropy loss.

A full description of the dataset as well as baseline experiments can be found in the following publication:

"Pl@ntNet-300K: a plant image dataset with high label ambiguity and a long-tailed distribution", Camille Garcin, Alexis Joly, Pierre Bonnet, Antoine Affouard, Jean-Christophe Lombardo, Mathias Chouet, Maximilien Servajean, Titouan Lorieul and Joseph Salmon, in Proc. of Thirty-fifth Conference on Neural Information Processing Systems, Datasets and Benchmarks Track, 2021.

Please cite the above reference for any publication using the dataset.

Utilities to load the data and train models with pytorch can be found here: https://github.com/plantnet/PlantNet-300K/

Wheat Plant Diseases Dataset 🌱

This dataset is designed to empower researchers and developers in creating robust machine learning models for classifying various wheat plant diseases. It offers a collection of high-resolution images showcasing real-world wheat diseases without the use of artificial augmentation techniques.

Content

• Number of Images: 14,155
• Image Quality: High-resolution image
• Disease Classes:
  • Aphid [Pest]
  • Mite [Pest]
  • Stem Fly [Pest]
  • Rusts
    • Black Rust / Stem Rust
    • Brown Rust / Leaf Rust
    • Yellow Rust / Stripe Rust
  • Smut (Loose, Flag)
  • Common Root Rot
  • Helminthosporium Leaf Blight / Leaf Blight
  • Wheat Blast
  • Fusarium head blight / Scab
  • Septoria Leaf Blotch
  • Spot Blotch
  • Tan Spot
  • Powdery Mildew

Disease Cause and Monitoring Information

• Wheat Disease Cause: Wheat plants can contract diseases from various sources, but some of the most common culprits fall into these categories:
  • Fungi: These microscopic organisms are a leading cause of wheat diseases. Examples include Fusarium (responsible for head blight and scab), Septoria (causing leaf blotch), and various rust fungi like stem rust, stripe rust, and leaf rust.
  • Bacteria: Bacterial diseases are less frequent but can still cause significant damage. Examples include bacterial streak and bunt.
  • Viruses: Wheat streak mosaic virus and Wheat dwarf virus are examples of viral diseases that can infect wheat.
  • Environmental Conditions: Stressful conditions like excessive moisture, drought, extreme temperatures, or lack of essential nutrients can weaken wheat plants, making them more susceptible to opportunistic pathogens.
• Disease Monitoring Using Images: This guide will provide valuable information on how to leverage visual inspection of wheat plants and the dataset images for disease identification. It will detail key visual characteristics (symptoms) to look for when diagnosing specific diseases in wheat.

Benefits

This dataset offers significant advantages for researchers and developers working on wheat disease classification with machine learning: - Large and Diverse: With over 14,155 high-quality images, the dataset provides a substantial foundation for training and testing machine learning models. - Real-world Images: The absence of artificial augmentation ensures the images represent natural variations of diseases, leading to improved model generalizability. - Comprehensive Disease Coverage: The inclusion of a wide range of wheat disease types allows for the development of models that can identify a broad spectrum of threats. - Monitoring Guidance: The additional information on disease causes and image-based monitoring empowers users to leverage the dataset for practical applications in real-world scenarios.

We demonstrate the outcomes of our graph-based diffusion method that employs random walk with restart on a multi-layered graph using the publicly available Pl@ntleaves (H. Go¨eau, P. Bonnet, A. Joly, N. Boujemaa, D. Barth´el´emy, J.-F. Molino, P. Birnbaum, E. Mouysset, M. Picard, The clef 2011 plant images classification task, Vol. 1177, 2011.) dataset. This dataset consists of 233 high-resolution leaf images captured in their natural surroundings. The images present various challenges for segmentation, including shadows, varying lighting conditions, and overlapping leaves. Our algorithm focuses on leaf portions by spreading intensity scores from foreground templates to image boundaries. By applying a threshold to the saliency maps generated through the diffusion process, we obtain binary masks that separate the leaves from the backgrounds. Ground truth images are provided to visually assess the effectiveness of our algorithm's performance. Folders description: * JPEGimages: Leaf color images. * masks: The ground truth binary masks that accurately delineat the leaf regions. * foreground_template: contains the bounding boxes that localize the leaves in blue and the foreground templates in red drawn on dataset images. * saliency_maps: contains saliency maps obtained by diffusing foreground queries within a multi-layer graph. * segmentation_results : contains the segmentation results obtained after thresholding the saliency maps. *MLG_Segmentation_results: a compressed folder containing the above folders.

