- Open Access
Hyperspectral band selection using genetic algorithm and support vector machines for early identification of charcoal rot disease in soybean stems
- Koushik Nagasubramanian†1,
- Sarah Jones†2,
- Soumik Sarkar3, 4,
- Asheesh K. Singh2, 4,
- Arti Singh2Email author and
- Baskar Ganapathysubramanian1, 3, 4Email author
© The Author(s) 2018
- Received: 22 November 2017
- Accepted: 16 September 2018
- Published: 3 October 2018
Charcoal rot is a fungal disease that thrives in warm dry conditions and affects the yield of soybeans and other important agronomic crops worldwide. There is a need for robust, automatic and consistent early detection and quantification of disease symptoms which are important in breeding programs for the development of improved cultivars and in crop production for the implementation of disease control measures for yield protection. Current methods of plant disease phenotyping are predominantly visual and hence are slow and prone to human error and variation. There has been increasing interest in hyperspectral imaging applications for early detection of disease symptoms. However, the high dimensionality of hyperspectral data makes it very important to have an efficient analysis pipeline in place for the identification of disease so that effective crop management decisions can be made. The focus of this work is to determine the minimal number of most effective hyperspectral wavebands that can distinguish between healthy and diseased soybean stem specimens early on in the growing season for proper management of the disease. 111 hyperspectral data cubes representing healthy and infected stems were captured at 3, 6, 9, 12, and 15 days after inoculation. We utilized inoculated and control specimens from 4 different genotypes. Each hyperspectral image was captured at 240 different wavelengths in the range of 383–1032 nm. We formulated the identification of best waveband combination from 240 wavebands as an optimization problem. We used a combination of genetic algorithm as an optimizer and support vector machines as a classifier for the identification of maximally-effective waveband combination.
A binary classification between healthy and infected soybean stem samples using the selected six waveband combination (475.56, 548.91, 652.14, 516.31, 720.05, 915.64 nm) obtained a classification accuracy of 97% for the infected class. Furthermore, we achieved a classification accuracy of 90.91% for test samples from 3 days after inoculation using the selected six waveband combination.
The results demonstrated that these carefully-chosen wavebands are more informative than RGB images alone and enable early identification of charcoal rot infection in soybean. The selected wavebands could be used in a multispectral camera for remote identification of charcoal rot infection in soybean.
- Charcoal rot
- Soybean disease
- Precision agriculture
- Band selection
- Genetic algorithm
- Support vector machines
Soybean [Glycine max (L.) Merr.] is the major oilseed crop grown in the United States . Soybean is also economically important as it is the second major crop overall produced by the United States . Soybean is used to produce biofuel, cooking oil, soy foods, and animal feed, among many other uses, but the crop is threatened by over 100 diseases with 35 believed to be important pathogens affecting soybean yield [2, 3].
Charcoal rot is an economically critical disease that affects soybean, as well as 500 other plant species worldwide, and is caused by the fungal pathogen Macrophomina phaseolina (Tassi) Goid [4–6]. Infection is favored by warm (30–35 °C), dry, drought-like conditions but can cause up to 50% yield loss even in irrigated environments [7–10]. Charcoal rot earned its common name from the gray-silver discoloration caused by microsclerotia formation in the vascular tissue and pith of lower stems and roots of infected plants [7, 11]. These microsclerotia are small dark survival structures that persist in the soil and plant debris after harvest and can act as an inoculum source for charcoal rot infection during the next growing season [3, 7, 12]. Symptoms generally become visible at the R5–R7 reproductive stages, or from early seed to early maturity, but can occasionally be seen earlier as reddish-brown lesions on the hypocotyl of seedlings [3, 7]. In more mature infected plants, a reddish-brown discoloration of the vascular tissue in the roots and lower stem generally precedes foliar symptom development . Following internal discoloration, diseased plants may yellow, then wilt, and prematurely senesce leaving dead leaves and petioles still attached to the stem [3, 7, 13]. Black microsclerotia on the above ground plant are first visible at the stem nodes and can be seen in the epidermal and sub epidermal tissue of plant stems as well as scattered on dry pods and seed of more mature plants [3, 7]. Management of charcoal rot has proven to be difficult as no fungicides are available for control and more work needs to be done to research the potential of seed treatments [3, 12]. In addition, crop rotation may not be a viable strategy to manage infection, because charcoal rot infects the United States’ major crops including corn, cotton, and sorghum [14, 15]. Furthermore, no commercial soybean varieties are considered resistant, though a few varieties demonstrate moderate resistance [8, 13, 16–19]. However, a genome wide association (GWA) study across both field and greenhouse environments recently reported a total of 19 single nucleotide polymorphisms (SNPs) associated with charcoal rot resistance in soybean . While over 800 soybean lines have been evaluated for charcoal rot resistance, identification of resistant genotypes has been limited due to a need for an accurate, rapid, and consistent method for disease assessment and classification [12, 13].
