- Email: [email protected]
Smart detection of leaf wilting by 3D image processing and 2D Fourier transform
Computers and Electronics in Agriculture 90 (2013) 68–75
Contents lists available at SciVerse ScienceDirect
Computers and Electronics in Agriculture journal homepage: www.elsevier.com/locate/compag
Smart detection of leaf wilting by 3D image processing and 2D Fourier transform X. Cai a, Y. Sun a,⇑, Y. Zhao b, L. Damerow c, P. Schulze Lammers c, W. Sun a, J. Lin b, L. Zheng a, Y. Tang a a
College of Information & Electrical Engineering, China Agricultural University, 100083 Beijing, China School of Technology, Beijing Forestry University, 100083 Beijing, China c Department of Agricultural Engineering, University of Bonn, 53115 Bonn, Germany b
a r t i c l e
i n f o
Article history: Received 21 June 2012 Received in revised form 19 October 2012 Accepted 13 November 2012
Keywords: 2D Fourier transform (2DFT) 3D image processing Laser scanner Wilting identification Zucchini
a b s t r a c t Wilting is a common symptom in plants responding to drought stress. Early wilting detection is of high importance for crop precision management. However, it is challenging to develop a reliable measurement technology. This study presents a sensing method based on three-dimensional (3D) images that are generated by a laser scanner. A leaf wilting index (LWI2DFT) is defined from the spectrum of a two-dimensional (2D) Fourier transform (FT), in which a leaf is equivalent to a mathematical surface in 3D space, and thus a wilting process refers to a series of the curved surfaces. To verify the effectiveness of the index LWI2DFT, we applied our method to zucchini plants (Cucurbita pepo L.) 17 days after they emerged from the soil. The laboratory experiments included two periods of 10 days each. From six leaves of three plants tested in the first period, it was observed that this index was capable of sensing the stress response of the zucchini from slight wilting to severe stress levels. The regression results between LWI2DFT and ambient temperature (Tair) fitted a linear equation with 0.814 6 R2 6 0.908, and those between LWI2DFT and the photosynthetically active radiation (LPAR) with 0.696 6 R2 6 0.856. In the second period we repeated the measurements with new samples to validate our results. A good correlation (0.609 6 R2 6 0.899 for Tair, 0.748 6 R2 6 0.892 for LPAR) confirmed the reliability of our proposed method. Ó 2012 Elsevier B.V. All rights reserved.
1. Introduction Plants rely on adequate amounts of water to maintain cell turgidity and leaf function. A number of factors, such as insufficient water content in soil, strong solar radiation, high temperature and salinity, may lead to plant wilting. Therefore, plant photosynthesis is reduced because CO2 absorption is reduced as leaf stomata gradually close in response to the environmental stress (Guerrero and Mullet, 1988). In arid and semiarid areas, plant wilting under drought stress is commonly observed, and is suggested as a visual indicator of soil water deficit (Bettina et al., 2007). Because the incipient wilting of most plants can vanish after re-watering at early stage, this indicator is beneficial for precision irrigation. For the identification of plant response to drought stress, diverse techniques have been developed and tested. A conventional method is to measure the leaf water potential using a pressure chamber, a psychrometer, or a cell pressure-probe (Zimmermann et al., 2008). However, these available techniques rely on contact measurements, and they don’t allow for maintaining the leaf freeform in stem. Moreover, they require a thorough understanding of plant–water relations (Bettina et al., 2007). An alternative method is based on the measurement of leaf tempera⇑ Corresponding author. Tel.: +86 10 62737416; fax: +86 10 82377326. E-mail address: [email protected] (Y. Sun). 0168-1699/$ - see front matter Ó 2012 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.compag.2012.11.005
ture when sensible heat on the leaf is not converted into latent heat by evaporative cooling effect of transpiration. Many studies applied infrared thermometry to determine leaf/canopy temperature or stomatal conductance (Hashimoto et al., 1984; Rahkonen and Jokela, 2003; Emekli et al., 2007; Gontia and Tiwari, 2008; Blonquist et al., 2009). Apart from these methods, machine vision has received increasing attention over the past two decades (Seginer et al., 1992; Ahmad and Reid, 1996; Kacira and Ling, 2001; Kacira et al., 2002a, 2002b; Foucher et al., 2004; Mizuno et al., 2007; Yang et al., 2008). A major advantage over measuring the leaf water potential is the non-contact nature. To extract wilting information from plant images, leaf color (Ahmad and Reid, 1996), physical morphology (Foucher et al., 2004), and the combination of both (Mizuno et al., 2007) have been considered. The color detection is relatively simple but physiologic damage for many plants may have already occurred when the leave color is changed. Thus, the morphological information of the entire plant or leaf may be more helpful to detect drought stress at the early wilting stage (Foucher et al., 2004). The detection of wilting is not only a problem associated with plant protection, but also represents an interdisciplinary challenge involving differential geometry, artificial intelligence, and pattern recognition. These efforts have lead to the definition of a number of indices, the crop water stress index (CWSI) and the top-projected canopy area (TPCA) (Kacira et al., 2002a, 2002b), the apparent leaf area index (ALAI) (Foucher et al., 2004) and the angle of leaves region (ALG)
69
X. Cai et al. / Computers and Electronics in Agriculture 90 (2013) 68–75
(Mizuno et al., 2007), which were evaluated under laboratory conditions. In brief, these indices are derived from 2D images, which are commonly acquired by charge coupled device (CCD) video cameras with an array of pixel sensors. When a plant in 3D space is projected onto a plane, it becomes a 2D image, which is equivalent to a contraction mapping of the corresponding function from 3D to 2D according to the theory of functional analysis in mathematics. As a result, the morphological information of plants is incomplete in the 2D images. Thus, Omasa et al. (2007) proposed 3D plant analysis because diverse techniques for the quick acquisition of 3D images, such as laser scanners and time-of-flight cameras, have become commercially available (Oggier et al., 2006; Ehlert et al., 2008). The spread and adaptation of these technologies and methods depend on their costs and applicability, which motivated this case study to identify plant wilting in 3D space. The boundary layer at the plant leaf surface is the interface for transition from liquid water to water vapor lost through the leaf stomata. In contrast to the morphological features of the entire plant, the leaf is more sensitive to drought stress. Additionally, for plants of the same species, there exist plant-to-plant variations in physical morphology due to site-specific field conditions, but they have an identical leaf form. Therefore, the leaf-based wilting index seems more reliable in practical use. In this study, rather than characterizing the morphological features of the entire plant or canopy structure within a field, the primary objective was to detect the leaf-based morphological response to drought stress using 3D scanning measurements.