Leaf area is one of the fundamental variables to quantify plant growth and physiological function and is therefore widely used to characterize genotypes and their interaction with the environment. We developed a smartphone app (PSM) to estimate projected leaf area (PLA), which is a good proxy for leaf area and biomass. The core of the application comprises different classification approaches to distinguish between foreground (plants or other targets) and background. We tested our approach in two case studies with banana and Eragrostis spec. and compared PLA results to ground truth data of leaf area and fresh weight. Using the same plants we also measured PLA with our SCREENHOUSE system. The SCREENHOUSE imaging system of IBG-2 is an automated greenhouse plant phenotyping platform, equipped with an imaging station for data acquisition under controlled light conditions. It is equipped with three RGB cameras (Grasshopper 2, Point Grey Research, 5MP) that image plants from three different view angles. Support vector machine (SVM) classification of foreground and background pixels is a supervised approach based on training data sets, which generally yields very good solutions for linear- and nonlinear separable data regarding stability and accuracy and which is robust against outliers in the data. The two genotypes of banana plantlets were obtained from University of Hohenheim - Institute of Crop Science (Crop Physiology of Specialty Crops), Germany. Khai Thong Ruang KTR (Musa AAA) is a drought-sensitive desert banana from Thailand, Saba (Musa ABB) is a drought-tolerant African plantain. In total we used 52 replicates, 27 KTR and 25 Saba. In the Eragrostis experiment we used two species, i.e. 100 replicates in Eragrostis tef (teff) and 40 replicates in Eragrostis pilosa. Each plant was imaged from 4 sides adding up to 208 images in banana and 560 images in Eragrostis. Projected leaf area (averaged over 4 views) was estimated with PSM and compared against SVM-classified images that were acquired and analyzed with the SCREENHOUSE imaging system. For the destructive measurements plant leaves where weighed with a high-accuracy lab balance (XS 205, Mettler Toledo, United States) and measured with a leaf area meter (LI-3100, Licor, United States) to obtain the true leaf area destructively. The data comes with a txt-file, which describes the experiment, the data folder structure and the imaging systems. The image collection contains the smartphone and SCREENHOUSE data. The SCREENHOUSE images come with image masks that were computed with support vecor machine classificaition. The PSM projected leaf area was computed with the ExGR greenness method using a threshold of 0.03 for Banana and -0.03 for Eragrostis. The data is completed by the destructive weight and leaf area measurements.

The Sugarcane Disease Image Dataset is a valuable resource designed to address critical challenges in sugarcane agriculture. This dataset comprises a comprehensive collection of images depicting various diseases and health conditions affecting sugarcane crops. Its significance lies in its potential to revolutionize disease detection, crop management, and yield optimization in the sugarcane industry. The primary problem addressed by this dataset is the early and accurate detection and management of sugarcane diseases. Farmers often face challenges in identifying diseases promptly, which can lead to increased pesticide usage and yield losses. By providing a diverse collection of annotated images, this dataset facilitates the development of AI-driven solutions that can help detect diseases at an early stage, enabling targeted interventions and ultimately contributing to higher sugarcane crop yields and improved sustainability in sugarcane agriculture.

The Herbarium Image Segmentation Dataset originates from the MNHN (Muséum National d’Histoire Naturelle) in Paris and includes 11 diverse plant families and genera, offering a rich variety within dicotyledons. The dataset comprises 2,277 RGB images, each paired with a corresponding segmentation mask. These images cover a range of genera: Amborella (91 images), Castanea (161), Desmodium (164), Ulmus (352), Rubus (184), Litsea (199), Eugenia (219), Laurus (250), Convolvulaceae (177), Magnolia (162), and Monimiaceae (318), showcasing significant morphological diversity.This dataset was generated by removing non-plant backgrounds to enhance the clarity of plant features. It is suitable for segmentation tasks in botanical research and supports studies on plant morphology, biodiversity, and conservation. The segmented images can improve accuracy in classification tasks, particularly in identifying plant morphological traits, and are intended to facilitate research in plant science, biodiversity, and conservation.

These files accompany a schools plant imaging workshop, created by Hannah Dee and colleagues as part of EPSRC Grant EP/LO17253/1.

The aim of the workshops is to introduce school-aged pupils to the concepts of plant imaging, and to image processing by building up a set of python programs that can open and investigate properties of images. The images in question are images of plants as they grow.

Workshop one concerns the creation of such datasets and can be found here: https://docs.google.com/document/d/1tXNXWhs49XUKWZl5s51mYF4Rpl44P00eih7Ulr8CHvw/edit#

Workshop two concerns investigating light changes in such datasets, introducing the concept of a "colour space" and can be found here: https://docs.google.com/document/d/1cAqEZQ-llwgvifhOiNROsFp91AaMjjU0qJOn5F7wb5o/edit#

These files allow students to skip workshop one and use our pre-recorded timelapse sequences. Further workshops are planned and will be linked from the google docs above when completed.