Current state of disease assessment and outlook
Multiple methods, which are predominantly visual, have been proposed for assessing charcoal rot severity of soybean plant canopies, roots, and stems in the field and indoor environments. These methods include evaluation of the intensity or length of stem and root discoloration caused by microsclerotia formation, evaluation of the percent chlorosis and necrosis of the plant canopy throughout the growing season, chlorosis and necrosis of foliage that remains attached to the plant at R7, calculation of colony forming unit index to quantify the microsclerotia content in the stem and root, and lesion length measurements of cut-stem inoculations on young plants [13, 19, 21, 22]. However, visual rating methods can be subjective and are susceptible to human error caused by rater ability, and inter/intra-rater reliability [23–28].
Furthermore, visual ratings only take advantage of visible wavelengths of the electromagnetic spectrum . Hyperspectral imaging can capture both spectral and spatial information from a wider range of the electromagnetic spectrum including the visible and near-infrared regions . Automating disease severity rating through hyperspectral imaging offers a potential solution to the standardization and reliability issues in current visual rating systems. While some hyperspectral systems do not incorporate imaging, but rather average all spectra obtained from a given area, the imaging aspect inherent in hyperspectral imaging techniques comparing to non-imaging hyperspectral systems offers many benefits for studying plant disease symptoms . Extraction of reflectance spectra from each pixel, enables one to relate changes in reflectance values to disease symptoms [31, 32]. Recent plant pathology and phenotyping studies have utilized hyperspectral imaging data to study the effect of different plant pathogens. Examples include approaches to identify differences in the reflectance patterns of resistant and susceptible barley genotypes inoculated with powdery mildew [30, 33] the content of charcoal rot (M. phaseolina) microsclerotia in ground root and stem tissue as a method for rating infection severity , and hyperspectral imaging to distinguish between the symptoms of Cercospora leaf spot, powdery mildew, and leaf rust at different developmental stages in sugar beet .
A key issue with utilizing hyperspectral imaging is that the resulting hyperspectral data cubes, or the 3-dimensional output of hyperspectral imaging comprised of 2 spatial dimensions and 1 wavelength dimension, are high dimensional and contain redundant information which reduces the ability to distinguish between different object classes in classification problem. . Using a hyperspectral camera on a drone for crop disease identification and phenotyping can also generate large quantities of data during the flight making it necessary to have a large on-board storage capacity and also substantially increases computational cost for any subsequent analysis. Therefore, there is a need to develop an analysis pipeline to reduce dimensionality of the data and to select optimal wavelengths that are most useful for phenotyping and disease identification. This serves as the motivation of this study.
Feature extraction and feature selection are two different methods for dimensionality reduction of hyperspectral data. Feature extraction methods such as Principal Component Analysis (PCA), Linear Discriminant Analysis (LDA), Independent Component Analysis (ICA) and Maximum Noise Fraction (MNF) project the original hyperspectral data into a new low-dimensional data by reducing the spectral dimension [36–39]. Feature extraction methods alter the physical meaning of the hyperspectral data during transformation to a new (and lower) dimensional space whereas feature selection methods preserve the original features . Feature selection essentially boils down to carefully selecting a subset of the available wavebands (i.e. waveband selection) that preserves certain traits of the full dataset . Feature selection methods are broadly classified into supervised or unsupervised methods . Supervised methods use input and desired output variables for training an algorithm whereas unsupervised methods use only the input data for training . Some supervised waveband selection methods use class separability metrics like Euclidean distance, transformed divergence, Bhattacharyya distance, Jeffreys–Matusita (JM) distance [44, 45]. A waveband selection method based on estimation of mutual information for classification of hyperspectral images was proposed by Guo et al. . Sequential search strategies like Sequential Forward Selection (SFS), Sequential Floating Forward Selection (SFSS), Sequential Backward Selection (SBS) and Sequential Backward Floating selection (SBSS) have also been used for waveband selection [47, 48]. These sequential search algorithms are simple and suboptimal. Evolutionary methods such as Particle Swarm Optimization (PSO) and genetic algorithms (GA) which can search for global optimal solutions have been found to be successful in effective waveband selection [49, 50]. In this study, we use an evolutionary method, specifically GA, as an optimizer along with Support Vector Machine (SVM)  as a classifier for effective waveband selection. GA-SVM based model have been successful in waveband selection for classification of remotely sensed hyperspectral images [49, 52–55]. Although computationally costly, evolutionary algorithms can give better optimal solution than sequential algorithms since the best feature combination is selected simultaneously .