(a pair of lead-acid batteries (12 V, 7 Ah) connected in series) and its control box was connected to a laptop via an RS-232 port to transmit data or exchange instruction codes. The measurement principle of the TRF (Lee and Shiou, 2011) is shown in Fig. 2. A laser diode projects a light beam onto the surface of the object of interest and a position sensitive detector (PSD) captures the reflected light location. According to this optical design, the relative distance z is determined by
z¼b
x sin a
ð1Þ
where x refers to the location of the image spot, a is the angle between the projected laser beam and the reflected light, and ß (dimensionless coefficient) is related to the inherent structure of the sensor. Detailed description of this sensor technique can be found in the literatures (Ehlert et al., 2008; Lee and Shiou, 2011). 2.2. Definition of the leaf wilting index
2. Materials and methods
In this study, assuming that a plant leaf without wilting appears as a planar surface, and the leaves with more or less bending/ crimping shapes refer to the wilting at different levels, the wilting dynamics of each leaf can be mathematically represented as a sequence of geometrical surfaces or functions (f (x, y)) in 3D space or represented as a stochastic event (f (x(t), y(t), t)) of a time series in four-dimensional (4D) space. Furthermore, it is presumed that the curvature at each point of f (x, y) changes as the leaf wilting progresses. Under these assumptions we perform a 2D Fourier transform on the function f (x, y) (which is a non-periodic and continuous function):
2.1. Laser scanner description
Fðu; vÞ ¼
Z Z
1
f ðx; yÞeJ2pðuxþvyÞ dxdy
ð2Þ
1
The laser scanner (Fig. 1) included a triangulation range finder (TRF) (M7L/400, resolution: ±1 mm, MEL Mikroelektronik GmbH, Germany), which was separately driven by two stepper-motors positioned along the x-axis (0–1.5 m, adjustable) or the y-axis (0–0.5 m, adjustable) of an aluminum frame. Within the x-y plane, the scanning span along each axis could be adjusted between 1 mm and 100 mm through the programmed parameters. For the performed experiment, the scanning span was set to 2 mm with a scanning velocity of 10 mm s1. The system operated at 24 VDC
where u and v are spatial frequencies along x and y directions, respectively, and F(u, v) is the 2D spectrum of f (x, y). Then a leaf wilting index (LWI2DFT) is defined as:
Laser diode
x
Lens
PSD
Lens Light beam Optical axis Z α z
Base line Y X
Fig. 1. The experimental scene of using a laser scanner for capturing 3D plants image.
Fig. 2. Measurement principle of the laser triangulation range finder.
70
X. Cai et al. / Computers and Electronics in Agriculture 90 (2013) 68–75
Table 1 The water loss of each pot in the two representative days of period-1 (May 24, a sunny day and May 27, a cloudy day). Plant
Diurnal water loss of experiment (g)
Plant-1 Plant-2 Plant-3
LWI2DFT ¼
A0 þ
A0 Pn
i¼1 Ai
May 24 (sunny)
May 27 (cloudy)
30.8 141.2 475.4
21.7 90.3 315.5
100
0.15 mg mg1, silt 0.67 mg mg1, and clay 0.18 mg mg1 (USDA Standard). Two weeks after young plants emerged out of the soil, we began to control the SWCs of these pots at three levels (relatively dry: 0.06–0.08 kg kg1, moderately dry: 0.17–0.20 kg kg1 and wet: 0.30–0.32 kg kg1) of mass water content. In order to daily monitor the variations of the SWC, each pot was weighted in the morning and evening so that the diurnal water loss were equally replenished in time. In the following discussion, plant-1 refers to that in the relatively dry soil, plant-2 to the one in moderately dry soil, and plant-3 to the one in wet soil.
ð3Þ
where Ao refers to the amplitude of the direct element and Ai to that of the i-th alternative element of F(u, v). In this study, Ao characterizes the planar degree of the leaf, whereas Ai reflects the curved or wrinkled degree of the same leaf. Thus, according to Eq. (3), the index LWI2DFT will decrease as the leaf wilting increases, and vice versa. 2.3. Experimental material and preparation For this proof-of-principle study, zucchini (Cucurbita pepo L.) was chosen as the experimental plant because: (i) it is sensitive to variations of solar radiation, ambient temperature (Tair) and soil water content (SWC); (ii) its leaves in full turgidity appear quite flat compared to their wilting shapes. These zucchinis were individually sown in pots (8 L) on April, 25, 2011 in a greenhouse. The textural compositions of the soil in the pots were: sand
2.4. Experimental procedure For these plant samples, the third (leaf-3) and the fourth (leaf-4) leaves from each plant were chosen as leaf samples, and each leaf sample was scanned periodically from 8:00 to 18:30. Meanwhile, we measured solar radiation with a pyranometer (AV-19Q, 0–3000 lmol m2 s1, accuracy: ±3%), Tair (EXTECH-392065, 50 to 150 °C, resolution: ±0.1 °C) and ambient relative humidity (RH) (PHQS, 0–100% RH, accuracy: ±3% RH) at an interval of 10 min. The Data Analysis of Matlab 7.12.0 was directly employed for the 2DFT spectrum analysis. The entire experiment consisted of two periods (period-1: May 21–30, 2011; period-2: June 4–13, 2011). The data of period-2 were used to validate those of period-1. For each period, three samples (i.e. plant-1, -2, -3) with three levels of controlled SWC were prepared. That is, six plant samples in total were tested.
Fig. 3. (A) The hour-scaled dynamics of LPAR and Tair in a sunny day of period-1 (24 May, 2011, mean of RH: 42.3%, mean of Tair: 36.5 °C, mean of LPAR: 429.2 lmol m2 s1). (B) Time series of leaf-3 LWI2DFT of the three plants at three levels of SWC (dry, moderate, wet soil) on 24 May, 2011. (C) Time series of leaf-4 LWI2DFT of the three plants at the three levels of SWC on 24 May, 2011.
X. Cai et al. / Computers and Electronics in Agriculture 90 (2013) 68–75
71
Fig. 4. Three wilting levels of leaf-4 from plant-2 (in the sunny day, May 24, 2011) are represented by 2D, 3D images and the spectrums of 2DFT.
