Korean Journal of Remote Sensing (대한원격탐사학회지)

Korean Society of Remote Sensing (대한원격탐사학회)
권호 표지
DOI QR코드

DOI QR Code

Selecting Significant Wavelengths to Predict Chlorophyll Content of Grafted Cucumber Seedlings Using Hyperspectral Images

  • Jang, Sung Hyuk (Institute for Agricultural Machinery & ICT Convergence, Chonbuk National University) ;
  • Hwang, Yong Kee (Institute for Agricultural Machinery & ICT Convergence, Chonbuk National University) ;
  • Lee, Ho Jun (Department of Agricultural Machinery Engineering, Graduate School, Chonbuk National University) ;
  • Lee, Jae Su (Department of Agricultural Machinery Engineering, Graduate School, Chonbuk National University) ;
  • Kim, Yong Hyeon (Institute for Agricultural Machinery & ICT Convergence, Chonbuk National University)
  • Received : 2018.08.12
  • Accepted : 2018.08.19
  • Published : 2018.08.31
Download PDF
⟨ Previous Next ⟩

Abstract

This study was performed to select the significant wavelengths for predicting the chlorophyll content of grafted cucumber seedlings using hyperspectral images. The visible and near-infrared (VNIR) images and the short-wave infrared images of cucumber cotyledon samples were measured by two hyperspectral cameras. A correlation coefficient spectrum (CCS), a stepwise multiple linear regression (SMLR), and partial least squares (PLS) regression were used to determine significant wavelengths. Some wavelengths at 501, 505, 510, 543, 548, 619, 718, 723, and 727 nm were selected by CCS, SMLR, and PLS as significant wavelengths for estimating chlorophyll content. The results from the calibration models built by SMLR and PLS showed fair relationship between measured and predicted chlorophyll concentration. It was concluded that the hyperspectral imaging technique in the VNIR region is suggested effective for estimating the chlorophyll content of grafted cucumber leaves, non-destructively.

Keywords

1. Introduction

Chlorophyll is closely related to the nutritional status of plants. In general, organic solvents such as acetone, ethanol, or dimethylsulfoxide (DMSO) are widely used to extract chlorophyll (Hiscox and Israelstam, 1979; McKinney, 1941). This method of extraction is not only destructive but is also time-consuming. A large variation in the measured value may occur due to loss of pigments during extraction or dilution (Nettoa et al., 2005). Furthermore, since the leaves are destroyed for extraction, the sample cannot remain intact after measurement.

Unlike conventional methods that are destructive, portable meter can determine the relative amount of chlorophyll content without sampling. The meter is effective for monocotyledonous plants whose leaves are long and narrow and needs a short length of time for measurement. However, measurement accuracy drops for low light intensity (Uddling et al., 2007).

Although there is no close correlation between chlorophyll content and photosynthesis capacity (Marini, 1986), the quantification of photosynthesis-related pigments is used as a key index of crop senescence (Hörtensteinera and Kräutler, 2011). In addition, the loss of chlorophyll is highly influenced by environmental stress, and thus variations in chlorophyll and carotenoid ratio are used as a stress index (Hendry and Price, 1993).

Light irradiated to plants is reflected from leaf surfaces, permeated into the canopy, or absorbed by substances such as leaf tissue moisture, metabolites, or chlorophyll. Therefore, the reflectance at the leaf surface indicates the physiological, morphological and biochemical state of the plant under given environmental conditions.

The remote sensing technique based on hyperspectral imaging, i.e., reflectance characteristics of visible and near-infrared light, is used for investigating vegetation (Jiang et al., 2006; Leprieur et al., 2000), moisture content of vegetable seedlings (Jang et al., 2018), biochemical components of leaves (Fourty et al., 1996), or physiological characteristics (Penuelas and Filella, 1998). Various indices for estimating these properties are presented (Daughtry et al., 2000; Gitelson et al., 2003).

Since hyperspectral images have a number of contiguous spectral bands with a narrow bandwidth, each pixel of an image has characteristic attribute of spectral reflectance information. In a hyperspectral image with three-dimensional data, the first and second data represent position information of a pixel, and the third data represents spectral information. Therefore, when a data processing technique is applied to a hyperspectral image with spectral information for each pixel, it will be possible to diagnose plant deformation, accumulation of specific compounds, and structural change depending on environmental stress as well as crop growth characteristics. Hyperspectral image data may have noise or defective pixels that ned to be removed by preprocessing operations (Rasti et al., 2018; Vidal and Amigo, 2012) such as correction, smoothing, and normalization.

