Original papersIntegrating spectral and image data to detect Fusarium head blight of wheat
Introduction
Wheat (Triticum aestivum L.) is one of the world’s leading food crops, making up a high percentage of calories consumed by the world’s population. Fusarium head blight (FHB), caused by the fungus Gibberella zeae, is a major disease in wheat worldwide. The disease can cause significant yield and quality reductions in the form of atrophy, weight reduction, and discoloration (Bauriegel et al., 2011). The fungus also produces mycotoxins including deoxynivalenol, nivalenol and zearalenones, which can adversely affect livestock and human health (Barbedo et al., 2015, Desjardin, 2006). Early and quick detection and monitoring of the development of the FHB disease are key to the control of this disease and the mycotoxins it produces in wheat.
Spectroscopy can potentially be a practical method for rapid detection of FHB in wheat and other plant diseases since it has many advantages in operation, including simple pretreatment and fast measurement. Hyperspectral imaging (HSI), an emerging spectra-based technology, is different from traditional computer vision technology. HSI combines spectroscopic and imaging techniques into one that can obtain a group of monochromatic images in hundreds of nearly continuous wavebands, in addition to providing both spectral and spatial information (Zhang et al., 2014). At present, the HSI technology has been widely used in crop growth detection and monitoring, especially in the areas of disease detection and crop damage in cereal crops (Singh et al., 2009, Zhang et al., 20122; Barbedo et al., 2017). Jiang et al. (2010) used hyperspectral images to identify healthy and yellow rust-infected wheat plants based on red edge and yellow edge positions. Alisaac et al. (2018) applied hyperspectral sensors to phenotype wheat varietal response to the infection of FHB and also investigated the correlations among disease severity, spike weight and spectral wavelengths. Bauriegel et al. (2011) used hyperspectral imaging technology to classify Fusarium infected and healthy ear tissues of wheat in the three spectral wavelength ranges with their results showing that application of spectral sub-ranges has a promising prospect. Furthermore, the data obtained from HSI also contains image data that can be used to identify wheat heads damaged by the Fusarium infection. Jin et al. (2018) applied a deep neural network (DNN) to the pixels of hyperspectral images to accurately discern the infected areas of wheat heads by FHB, illustrating that DNN is an excellent classification algorithm and that image information of hyperspectral data has the potential to classify healthy and diseased wheat ears. It can be seen that hyperspectral imaging system can not only provide spatial information but also provide spectral information of each pixel in the image, which enables hyperspectral technology to detect the intrinsic chemical and molecular information of the object and obtain the external features of the object (Wu and Sun, 2013). However, most of the previous studies are use only the spectral information without using the image information or vice versa. At present, there are some reports showing that integrating spectral and image information can have advantages over them being used alone to detect wheat FHB. Yang et al. (2015) classified corn seed varieties by using spectral and image features extracted from hyperspectral images and obtained an accuracy of over 96% based on support vector machines model, which was better in accuracy than spectral or image information used alone. Ropelewska and Zapotoczny (2018) studied the effect of the ventral and dorsal sides of wheat kernels on classifying Fusarium-infected samples; their results showed that due to both sides containing different information, there was a difference in the model performance between the ventral and dorsal sides. Little research has been conducted on the use of the front side (side A) and reverse side (side B) information to detect FHB disease in wheat. Thus, in the current study, we evaluated infected wheat heads using fusion features of spectral and image information from both the side A and side B of wheat heads to develop models that can detect wheat FHB. To investigate the optimal features dataset required to detect wheat FHB, the random forest model was built based on spectral, image and fusion features of both spectral and image features of wheat heads. A deep convolutional neural network (DCNN) model was finally constructed based on double-sided comprehensive information to develop a method that can accurately detect and predict the FHB disease in wheat.
Deep learning method is derived from the artificial neural network Multi-Layer Perceptron. DCNN has been developed rapidly and used widely in processing images, face detection, and so on (Delgado et al., 2013, Li et al., 2015). At present, DCNN has been introduced into hyperspectral image analysis with excellent performance. Hu et al. (2015) demonstrated that using DCNN to classify hyperspectral images in the spectral domain can achieve better performance than the traditional methods evaluated. Similarly, Li et al. (2017) developed a hyperspectral imaging reconstruction model based on DCNN to improve spatial features and concluded that DCNN-based framework can achieve desirable performance. Jin et al. (2018) applied a deep neural network to the pixels of hyperspectral images to classify healthy and diseased wheat heads in the field with excellent results on the detection of healthy and Fusarium-infected wheat heads.
The use of hyperspectral imaging on wheat FHB analysis depends on professional knowledge and skills. However, there is a potential to avoid such need by using machine learning or deep learning methods. Random forest, partial least squares regression, support vector regression and convolutional neural network (CNN) were often used to develop an analysis model for hyperspectral imaging (Adam et al., 2017, Tekle et al., 2015, Chu et al., 2017, Lee and Kwon, 2017). The first three models could be regarded as shallow models, which may not be as effective as CNN.
In this study, we compared those four models with the objective to develop a robust and generalized model based on hyperspectral imaging to predict the FHB disease in wheat (Fig. 1).
