determination of peanut pod maturity by near‐infrared ... - PubAg - USDA

DETERMINATION OF PEANUT POD MATURITY BY NEAR‐INFRARED REFLECTANCE SPECTROSCOPY W. R. Windham, C. V. K. Kandala, J. Sundaram, R. C. Nuti

ABSTRACT. Peanuts are indeterminate crops and do not mature evenly. Thus, it is difficult to decide the optimal time of harvest. Pod maturity is currently determined by the hull scrape method in conjunction with the maturity profile board (MPB) for estimating the “days until digging”. The method is based on the known correlation between maturity level and pod mesocarp color, which is a subjective assessment. The objective of this research was to develop visible (Vis; 400 to 750 nm) and/or visible plus shortwave near‐infrared (Vis/NIR; 400 to 1100 nm) reflectance models to predict maturity classes on a pod‐by‐pod basis. This would allow estimation of the optimum days to dig the crop. Peanuts (`Georgia Green') were harvested on five dates in 2008 and analyzed by the hull scrape method and Vis/NIR reflectance spectroscopy. Spectra from the side of the pod basal segment (N = 754) and from the saddle of the dorsal segment (N = 625) of each pod were analyzed in the Vis and Vis/NIR regions. Partial least squares regression was used to regress MPB maturity column numbers on spectra of pods from three sampling dates. Calibration regression error for MPB class column number was higher for saddle spectra. Based on the calibrations, placement of the pods on the profile board could differ by 1.5 columns, which could alter the estimate of days until digging by +3 days. Validations of all spectroscopic models were equal to and/or 3 to 7 days longer than the corresponding MPB estimates of days until digging. Separation of the maturity classes was primarily due to the broad and increasing absorbance at 640 nm as the pod matures. Keywords. Hull scrape, Peanut maturity, Profile board, Visible/NIR spectroscopy.

D

etermining when to dig peanuts (Arachis hypogea L.) is complicated since the reproductive period is indeterminate. Peanuts dug at any time during normal harvest dates would have a wide distribution of pod maturity. Over the years, many researchers have reported on the correlation between peanut maturity and the optimum time to dig the crop, yield, and the economic value of the kernels (Sanders et al., 1982; Sanders et al., 1989; Mozingo et al., 1991; Sanders and Bett, 1995). In general, digging the crop when pods are at optimum maturity increases yield and improves market grades (Jordan, 2006). Therefore, knowledge of pod maturity distributions can lead to an estimate of the optimum time to harvest the crop, which is important to the grower and processor. The hull scrape method developed by Williams and Drexler (1981) is currently the accepted method for determining peanut maturity distributions of the crop. The maturity distributions are based on the changes in color and structure of the pod mesocarp. The method has remained

Submitted for review in August 2009 as manuscript number FPE 8179; approved for publication by the Food & Process Engineering Institute Division of ASABE in March 2010. Presented at the 2009 ASABE Annual Meeting as Paper No. 095926. The authors are William R. Windham, ASABE Member, Animal Scientist, USDA‐ARS Richard B. Russell Research Center, Athens, Georgia; and Chari V. Kandala, Agricultural Engineer, Jaya Sundaram, ASABE Member, Agricultural Engineer, and Russell C. Nuti, Plant Physiologist, USDA‐ARS National Peanut Research Laboratory, Dawson, Georgia. Corresponding author: William R. Windham, USDA‐ARS Richard B. Russell Research Center, P.O. Box 5677, Athens, GA 30604; phone: 706‐546‐3513; fax: 706‐546‐3607; e‐mail: Bob.windham@ars. usda.gov.