Cucumber is one of the major vegetable crops. Demand of grafted seedlings is still increasing because grafting is to enhance the water and nutrient uptake, to increase the tolerance to low temperature and to soil salinity, and to reduce the incidence of soil-borne disease. Grafted seedlings, greatly influenced by physical environment conditions in the process of grafting scion and rootstock, experience moisture stress. Moreover, since low light intensity is given in this process, the photosynthetic rate of the grafted seedlings is low, and the changes in the chlorophyll content and substrates are evident. This study attempts to quantitatively identify the substrate changes according to the environmental changes in the process of healing grafted seedlings. The purpose of this study is to select significant wavelengths for estimating the chlorophyll content of the leaves of the cucumber seedlings in a non-destructive manner according to the hyperspectral imaging with preprocessing methods.

2. Materials and Methods

1) Variety and Grafting

The scion for seedlings used in this experiment was cucumber, Cucumis sativus L. cv. (Joeun Baekdadaki, Dongbuhitek, Korea) and the rootstock was figleaf gourd, Cucurbita ficifolia cv. (Heukjong, Monsanto, Korea). Peat moss (BM1, Berger Peat Moss, Canada) was used for seedling mixture.

Seedlings grown for 9-10 days in the 128-hole plug trays were grafted into scion-rootstock union. The scion, cut at an angle of 45° from 1 cm below the cotyledon, was grafted with the stock whose root, but one cotyledon, was removed. At this time, a grafting clip fixed both scion and rootstock whose cutting face could align with each other for smooth union. The pH and EC of the culture were adjusted to 5.5-6.0 and 1.5 mS·cm-1, respectively, using a portable meter (HD98569, Delta Ohm, Italy).

2) Grafting Environment for Seedlings

The grafted cucumber seedlings were healed in a closed seedling production system (‘closed system’)capable of controlling temperature, relative humidity and air velocity. Inside the closed system, a 4-shelf aluminum profile of 1,200 × 600 × 2,000(W× D × H)mm was set up. Each shelf occupied with 160 cucumber seedlings grafted in 4-40 plug trays had LED lamps with white, red, blue, and red-blue mixed.

The light intensity of the LED lamp was adjusted using a lamp controller (V1.0, ODTech, Korea) and a power supply (SE-450-36, Mean Well, Taiwan). Light intensity at each shelf was maintained at 50 μmol·m-2·s-1 photosynthetic photon flux (PPF) with a quantum sensor (SKP215, Skye, UK). Temperature and relative humidity in the closed system were adjusted to 25°C and 90%, respectively, and the photoperiod of the LED lamp was 16 h&middt;d-1. For a healthy graft, the PPF in each treatment shelf was raised to 150 μmol·m-2·s-1, while the relative humidity was reduced to 70% on the 7th day after the grafting.

3) Hyperspectral Image Acquisition and Preprocessing

(1) System for Hyperspectral Image Acquisition

In this study, two hyperspectral imaging systems were used to obtain hyperspectral images of cucumber seedlings: one for the visible and near-infrared (VNIR) images and the other for short-wave infrared (SWIR) images.

The VNIR system consists of an image acquisition unit, a sample transfer unit, and a light source unit. The image acquisition unit is composed of an imaging spectrograph (VNIR, Headwall photonics, USA), an electron-multiplying charge-coupled device (EMCCD), and a camera (Luca DL-604M, Andor Technology, USA). The sample transfer unit has a step-motor(Vexta PK264, Oriental Motor, Japan) and a motor controller (VXM-1, Velmex, USA). The light source unit has two 150W halogen lamps (93638EKE, Osram, Germany), fiber optics (DC950, Dolan-Jenner Industries, USA) and line light cables(QF5548, Dolan-JennerIndustries, USA).