3. Results and discussion 3.1. Temporary variations of LWI2DFT and 2DFT interpretation To exemplify the sensitivity of LWI2DFT, we present two sets of experimental results as representatives from two typical days (a sunny day and a cloudy day) in period-1. Table 1 lists the water loss of each pot during these days. Fig. 3 shows the temporary traces of Tair, LPAR and LWI2DFT (Fig. 3b and c refers to LWI2DFT of the third and that of the fourth leaf in each plant, respectively) over
time on the sunny day (May 24, 2011). It is clear to see that no significant variation of LWI2DFT for each plant was observed until Tair increased to 33 °C and LPAR up to 278 lmol m2 s1. That is, the wilting symptom for all samples did not appear in the morning. Fig. 4A provides the additional evidences with 2D and 3D images (obtained at 8:30) of the fourth leaf of plant-2 to illustrate this state. The values of each LWI2DFT began going down around 11:00, indicating that the initial wilting appeared. The slight wilting images of plant-2 in Fig. 4B confirmed this trend and implied that the leaf stomata began closing. The values of LWI2DFT reached
72
X. Cai et al. / Computers and Electronics in Agriculture 90 (2013) 68–75
Fig. 5. (A) The hour-scaled dynamics of LPAR and Tair in a cloudy day of period-1 (27 May, 2011, mean of RH: 52.1%, mean of Tair: 34.6 °C, mean of LPAR: 245.8 lmol m2 s1). (B) Time series of leaf-3 LWI2DFT of the three plants at the three levels of SWC on 27 May, 2011. (C) Time series of leaf-4 LWI2DFT of the three plants at the three levels of SWC on 27 May, 2011.
Fig. 7. Regressions associating the leaf-4 LWI2DFT with Tair for each plant in the sunny day (May 24) of period-1.
Fig. 6. Tair, LPAR and the daily LWI2DFT variations of leaf-4 between 15:00 and 15:30 for all plants throughout period-1 (May 21– 30, 2011).
a minimum during 15:00 – 15:30, indicating that the wilting of each plant became severe, which also agreed with the 2D and 3D images of plant-2 obtained at this time (Fig. 4C). During the late
afternoon hours (between 16:00 and 17:00), the gradual increase of LWI2DFT revealed that the wilting symptom of each plant relieved as Tair declined below 35 °C. Before 17:30, LWI2DFT of all plants returned to the highest values, indicating that these plants completely recovered from their severely wilted state. More significant results can be found from the 2DFT spectra in Fig. 4. The direct element (standing at center) monotonously decreased as the leaf wilting increased. That is, the trend of LWI2DFT varied as we expected by decomposing the spectra of 2DFT in Eq. (3). Alternative responses of these plants to the variations of Tair and LPAR on the cloudy day (May 27, 2011) are shown in Fig. 5. The temporary trace of LWI2DFT for plant-1 indicated that its wilting occurred between 12:00 and 17:00. The lowest value of LWI2DFT for plant-1 was around 15:30 at Tair=36.1 °C and LPAR = 418.52 lmol m2 s1. In contrast, plant-2 appeared slightly wilted around 15:30 but plant-3 maintained wilt-less throughout this day. Clearly, the morphological responses of these plants agreed to the differences of SWC in their pots.
X. Cai et al. / Computers and Electronics in Agriculture 90 (2013) 68–75
73
statistically fitted linear approximations. The values of R2 (0.814 6 R2 6 0.908) in Fig. 7 were slightly higher than those (0.696 6 R2 6 0.856) in Fig. 8. This may result from fluctuations of the solar radiation in nature, which is more susceptible to atmospheric interference. Despite the existence of hysteresis effect (the wilting development could not catch up with the variations of Tair and LPAR inphase), the good correlations prove that LWI2DFT was sensitive to either Tair or LPAR for these plants.
3.3. Validation of LWI2DFT in period-2 For validation purpose, measurements were repeated with new plants in period-2. To make the results of period-2 comparable with those of period-1 as shown in Fig. 3, we chose a set of data acquired on June 6 in period-2 because the temporary variations of Tair and LPAR (Fig. 9A) of this day were quite similar to those (Fig. 3A) on May 24 in period-1. Comparing Fig. 3B and C to Fig. 9B and C, the temporally wilting processes of the leaves monitored were similar as well. These subfigures illustrate that the severe wilting occurred around 15:30 and disappeared around 17:30. However, the initial wilting for all samples on May 24 occurred around 11:00, whereas those on June 6 did not appear until 13:00. This can be explained from the traces of Tair, which increased to 35 °C around 11:00 on May 24 (Fig. 3A), but to this level not before 13:00 on June 6 (Fig. 9A). In order to validate the applicability of the regression equations presented in Fig. 7, each subfigure of Fig. 10 shows a bisector (1:1) to
Fig. 8. Regressions associating the leaf-4 LWI2DFT with LPAR for each plant in the sunny day (May 24) of period-1.
Since both Fig. 3 (the sunny day) and Fig. 5 (the cloudy day) show that the lowest values of‘ LWI2DFT were reached during 15:00–15:30, we provide Fig. 6 to show the wilting states of these samples (Fig. 6b) conjoined with the variations of Tair and LPAR (Fig. 6a) at this time throughout period-1. The minimum of LWI2DFT (=0.246) related to the maximal Tair (=42.1 °C) on May 24, but no wilting symptoms were perceived on May 22, 26 and 30. In general, by means of LWI2DFT the different wilting states of these samples were well characterized. 3.2. Statistical analysis relating LWI2DFT to Tair or LPAR The three subfigures of Fig. 7 provide the statistical results relating LWI2DFT to Tair for the fourth leaf of each sample monitored on May 24. Fig. 8 provides similar results of the fourth leaf associating LWI2DFT with LPAR. Both figures show that these results
Fig. 9. (A) The hour-scaled dynamics of LPAR and Tair in the sunny day of period-2 (6 June, 2011, mean of RH: 43.2%, mean of Tair: 35.0 °C, mean of LPAR: 320.9 lmolm2s1). (B) Time series of leaf-3 LWI2DFT at the three levels of SWC on 6 June, 2011. (C) Time series of leaf-4 LWI2DFT at the three levels of SWC on 6 June, 2011.
