Hyperspectral and Chlorophyll Fluorescence Imaging for Early Detection of Plant Diseases

Abstract

In recent years, market pressures have reinforced the demand to solve the problem of an increased occurrence of Fusarium head blight (FHB) in cereal production, especially in wheat. The symptoms of this disease are clearly detectable by means of image analysis. This technique can therefore be used to map occurrence and extent of Fusarium infections. From this perspective, a separate harvest in the field can be taken into consideration. Based on the application of chlorophyll fluorescence and hyperspectral imaging, characteristics, requirements and limitations of Fusarium detection on wheat, both in the field and in the laboratory, are discussed. While the modification of spectral signatures due to fungal infection allows its detection by hyperspectral imaging, the decreased physiological activity of tissues resulting from Fusarium impacts provides the base for CFI analyses. In addition, the two methods are compared in view of their usability for the detection of Fusarium, and different approaches for data analysis are presented.

Graphical Abstract

1. Introduction

The aim of modern agriculture is it not only to increase and optimize production but also to produce safe and healthy food and feed of high quality. In this regard, especially Fusarium infections on cereals represent an important, increasingly growing problem. Fusarium promoting cultivation systems such as intensified maize production have considerably aggravated the problem in recent years. Consequently, the extent of Fusarium infection has increased worldwide. Yield losses of up to 30% may highlight the tremendous impact of this disease.

Typical early and externally visible symptoms of Fusarium infections are the bleaching of individual spikelets and the partial dieback of the ear or head well before maturity. These symptoms are eponymous for the name of this fungal disease: “head blight”. The direct consequence of Fusarium infection is a massive crop failure due to the final development of shrunken lowmass tombstone kernels. The worst problem of the disease, however, is the potential toxic side effect due to the production of mycotoxins. Highly contaminated lots of grain are evidentially harmful and dangerous to humans and livestock. Fusarium generates mycotoxins such as deoxinivalenol (DON), zearalenone and fumonisins to different degrees , which can cause vomiting, mass loss, kidney failure, miscarriage, false pregnancy and cancer.

Therefore, infected grains should always be excluded from the human food cycle or livestock feed. In this context, identification of Fusarium infestations in the field with simple and rapid methods would be a substantial progress in food and feed safety. This might offer producers the opportunity to separately harvest infected and healthy grains, which, in turn avoids the risk of mixing contaminated and non-infected lots for storage. Separate harvest is also appropriate because spread of Fusarium fungi and synthesis of mycotoxins can proceed under certain conditions in storage. In addition, the risk of mycotoxin toxication of food can be decreased and the rational use of the Fusarium infected grain facilitated.

Until now, rapid and comprehensive Fusarium detection methods are not available in practice. It is still common to test grain lots for a possible mycotoxin contamination namely to determine their content of mycotoxin only at harvest or post-harvest. For this purpose, time consuming and expensive laboratory tests are necessary (HPLC, serological rapid tests, Fast-DON-ELISA-test, counting method ).

Along the entire cereal production chain, there are various options and management practices to avoid Fusarium infection and, thus, the occurrence of head blight (Figure 1). First of all, Fusarium tolerant wheat cultivars should be chosen for cultivation. In crop rotation, cereals such as maize, wheat, barley of durum, which readily spread Fusarium may be exclude, if possible. Furthermore, tillage may eliminate infected grain stubble or straw, which may function as inoculums the next year. This also prevents hibernation of fungi spores and, hence, a rapid distribution in the following spring. If disease pressure is high, e.g., due to unfavourable climatic conditions, fungicide must be applied shortly prior to flowering. At this time, azoles are often prophylactically sprayed on the entire field to assure prevention of Fusarium infection. Without doubt, closely targeted application of fungicides may be economically and ecologically advantages. This, however, requires routinely pre harvest acquisition of exact knowledge about the true, site-specific state of Fusarium infection.

Thus, automated reliable Fusarium detection is urgently needed because the frequency of infections is increasing, and, as a consequence, the legal provisions are intensified. Producers are waiting for innovative detection methods and appliances. Although currently not applied by default in crop production, such knowledge may be yet obtained by field monitoring with several recent imaging techniques. This review introduces the means of both chlorophyll fluorescence (CFI) and hyperspectral imaging for rapid site-specific on-field detection of head blight. It also presents recent advances in these techniques and discusses their capabilities and limits for practical applications of these methods.

Figure 1. Options to prevent Fusarium infection in the cereal production chain.