The light from the halogen lamp, whose irradiation angle was adjusted by the guide, maintained an angle of about 45° to minimize the non-uniformity of the reflected light. The VNIR system was controlled by Visual Basic 6.0. When the exposure time of the lens s set, the camera can automatically calculate the travel distance and speed of the step-motor. The image projected on the lens is line-scanned through a slit with a thickness of 25 μm and a spectrograph produces spectra by each wavelength. The spectral information is amplified by the EMCCD camera and stored as an image. In this study, hyperspectral images of grafted cucumber seedlings were obtained with the exposure time of 5.0 ms, and the step motor movement of 0.64 mm interval with 313 steps; a total movement of 20 cm.

The SWIR system consists of an image acquisition unit, a camera fixing frame, and a light source unit. The image acquisition unit has a camera with a spectrograph (SWIR 3.0, Specim, Finland), a detector (mercury cadmium telluride, MCT) and a lens. The light passes through the lens and the slit of 30 μm thickness. Spectra were produced by spectrograph, and detected by detector by each wavelength, and converted into three-dimensional data for storage. Two 650 W halogen lamps (64540BVM, Osram, Germany) were used in the light source unit. In the measurement process, the focus adjustment and the image acquisition were performed in the same way as the VNIR imaging system.

(2) Hyperspectral Image Acquisition

The 560 hyperspectral images in the wavelengths of VNIR and SWIR were measured, for 80 seedlings(20 seedlings × 4 treatments) per day, 7 days from the 2nd to 8th day after the initiation of the grafted cucumber seedlings. At this time, a white reference plate (SRT-99100, Labphere, USA) with a reflectance of 99% was placed beside the grafted seedlings.

The range of the measured VNIR image is 378.1-1,002.2 nm where the interval of 4.7 nm per band resulted in 133 bands. The range of SWIR image is 904.4-2,515.9 nm where the interval of 5.6 nm per band resulted in 288 bands. Analysis excluded wavelengths at the border region of VNIR and SWIR because the output signal of sensor is low for that peculiar region. Therefore, the range of the analyzed image is 400-1,000 nm for VNIR and 1,000-2,400 nm for SWIR.

(3) Hyperspectral Image Preprocessing

The flat field correction (FFC), shown in equation (1), was applied to correct the brightness of the original image acquired using the hyperspectral imaging system (Nieuwenhove et al., 2015).

\(\text { Reflectance }=\frac{R_{\text {raw }}-R_{\text {dark }}}{R_{\text {white }}-R_{\text {dark }}}\)       (1)

where, Rraw is the original spectrum obtained through the hyperspectral imaging system, Rdark is the dark spectrum measured with the lens closed, and Rwhite is the spectrum of the white reference plate with the reflectance of 99%.

The spectral data of the brightness correction was preprocessed by applying the Savitzky-Golay filter for Smoothing (SM), determining the spectral average and standard deviation fo Standard Normal Variate (SNV), and by employing ideal spectrum data to correct spectrum of samples for Multiplicative Scatter Correction (MSC), and by sing the First Derivative (FD). Altogether, 7 different processes(FFC, SM, SNV, SNV + SM, MSC, MSC + SM, FD) were applied in this study (Fig. 1). MATLAB (2017a, MathWorks, USA) were used for all preprocessing tasks.

OGCSBN_2018_v34n4_681_f0001.png 이미지

Fig. 1. The process of spectral preprocessing applied in the study.

(4) Extraction of Region of Interest

A monochrome binary image was made to separate the cucumber leaves from the background of the preprocessed hyperspectral image. Next, an erosion mask was applied to the binary image to extract the region of interest of the leaf.

4) Measuring Chlorophyll Content

A portable chlorophyll meter (SPAD-502, Minolta Co., Japan) was used to measure the chlorophyll content of the cucumber seedlings. The chlorophyll content was measured at the left, right and center of the cotyledon surface. The mean value of chlorophyll content measured in 5 replicates at each site was used for statistical model development and validation.