Dataset used in the manuscript 'Digitally Deconstructing Leaves in 3D Using X-ray microcomputed Tomography and Machine Learning'. Please cite the paper presenting this dataset:

Citation: Théroux-Rancourt, G., M. R. Jenkins, C. R. Brodersen, A. McElrone, E. J. Forrestel, and J. M. Earles. 2020. Digitally deconstructing leaves in 3D using X-ray microcomputed tomography and machine learning. Applications in Plant Sciences 8(7): .

Description of the dataset

A 'Cabernet Sauvignon' grapevine (Vitis vinifera L.) leaf from a plant of the BOKU experimental vineyard in Tulln, Austria, was scanned using microCT at the Swiss Light Source. The original reconstructions of the scans are using the gridrec (Gridrec_reconstruction_downsized.zip) and the paganin, or phase-contrast, algortithm (Phase_contrast_reconstruction_downsized.zip). To facilitate automated segmentation, the size of the image in the x and y dimensions have been halved, so that the size of the pixels is 0.325 µm in those dimensions, but 0.1625 µm in the z (slices) dimension.

A binary image segmenting the leaf cells and the airspace for each gridrec and phase-contrast stacks are created, and both are combined together (Binary_stack_for_local_thickness.zip), a map of the local thickness is created (Local_thickness_map.zip). This map gives information on the largest diameter of the pixels labeled as cells in the binary stack.

Hand-labeled slices or ground truths were drawn on the following slices: 80, 140, 200, 260, 340, 400, 440, 540, 620, 740, 800, 860, 940, 1060, 1140, 1240, 1300, 1400, 1480, 1540, 1600, 1690, 1740, 1840 (Hand_labelled_slices.tif).

Using the hand-labeled slices and the different images, a random-forest model was trained, which allowed to automatically segment the remaining slices of the stack (Fullstack_Prediction_Example-6_training_slices-6...).

The source code for the segmentation program is available here, and the source code for the testing used in the paper is available here.

The potato crop is found at fourth as a major crop in the world, after rice, wheat, and maize. Nonetheless, in Ethiopia, the yield per unit area of potato is very low compared to other countries. There are a plethora of reasons, one of them is the disease. The major disease, which affects the major potato production area is Late Blight, according to researchers the disease caused 100% crop loss on the unimproved local cultivar and 67.1% on a susceptible variety.

Not to take early Late Blight disease management would destroy the whole farm within a few days. For decades many researchers have experimented on plant disease detection and classification using computer vision via different approaches and algorithms. Many researchers used traditional machine learning algorithms which require a handcrafted feature extraction to detect a given image. Besides, the data collected were under a laboratory setup which makes it less reliable while testing in real cultivation farms captured images of the potato.

Heterogeneous image datasets were collected from (Holeta, Ethiopia) potato farm with the help of two plant pathologists. The dataset correctly labeled with two classes as ‘Healthy’ and ‘Late Blight’, and the image has variety meaning some of the images captured with less noisy background image and others with a highly noisy environment. Under ‘Late Blight’ class 63 images were collected and under ‘Healthy’ class 363 images were collected. Finally, the prepared dataset could be used for different researches that aimed at plant disease detection and classification.

Emerging in the realm of bioinformatics, plant bioinformatics integrates computational and statistical methods to study plant genomes, transcriptomes, and proteomes. With the introduction of high-throughput sequencing technologies and other omics data, the demand for automated methods to analyze and interpret these data has increased. We propose a novel explainable gradient-based approach EG-CNN model for both omics data and hyperspectral images to predict the type of attack on plants in this study. We gathered gene expression, metabolite, and hyperspectral image data from plants afflicted with four prevalent diseases: powdery mildew, rust, leaf spot, and blight. Our proposed EG-CNN model employs a combination of these omics data to learn crucial plant disease detection characteristics. We trained our model with multiple hyperparameters, such as the learning rate, number of hidden layers, and dropout rate, and attained a test set accuracy of 95.5%. We also conducted a sensitivity analysis to determine the model’s resistance to hyperparameter variations. Our analysis revealed that our model exhibited a notable degree of resilience in the face of these variations, resulting in only marginal changes in performance. Furthermore, we conducted a comparative examination of the time efficiency of our EG-CNN model in relation to baseline models, including SVM, Random Forest, and Logistic Regression. Although our model necessitates additional time for training and validation due to its intricate architecture, it demonstrates a faster testing time per sample, offering potential advantages in real-world scenarios where speed is paramount. To gain insights into the internal representations of our EG-CNN model, we employed saliency maps for a qualitative analysis. This visualization approach allowed us to ascertain that our model effectively captures crucial aspects of plant disease, encompassing alterations in gene expression, metabolite levels, and spectral discrepancies within plant tissues. Leveraging omics data and hyperspectral images, this study underscores the potential of deep learning methods in the realm of plant disease detection. The proposed EG-CNN model exhibited impressive accuracy and displayed a remarkable degree of insensitivity to hyperparameter variations, which holds promise for future plant bioinformatics applications.