2. Imaging Techniques to Detect Head Blight Symptoms

During initial successful infection, Fusarium induces various internal changes and host-specific responses in inoculated plants. The resulting disease symptoms do normally not occur immediately but become externally visible only after approx. 7–11 days after inoculation. Only then, these symptoms can be detected and analyzed with spectro-optical reflectance measurements in the visible (VIS) and near-infrared (NIR) range but also with the aid of fluorescence spectroscopy. External, and to a certain degree also host-internal biochemical changes of cuticle, cell-walls, epidermis cells etc. changes may be evaluated by fluorescence and NIR measurements. In addition, fungal impacts on specific tissue properties such as composition and overall contents of leaf pigments or changes in cell water, sugar or protein contents may be investigated by remission measurements in the VIS and the NIR range, respectively. On the other hand, Fusarium effects on hosts’ metabolic competence at the cellular level at progressing infection not necessarily develop externally visible symptoms. These plant responses can be, not least, comprehensively monitored by analyzing chlorophyll fluorescence transients, which, among others, reflect the integrity of the photosynthetic apparatus .

2.1. Chlorophyll Fluorescence Imaging for Evaluation of Fungal Infections

Chlorophyll fluorescence analysis (CFA; e.g., and chlorophyll fluorescence imaging (CFI; e.g.) are well-established, effective tools for a comprehensive examination of development and effects of bacterial, fungal and viral infections on leaves of many cultivated plants . It can be used for entire intact plants , detached leaves and also leaf disks punched out from infected plant material.

On wheat, CFI was applied to determine, among others, effects of drought and heat stress, of limited supply of nutrients and of various diseases such as leaf rust, leaf and glume blotch or powdery mildew .

For disease detection, the empirical fluorescence parameter F_v/F₀ has been proposed for use on dark adapted plants. Although a clear physiological derivation is still lacking, F_v/F₀ presumably represents the maximum quantum yield of fluorescence . This parameter has been used as an indicator of the photosystem II (PSII) status and may estimate rates of energy transport from PSII to PSI in low-temperature fluorescence (−196 °C).

In addition, the potential maximum quantum yield of electron flow through completely open PSII, F_v/F_m is often used for evaluation of microbial diseases. F_v/F_m reflects the maximum photochemical efficiency and the interference with different environmental factors and may also indicate potential pathogen-related functional disturbance of the photosynthetic apparatus. Fusarium infections immediately impact F_v/F_m because the fungi rapidly and strongly impair metabolism and, thus, photosynthetic processes of contaminated spikelets or head parts of the host plants. The mycotoxins produced by the fungi may also contribute to the complete reduction of photosynthetic performance as has been found for infection of maize and banana by Colletotrichum musae and Fusarium moniliforme, respectively. Hence, it has been shown that the reduction of F_v/F_m, and also F_v/F₀, correlates closely with the degree of infection and is, therefore, a suitable parameter for detection of head blight and other fungal diseases .

2.1.1. Advantages of Image Analyses

Standard handheld Fluorometer, available from many various companies (e.g., Heinz Walz GmbH, Effeltrich, Germany; Hansatech Instruments Ltd., Norfolk, UK; Photon Systems Instruments, Brno, Czech Republic; Opti-Sciences Inc., Hudson, NY, USA; EARS Holding B.V., Delft, The Netherlands) provide average chlorophyll fluorescence values measured at a certain point of the examined object. In an attempt to identify tulip breaking virus (TBV) infections on three tulip cultivars with different leaf colour patterns, analyzed average Fv/Fm values, taken as precisely and as early as possible under laboratory conditions. Nonetheless, disease responses obtained by averaging Fv/Fm over the entire leaves differed up to 46% from the degree of disease determined by ELISA-tests. It seems obvious that both spot measurements and averaging of the fluorescence parameter might level off most of the disease-related differences in photosynthetic responses .

Applying chlorophyll fluorescence imaging , analyzed the spatial variation of disease development within infected wheat heads instead of averaging the chlorophyll fluorescence values over the entire object. This approach allows the close evaluation of even minor changes in the typical infection pattern of head blight. During disease development, the relative area of the heads with impacted spikelets and, hence, low Fv/Fm (<0.3) increased, while that of healthy spikelets with high Fv/Fm (>0.3) declined. Hence, spatial distribution patterns and not average chlorophyll fluorescence parameters were used to evaluate changes in the degree of disease during progressing infection. Pixel-wise classification of all Fv/Fm in relatively small value class and class-wise accumulation of all Fv/Fm, starting from 0 up to a value of 0.3 facilitated the detection of the head blight disease. The degree of disease was even differentiable in 10%-steps. Instead of leveling the infection-related differences by averaging the Fv/Fm values, considerable differences between healthy and infected plants emerged by application of this approach. In addition, mathematical operations were easy to program and numerical efficient, which is essential for field application.

2.1.2. The Timeframe of Detection

Because it closely reflects the physiology of photosynthesis, chlorophyll fluorescence imaging enables the early detection of Fusarium infection-related tissue damage. It is, indeed, well established that variations in photosynthetic activity and in the chlorophyll fluorescence patterns are detectable at early stages of infection . However, the earliest changes on a cellular level had only a minor effect on PSII . Only when the integrity of cellular structures of host plants was damaged by the fungi, the photosynthetic system was impaired. With light and electron microscopy, these authors showed that the dominating fungal hyphae caused pronounced cellular changes (e.g., degeneration of cytoplasm, destruction of cell organelles, disintegration of cell walls and deposition of cell wall material at the walls of vascular elements of the diseased head) only 4 to 5 days after infection.