5) Statistical Model Development and Verification

(1) Selection of Significant Wavelengths

In order to determine the significant wavelengths affecting the estimation of chlorophyll content in cucumber seedlings from preprocessed spectrum data, a correlation coefficient spectrum (CCS), a stepwise multiple linear regression (SMLR), and a partial least squares(PLS) were used. For model development, 560 datasets were divided into two groups: 420 datasets for calibration and 140 datasets or validation. They were analyzed using statistical software SAS (V9.4, SAS Institute Inc., USA). In the CCS analysis, the absolute value of the correlation coefficient (r) between the reflectance per wavelength and the SPAD value of the cucumber seedlings, that is, the wavelength corresponding to |r|>0.5 was determined. In the SMLR analysis, the corresponding wavelength of the spectral reflectance required to describe the chlorophyll content is selected by the statistical significance level. In the PLS regression analysis, the coefficient β was determined using X-loading, X-weight, and Y-weight calculated in the analysis process. In this case, the β value indicates the importance of each independent variable. The larger the absolute of β value, the greater the influence of the wavelength on the PLS regression model (Min and Lee, 2005). In this study, the absolute value criterion of coefficient β was set to 1.5. In the PLS analysis, a model with a minimum value of the predicted residual error sum of squares(PRESS) in the cross-validation process was used to determine the appropriate number of factors. PRESS is the sum of squares of the residuals between the observed values and the predicted values of the PLS regression equation. The smaller the PRESS value, the better the predictive power of the model (Osten, 1988). Ultimately, the wavelengths commonly included in the results of CCS, SMLR, and PLS analyzes were selected as the significant wavelengths required for chlorophyll content prediction

(2) Statistical Model Development and Verification

PLS and SMLR methods were used to develop calibration models in this study. The model developed using 140 validation data was verified and the coefficient of determination (R2), which indicates the linear fitting of the variables, was used to evaluate the fitness of the model. The standard error of calibration (SEC), standard error of validation (SEV), and root mean square difference (RMSD) were calculated to verify the measurement error and the prediction error of the developed model. These can be expressed as follows.

\(\mathrm{SEC}=\sqrt{\frac{1}{n-p-1} \sum_{i=1}^{n} e_{i}^{2}}\)       (2)

\(\mathrm{SEV}=\sqrt{\frac{1}{n-p-1} \sum_{i=1}^{n}\left(e_{i}-\bar{e}\right)^{2}}\)       (3)

\(\mathrm{RMSD}=\sqrt{\frac{1}{n} \sum_{i=1}^{n} e_{i}^{2}}\)       (4)

where, n is the number of samples, p is the number of independent variables, ei is the difference between the predicted value and the measured value, and ē s the average of ei.

3. Results and Discussion

Out of 560 data measured by SPAD meter, calibration dataset (420 measurements) whose mean and standard deviation of SPAD values were 64.5 and 4.2, respectively, was used to develop a prediction model for chlorophyll content in the cucumber seedling laves. In the meantime, the mean and standard deviation of SPAD values using validation dataset(140 measurements) were 64.4 and 4.0, respectively, similar to the distribution of calibration dataset.

Fig. 2, 3, and 4 demonstrate the simple correlation of chlorophyll content measured by SPAD vs. reflectance at each wavelength. The wavelengths at which the absolute value of correlation coefficient is above a certain level, 0.5 for this research, was selected as significant. For VNIR range of 400-1000 nm, the correlation coefficient between the SPAD value and the spectral reflectance at each wavelength was higher with SNV and SNV+SM than with the other preprocessing methods (Fig. 2). The highest correlation coefficient was -0.68 at 723 nm with SNV preprocessing, and -0.67 at 718 nm with combined preprocessing, SNV + SM. The wavelength ranges at which |r| was higher than 0.5 were 501-510 nm, 543-548 nm, 619-661 nm and 704-727 nm with SNV (Fig. 2a), and 496-505 nm, 538-548 nm, 619-680 nm and 699-727 nm with SNV + SM (Fig. 2b). The highest correlation coefficient was 0.59 at 576 nm with FD, the first derivative preprocessing. The wavelengths at which |r| was higher than 0.5 were at 524-529 nm, 557-590 nm and 666-680 nm with FD (Fig. 3). For SWIR range of 1,000-2,400 nm, on the other hand, the correlation coefficient between the spectral reflectance at each wavelength and the SPAD value was lower than 0.4 with all preprocessing methods applied (Fig. 4).

Fig. 2. Correlation coefficients between reflectance at each wavelength and SPAD value of calibration data set with (a) SNV and (b) SNV+SM preprocessing.

OGCSBN_2018_v34n4_681_f0003.png 이미지

Fig. 3. Correlation coefficients between reflectance at each wavelength and SPAD value of calibration data set with FD preprocessing.