unchanged over the years, with the exception of the removal of the pod exocarp and the evolution of the maturity profile board (MPB). The current use of the MPB and the method by which peanut maturity distributions are determined to estimate the “days until digging” has been described by Rowland et al. (2006). The MPB is divided into the color classes white, yellow 1, yellow 2, orange, brown, and black. These color classes are further divided into individual columns that represent successive stages of pod maturity development. Maturity distributions from the MPB have been compared to kernel maturity assessment and correlated with changes in the kernel physical and chemical attributes. Kernel moisture is high in immature kernels and decreases with maturity as mesocarp color progresses from the yellow 1 to black classes on the MPB (Kim and Hung, 1991). Kernel size and oil content increase significantly from the yellow 2 to the orange classes (Kim and Hung, 1991). As the peanut matures into the brown and black classes, there is little change in kernel moisture and oil content. A limited number of instrumental techniques have been investigated to predict peanut crop maturity. The majority measured the kernel versus the pod, but all used the maturity distribution classes from the hull scrape method as the reference standard. Shahin et al. (2000) reported the development of a fuzzy logic model to predict maturity based on kernel NMR signal. Compared to the hull scrape method, the model predictions were 45%, 63%, and 73% accurate when maturity was classified into 6, 5, and 3 classes, respectively. Whitaker et al. (1987) correlated maturity of kernels with near‐infrared (NIR) reflectance spectroscopy. The standard error of prediction and r2 were 0.60 and 0.96, respectively, for NIR‐predicted classes 1 through 5 compared

Transactions of the ASABE Vol. 53(2): 491-495

2010 American Society of Agricultural and Biological Engineers ISSN 2151-0032

491

to the hull scrape method. They concluded that NIR reflectance techniques could be used to measure peanut maturity. Machine vision has also been investigated to measure kernel maturity (Ghate et al., 1993) and pod ripeness color (Williams and Adams, 1990), after the exocarp of the pod had been removed. Both systems measured surface texture characteristics and gray‐level gradients and appeared to be promising techniques for an automated peanut maturity detection system. Finally, Rowland et al. (2006) demonstrated that degree‐day models could be used to predict peanut maturity in the southeastern U.S. They also proposed to estimate the optimum week to dig from the distribution of only the brown and black pods. The limitations of the hull scrape method have been well documented in the literature (Carley et al., 2008; Jordan et al., 2005; Rowland et al., 2006; Shahin et al., 2000; Tollner et al., 1998). These limitations include: (1) subjective assessment and judgment of the observer to separate color categories, (2) time and labor to remove the exocarp, (3) time and labor to categorize the pods on the MPB, and (4) the method may or may not applicable to late‐maturing cultivars. Therefore, a faster and easier maturity measurement would be advantageous for the peanut industry. The objective of this research was to develop visible and/or visible plus shortwave near‐infrared reflectance models to predict MPB classes on a pod‐by‐pod basis and then estimate the optimum days to dig the crop.

MATERIALS AND METHODS PEANUT SAMPLES AND MATURITY PROFILE BOARD Cultivar `Georgia Green' (Branch, 1996) peanuts were planted and grown under irrigation in 2008 near the USDA‐ ARS National Peanut Research Laboratory in Dawson, Georgia. Best management practices for peanut production were followed, as recommended by the University of Georgia. Five sample sets for maturity were dug on 10 September, 11 September, 7 October, 8 October, and 10 October 2008. Plants with attached pods were transported to the National Peanut Research Laboratory for hull scrape maturity rating and visible/near‐infrared spectroscopic analysis. Approximately 150 to 200 pods were removed from the plants in each sample. The exocarp was removed through pressure washing using a rotary turbo nozzle. Washing was stopped after approximately 30 s to remove the immature white and yellow pods before they disintegrated. Blasted pods from each sample set were placed on the maturity profile board (MPB) and separated into maturity classes based on the mesocarp color by a single observer of the National Peanut Research Laboratory (Rowland et al., 2006; Williams and Drexler, 1981). The MPB consists of 25 columns representing various maturity classes based on the transition of mesocarp color from white to black and pod texture. The color/maturity classes on the MPB are: (1) white; (2) yellow 1a, yellow 1b, yellow 1c, yellow 1d; (3) yellow 2a, yellow 2b, yellow 2c, yellow 2d; (4) orange a, orange b orange c, orange d; (5) brown a, brown b, brown c; and (6) black a to black f. After placing the pods on the MPB, the pods were grouped into six maturity classes. Pods in columns 1‐4 were white, columns 5‐8 were yellow 1, columns 9‐12 were yellow 2, columns 13‐16 were orange, columns 17‐19 were brown, and columns