74
X. Cai et al. / Computers and Electronics in Agriculture 90 (2013) 68–75
Fig. 11. Statistical comparisons between the determined LWI2DFT and that predicated through LPAR measurements on 6 June, 2011. Fig. 10. Statistical comparisons between the determined LWI2DFT and that predicated through Tair measurements on 6 June, 2011.
compare the LWI2DFT values determined by Eq. (3) with‘ the results predicated through the Tair measurements and this regression equation. Likewise, Fig. 11 compares the same data of LWI2DFT with the results predicated through the LPAR measurements and the regression equations presented in Fig. 8. The levels of RMSE and R2 in Figs. 10 and 11 suggest that most of the regression equations have good accuracies for predicating LWI2DFT. Similar to the results obtained in period-1 (see Fig. 6), Fig. 12 shows that in period-2 the wilting of all samples occurred between 15:00 and 15:30. In contrast, the severe wilting in period-2 appeared relatively frequent. This can be indirectly explained when the averaged day temperatures (8:00–20:00) are compared (period-1: 33.6 °C; period-2: 36.5 °C).
4. Summary and conclusions Using the proposed method, the temporary wilting of zucchini leaves has been satisfactorily identified from the 3D images laser-scanned and by a wilting index defined from 2D Fourier transform. Because the 2DFT spectrum-based index mainly depends on the curvatures of the scanned points distributed on the leaf surface,
Fig. 12. Tair, LPAR and the daily LWI2DFT variations of leaf-4 between 15:00 and 15:30 for all plants throughout period-2 (June 4 – 13, 2011).
X. Cai et al. / Computers and Electronics in Agriculture 90 (2013) 68–75
the interference of leaf inclination has been effectively reduced, and thus the smart identification of early/slightly wilting is feasible. Additionally, this index is also promising as a wilting indicator extended to other plants of Cucurbita pepo, which have the leaf shapes and the responses to drought stress similar to those of zucchini. On the other hand, due to the diversity of plants with miscellaneous shapes of leaf and wilting morphologies, one can not expect a universal index, such like Eq. (3) in this study, suitable to identify the wilting for all species. Nevertheless, this proof-ofprinciple study provides a novel application of 2DFT and artificial intelligence in sensing plants, which enlightens us to explore different mathematical-tools for extracting the physiologic information of leaf morphology for various plants in 3D space. Acknowledgments We wish to thank (i) the Chinese-German Center for Scientific Promotion (Chinesisches-Deutsches Zentrum fuer Wissenschaftsfoerderung) under the project of Sino-German Research Group (GZ494), (ii) the National Nature Science Foundation of China (31171458) (iii) the international cooperation fund of Ministry of Science and Technology, China (2010DFA34670) for their financial support, and (iv) The specific fund of China Agricultural University for Ph.D. students innovation. References Ahmad, I.S., Reid, J.F., 1996. Evaluation of color representations for maize images. Journal of Agricultural Engineering Research 63, 185–196. Bettina, M.J.E., Melvin, T.T., T, A.K., 2007. Visual assessment of wilting as a measure of leaf water potential and seedling drought survival. Journal of Tropical Ecology 23, 497–500. Blonquist, J.J.M., Norman, J.M., Bugbee, B., 2009. Automated measurement of canopy stomatal conductance based on infrared temperature. Agricultural and Forest Meteorology 149, 1931–1945. Ehlert, D., Horn, H., Adamek, R., 2008. Measuring crop biomass density by laser triangulation. Computers and Electronics in Agriculture 61, 117–125.
75
Emekli, Y., Bastug, R., Buyuktas, D., Emekli, N.Y., 2007. Evaluation of a crop water stress index for irrigation scheduling of bermudagrass. Agricultural Water Management 90, 205–212. Foucher, P., Revollon, P., Vigouroux, B., Chassériaux, G., 2004. Morphological image analysis for the detection of water stress in potted Forsythia. Biosystems Engineering 89, 131–138. Guerrero, F.D., Mullet, J.E., 1988. Reduction of turgor induces rapid changes in leaf translatable RNA. Plant Physiology 88, 401–408. Gontia, N.K., Tiwari, K.N., 2008. Development of crop water stress index of wheat crop for scheduling irrigation using infrared thermometry. Agricultural Water Management 95, 1144–1152. Hashimoto, Y., Ino, T., Kramer, P.J., Naylor, A.W., Strain, B.R., 1984. Dynamicanalysis of water stress of sunflower leaves by means of thermal image processing system. Plant Physiology 76, 266–269. Kacira, M., Ling, P.P., 2001. Design and development of an automated and noncontact sensing system for continuous monitoring of plant health and growth. Transactions of the ASAE 44, 989–996. Kacira, M., Ling, P.P., Short, T.H., 2002a. Establishing crop water stress index (CWSI) threshold values for early non-contact detection of plant water stress. Transactions of the ASAE 45, 775–780. Kacira, M., Ling, P.P., Short, T.H., 2002b. Machine vision extracted plant movement for early detection of plant water stress. Transactions of the ASAE 45, 1147– 1153. Lee, R.T., Shiou, F.J., 2011. Multi-beam laser probe for measuring position and orientation of freeform surface. Measurement 44, 1–10. Mizuno, S., Noda, K., Ezaki, N., Takizawa, H., Yamamoto, S., 2007. Detection of wilt by analyzing color and stereo vision data of plant. Lecture notes in computer science, Springer-Verlag, vol. 4418, pp. 400–411. Oggier, T., Lustenberger, F., Blanc, N., 2006. Miniature 3D TOF camera for real-time imaging. Lecture notes in computer science, Springer-Verlag, vol. 4021, pp. 212–216. Omasa, K., Hosoi, F., Konish, A., 2007. 3D lidar imaging for detecting and understanding plant responses and canopy structure. Journal of Experimental Botany 58, 881–898. Rahkonen, J., Jokela, H., 2003. Infrared radiometry for measuring plant leaf temperature during thermal weed control treatment. Biosystems Engineering 86, 257–266. Seginer, I., Elster, R.T., Goodrum, J.W., Rieger, M.W., 1992. Plant wilt detection by computer–vision tracking of leaf tips. Transactions of the ASAE 35, 1563–1567. Yang, Y., Ling, P., Fleisher, D.H., Timlin, D.J., Reddy, V.R., 2008. Non-contacting techniques for plant drought stress detection. Transactions of the ASAE 51, 1483–1492. Zimmermann, D., Reuss, R., Westhoff, M., Geßner, P., Bauer, W., Bamberg, E., Bentrup, F.W., Zimmermann, U., 2008. A novel, non-invasive, onlinemonitoring, versatile and easy plant-based probe for measuring leaf water status. Journal of Experimental Botany 59, 3157–3167.