OGCSBN_2018_v34n4_681_f0004.png 이미지

Fig. 4. Correlation coefficients between reflectance at each wavelength and SPAD value of calibration data set (a) SNV+SM and (b) FD preprocessing.

Table 1 is the result of SMLR analysis by which the significant wavelengths were selected after calibration dataset was preprocessed with all methods suggested in the materials and methods. No common wavelength was selected regardless of preprocessing methods. However, the wavelength 619 nm was commonly included in four (FFC, SM, MSC+SM, and FD) preprocessing methods, 709 nm in three (FFC, SM and MSC)methods, 960 nm by three (MSC, MSC+SM and FD)methods. In addition, 444, 633, 642, 675, 694, 704, 832, 846, 860, and 926 nm were common wavelengths selected by a pair of preprocessing methods.

Table 1. Significant wavelengths selected by stepwise multiple linear regression (SMLR) analysis to predict chlorophyll content of grafted cucumber seedlings

OGCSBN_2018_v34n4_681_t0001.png 이미지

Fig. 5 is the result of PSL by which β coefficient was determined with SNV preprocessing (Fig. 5a) and with SNV+SM (Fig. 5b) in the VNIR region. Abrupt changes in β value with SNVwere remarkably reduced, or smoothed with SNV+SM. Wavelengths corresponding to |β|>1.5 were at 406, 477, 510, 638, 694, 718-732, 751, 959, 978, 983 and 992 nm for SNV preprocessing, and at 401, 406, 444, 449, 496, 501, 505, 510, 515 and 543 nm for SNV+SM.

OGCSBN_2018_v34n4_681_f0005.png 이미지

Fig. 5. B-matrix determined from calibration dataset using partial least square (PLS) analysis with (a) SNV and (b) SNV+SM preprocessing.

Accordingly, VNIR wavelengths at 501, 505, 510, 543, 548, 619, 718, 723, and 727 nm were commonly included when two out of three (CCS, SMLR and PLS) analyses employed in this study.

There has been no report that hyperspectral imaging was used to predict the chlorophyll content of the cucumber seedlings. The reflectance for VNIR range is closely related to the contents of chlorophyll, xanthophyll or moisture in the plant leaf (Penuelas and Filella, 1998). Estimation of chlorophyll content using reflectance index has mainly been done on food crops or trees (Gitelson et al., 2003; Kochubey and Kazantsev, 2007). An analysis of the relationship between the chlorophyll content and the reflectance characteristics in the VNIR of Eucalyptus leaves was reported that 710 nm was the most sensitive to chlorophyll content changes (Datt, 1999). Reflectance at 675 nm was used to determine the chlorophyll content of orange leaves non-destructively (Wallihan, 1973). An attempt has also been made to determine the chlorophyll content of pine trees using reflectance at the red edge, or 680-730 nm (Ferns et al., 1984).

The leaf color of vegetables is greatly affected by chlorophyll content. Therefore, leaf color is a major factor for vegetable quality, but reports on the relation between chlorophyll content and reflectance are not sufficient (Jones et al., 2007). Thomas and Gaussman (1977) presented a correlation between reflectance and chlorophyll content of melon, cucumber and lettuce at 550 nm. In order to estimate the chlorophyll content of lettuce and spinach where different concentrations of nitrogen were applied, indices using reflectance at 720 and 740 nm were suggested (Xue and yang, 2009). On the other hand, a portable chlorophyll meter that can be conveniently used in the crop field determines the chlorophyll content by measuring the absorption at two wavelengths of 650 nm and 940 nm (Ling et al., 2011). As a result, it can be seen that the significant wavelengths affecting chlorophyll content changes are not the same depending on the type of crop or leaf structure.

Fig. 6 demonstrates the typical reflectance characteristics of grafted cucumber seedlings with different SPAD value in the NIR and SWIR region. Spectral reflectance of the cucumber seedlings was high for higher SPAD value in the VNIR region (Fig. 6a). Spectral reflectance was not decisive in the SWIR region where reflectance was high for higher SPAD at 1,407-2,332 nm, but low at 1,000-1,136 nm.

OGCSBN_2018_v34n4_681_f0006.png 이미지

Fig. 6. Spectral reflectance for the leaf of grafted cucumber seedlings with ow and high SPAD value in the (a) VNIR and (b) SWIR region.