Section snippets
Experimental setup
Wheat head samples were collected from the experimental field of Anhui Academy of Agricultural Sciences in Anhui province, China (31°89′N, 117°13′E). The field experiment was conducted in an approximate 20 m × 20 m area that was divided into two equal sections: one section was inoculated with the Fusarium head blight pathogen (F. graminearum) and the other section as the control. The Fusarium fungus spores were evenly sprayed onto wheat plants (cv ‘Huaimai35′, susceptible to the FHB disease) in
Spectral response differences of wheat head samples
The original average spectra of healthy and Fusarium-infected wheat heads on both sides A and B in the 400–1000 nm range are shown in Fig. 5. First, there was a clear difference in spectral reflectance between both sides of the wheat heads. This was due to the three-dimensional structure of the wheat heads, and the difference in the degree of infection observed on the head surfaces. A similar trend of spectral response between side A and side B was observed on all samples evaluated. Secondly,
Conclusion
Developing early and quick detection and prediction of wheat FHB is crucial to the control of this important disease. In this study, the spectral and image data from Fusarium-infected wheat heads were collected by hyperspectral system and used to develop models to compare their abilities to detect the FHB disease in wheat. The models that were developed based on the integration of spectral and image features could provide better performance in predicting the FHB disease than the models that
Funding
The study was supported by the National Natural Science Foundation of China (Grant No. 41771463 and 41771469), Anhui Provincial Major Science and Technology Projects (Grant No. 18030701209), Team Project of Anhui Academy of Agricultural Sciences (Grant No. 2020YL073) and the open funding of National Engineering Research Center for Agro-Ecological Big Data Analysis & Application (Grant No. AE2018010).
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
References (55)
- et al.
Convolutional neural networks for vibrational spectroscopic data analysis
Anal. Chim. Acta
(2017) - et al.
Early detection of Fusarium infection in wheat using hyper-spectral imaging
Comput. Electr. Agric.
(2011) - et al.
Detecting Fusarium head blight in wheat kernels using hyperspectral imaging
Biosyst. Eng.
(2015) - et al.
Deoxynivalenol screening in wheat kernels using hyperspectral imaging
Biosyst. Eng.
(2017) - et al.
Random forest in remote sensing: A review of applications and future directions
Isprs J Photogramm.
(2016) - et al.
Detecting skin in face recognition systems: A colour spaces study
Digit Signal Process.
(2010) - et al.
Detection of aflatoxin B1 (AFB1) in individual maize kernels using short wave infrared (SWIR) hyperspectral imaging
Biosyst. Eng.
(2017) - et al.
Discrimination of gluten-free oats from contaminants using near infrared hyperspectral imaging technique
Food Control
(2017) - et al.
Winter wheat biomass estimation based on spectral indices, band depth analysis and partial least squares regression using hyperspectral measurements
Comput. Electr. Agric.
(2014) Complex wavelets for shift invariant analysis and filtering of signals
Appl. Comput. Harmon. A
(2001)
Hyperspectral image reconstruction by deep convolutional neural network for classification
Pattern Recogn.
Comparison of random forest, support vector machine and back propagation neural network for electronic tongue data classification: Application to the recognition of orange beverage and Chinese vinegar
Sensor Actuat B-Chem.
Image segmentation from scale and rotation invariant texture features from the double dyadic dual-tree complex wavelet transform
Image Vis. Comput.
Comparing the analytical performances of Micro-NIR and Ft-NIR spectrometers in the evaluation of acerola fruit quality, using PLS and SVM regression algorithms
Talanta
The use of Vis/NIRS and chemometric analysis to predict fruit defects and postharvest behaviour of 'Nules Clementine' mandarin fruit
Food Chem.
Variable weighted convolutional neural network for the nitrogen content quantization of Masson pine seedling leaves with near-infrared spectroscopy
Spectrochim. Acta. A
Identification of nine Iranian wheat seed varieties by textural analysis with image processing
Comput. Electron. Agric.
Feature selection based on artificial bee colony and gradient boosting decision tree
Appl. Soft Comput.
Classification of contaminants from wheat using near-infrared hyperspectral imaging
Biosyst. Eng.
Machine learning predictive models for mineral prospectivity: An evaluation of neural networks, random forest, regression trees and support vector machines
Ore Geol. Rev.
Detection of insect-damaged wheat kernels using near-infrared hyperspectral imaging
J. Stored Prod. Res.
Estimation of biomass in wheat using random forest regression algorithm and remote sensing data
Crop J.
A comparative assessment of support vector regression, artificial neural networks, and random forests for predicting and mapping soil organic carbon stocks across an Afromontane landscape
Ecol. Ind.
Hyperspectral measurements of yellow rust and fusarium head blight in cereal crops: Part 1: Laboratory study
Biosyst. Eng.
Principles, developments and applications of computer vision for external quality inspection of fruits and vegetables: a review
Food Res. Int.
Detecting the early stage of phaeosphaeria leaf spot infestations in maize crop using in situ hyperspectral data and guided regularized random forest algorithm
J. Spectrosc.
Hyperspectral quantification of wheat resistance to Fusarium head blight: comparison of two Fusarium species
Eur. J. Plant Pathol.
Cited by (39)
Macro-micro exploration on dynamic interaction between aflatoxigenic Aspergillus flavus and maize kernels using Vis/NIR hyperspectral imaging and SEM technology
2024, International Journal of Food MicrobiologyEnhancing fusarium head blight detection in wheat crops using hyperspectral indices and machine learning classifiers
2024, Computers and Electronics in AgricultureA research review on deep learning combined with hyperspectral Imaging in multiscale agricultural sensing
2024, Computers and Electronics in AgricultureA review on the combination of deep learning techniques with proximal hyperspectral images in agriculture
2023, Computers and Electronics in AgricultureInvestigation into maize seed disease identification based on deep learning and multi-source spectral information fusion techniques
2023, Journal of Food Composition and Analysis