Contents lists available at SciVerse ScienceDirect
Computers and Electronics in Agriculture journal homepage: www.elsevier.com/locate/compag
Smart detection of leaf wilting by 3D image processing and 2D Fourier transform X. Cai a, Y. Sun a,⇑, Y. Zhao b, L. Damerow c, P. Schulze Lammers c, W. Sun a, J. Lin b, L. Zheng a, Y. Tang a a
College of Information & Electrical Engineering, China Agricultural University, 100083 Beijing, China School of Technology, Beijing Forestry University, 100083 Beijing, China c Department of Agricultural Engineering, University of Bonn, 53115 Bonn, Germany b
a r t i c l e
i n f o
Article history: Received 21 June 2012 Received in revised form 19 October 2012 Accepted 13 November 2012
Keywords: 2D Fourier transform (2DFT) 3D image processing Laser scanner Wilting identification Zucchini
a b s t r a c t Wilting is a common symptom in plants responding to drought stress. Early wilting detection is of high importance for crop precision management. However, it is challenging to develop a reliable measurement technology. This study presents a sensing method based on three-dimensional (3D) images that are generated by a laser scanner. A leaf wilting index (LWI2DFT) is defined from the spectrum of a two-dimensional (2D) Fourier transform (FT), in which a leaf is equivalent to a mathematical surface in 3D space, and thus a wilting process refers to a series of the curved surfaces. To verify the effectiveness of the index LWI2DFT, we applied our method to zucchini plants (Cucurbita pepo L.) 17 days after they emerged from the soil. The laboratory experiments included two periods of 10 days each. From six leaves of three plants tested in the first period, it was observed that this index was capable of sensing the stress response of the zucchini from slight wilting to severe stress levels. The regression results between LWI2DFT and ambient temperature (Tair) fitted a linear equation with 0.814 6 R2 6 0.908, and those between LWI2DFT and the photosynthetically active radiation (LPAR) with 0.696 6 R2 6 0.856. In the second period we repeated the measurements with new samples to validate our results. A good correlation (0.609 6 R2 6 0.899 for Tair, 0.748 6 R2 6 0.892 for LPAR) confirmed the reliability of our proposed method. Ó 2012 Elsevier B.V. All rights reserved.
1. Introduction Plants rely on adequate amounts of water to maintain cell turgidity and leaf function. A number of factors, such as insufficient water content in soil, strong solar radiation, high temperature and salinity, may lead to plant wilting. Therefore, plant photosynthesis is reduced because CO2 absorption is reduced as leaf stomata gradually close in response to the environmental stress (Guerrero and Mullet, 1988). In arid and semiarid areas, plant wilting under drought stress is commonly observed, and is suggested as a visual indicator of soil water deficit (Bettina et al., 2007). Because the incipient wilting of most plants can vanish after re-watering at early stage, this indicator is beneficial for precision irrigation. For the identification of plant response to drought stress, diverse techniques have been developed and tested. A conventional method is to measure the leaf water potential using a pressure chamber, a psychrometer, or a cell pressure-probe (Zimmermann et al., 2008). However, these available techniques rely on contact measurements, and they don’t allow for maintaining the leaf freeform in stem. Moreover, they require a thorough understanding of plant–water relations (Bettina et al., 2007). An alternative method is based on the measurement of leaf tempera⇑ Corresponding author. Tel.: +86 10 62737416; fax: +86 10 82377326. E-mail address: [email protected] (Y. Sun). 0168-1699/$ - see front matter Ó 2012 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.compag.2012.11.005
ture when sensible heat on the leaf is not converted into latent heat by evaporative cooling effect of transpiration. Many studies applied infrared thermometry to determine leaf/canopy temperature or stomatal conductance (Hashimoto et al., 1984; Rahkonen and Jokela, 2003; Emekli et al., 2007; Gontia and Tiwari, 2008; Blonquist et al., 2009). Apart from these methods, machine vision has received increasing attention over the past two decades (Seginer et al., 1992; Ahmad and Reid, 1996; Kacira and Ling, 2001; Kacira et al., 2002a, 2002b; Foucher et al., 2004; Mizuno et al., 2007; Yang et al., 2008). A major advantage over measuring the leaf water potential is the non-contact nature. To extract wilting information from plant images, leaf color (Ahmad and Reid, 1996), physical morphology (Foucher et al., 2004), and the combination of both (Mizuno et al., 2007) have been considered. The color detection is relatively simple but physiologic damage for many plants may have already occurred when the leave color is changed. Thus, the morphological information of the entire plant or leaf may be more helpful to detect drought stress at the early wilting stage (Foucher et al., 2004). The detection of wilting is not only a problem associated with plant protection, but also represents an interdisciplinary challenge involving differential geometry, artificial intelligence, and pattern recognition. These efforts have lead to the definition of a number of indices, the crop water stress index (CWSI) and the top-projected canopy area (TPCA) (Kacira et al., 2002a, 2002b), the apparent leaf area index (ALAI) (Foucher et al., 2004) and the angle of leaves region (ALG)
69
X. Cai et al. / Computers and Electronics in Agriculture 90 (2013) 68–75
(Mizuno et al., 2007), which were evaluated under laboratory conditions. In brief, these indices are derived from 2D images, which are commonly acquired by charge coupled device (CCD) video cameras with an array of pixel sensors. When a plant in 3D space is projected onto a plane, it becomes a 2D image, which is equivalent to a contraction mapping of the corresponding function from 3D to 2D according to the theory of functional analysis in mathematics. As a result, the morphological information of plants is incomplete in the 2D images. Thus, Omasa et al. (2007) proposed 3D plant analysis because diverse techniques for the quick acquisition of 3D images, such as laser scanners and time-of-flight cameras, have become commercially available (Oggier et al., 2006; Ehlert et al., 2008). The spread and adaptation of these technologies and methods depend on their costs and applicability, which motivated this case study to identify plant wilting in 3D space. The boundary layer at the plant leaf surface is the interface for transition from liquid water to water vapor lost through the leaf stomata. In contrast to the morphological features of the entire plant, the leaf is more sensitive to drought stress. Additionally, for plants of the same species, there exist plant-to-plant variations in physical morphology due to site-specific field conditions, but they have an identical leaf form. Therefore, the leaf-based wilting index seems more reliable in practical use. In this study, rather than characterizing the morphological features of the entire plant or canopy structure within a field, the primary objective was to detect the leaf-based morphological response to drought stress using 3D scanning measurements.