Table 2 shows the results of the SMLR model performance using validation dataset vs. calibration dataset with the selected number of wavelengths, according to the different preprocessing methods, listed in the Table 1 to estimate the chlorophyll content of the cucumber seedlings. SM preprocessing method showed that the best R2 for the validation dataset was 0.72 and RMSD was 2.14. SNV+SM preprocessing with the selected number of wavelengths, 5 from Table 1 resulted in no difference in standard errors (SEC, SEV). In the meantime, the differences in R2 and RMSD between calibration and validation datasets were negligible.

Table 2. Results of stepwise multiple linear regression (SMLR) analysis using VNIRdata with different preprocessing techniques

OGCSBN_2018_v34n4_681_t0002.png 이미지

Table 3 shows the results of PLS model performance using validation dataset vs. calibration dataset according to the different preprocessing methods. The number of factors in Table 3 was the result from the cross validation process based on the lowest PRESS value. Most preprocessing generated 12 factors, except 11 for SNV+SM. In the VNIR region with PLS regression model, R2 using validation dataset with FFC and SNV preprocessing was 0.72, the highest performance. RMSD was 2.12 for FFC preprocessing and 2.13 for SNV, respectively. In the PLSmodel, when R2 and SEV using validation dataset was compared with R2 and SEC using calibration dataset, the former was consistently smaller than the latter. But PLS performance inquiry with RMSD values according to the different preprocessing method revealed inconclusive between the two datasets. This consistency in the PLS results was also found in the SMLR results. Namely, model performance of both SMLRand PLS looks very similar despite of their different analysis method. This result encourages us to believe that both SMLR and PLS models are not exclusive each other, and meet the purpose of this study to select common, significant wavelengths.

Table 3. Results of partial least square (PLS) regression analysis using VNIR data with different preprocessing techniques

OGCSBN_2018_v34n4_681_t0003.png 이미지

Table 4 is the results of PLS using SWIR data. The number of factors determined by PRESS value when PLS regression model was run with different preprocessing method was 11-15. Despite of this result of a slight increase from the result with VNIR data, the validation performance of PLS regression model with SWIR region was not suitable due to lower 2 and higher SEV and RMSD than that with VNIR region. Lower R2 may be explained by over-fitting of the PLS prediction model.

Table 4. Results of partial least square (PLS) regression analysis using SWIR data with different preprocessing techniques

OGCSBN_2018_v34n4_681_t0004.png 이미지

Vegetable grafting is to improve plant production, to reduce disease susceptibility, and o increase plant vigor. Commercial production and demand for grafted vegetable plants continues to increase. However, there was no report on estimating the chlorophyll content of grafted seedlings. Hyperspectral imaging with different preprocessing methods is a useful tool to estimate the chlorophyll content of other crops as well as vegetable seedlings. Significant wavelengths and the calibration models obtained in this study can be applied for developing new index to predict the chlorophyll content of grafted seedlings in nursery.

In order to improve the accuracy of the prediction model in the future, it is preferable to determine the chlorophyll content of the cucumber seedlings by the extraction method instead of using the portable chlorophyll meter. In addition, increasing the number of samples used in model development and verification will also improve the accuracy of the model.

4. Conclusion

This study was conducted to select significant wavelengths required for estimating chlorophyll content of leaf, with different combination of preprocessing methods and statistical analyses after hyperspectral images were obtained from grafted cucumber seedlings. A correlation coefficient spectrum (CCS), a stepwise multiple linear regression (SMLR), and partial least squares (PLS) regression were used to determine significant wavelengths. A model for estimating chlorophyll content, developed using those selected wavelengths, was tested using validation dataset. Soe wavelengths at 501, 505, 510, 543, 548, 619, 718, 723, 727 nm were selected as significant common wavelengths as required to estimate the chlorophyll content. In the VNIR region, SMLR analysis model with SNV+SM or SNV preprocessing needed less number (5 and 7 wavelengths) of selected wavelengths, comparing with other preprocessings (9-11 wavelengths), to achieve competitive SMLR model performance, or, became relatively simple. The results from the calibration models built by SMLRand PLS showed fair relationship between measured and predicted chlorophyll concentration. With the best calibration model built on wavelengths selected by SMLR with SM preprocessing for VNIR data, R2 for the validation dataset was 0.72, and RMSD was 2.14. The calibration model built by PLS on the full spectral region had an R2 of 0.72 for the validation dataset with FFC and SNV preprocessings. RMSD was 2.12 for FFC preprocessing and 2.13 for SNV, respectively. When PLS analysis was applied in the VINIR, the number of influential factors based on PRESS value was 12 for all except SNV+SM preprocessing method. In the SWIR, when PLS analysis was applied, more influential factors based on PRESS value were selected than in the VNIR. But lower R2 , and higher SEV, and RMSD indicate that analysis with the wavelengths in the SWIR region turns out to be inadequate for developing a model for estimating the chlorophyll content of the cucumber leaf. Therefore, the hyperspectral imaging technique in the VNIR region, not SWIR, is suggested effective for estimating the chlorophyll content of grafted cucumber leaves, non-destructively.