492

20‐25 were black (Rowland et al., 2006). Although the white maturity class has four columns, all white pods were placed in one column and assigned the value of 4. Based on the maturity class distribution, the optimum digging date was determined for each sample set. VISIBLE/NIR SPECTROSCOPIC SIGNAL ACQUISITION For each sample set, peanut pods from each column within each maturity class were removed from the MPB, kept moist, and placed in labeled plastic bags. Each pod within a maturity class was scanned with a FOSS XDS Rapid Content analyzer (FOSS North America, Eden Prairie, Minn.). Spectra were recorded from 400 to 2500 nm. Reference reflectance values were obtained at every tenth sample using a ceramic standard. Spectra were collected from the side of the pod basal segment and from the saddle of the dorsal segment of each pod. Some peanut samples contained only one kernel and therefore had no saddle to scan. As such, the total number of saddle spectra was less than the number of pods. The instrument's aperture was adjusted to fit the size of each pod. Each sample was scanned 32 times, averaged, and transformed to log(1/reflectance). SPECTROSCOPIC ANALYSIS A commercial spectral analysis program (NIRS3, Infrasoft International, Inc., Port Matilda, Pa.) was used for multivariate analysis. Visible (Vis; 400 to 750 nm) and visible plus shortwave near‐infrared (Vis/NIR; 400 to 1100 nm) spectral regions were used for calibration. Sample sets 1, 3, and 4 were chosen for calibration because of the range of maturity classes. Pod (N = 476) and saddle (N = 408) spectral data were first transformed with multiplicative scatter correction (MSC) (Isaksson and Naes, 1988) and then by two separate data transformations: 10 nm smoothing interval and first‐derivative processing (gap = 10 nm, smoothing interval = 10 nm). Partial least squares regression (PLSR; Naes et al., 2002) was used to regress MPB maturity column numbers on log(1/reflectance) spectra of pods. The optimum number of PLS factors was determined by cross‐validation (Martens and Naes, 1989). During cross‐validation, one‐fourth of the calibration samples at a time were temporarily removed from the calibration set and used for prediction. The optimum number of factors for the MPB column number was that which produced a minimum overall error between modeled and reference values (standard error of cross‐validation, SECV). Second‐derivative preprocessing transformations and altering the first‐derivative gap and smoothing interval did not improve the SECV. Calibration models were validated with sample sets 2 N = 116) and 5 (N = 168). Model performance was reported as the coefficient of determination (r2), standard error of prediction (SEP), and average difference between MPB column number and model values (bias). In addition, histograms were plotted with predicted MPB column numbers to mimic the MPB, and model performance was reported as “days until digging”.

RESULTS AND DISCUSSION The optimal digging date was estimated by placing 121, 116, 165, 190, and 168 pods for sets 1 through 5, respectively, on the MPB according to the relative degree that the primary

TRANSACTIONS OF THE ASABE

color had progressed over the previous color. The profile board has a slope and a harvest projection line. The slope line represents the typical rate that pods are set, and the projection line is set at a height of three pods. The projection line represents the balance point for maturity risk management. On the x‐axis of the board is a time scale that correlates days until digging with the color of the brown and black classes. The profile board for each sample set was evaluated to find where the leading slope (from right to left of the sample profile) crosses the projection line, and then days until digging were read on the x‐axis. This estimate is ±3.5 days from the sampling date plus the estimate. Days until digging for samples sets 1, 2, 3, 4, and 5 were 10, 3‐7, 7, 0, and 10‐14, respectively. Sample sets 1, 3, and 4 were chosen for calibration because of the range of maturity classes. Statistics for pod and saddle calibration using visible spectra are shown in table 1. To scan the basal pod, the aperture of the instrument was adjusted for each pod to accommodate size differences. In order to scan the dorsal saddle, the aperture opening was generally larger than the saddle region. As a result, the saddle spectra possibly included portions of the basal and apical segments of the pod. The standard error of cross‐validation (SECV) and R2 were lower for the pod spectra. This result was unexpected since the saddle is the location of initial color transition in the mesocarp (Williams and Drexler, 1981). The higher SECV for the saddle was possible due to the inability to focus the energy on the saddle region only. Multiplicative scatter correction significantly reduced SECV versus no scatter correction (data not shown). However, first‐ derivative data pretreatment and the level of gap and smoothing did not significantly reduce the SECV. Statistics for pod and saddle calibration using Vis/NIR spectra are shown in table 2. There was no difference in the pod SECV and R2 by including shortwave NIR (750 to 1100nm) spectra compared to only visible spectra (table 1). However, for the dorsal saddle, the SECV was lower and the R2 was higher from calibration with Vis/NIR spectra. The SECV data in both tables 1 and 2 represent the error of placing pods in a given column on the MPB. This means that, on the average, placement of the pods on the profile board could differ by 1.5 columns compared to the observer's