The objectives of this study were (1) hyperspectral imaging enabled early identification of charcoal rot disease and (2) to determine the most effective minimum number of wavebands for discrimination of healthy and charcoal rot infected stems. This study shows that a genetic algorithm-support vector machine based model can be used in selecting the most effective waveband combination for early detection of charcoal rot disease in soybeans. Additionally, using F1-Score as an optimization metric instead of classification accuracy can overcome the skewness of classification accuracy metric for the dominant class of an imbalanced dataset (number of healthy samples more than the number of infected samples) .
Four soybean genotypes, Pharaoh (susceptible), PI479719 (susceptible), DT97-4290 (moderately resistant), and PI189958 (moderately resistant) were included in this study. Two seed of each genotype were planted in a commercial soil substrate (Sungro horticulture professional growing mix) in 8 oz styrofoam cups in a growth chamber at 30 °C day/21 °C night with a 16-h photoperiod. Each styrofoam cup was supplemented with 1/8tsp (0.65 g) of osmocote 15-9-12 at planting. Ten days after planting, plants were thinned down to one plant per pot choosing the most vigorous plant. Plants were arranged in a randomized complete block design with four replications. The two treatments were inoculation and mock-inoculation. Data collection was completed within 15 days after inoculation (DAI). Replication 1 was planted in the growth chamber in September 2016. Lesion lengths and hyperspectral images were collected at 3, 6, 12, and 15 DAI to study the earlier and then later time points post inoculation. Replications 2–4 were planted together in November 2016. Lesion length ratings and data cubes were collected at 3, 6 and 9DAI in replications 2–4 focusing on the earlier disease development time points.
Culture and inoculation of M. phaseolina
The pathogen M. phaseolina 2013X, originally collected from the field in Iowa in 2013, was re-isolated from inoculated stems of soybean plants grown in the growth chamber. Inoculation was performed 3 weeks (21 days) after planting of seeds. In order to prepare for inoculation, cultures of M. phaseolina were started in the lab, 17 days after planting (i.e. 4 days before inoculation). This culture preparation consisted of transferring 0.5 cm plugs of M. phaseolina to Potato Dextrose Agar (PDA) plates which were then stored in the dark at 30 °C for 4 days. Twenty-one days after planting, the four soybean genotypes were inoculated according to the cut-stem inoculation technique . Sterile 200 µl pipette tips were placed open end down into the media around the leading edge of the fungal colony cutting a small disk of media and fungal hyphae from the plate. Each soybean stem was severed exactly 40 mm above the unifoliate node. A pipette tip was removed from the culture plate ensuring that it carried a disk of PDA media + M. phaseolina mycelia for the inoculation treatment. The pipette tip was pushed onto the cut stem, like a hat, and the open wound imbedded in the media. The same protocol was carried out for the mock-inoculation treatment using uncontaminated plates of PDA media. Three days after inoculation, pipette tips were removed from all plants.
Hyperspectral image acquisition
Pika XC hyperspectral line scanning imager (Resonon, Bozeman, MT) was used to construct hyperspectral data cubes of soybean stems. The Pika XC imager has a spectral resolution of 2.5 nm, with 240 spectral channels covering a spectral range from 382 to 1032 nm. Hyperspectral images of healthy and charcoal rot infected stems were collected at different time points, as explained previously, for classification.
The imaging system also includes a mounting tower, linear translation stage, and a computer pre-loaded with SpectrononPro software (Resonon, Bozeman, MT). Illumination was provided by two 70-watt quartz-tungsten-halogen Illuminator lamps (ASD Inc., Boulder, CO) which provide stable illumination over a 350–2500 nm range. The distance between the lamps and the plant stem being imaged was 54 cm with lights pointed towards the sample at a 45-degree angle. Prior to imaging, the ASD pro-lamps were turned on and warmed up for at least 20 min to produce a stable light source.
Plant stems were destructively imaged at different time points after inoculation (3, 6, 9, 12 and 15 DAI). All leaves were removed from the plant stem and the stem severed at the soil surface immediately prior to hyperspectral data cube collection. Stems were placed on the linear translation stage for imaging. Growth patterns of stem lesions often resulted in irregular lesion boundaries. So, stems were positioned on the linear translation stage so that the longest edge of the lesion was facing the camera lens. Following calibration, a data cube was collected from each stem. The hyperspectral data cubes and corresponding RGB images were saved on an external hard drive.