(a pair of lead-acid batteries (12 V, 7 Ah) connected in series) and its control box was connected to a laptop via an RS-232 port to transmit data or exchange instruction codes. The measurement principle of the TRF (Lee and Shiou, 2011) is shown in Fig. 2. A laser diode projects a light beam onto the surface of the object of interest and a position sensitive detector (PSD) captures the reflected light location. According to this optical design, the relative distance z is determined by
z¼b
x sin a
ð1Þ
where x refers to the location of the image spot, a is the angle between the projected laser beam and the reflected light, and ß (dimensionless coefficient) is related to the inherent structure of the sensor. Detailed description of this sensor technique can be found in the literatures (Ehlert et al., 2008; Lee and Shiou, 2011). 2.2. Definition of the leaf wilting index
2. Materials and methods
In this study, assuming that a plant leaf without wilting appears as a planar surface, and the leaves with more or less bending/ crimping shapes refer to the wilting at different levels, the wilting dynamics of each leaf can be mathematically represented as a sequence of geometrical surfaces or functions (f (x, y)) in 3D space or represented as a stochastic event (f (x(t), y(t), t)) of a time series in four-dimensional (4D) space. Furthermore, it is presumed that the curvature at each point of f (x, y) changes as the leaf wilting progresses. Under these assumptions we perform a 2D Fourier transform on the function f (x, y) (which is a non-periodic and continuous function):
2.1. Laser scanner description
Fðu; vÞ ¼
Z Z
1
f ðx; yÞeJ2pðuxþvyÞ dxdy
ð2Þ
1
The laser scanner (Fig. 1) included a triangulation range finder (TRF) (M7L/400, resolution: ±1 mm, MEL Mikroelektronik GmbH, Germany), which was separately driven by two stepper-motors positioned along the x-axis (0–1.5 m, adjustable) or the y-axis (0–0.5 m, adjustable) of an aluminum frame. Within the x-y plane, the scanning span along each axis could be adjusted between 1 mm and 100 mm through the programmed parameters. For the performed experiment, the scanning span was set to 2 mm with a scanning velocity of 10 mm s1. The system operated at 24 VDC
where u and v are spatial frequencies along x and y directions, respectively, and F(u, v) is the 2D spectrum of f (x, y). Then a leaf wilting index (LWI2DFT) is defined as:
Laser diode
x
Lens
PSD
Lens Light beam Optical axis Z α z
Base line Y X
Fig. 1. The experimental scene of using a laser scanner for capturing 3D plants image.
Fig. 2. Measurement principle of the laser triangulation range finder.
70
X. Cai et al. / Computers and Electronics in Agriculture 90 (2013) 68–75
Table 1 The water loss of each pot in the two representative days of period-1 (May 24, a sunny day and May 27, a cloudy day). Plant
Diurnal water loss of experiment (g)
Plant-1 Plant-2 Plant-3
LWI2DFT ¼
A0 þ
A0 Pn
i¼1 Ai
May 24 (sunny)
May 27 (cloudy)
30.8 141.2 475.4
21.7 90.3 315.5
100
0.15 mg mg1, silt 0.67 mg mg1, and clay 0.18 mg mg1 (USDA Standard). Two weeks after young plants emerged out of the soil, we began to control the SWCs of these pots at three levels (relatively dry: 0.06–0.08 kg kg1, moderately dry: 0.17–0.20 kg kg1 and wet: 0.30–0.32 kg kg1) of mass water content. In order to daily monitor the variations of the SWC, each pot was weighted in the morning and evening so that the diurnal water loss were equally replenished in time. In the following discussion, plant-1 refers to that in the relatively dry soil, plant-2 to the one in moderately dry soil, and plant-3 to the one in wet soil.
ð3Þ
where Ao refers to the amplitude of the direct element and Ai to that of the i-th alternative element of F(u, v). In this study, Ao characterizes the planar degree of the leaf, whereas Ai reflects the curved or wrinkled degree of the same leaf. Thus, according to Eq. (3), the index LWI2DFT will decrease as the leaf wilting increases, and vice versa. 2.3. Experimental material and preparation For this proof-of-principle study, zucchini (Cucurbita pepo L.) was chosen as the experimental plant because: (i) it is sensitive to variations of solar radiation, ambient temperature (Tair) and soil water content (SWC); (ii) its leaves in full turgidity appear quite flat compared to their wilting shapes. These zucchinis were individually sown in pots (8 L) on April, 25, 2011 in a greenhouse. The textural compositions of the soil in the pots were: sand
2.4. Experimental procedure For these plant samples, the third (leaf-3) and the fourth (leaf-4) leaves from each plant were chosen as leaf samples, and each leaf sample was scanned periodically from 8:00 to 18:30. Meanwhile, we measured solar radiation with a pyranometer (AV-19Q, 0–3000 lmol m2 s1, accuracy: ±3%), Tair (EXTECH-392065, 50 to 150 °C, resolution: ±0.1 °C) and ambient relative humidity (RH) (PHQS, 0–100% RH, accuracy: ±3% RH) at an interval of 10 min. The Data Analysis of Matlab 7.12.0 was directly employed for the 2DFT spectrum analysis. The entire experiment consisted of two periods (period-1: May 21–30, 2011; period-2: June 4–13, 2011). The data of period-2 were used to validate those of period-1. For each period, three samples (i.e. plant-1, -2, -3) with three levels of controlled SWC were prepared. That is, six plant samples in total were tested.
Fig. 3. (A) The hour-scaled dynamics of LPAR and Tair in a sunny day of period-1 (24 May, 2011, mean of RH: 42.3%, mean of Tair: 36.5 °C, mean of LPAR: 429.2 lmol m2 s1). (B) Time series of leaf-3 LWI2DFT of the three plants at three levels of SWC (dry, moderate, wet soil) on 24 May, 2011. (C) Time series of leaf-4 LWI2DFT of the three plants at the three levels of SWC on 24 May, 2011.
X. Cai et al. / Computers and Electronics in Agriculture 90 (2013) 68–75
71
Fig. 4. Three wilting levels of leaf-4 from plant-2 (in the sunny day, May 24, 2011) are represented by 2D, 3D images and the spectrums of 2DFT.