Acknowledgment

This work was supported by Korea Institute of Planning and Evaluation for Technology in Food, Agriculture and Forestry through Agriculture, Food and Rural Affairs Research Center Support Program, funded by Ministry of Agriculture, Food and Rural Affairs (No. 717001-07-2-SB210).

References

  1. Datt, B., 1999. Visible/near infrared reflectance and chlorophyll content in Eucalyptus leaves, International Journal of Remote Sensing, 20(14): 2741-2759. https://doi.org/10.1080/014311699211778
  2. Daughtry, C.S.T., C.L. Walthall, M.S. Kim, E.B. Colstoun, and J.E. McMurtrey III, 2000. Estimating corn leaf chlorophyll concentration from leaf and canopy reflectance, Remote Sensing of Environment, 74(2): 229-239. https://doi.org/10.1016/S0034-4257(00)00113-9
  3. Ferns, D.C., S.J. Zara, and J.Barber, 1984. Application of high resolution spectroradiometry to vegetation, Photogrammetric Engineering and Remote Sensing, 50(12): 1725-1735.
  4. Fourty, T., F. Baret, S. Jacquemoud, G. Schmuck, and J. Verdebout, 1996. Leaf optical properties with explicit description of its biochemical composition: Direct and inverse problems, Remote Sensing of Environment, 56(2): 104-117. https://doi.org/10.1016/0034-4257(95)00234-0
  5. Gitelson, A.A., Y. Gritz, and M.N. Merzlyak, 2003. Relationships between leaf chlorophyll content and spectral reflectance and algorithms for nondestructive chlorophyll assessment in higher plant leaves, Journal of Plant Physiology, 160(3): 271-282. https://doi.org/10.1078/0176-1617-00887
  6. Hendry, G.A.F. and A.H. Price, 1993. Stress indicators: chlorophylls and carotenoids, In: Hendry, G.A.F., Grime, J.P. (Eds.), Methods in Comparative Plant Ecology, Chapman & Hall, London, UK, pp.148-152.
  7. Hiscox, J.D. and G.F. Israelstam, 1979. A method for the extraction of chlorophyll from leaf tissue without maceration, Canadian Journal of Botany, 57(12): 1332-1334. https://doi.org/10.1139/b79-163
  8. Hortensteinera, S. and B. Krautler, 2011. Chlorophyll breakdown in higher plants, Biochimica et Biophysica Acta Bioenergetics, 1807(8): 977-988. https://doi.org/10.1016/j.bbabio.2010.12.007
  9. Jang, S.H., H.J. Lee, J.S. Lee, and Y.H. Kim, 2018. Prediction of moisture content of grafted cucumber seedlings using hyperspectral image, Proc. of the 9th International Symposium on Machinery and Mechatronics for Agriculture and Biosystems Engineering,Jeju, Korea, May 28-30, p.201.
  10. Jiang, Z., A.R, Huete, J. Chen, Y. Chen, J. Li, G. Yan, and X. Zhang, 2006. Analysis of NDVI and scaled difference vegetation index retrievals of vegetation fraction, Remote Sensing of Environment, 101(3): 366-378. https://doi.org/10.1016/j.rse.2006.01.003
  11. Jones, C.L., P.R. Weckler, N.O. Maness, R. Jayasekara, M.L. Stone, and D.Chrz, 2007.Remote sensing to estimate chlorophyll concentration in spinach using multi-spectral plant reflectance, Transactions of the ASABE, 50(6): 2267-2273. https://doi.org/10.13031/2013.24079
  12. Kochubey, S.M. and T.A. Kazantsev, 2007.Changes in the first derivatives of leaf reflectance spectra of various plants induced by variations of chlorophyll content, Journal of Plant Physiology, 164(12): 1648-1655. https://doi.org/10.1016/j.jplph.2006.11.007