Charcoal rot rating protocol
Genetic algorithm-support vector machine based feature selection
Support vector machine
Support Vector Machine (SVM) is a kernel-based discriminative supervised learning algorithm for classification [51, 58]. SVM is one approach for constructing a classifier that maps an input data (of N waveband information) to a class (healthy vs infected). SVM has been used with significant success in identification of variety of plant stresses [43, 59]. Formally, SVM projects data which are not separable linearly into a higher dimensional space using a kernel and separates the classes with an optimal hyperplane that maximizes the margin between the classes . In this study, we used Radial Basis Function (RBF)  kernel to learn the non-linear classifier. SVM has been used as a classifier in wrapper based feature selection methods for classification of hyperspectral images [49, 52, 54, 55, 62–65]. After trial and error, the two Radial Basis Function (RBF) kernel parameters C and γ were set to 1000 and 1, respectively.
Genetic algorithms are population based stochastic search optimization techniques inspired by natural selection and natural genetics principles . The population of candidate solutions (i.e. wavebands) is represented as a long string of bits and is called ‘chromosome’. Each of these chromosomes is assigned a score using a fitness function for evaluation . In this case, the fitness function evaluates how well the chromosome (i.e. that particular selection of wavebands) performs to distinguish between diseased and healthy specimens. These chromosomes are evolved in successive generations using selection, mutation and crossover genetic operators for exploring the solution space until a best solution is obtained, or termination criteria is encountered. Selection of chromosomes for reproduction can be done in diverse ways . One of the ways is to choose the pair of chromosomes in the population that provides relatively good fitness scores to perform crossover. Crossover operator randomly combines genetic information of two chromosomes. Mutation operator modifies some component of a chromosome to form random new populations in the search space which prevents GA from choosing local optimal solutions. The “elite” is a GA hyperparameter decides the number of most-fit individuals passed from one generation to the next generation without changing. This process of selection, mutation and crossover is repeated for multiple generations to improve the population fitness  (Fig. 3).
Confusion matrix definition
True Positive (TP)
False Negative (FN)
False Positive (FP)
True Negative (TN)
The termination criteria depend on the average change in fitness value for 50 continuous generations or the maximum number of generations allowed which were 100 in our study. The last generation of GA iteration will contain the most optimal solution.
Implementation details of genetic algorithm
Number of genetic algorithm iterations
Maximum number of generations
Binary selection tournament
Average change in best fitness value is less than 10−6 for 50 generations or number of generations = 100
Mean and standard error of the mean for lesion length
Number of samples
Standard error mean
Exterior lesion length
Interior lesion length
Dead lesion length
Confusion matrix of test samples from 3, 6, 9, 12 and 15 DAI
TP = 17
FP = 8
FN = 1
TN = 13
TP = 18
FP = 1
FN = 0
TN = 20
Classification results of test samples from 3, 6, 9, 12 and 15 DAI
Overall accuracy (%)
The RGB wavelengths alone did not perform well, which might be because of their inability to differentiate between the reflectance values of a healthy stem and charcoal rot infected stem. The classification accuracy and F1 score of the selected 6 waveband combinations indicate that they were good at distinguishing between healthy and charcoal rot infected samples.
Early disease detection for 3-DAI samples
Classification results for 3-DAI samples
Overall accuracy (%)
TP = 5
FP = 2
FN = 0
TN = 4
TP = 5
FP = 1
FN = 0
TN = 5
Disease length prediction
Hyperspectral images of four different soybean genotypes (two susceptible and two moderately resistant), half healthy and half infected with charcoal rot disease were collected at 5 different time points post infection. The main objectives of this study were to identify the most effective minimal number of wavebands from a set of 240 hyperspectral wavebands that are required for identification of charcoal rot disease and to analyze the performance of these wavebands in early detection of the disease.
The study used both spectral and spatial information (mean value of reflectance from different wavelengths) for disease identification. Due to imbalanced dataset of healthy and infected stems used in our study, the SVM classification performance which was optimized using GA for optimal waveband selection was evaluated for maximizing the F1 score value of the infected class instead of overall classification accuracy.