3. Results and discussion 3.1. Temporary variations of LWI2DFT and 2DFT interpretation To exemplify the sensitivity of LWI2DFT, we present two sets of experimental results as representatives from two typical days (a sunny day and a cloudy day) in period-1. Table 1 lists the water loss of each pot during these days. Fig. 3 shows the temporary traces of Tair, LPAR and LWI2DFT (Fig. 3b and c refers to LWI2DFT of the third and that of the fourth leaf in each plant, respectively) over
time on the sunny day (May 24, 2011). It is clear to see that no significant variation of LWI2DFT for each plant was observed until Tair increased to 33 °C and LPAR up to 278 lmol m2 s1. That is, the wilting symptom for all samples did not appear in the morning. Fig. 4A provides the additional evidences with 2D and 3D images (obtained at 8:30) of the fourth leaf of plant-2 to illustrate this state. The values of each LWI2DFT began going down around 11:00, indicating that the initial wilting appeared. The slight wilting images of plant-2 in Fig. 4B confirmed this trend and implied that the leaf stomata began closing. The values of LWI2DFT reached
72
X. Cai et al. / Computers and Electronics in Agriculture 90 (2013) 68–75
Fig. 5. (A) The hour-scaled dynamics of LPAR and Tair in a cloudy day of period-1 (27 May, 2011, mean of RH: 52.1%, mean of Tair: 34.6 °C, mean of LPAR: 245.8 lmol m2 s1). (B) Time series of leaf-3 LWI2DFT of the three plants at the three levels of SWC on 27 May, 2011. (C) Time series of leaf-4 LWI2DFT of the three plants at the three levels of SWC on 27 May, 2011.
Fig. 7. Regressions associating the leaf-4 LWI2DFT with Tair for each plant in the sunny day (May 24) of period-1.
Fig. 6. Tair, LPAR and the daily LWI2DFT variations of leaf-4 between 15:00 and 15:30 for all plants throughout period-1 (May 21– 30, 2011).
a minimum during 15:00 – 15:30, indicating that the wilting of each plant became severe, which also agreed with the 2D and 3D images of plant-2 obtained at this time (Fig. 4C). During the late
afternoon hours (between 16:00 and 17:00), the gradual increase of LWI2DFT revealed that the wilting symptom of each plant relieved as Tair declined below 35 °C. Before 17:30, LWI2DFT of all plants returned to the highest values, indicating that these plants completely recovered from their severely wilted state. More significant results can be found from the 2DFT spectra in Fig. 4. The direct element (standing at center) monotonously decreased as the leaf wilting increased. That is, the trend of LWI2DFT varied as we expected by decomposing the spectra of 2DFT in Eq. (3). Alternative responses of these plants to the variations of Tair and LPAR on the cloudy day (May 27, 2011) are shown in Fig. 5. The temporary trace of LWI2DFT for plant-1 indicated that its wilting occurred between 12:00 and 17:00. The lowest value of LWI2DFT for plant-1 was around 15:30 at Tair=36.1 °C and LPAR = 418.52 lmol m2 s1. In contrast, plant-2 appeared slightly wilted around 15:30 but plant-3 maintained wilt-less throughout this day. Clearly, the morphological responses of these plants agreed to the differences of SWC in their pots.
X. Cai et al. / Computers and Electronics in Agriculture 90 (2013) 68–75
73
statistically fitted linear approximations. The values of R2 (0.814 6 R2 6 0.908) in Fig. 7 were slightly higher than those (0.696 6 R2 6 0.856) in Fig. 8. This may result from fluctuations of the solar radiation in nature, which is more susceptible to atmospheric interference. Despite the existence of hysteresis effect (the wilting development could not catch up with the variations of Tair and LPAR inphase), the good correlations prove that LWI2DFT was sensitive to either Tair or LPAR for these plants.
3.3. Validation of LWI2DFT in period-2 For validation purpose, measurements were repeated with new plants in period-2. To make the results of period-2 comparable with those of period-1 as shown in Fig. 3, we chose a set of data acquired on June 6 in period-2 because the temporary variations of Tair and LPAR (Fig. 9A) of this day were quite similar to those (Fig. 3A) on May 24 in period-1. Comparing Fig. 3B and C to Fig. 9B and C, the temporally wilting processes of the leaves monitored were similar as well. These subfigures illustrate that the severe wilting occurred around 15:30 and disappeared around 17:30. However, the initial wilting for all samples on May 24 occurred around 11:00, whereas those on June 6 did not appear until 13:00. This can be explained from the traces of Tair, which increased to 35 °C around 11:00 on May 24 (Fig. 3A), but to this level not before 13:00 on June 6 (Fig. 9A). In order to validate the applicability of the regression equations presented in Fig. 7, each subfigure of Fig. 10 shows a bisector (1:1) to
Fig. 8. Regressions associating the leaf-4 LWI2DFT with LPAR for each plant in the sunny day (May 24) of period-1.
Since both Fig. 3 (the sunny day) and Fig. 5 (the cloudy day) show that the lowest values of‘ LWI2DFT were reached during 15:00–15:30, we provide Fig. 6 to show the wilting states of these samples (Fig. 6b) conjoined with the variations of Tair and LPAR (Fig. 6a) at this time throughout period-1. The minimum of LWI2DFT (=0.246) related to the maximal Tair (=42.1 °C) on May 24, but no wilting symptoms were perceived on May 22, 26 and 30. In general, by means of LWI2DFT the different wilting states of these samples were well characterized. 3.2. Statistical analysis relating LWI2DFT to Tair or LPAR The three subfigures of Fig. 7 provide the statistical results relating LWI2DFT to Tair for the fourth leaf of each sample monitored on May 24. Fig. 8 provides similar results of the fourth leaf associating LWI2DFT with LPAR. Both figures show that these results
Fig. 9. (A) The hour-scaled dynamics of LPAR and Tair in the sunny day of period-2 (6 June, 2011, mean of RH: 43.2%, mean of Tair: 35.0 °C, mean of LPAR: 320.9 lmolm2s1). (B) Time series of leaf-3 LWI2DFT at the three levels of SWC on 6 June, 2011. (C) Time series of leaf-4 LWI2DFT at the three levels of SWC on 6 June, 2011.
