CHARACTERISTICS OF ACTIVE SPECTRAL SENSOR FOR PLANT SENSING Y. Kim, D. M. Glenn, J. Park, H. K. Ngugi, B. L. Lehman
ABSTRACT. Plant stress has been estimated by spectral signature using both passive and active sensors. As optical sensors measure reflected light from a target, changes in illumination conditions critically affect sensor response. Active spectral sensors minimize the illumination effects by producing their own illumination, which is reflected from the target and measured by the detector. Although active sensors use modulated radiation that can be differentiated from ambient illumination, sensor performance characteristics must be well understood and examined in different target conditions of plant foliage in order to validate the data and increase the accuracy. In this article, the performance of a commercial active spectral sensor, GreenSeeker, was evaluated to study the effects of: partial canopy coverage, target off‐center, standoff distance, target surface tilting, wetness of target surface, illumination and temperature, bidirectional solar angle, and diurnal solar radiation. Experiments examined a valid range of sensor responses and identified a major effect of relative humidity that was amplified by moistened surfaces, resulting in an increase of NDVI response up to 41%. These evaluations illustrate the potentials and limitations of active spectral sensors for plant sensing and provide a guideline to understanding sensor performance in order to improve measurement accuracy. Keywords. Illumination, Infrared radiation, Precision agriculture, Sensors, Spectral analysis, Vegetation indices.
S
pectral sensing technology provides an assessment of the spectral signature reflected from a target in a wide range of spectral wavelengths beyond human vision, and it has been applied in precision agricul‐ ture for nondestructive estimation of plant and soil properties to improve quality and productivity. Biological and chemical properties of plant and soil are assessed by observing the spectral responses of leaf cuticles and soil particles via reflec‐ tance or transmittance. Leaf transmittance is observed with devices such as a chlorophyll meter (SPAD, Minolta Co., Ja‐ pan) that measures the light transmitted through a small por‐ tion of a leaf from two light‐emitting diodes at 650 nm and 940 nm. Leaf reflectance, in which light reflected from leaves within a plant canopy is measured with a sensor, is more commonly used for plant spectral sensing and allows for a substantially larger number of plants or leaf area to be monitored, thereby potentially reducing variability. Spectral reflectance is associated with plant physiological characteristics such that plant stress results in an increase in visible reflectance at 400‐700 nm wavelengths and a de‐ crease in near‐infrared (NIR) reflectance at 700‐1300 nm
Submitted for review in March 2011 as manuscript number IET 9084; approved for publication by the Information & Electrical Technologies Division of ASABE in December 2011. The authors are Yunseop “James” Kim, ASABE Member, Research Scientist, Department of Electrical and Computer Engineering, Purdue University, West Lafayette, Indiana; David M. Glenn, ASABE Member, Plant Physiologist, USDA‐ARS Appalachian Fruit Research Station, Kearneysville, West Virginia; Johnny Park, Research Assistant Professor, Department of Electrical and Computer Engineering, Purdue University, West Lafayette, Indiana; and Henry K. Ngugi, Assistant Professor, and Brian L. Lehman, Research Technician, Department of Plant Pathology, Pennsylvania State University, Biglerville, Pennsylvania. Corresponding author: Yunseop Kim, Department of Electrical and Computer Engineering, Purdue University, 465 Northwestern Avenue, West Lafayette, IN 47907; phone: 765‐414‐7933; fax: 765‐494‐0616; e‐mail:
[email protected].
wavelengths (McMurtrey et al., 1994; Suarez et al., 2009). Differences in spectral reflectance in the visible and NIR wavelengths are used to calculate a normalized difference vegetation index (NDVI) and have been used for many ground‐based, on‐the‐go mapping applications using active spectral sensors. A spectral sensor was introduced by Beck and Vyse (1995) to discriminate plant from soil and was fur‐ ther investigated by Stone et al. (1996a, 1996b) for the detec‐ tion of nitrogen deficiency in winter wheat. An active spectral sensor, GreenSeeker (NTech Industries, Inc., Ukiah, Cal.), was proposed by Stone et al. (2003) for real‐time opti‐ cal sensing and application of variable‐rate nitrogen (N) fer‐ tilization. Other researchers continued examining the potential for increasing or maintaining yield with decreased N by site‐specific application based on active sensors in both large field‐scale plots (Solie et al., 2002) and small (1 m2) re‐ search plots (Raun et al., 2002). GreenSeeker was further used for mapping NDVI for various purposes in different crops: yield estimation in corn (Freeman et al., 2007; Martin et al., 2007), chlorophyll concentration in spinach (Jones et al., 2007), biomass in pasture (Flynn et al., 2008), and plant nitrogen estimation in cotton (Bronson et al., 2005). Howev‐ er, little research has examined how the sensor responded un‐ der varying target and environmental conditions, such as partial canopy coverage, standoff distance between sensor and target, tilted target surface, wetness of target surface, and diurnal solar radiation. Spectral sensing technology has been applied to many dif‐ ferent plant and soil sensing applications, such as the effect of leaf water content (Perez‐Priego et al., 2005; Sonmez et al., 2008), chlorophyll content (Sui et al., 2005; Kim and Reid, 2006), disease severity (Bravo et al., 2003; West et al., 2003), fruit quality (Lu and Ariana, 2002), yield estimation (Yang et al., 2006), and soil property characterization (Morra et al., 1991). These applications were demonstrated with var‐ ious sensing platforms, from indoor laboratory sensing to outdoor satellite remote sensing. Although these studies