An effective six waveband combination for discrimination of healthy and charcoal rot infected stems was found. Early identification of charcoal rot disease at 3 days after inoculation was possible using the selected waveband combinations. The GA-SVM model obtained F1-score of 0.97 and classification accuracy of 97% using selected 6 hyperspectral waveband combinations for complete test data (samples from 3, 6, 9, 12 and 15 DAI). These results were 26.1% better than those obtained using only the visible RGB wavelengths highlighting the importance of including the additional non-visible wavelengths for disease detection. The F1-score and classification accuracy for early detection (3-DAI samples) samples were 0.90 and 90.91% respectively using the selected 6 wavebands. Two out of the three wavelengths selected (720.05 nm, 915.64 nm) along with the RGB wavebands in the six waveband combinations were selected in the near-infrared region and one was selected in the visible region (516.31 nm) indicating that both near infrared region and visible region were useful in early identification of charcoal rot disease. This relationship between the stem reflectances and charcoal rot disease is along the lines of the results of a previous study . Genotypes with susceptible and moderately resistant responses to charcoal rot were used in this study. The length of disease progression (mm) in each stem was measured to understand the severity of the disease spread among different genotypes. Using hyperspectral imaging combined with GA-SVM enabled waveband selection resulting in a higher classification accuracy compared to visible wavelengths alone. However, this study focused on indoor imaging so future work should utilize field inoculations and evaluations to expand this technology into the field. Furthermore, field inoculations of diverse soybean genotypes will be imaged using a multispectral camera with the selected wavebands from the GA-SVM model for early identification of charcoal rot disease to understand the disease resistance of specific genotypes. Also, the length of disease progression in different genotypes will be studied with larger sample size to characterize their disease resistance. In conclusion, this study provides an efficient methodology for selecting the most effective wavebands from hyperspectral data to be used for early disease detection of charcoal rot in soybean stems.
BG, AS, SS and AKS formulated research problem and designed approaches. SJ and AS collected data. KN, SJ, BG and SS developed processing workflow and performed data analytics. All authors contributed to the writing and development of the manuscript. All authors read and approved the final manuscript.
We thank Jae Brungardt, Brian Scott, and Hsiang Sing Naik for support during experimentation.
The authors declare that they have no competing interests.
Availability of data and materials
The datasets used and/or analyzed during the current study available from the corresponding author on reasonable request.
This work was funded by Iowa Soybean Association (AS), USDA National Institute of Food and Agriculture (NIFA) - Grant# 2017-67007-26151 (SS, BG, AS, AKS), ISU Research grant through the PIIR award (AS, AKS, SS, BG), R F Baker Center for Plant Breeding (AKS), Monsanto Chair in Soybean Breeding at Iowa State University (AKS) and PSI Faculty Fellow award (BG, SS, AKS).
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.
- USDA NASS. Acreage [Internet]. 2016. http://www.usda.gov/nass/PUBS/TODAYRPT/acrg0616.pdf. Accessed 4 July 2017.
- Boerema A, Peeters A, Swolfs S, Vandevenne F, Jacobs S, Staes J, et al. Soybean trade: balancing environmental and socio- economic impacts of an intercontinental market.(Report). PLoS ONE. 2016;11:e0155222.View ArticleGoogle Scholar
- Hartman GL, Rupe JC, Sikora EF, Domier LL, Davis JA, Steffey KL. Compendium of soybean diseases and pests. Am Phytopath Soc. 2015;5:67–9.Google Scholar
- Doupnik B Jr. Soybean production and disease loss estimates for north central United States from 1989 to 1991. Plant Dis. 1993;77:1170–1.Google Scholar
- Hartman GL, Chang HX, Leandro LF. Research advances and management of soybean sudden death syndrome. Crop Prot. 2015;73:60–6.View ArticleGoogle Scholar
- Koenning SR, Wrather JA. Suppression of soybean yield potential in the continental United States by plant diseases from 2006 to 2009 [Internet]. Online. Plant Health Progress; 2010. http://www.plantmanagementnetwork.org/pub/php/research/2010/yield/. Accessed 1 Sept 2016.