74
X. Cai et al. / Computers and Electronics in Agriculture 90 (2013) 68–75
Fig. 11. Statistical comparisons between the determined LWI2DFT and that predicated through LPAR measurements on 6 June, 2011. Fig. 10. Statistical comparisons between the determined LWI2DFT and that predicated through Tair measurements on 6 June, 2011.
compare the LWI2DFT values determined by Eq. (3) with‘ the results predicated through the Tair measurements and this regression equation. Likewise, Fig. 11 compares the same data of LWI2DFT with the results predicated through the LPAR measurements and the regression equations presented in Fig. 8. The levels of RMSE and R2 in Figs. 10 and 11 suggest that most of the regression equations have good accuracies for predicating LWI2DFT. Similar to the results obtained in period-1 (see Fig. 6), Fig. 12 shows that in period-2 the wilting of all samples occurred between 15:00 and 15:30. In contrast, the severe wilting in period-2 appeared relatively frequent. This can be indirectly explained when the averaged day temperatures (8:00–20:00) are compared (period-1: 33.6 °C; period-2: 36.5 °C).
4. Summary and conclusions Using the proposed method, the temporary wilting of zucchini leaves has been satisfactorily identified from the 3D images laser-scanned and by a wilting index defined from 2D Fourier transform. Because the 2DFT spectrum-based index mainly depends on the curvatures of the scanned points distributed on the leaf surface,
Fig. 12. Tair, LPAR and the daily LWI2DFT variations of leaf-4 between 15:00 and 15:30 for all plants throughout period-2 (June 4 – 13, 2011).
X. Cai et al. / Computers and Electronics in Agriculture 90 (2013) 68–75
the interference of leaf inclination has been effectively reduced, and thus the smart identification of early/slightly wilting is feasible. Additionally, this index is also promising as a wilting indicator extended to other plants of Cucurbita pepo, which have the leaf shapes and the responses to drought stress similar to those of zucchini. On the other hand, due to the diversity of plants with miscellaneous shapes of leaf and wilting morphologies, one can not expect a universal index, such like Eq. (3) in this study, suitable to identify the wilting for all species. Nevertheless, this proof-ofprinciple study provides a novel application of 2DFT and artificial intelligence in sensing plants, which enlightens us to explore different mathematical-tools for extracting the physiologic information of leaf morphology for various plants in 3D space. Acknowledgments We wish to thank (i) the Chinese-German Center for Scientific Promotion (Chinesisches-Deutsches Zentrum fuer Wissenschaftsfoerderung) under the project of Sino-German Research Group (GZ494), (ii) the National Nature Science Foundation of China (31171458) (iii) the international cooperation fund of Ministry of Science and Technology, China (2010DFA34670) for their financial support, and (iv) The specific fund of China Agricultural University for Ph.D. students innovation. References Ahmad, I.S., Reid, J.F., 1996. Evaluation of color representations for maize images. Journal of Agricultural Engineering Research 63, 185–196. Bettina, M.J.E., Melvin, T.T., T, A.K., 2007. Visual assessment of wilting as a measure of leaf water potential and seedling drought survival. Journal of Tropical Ecology 23, 497–500. Blonquist, J.J.M., Norman, J.M., Bugbee, B., 2009. Automated measurement of canopy stomatal conductance based on infrared temperature. Agricultural and Forest Meteorology 149, 1931–1945. Ehlert, D., Horn, H., Adamek, R., 2008. Measuring crop biomass density by laser triangulation. Computers and Electronics in Agriculture 61, 117–125.
75
Emekli, Y., Bastug, R., Buyuktas, D., Emekli, N.Y., 2007. Evaluation of a crop water stress index for irrigation scheduling of bermudagrass. Agricultural Water Management 90, 205–212. Foucher, P., Revollon, P., Vigouroux, B., Chassériaux, G., 2004. Morphological image analysis for the detection of water stress in potted Forsythia. Biosystems Engineering 89, 131–138. Guerrero, F.D., Mullet, J.E., 1988. Reduction of turgor induces rapid changes in leaf translatable RNA. Plant Physiology 88, 401–408. Gontia, N.K., Tiwari, K.N., 2008. Development of crop water stress index of wheat crop for scheduling irrigation using infrared thermometry. Agricultural Water Management 95, 1144–1152. Hashimoto, Y., Ino, T., Kramer, P.J., Naylor, A.W., Strain, B.R., 1984. Dynamicanalysis of water stress of sunflower leaves by means of thermal image processing system. Plant Physiology 76, 266–269. Kacira, M., Ling, P.P., 2001. Design and development of an automated and noncontact sensing system for continuous monitoring of plant health and growth. Transactions of the ASAE 44, 989–996. Kacira, M., Ling, P.P., Short, T.H., 2002a. Establishing crop water stress index (CWSI) threshold values for early non-contact detection of plant water stress. Transactions of the ASAE 45, 775–780. Kacira, M., Ling, P.P., Short, T.H., 2002b. Machine vision extracted plant movement for early detection of plant water stress. Transactions of the ASAE 45, 1147– 1153. Lee, R.T., Shiou, F.J., 2011. Multi-beam laser probe for measuring position and orientation of freeform surface. Measurement 44, 1–10. Mizuno, S., Noda, K., Ezaki, N., Takizawa, H., Yamamoto, S., 2007. Detection of wilt by analyzing color and stereo vision data of plant. Lecture notes in computer science, Springer-Verlag, vol. 4418, pp. 400–411. Oggier, T., Lustenberger, F., Blanc, N., 2006. Miniature 3D TOF camera for real-time imaging. Lecture notes in computer science, Springer-Verlag, vol. 4021, pp. 212–216. Omasa, K., Hosoi, F., Konish, A., 2007. 3D lidar imaging for detecting and understanding plant responses and canopy structure. Journal of Experimental Botany 58, 881–898. Rahkonen, J., Jokela, H., 2003. Infrared radiometry for measuring plant leaf temperature during thermal weed control treatment. Biosystems Engineering 86, 257–266. Seginer, I., Elster, R.T., Goodrum, J.W., Rieger, M.W., 1992. Plant wilt detection by computer–vision tracking of leaf tips. Transactions of the ASAE 35, 1563–1567. Yang, Y., Ling, P., Fleisher, D.H., Timlin, D.J., Reddy, V.R., 2008. Non-contacting techniques for plant drought stress detection. Transactions of the ASAE 51, 1483–1492. Zimmermann, D., Reuss, R., Westhoff, M., Geßner, P., Bauer, W., Bamberg, E., Bentrup, F.W., Zimmermann, U., 2008. A novel, non-invasive, onlinemonitoring, versatile and easy plant-based probe for measuring leaf water status. Journal of Experimental Botany 59, 3157–3167.