Transactions of the ASABE Vol. 55(1): 293-301
2012 American Society of Agricultural and Biological Engineers ISSN 2151-0032
293
proved the concept and potential use of spectral sensing, the experiments were performed under limited conditions, such as clear sky and a short period. Thus, improvements are still required, for example, to cope with illumination changes and solar diurnal effects. Proper lighting is a critical factor for spectral sensing, since sensors collect incoming light energy through lenses and convert it to a spectral signature of the target. Several field studies have documented the effects of changing sun angle on the reflectance of vegetation canopy. Ranson et al. (1985) investigated the effects of the solar direction and sen‐ sor view angle on reflectance factors from corn canopies and found a strong effect of solar zenith angle on leaf reflectance. Tumbo et al. (2002) reported that the correlation between spectral reflectance and chlorophyll content of corn plants significantly decreased with varying solar irradiance. Kim and Reid (2007) examined ambient illumination effects on a spectral image sensor and derived a compensation for the non‐linearity of the solar irradiation under diurnal changes of the solar zenith angle. These prior studies found that lighting conditions of solar irradiance and angle strongly affect the re‐ flectance response of passive spectral sensors. None of the prior published research has characterized the effect of light conditions on the performance of active spectral sensors. Active spectral sensors transmit modulated light in certain wavebands and detect only the modulated radiation reflected from the target. Thus, these sensors are designed to measure the spectral signature consistently regardless of ambient lighting conditions. However, although active sensors use modulated radiation that can be differentiated from ambient illumination, the sensor characteristics should be examined under possible varying target conditions of plant leaves in or‐ der to assess the reliability and accuracy of the spectral mea‐ surements. The objectives of this article are to evaluate the performance of an active spectral sensor and investigate the effects of various target conditions including illumination changes. These evaluations provide a valid range of the sen‐ sor measurements and guidelines to improve the measure‐ ment accuracy.
MATERIALS AND METHODS The GreenSeeker active spectral sensor (NTech Indus‐ tries, Inc., Ukiah, Cal.) is a radiometer that actively transmits light at two specific wavelengths: red at 660 nm ±12 nm and near‐infrared (NIR) at 770 nm ±12 nm. It then measures the
light reflected from the target and calculates the normalized difference vegetation index (NDVI) as: NDVI =
ρNIR − ρVIS ρNIR + ρVIS
(1)
where ρΝΙΡ and ρVIS are the spectral reflectance of the NIR and visible red wavebands, respectively. As illustrated in fig‐ ure 1, the sensor is designed to be located at a distance of 81 to 122 cm from the target with a line scan width of about 61cm (NTech, 2007). The calculated data from the sensor are transmitted via RS‐232 serial communication to a computer with a scan rate of up to 50 Hz. The sensor was reported to be unaffected by moderate contamination and partially offset by heavy contamination when soil was dusted on the lens sur‐ face (OSU, 2003). The performance of the active spectral NDVI sensor was evaluated to study the effects of: (1) partial canopy coverage, (2) target off‐center, (3) standoff distance, (4) target surface tilting, (5) wetness of target surface, (6) illumination and temperature, (7) bidirectional solar angle, and (8) changes in diurnal solar radiation. The experiments were carried out at the Pennsylvania State University (PSU) Fruit Research and Extension Center (FREC) in Biglerville, Pennsylvania, in 2009‐2010 for all experiments except experiment 6, which was carried out in a growth chamber at the USDA‐ARS Ap‐ palachian Fruit Research Station in Kearneysville, West Vir‐ ginia, in 2009. Each measurement had the sensor running for 120 s at 10 Hz acquisition rate. The data were averaged into a single measurement point at each setting for all experiments except experiments 5 and 8, which monitored a sequence of data continuously at 1 Hz acquisition rate. Target leaves were taken from young apple trees (cv. `Gale Gala') and attached to a 100 cm × 10 cm board with 50% overlap between leaves. The target board was carefully positioned to ensure that the red linear beam from the sensor was positioned on the middle of the board under dim light for the indoor experiment and un‐ der a temporary or permanent coverage of a tarp for the out‐ door experiment. Experiments 1 to 5 were conducted in a room at 18°C tem‐ perature and 58% relative humidity with the target board positioned in front of the NDVI sensor (fig. 2a). The sensor was mounted vertically with a side‐view of the target leaves for experiments 1 to 4. An acceptable range of target condi‐ tions for experiments 1 to 4 was selected based on the stan‐ dard deviation (