- Gupta GKG, Sharma SSK, Ramteke R. Biology, epidemiology and management of the pathogenic fungus Macrophomina phaseolina (Tassi) goid with special reference to charcoal rot of soybean (Glycine max (L.) Merrill). J Phytopathol. 2012;160:167–80.View ArticleGoogle Scholar
- Mengistu A, Arelli PA, Bond JP, Shannon GJ, Wrather AJ, Rupe JB, et al. Evaluation of soybean genotypes for resistance to charcoal rot. Plant Health Prog. 2011;10:1–26.Google Scholar
- Meyer WA. Factors affecting charcoal rot of soybean seedlings. Phytopathology. 1974;64:845.View ArticleGoogle Scholar
- Wyllie TD, Scott DH. Soybean diseases of the north central region. USA: APS Press; 1988.Google Scholar
- Wrather JA, Shannon JG, Carter TE, Bond JP, Rupe JC, Almeida AMR. Reaction of drought-tolerant soybean genotypes to Macrophomina phaseolina. Plant Health Prog. 2008. https://doi.org/10.1094/PHP-2008-0618-01-RS.View ArticleGoogle Scholar
- Romero Luna MP, Mueller D, Mengistu A, Singh AK, Hartman GL, Wise KA. Advancing our understanding of charcoal rot in soybeans. J Integr Pest Manag. 2017;8(1). Google Scholar
- Mengistu A, Ray JD, Smith JR, Paris RL. Charcoal rot disease assessment of soybean genotypes using a colony-forming unit index. Crop Sci. 2007;47:2453–61.View ArticleGoogle Scholar
- Short GE. Survival of Macrophomina phaseolina in soil and in residue of soybean. Phytopathology. 1980;70:13.View ArticleGoogle Scholar
- Su G, Suh SO, Schneider RW, Russin JS. Host specialization in the charcoal rot fungus, Macrophomina phaseolina. Phytopathology. 2001;91:120.View ArticleGoogle Scholar
- Mengistu A, Arelli P, Bond J, Nelson R, Rupe J, Shannon G, et al. Identification of soybean accessions resistant to Macrophomina phaseolina by field screening and laboratory validation. Plant Health Prog. 2013;14(1):25.View ArticleGoogle Scholar
- Paris RL. Registration of soybean germplasm line DT97-4290 with moderate resistance to charcoal rot. Crop Sci. 2006;46:2324–5.View ArticleGoogle Scholar
- Pawlowski ML, Hill CB, Hartman GL. Resistance to charcoal rot identified in ancestral soybean germplasm. Crop Sci. 2015;55(3):1230–5.View ArticleGoogle Scholar
- Smith G, Carvil O. Field screening of commercial and experimental soybean cultivars for their reaction to Macrophomina phaseolina. Plant Dis. 1997;81:363–8.View ArticleGoogle Scholar
- Coser SM, Chowda Reddy RV, Zhang J, Mueller DS, Mengistu A, Wise KA, et al. Genetic architecture of charcoal rot (Macrophomina phaseolina) Resistance in Soybean revealed using a diverse panel. Front Plant Sci. 2017;8:1626.View ArticleGoogle Scholar
- Barratt RW, Horsfall JG. An improved grading system for measuring plant disease. Phytopathology. 1945;35:655.Google Scholar
- Twizeyimana M, Hill CB, Pawlowski M, Paul C, Hartman GL. A cut-stem inoculation technique to evaluate soybean for resistance to Macrophomina phaseolina. Plant Dis. 2012;96:1210–5.View ArticleGoogle Scholar
- Bock CH, Nutter Jr FW. Detection and measurement of plant disease symptoms using visible-wavelength photography and image analysis. Plant Sci Rev. 2012; p. 73Google Scholar
- Bock CH, Poole GH, Parker PE, Gottwald TR. Plant disease severity estimated visually, by digital photography and image analysis, and by hyperspectral imaging. CRC Crit Rev Plant Sci. 2010;29:59–107.View ArticleGoogle Scholar
- James WC. Assessment of plant diseases and losses. Annu Rev Phytopathol. 1974;12:27–48.View ArticleGoogle Scholar
- Nutter JFW. Assessing the accuracy, intra- rater repeatability, and inter- rater reliability of disease assessment systems. Phytopathology. 1993;83:806.View ArticleGoogle Scholar
- Ghosal S, Blystone D, Singh AK, Ganapathysubramanian B, Singh A, Sarkar S. An explainable deep machine vision framework for plant stress phenotyping. PNAS. 2018;115(18):4613–8.PubMedGoogle Scholar
- Singh AK, Ganapathysubramanian B, Sarkar S, Singh A. Deep learning for plant stress phenotyping: trends and future perspectives. Trends Plant Sci. 2018;. https://doi.org/10.1016/j.tplants.2018.07.004.View ArticlePubMedGoogle Scholar
- Mahlein A. Plant disease detection by imaging sensors-parallels and specific demands for precision agriculture and plant phenotyping. Plant Dis. 2016;100:241–51.View ArticleGoogle Scholar