This dataset is a collection of 5,049 images of cotton leaf surfaces acquired with a hand-held microscope to develop deep learning models for leaf hairiness and assist Cotton breeders in their variety selection efforts. These images were collected from two populations (A: 3,276 images; B: 1,773 images) over the 2021-2022 season in a field located at Australian Cotton Research Institute, -30.21, 149.60, Narrabri, NSW, Australia. Populations and genotypes have been anonymized to protect germplasm Intellectual Property.

This dataset is being released together with our HairNet2 paper (Farazi et al 2024). See below for links to related Datasets and Publications.

Lineage: Plant genotypes and growth conditions: Two cotton populations called A and B, were selected for their heterogeneous leaf hairiness, with population A being generally less hairy than population B. Both populations were planted in the summer growing season of 2021-22 at ACRI. Seeds of each genotype were planted in a field on the 23rd of October 2021 at a planting density of 10-12 plants/m2 in rows spaced at 1m. Each genotype was grown in a single 13m plot.

Leaf selection and imaging Leaf samples from these plant populations were collected on the 2nd and 6th of March 2022 (at 19 weeks, first open boll stage). Leaf 3 was harvested from 10 plants per genotype, placed in a paper bag and imaged the same day using the same protocol and equipment as in Rolland, Farazi et al 2022, with the following distinctions: - for population A, two images were collected per leaf: one along the central midvein and one on the leaf blade. - for population B, one image was collected per leaf: along the central midvein. The abaxial side of leaves were imaged at a magnification of about 31x with a portable AM73915 Dino-lite Edge 3.0 (AnMo Electronics Corporation, Taiwan) microscope equipped with a RK-04F folding manual stage (AnMo Electronics Corporation, Taiwan) and connected to a digital tablet running DinoCapture 2.0 (AnMo Electronics Corporation, Taiwan). The exact angle of the mid-vein in each image was not fixed. However, either end of the mid-vein was always cut by the left and right borders of the field of view, and never by the top and bottom ones.

Visual scoring of images by human expert A human expert scored all CotLeaf-X images using arbitrary ordinal scales (0 − 5 for population A and 2 − 5.5 for population B), where higher numbers corresponded to images with more trichomes.

Together with the Agriculture University, we compiled a database of plant images and omics data. The dataset contains images of four distinct plant maladies, including powdery mildew, rust, leaf spot, and blight, as well as gene expression and metabolite data. Using a high resolution camera in a controlled environment at the facility of the Agriculture University of Peshawar, we captured 8,000 images of plants, with 2,000 images for each disease type. Each image was labeled with the disease type corresponding to it. The images were preprocessed by resizing them to 224x224 pixels and standardizing the pixel values. The dataset was divided into 70:15:15 training, validation, and testing sets, correspondingly. In addition to collecting images of the same plants, we also collected gene expression and metabolite data. We extracted RNA from the plant leaves using a commercial reagent and sequenced it on an Illumina HiSeq 4000 platform. The average length of the 100 million paired-end readings obtained was 150 base pairs. The unprocessed reads were trimmed with Trimmomatic and aligned with STAR against the reference genome. We counted the number of reads that mapped to each gene using featureCounts, and then identified differentially expressed genes between healthy and diseased plants using the DESeq2 package in R. Using gas chromatography-mass spectrometry (GC-MS), we gathered additional metabolite information. Using a methanol-water extraction protocol, we extracted metabolites from the plant leaves and analyzed the extracts using GC-MS. We obtained 500 metabolite characteristics, including amino acids, organic acids, and sugars.If you use the dataset mentioned here, please make sure to give credit to the researchers by citing their paper titled 'Deep Learning for Plant Bioinformatics: An Explainable Gradient-Based Approach for Disease Detection.'ReferenceShoaib, M., Shah, B., Sayed, N., Ali, F., Ullah, R., & Hussain, I. (2023). Deep learning for plant bioinformatics: an explainable gradient-based approach for disease detection. Frontiers in Plant Science, 14(October), 1–17. https://doi.org/10.3389/fpls.2023.1283235