  13. Leprieur, C., Y.H. Kerr, S. Mastorchio, and J.C. Meunier, 2000. Monitoring vegetation cover across semi-arid regions:Comparison of remote observations from various scales, International Journal of Remote Sensing, 21(2): 281-300. https://doi.org/10.1080/014311600210830
  14. Ling, Q., W. Huang, and P. Jarvis, 2011. Use of a SPAD-502 meter to measure leaf chlorophyll concentration in Arabidopsis thaliana, Photosynthesis Research, 107(2): 209-214. https://doi.org/10.1007/s11120-010-9606-0
  15. Marini, R.P., 1986. Do net gas exchange rates of green and red peach leaves differ?, HortScience, 21: 118-120.
  16. McKinney, G., 1941. Absorption of light by chlorophyll solutions, Journal of Biological Chemistry, 140: 315-322. https://doi.org/10.1016/S0021-9258(18)51320-X
  17. Min, M. and W.S. Lee, 2005. Determination of significant wavelengths and prediction of nitrogen content for citrus, Transactions of the ASAE, 48(2): 455-461. https://doi.org/10.13031/2013.18308
  18. Nettoa, A.T., E. Campostrinib, J.G. Oliveirab, and R.E. Bressan-Smith, 2005. Photosynthetic pigments, nitrogen, chlorophyll a fluorescence and SPAD-502 readings in coffee leaves, Scientia Horticulturae, 104(2): 199-209. https://doi.org/10.1016/j.scienta.2004.08.013
  19. Nieuwenhove, V.V., J.D. Beenhouwer, F.D. Carlo, L. Mancini, F. Marone, and J. Sijbers, 2015. Dynamic intensity normalization using eigen flat fields in X-ray imaging, Optics Express, 23(21): 27975-27989. https://doi.org/10.1364/OE.23.027975
  20. Osten, D.W.,1988. Selection of optimal regression models via cross-validation, Journal of Chemometrics, 2(1): 39-48. https://doi.org/10.1002/cem.1180020106
  21. Penuelas, J. and I. Filella, 1998. Visible and near-infrared reflectance techniques for diagnosing plant physiological status, Trends in Plant Science, 3(4): 151-156. https://doi.org/10.1016/S1360-1385(98)01213-8
  22. Rasti, B., P. Scheunders, P. Ghamisi, G. Licciardi, and J. Chanussot, 2018. Noise Reduction in Hyperspectral Imagery: Overview and Application, Remote Sensing, 10(3): 482-509. https://doi.org/10.3390/rs10030482
  23. Thomas, J.R. and H.W. Gaussman, 1977. Leaf reflectance vs. leaf chlorophyll and carotenoid concentration for eight crops, Agronomy Journal, 69(5): 799-802. https://doi.org/10.2134/agronj1977.00021962006900050017x
  24. Uddling, J., J. Gelang-Alfredsson, K. Piikki, and H. Pleijel, 2007. Evaluating the relationship between leaf chlorophyll concentration and SPAD-502 chlorophyll meter readings, Photosynthesis Research, 91(1): 37-46. https://doi.org/10.1007/s11120-006-9077-5
  25. Vidal, M. and J.M. Amigo, 2012. Pre-processing of hyperspectral images, Essential steps before image analysis, Chemometrics and Intelligent Laboratory Systems, 117(1): 138-148. https://doi.org/10.1016/j.chemolab.2012.05.009
  26. Wallihan, E.F., 1973. Portable reflectance meter for estimating chlorophyll concentrations in leaves, Agronomy Journal, 65(4): 659-662. https://doi.org/10.2134/agronj1973.00021962006500040039x
  27. Xue, L. and L. Yang, 2009. Deriving leaf chlorophyll content of green-leafy vegetables from hyperspectral reflectance, Journal of Photogrammetry and Remote Sensing, 64(1): 97-106. https://doi.org/10.1016/j.isprsjprs.2008.06.002