- Mirwaes W, Anne-Katrin M, Christian B, Ulrike S, Erich-Christian O, Kristian K. Plant phenotyping using probabilistic topic models: uncovering the hyperspectral language of plants. Sci Rep. 2016;6:22482.View ArticleGoogle Scholar
- Mahlein A-K, Oerke E-C, Steiner U, Dehne H-W. Recent advances in sensing plant diseases for precision crop protection. Eur J Plant Pathol. 2012;133:197–209.View ArticleGoogle Scholar
- Mahlein A-K, Steiner U, Hillnhütter C, Dehne H-W, Oerke E-C. Hyperspectral imaging for small- scale analysis of symptoms caused by different sugar beet diseases. Plant Methods. 2012;8:3.View ArticleGoogle Scholar
- Kuska M, Wahabzada M, Leucker M, Dehne H-W, Kersting K, Oerke E-C, et al. Hyperspectral phenotyping on the microscopic scale: towards automated characterization of plant-pathogen interactions. Plant Methods. 2015;11:28.View ArticleGoogle Scholar
- Fletcher RS, Smith JR, Mengistu A, Ray JD. Relationships between microsclerotia content and hyperspectral reflectance data in soybean tissue infected by Macrophomina phaseolina. Am J Plant Sci. 2014;5:3737–44.View ArticleGoogle Scholar
- Su H, Du Q, Chen G, Du P. Optimized hyperspectral band selection using particle swarm optimization. IEEE J Sel Top Appl Earth Obs Remote Sens. 2014;7:2659–70.View ArticleGoogle Scholar
- Bandos TVV, Bruzzone L, Camps-Valls G. Classification of hyperspectral images with regularized linear discriminant analysis. IEEE Trans Geosci Remote Sens. 2009;47:862–73.View ArticleGoogle Scholar
- Liu X, Zhang B, Gao LR, Chen DM. A maximum noise fraction transform with improved noise estimation for hyperspectral images. Sci China Ser F Inf Sci. 2009;52:1578–87.View ArticleGoogle Scholar
- Tyo JS, Konsolakis A, Diersen DI, Olsen RC. Principal-components-based display strategy for spectral imagery. IEEE Trans Geosci Remote Sens. 2003;41:708–18.View ArticleGoogle Scholar
- Villegas-Fernández AM, Sillero JC, Emeran AA, Flores F, Rubiales D. Multiple-disease resistance in Vicia faba: multi-environment field testing for identification of combined resistance to rust and chocolate spot. Field Crops Res. 2011;124:59–65.View ArticleGoogle Scholar
- Yang C, Lee WS, Gader P. Hyperspectral band selection for detecting different blueberry fruit maturity stages. Comput Electron Agric. 2014;109:23–31. https://doi.org/10.1016/j.compag.2014.08.009.View ArticleGoogle Scholar
- Dash’ M, Liu H. Feature selection for classification. Intell Data Anal. 1997;1:131–56.View ArticleGoogle Scholar
- Guyon I, Elisseeff A. An introduction to variable and feature selection. J Mach Learn Res. 2003;3:1157–82.Google Scholar
- Singh A, Ganapathysubramanian B, Singh AK, Sarkar S. Machine learning for high-throughput stress phenotyping in plants. Trends Plant Sci. 2016;21:110.View ArticleGoogle Scholar
- Keshava N. Distance metrics and band selection in hyperspectral processing with applications to material identification and spectral libraries. IEEE Trans Geosci Remote Sens. 2004;42(7):1552–65.View ArticleGoogle Scholar
- Yang H, Du Q, Su H, Sheng Y. An efficient method for supervised hyperspectral band selection. IEEE Geosci Remote Sens Lett. 2011;8:138–42.View ArticleGoogle Scholar
- Guo B, Gunn SSR, Damper RI, Nelson JDB. Band selection for hyperspectral image classification using mutual information. Geosci Remote Sens Lett IEEE. 2006;3:522–6.View ArticleGoogle Scholar
- Pudil P, Novovičová J, Kittler J. Floating search methods in feature selection. Pattern Recognit Lett. 1994;15:1119–25.View ArticleGoogle Scholar
- Serpico SB, Bruzzone L. A new search algorithm for feature selection in hyperspectral remote sensing images. IEEE Trans Geosci Remote Sens. 2001;39:1360–7.View ArticleGoogle Scholar
- Li S, Wu H, Wan D, Zhu J. An effective feature selection method for hyperspectral image classification based on genetic algorithm and support vector machine. Knowl Based Syst. 2011;24:40–8. https://doi.org/10.1016/j.knosys.2010.07.003.View ArticleGoogle Scholar
- Yang H, Zhang S, Deng K, Du P. Research into a feature selection method for hyperspectral imagery using PSO and SVM. J China Univ Min Technol. 2007;17:473–8.View ArticleGoogle Scholar