However, the capability of chlorophyll fluorescence analysis and imaging, respectively, to detect fungal infections does not only depend on the period after inoculation; it is also considerably influenced by the object to examine itself. Infections by Pseudomonas syringae in Arabidopsis thaliana are detectable within few hours after inoculation . In contrast, Fv/F0 decreased two to three days before leaf rust and mildew infections, respectively, became visible on leaves of winter wheat . Pustules of leaf rust appeared 6 days after inoculation, while symptoms of mildew were visible approx. from the 9th dai onward. With leaf rust infection, Fv/F0 significantly declined by about 0.4 relative units after the 6th dai onward, while decreases in Fv/Fm were considerably smaller (0.02 relative unit) compared to uninfected controls. First symptoms of Venturia inaequalis infection on apple seedlings could be detected at the 7th onward. In seedlings of common spruce (Picea abies), needle rust (Chrysomyxa rhododendri) infections could be identified with CFA only three weeks after infection but not at earlier stage.

Time range for the meaningful detection of Fusarium infections of wheat with CFA and CFI is also limited. With the onset of wheat head maturation, chlorophyll content of the spikelets inevitably declines, and, concomitantly, also Fv/Fm decreases irrespective of whether infected or not (Figure 2). Consequently, relative cumulative Fv/Fm rapidly increases at low Fv/Fm classes (Figure 2a). Logically, this parameter is no longer suitable for biunique disease detection on fully mature wheat heads (cf. dai 43 = BBCH 89) after reaching the final grain developmental stage (BBCH 79; developed by Biologische Bundesanstalt für Land- und Forstwirtschaft, Bundessortenamt und Chemische Industrie, this system comprehensively describes development states of crops according to Eucarpia Codes, EC).

2.1.3. Detection Accuracy of CFI on Wheat Plants with Different Degrees of Fusarium Infection

The potential maximum photochemical efficiency Fv/Fm readily indicates damage of photosynthetic apparatus by Fusarium culmorum. Very high detection accuracy (10% RMSE) could be achieved with the application of the relative cumulative Fv/Fm (rcFv/Fm) at a threshold of 0.3 (Figure 2b). Relative cumulative Fv/Fm values in the lower efficiency classes gradually increased from <0.1% at 2%–3% of visually evaluated degrees of Fusarium infection (degrees of infection, doi) to 10% at doi of 10%–20%, hold approx. 30% at degrees of infection of 40%–60% and reach 85% at an infection rate of 90%. In the above investigation, averaging Fv/Fm over the analysed heads indeed levelled all these differences. However, the application of distinct Fv/Fm value classes precisely monitored the damaging effects of Fusarium on photosynthesis. Beyond a minimum doi of 5%, the respective degrees of infection could be readily distinguished in steps of 10% with this method [38]. In other investigations , where all efficiency classes were considered by averaging, changes in Fv/Fm with progressive infection remained small and could be only insufficiently resolved.

For practical application, the true degree of infection may be easily evaluated by remote CFI and use of relative cumulative Fv/Fm. With the threshold relative cumulative Fv/Fm of 0.3, the degree of infection contaminated wheat plants can be reliable estimated with only an error of 5.8% (doi (%) = 4.00 + 1.073 × rcFv/Fm; R2 = 0.98).

2.2. Hyperspectral Imaging in the VIS (400–700 nm) and NIR-Range (700–3000 nm)

Hyperspectral imaging (Figure 5) largely enhances the possibilities of multispectral image analysis. Major advantage of hyperspectral imaging is the pixel-wise incorporation of a continuous spectral signature of hundreds of wavelengths into a two-dimensional image of the object under inspection . This comprises the free choice to calculate ratios of different desired wavelengths (ranges) and of different indices and finally, obtaining multiple wavelengths optimal for evaluation of specific problems under consideration.

Figure 5. (a) Hyperspectral image scanner; (b) Reflexion images of a Fusarium-infected wheat sample at different spectra channels (from left to right: red: 550 nm, green: 685 nm, blue: 765 nm).

By applying six different vegetation indices, calculated from spectra of sugar beet leaves infected with Cercospora beticola, Erysiphe betae and Uromyces betae, it was possible to distinguish infected and healthy leaves during the course of three weeks. Using discriminant analysis and neuronal networks, 10 optimal wavelengths, out of 167 originally recorded both in the VIS and NIR range, were identified for effective weed detection in sugar beet . For the distinction of various rust pathogens on wheat, different vegetation indices were calculated from spectra recorded between 400 and 900 nm.