- Cortes C, Vapnik V. Support-vector networks. Mach Learn. 1995;20:273–97.Google Scholar
- Bazi Y, Melgani F. Toward an optimal SVM classification system for hyperspectral remote sensing images. IEEE Trans Geosci Remote Sens. 2006;44:3374–85.View ArticleGoogle Scholar
- Chen Y, Zhao X, Lin Z. Optimizing subspace SVM ensemble for hyperspectral imagery classification. IEEE J Sel Top Appl Earth Obs Remote Sens. 2014;7:1295–305.View ArticleGoogle Scholar
- Pal M. Support vector machine- based feature selection for land cover classification: a case study with DAIS hyperspectral data. Int J Remote Sens. 2006;27:2877–94.View ArticleGoogle Scholar
- Zhuo L, Zheng J, Wang F, Li X, Ai B, Qian J. A genetic algorithm based wrapper feature selection method for classification of hyperspectral images using support vector machine. Int Arch Photogramm Remote Sens Spat Inf Sci. 2008;37:397–402.Google Scholar
- Simon D. Evolutionary optimization algorithms. Hoboken: Wiley; 2013.Google Scholar
- He H, Garcia EA. Learning from imbalanced data. IEEE Trans Knowl Data Eng. 2009;21:1263–84.View ArticleGoogle Scholar
- Burges CCJCJCJC. A tutorial on support vector machines for pattern recognition. Data Min Knowl Discov. 1998;2:121–67.View ArticleGoogle Scholar
- Naik HS, Zhang J, Lofquist A, Assefa T, Sarkar S, Ackerman D, et al. A real-time phenotyping framework using machine learning for plant stress severity rating in soybean. Plant Methods. 2017;13(1):1–23.View ArticleGoogle Scholar
- Boser BE, Guyon IM, Vapnik VN. A training algorithm for optimal margin classifiers. In: Proceedings of the fifth annual workshop on Computational learning theory [Internet]. ACM; 1992. p. 144–152. http://portal.acm.org/citation.cfm?doid=130385.130401. Accessed 1 Sept 2016.
- Buhmann MD. Radial basis functions. Acta Numer. 2000;9:1–38.View ArticleGoogle Scholar
- Bor-Chen K, Hsin-Hua H, Cheng-Hsuan L, Chih-Cheng H, Jin-Shiuh T. A kernel-based feature selection method for SVM with RBF kernel for hyperspectral image classification. IEEE J Sel Top Appl Earth Obs Remote Sens. 2014;7:317–26.View ArticleGoogle Scholar
- Samadzadegan F, Hasani H, Schenk T. Simultaneous feature selection and SVM parameter determination in classification of hyperspectral imagery using ant colony optimization. Can J Remote Sens. 2012;38:139–56.View ArticleGoogle Scholar
- Vaiphasa C, Skidmore AK, de Boer WF, Vaiphasa T. A hyperspectral band selector for plant species discrimination. ISPRS J. Photogramm. Remote Sens. 2007;62:225–35.View ArticleGoogle Scholar
- Covert SF, Aoki T, O’Donnell K, Starkey D, Holliday A, Geiser DM, et al. Sexual reproduction in the soybean sudden death syndrome pathogen Fusarium tucumaniae. Fungal Genet Biol. 2007;44:799–807.View ArticleGoogle Scholar
- Goldberg DE, Holland JH. Genetic algorithms and machine learning. Mach Learn. 1988;3:95–9.View ArticleGoogle Scholar
- Holland JH, Goldberg D. Genetic algorithms in search, optimization and machine learning. Massachusetts: Addison-Wesley; 1989.Google Scholar
- Goldberg DE, Deb K. A Comparative Analysis of Selection Schemes Used in Genetic Algorithms. Found. Genet. Algorithms. 1991;1:69–93.Google Scholar
- Mohanty SP, Hughes D, Salathe M. Inference of plant diseases from leaf images through deep learning. 2016;1–6. arXiv:1604.03169 [cs].
- Amara J, Bouaziz B, Algergawy A. A deep learning-based approach for banana leaf diseases classification. BTW Workshops. 2017; p. 79–88.Google Scholar
- Forman G. An extensive empirical study of feature selection metrics for text classification. J Mach Learn Res. 2003;3:1289–305.Google Scholar
- Powers DMW. Evaluation: from precision, recall and f-measure to roc, informedness, markedness and correlation. J Mach Learn Technol. 2011;2:37–63.Google Scholar
- Deep K, Thakur M. A new crossover operator for real coded genetic algorithms. Appl Math Comput. 2007;188:895–911.Google Scholar
- Deep K, Thakur M. A new mutation operator for real coded genetic algorithms. Appl Math Comput. 2007;193:211–30.Google Scholar
- Krizhevsky A, Sutskever I, Hinton GE. Imagenet classification with deep convolutional neural networks. In: Advances in neural information processing systems. 2012; p. 1097–105.Google Scholar