ability to discriminate color classes. An error of 1.5 columns could alter the estimate of days until digging by ±3 days. Basal pod spectra in each of the six maturity classes from the MPB were averaged for sample set 1 (fig. 1). The one‐ fourth classes of yellow 1 and yellow 2 were averaged into one spectrum each. All other maturity classes were averaged within their sub‐divisions. In general, absorbance increased as pods matured, especially in the region of 550 to 650 nm, where yellow and orange absorb (Carley et al., 2008). Figure2 shows the significant wavelengths from the loadings of the first three principal components (PC) for the basal pod three‐factor model (table 2). Loadings are the regression coefficients of each variable (wavelength) for each PC. Loadings often resemble the spectra of samples and the spectra of constituents and thus offer interpretation to known absorbance bands (Cowe and McNicol, 1985). The loading plots indicate how the variance is accounted for in a PC across the wavelength scale. Numerically higher (±) weights indicate a relative high contribution of the wavelength area to that PC. The shape of the plot (fig. 2) of the first PC loading showed broad and increasing absorbance as the pod matured, with a maximum weight at 640 nm. Loading 2 had large intensities at 472 nm, related to a shift in absorbance in the blue region from 420 nm for the white and yellow 1 classes to 472 nm as pod maturity progressed to the orange class (fig. 1). Loading2 had an additional large intensity at 976 nm due to

Table 1. Calibration statistics for MPB column number using visible (400 to 750 nm) reflectance spectra. Pod Part Pretreatment Factors SECV[a] R2

Figure 1. Average absorbance (log 1/reflectance) pod spectra of MPB maturity classes and correlation with MPB column number.

Basal pod

10 nm smooth + MSC[b] 10 nm 1st deriv. + MSC

5 4

1.51 1.49

0.93 0.94

Dorsal saddle

10 nm smooth + MSC 10 nm 1st deriv. + MSC

5 4

1.95 1.82

0.87 0.89

[a] [b]

Standard error of cross‐validation. MSC = multiplicative scatter correction.

Table 2. Calibration statistics for MPB column number using visible/NIR (400 to 1100 nm) reflectance spectra. Pod Part Pretreatment Factors SECV[a] R2 Basal pod

10 nm smooth + MSC[b] 10 nm 1st deriv. + MSC

3 5

1.46 1.39

0.93 0.94

Dorsal saddle

10 nm smooth + MSC 10 nm 1st deriv. + MSC

5 5

1.58 1.51

0.92 0.92

[a] [b]

Standard error of cross‐validation. MSC = multiplicative scatter correction.

Vol. 53(2): 491-495

Figure 2. Loading weights of principal components 1, 2, and 3 for PLS calibration of MPB column number.

493

Table 3. Validation statistics for the prediction of MPB column number by visible and visible plus shortwave NIR spectroscopic calibrations. Calibration Spectral Region (nm) Validation SEP[a] Bias[b] r2 Slope Basal pod Dorsal saddle Basal pod Dorsal saddle [a] [b] [c] [d]

400 to 750 400 to 750 400 to 1100 400 to 1100

Basal pod[c] Dorsal saddle[d] Basal pod Dorsal saddle

1.70 2.20 1.61 1.50

0.42 0.48 0.46 0.56

0.89 0.79 0.89 0.91

1.02 0.99 0.98 0.99

Standard error of prediction. Mean PPB column number minus mean predicted column number. Sample sets 2 and 5 (N = 278). Sample sets 2 and 5 (N = 217).

the absorbance of moisture (Williams and Norris, 1987). The average spectrum at 976 nm shows that the moisture content decreased as the pod matured (fig. 1). There was a large decrease in absorbance from the white to yellow 2 classes. However, there was very little difference in the absorbance due to moisture in the brown and black classes. Kim and Hung (1991) reported a 30% decrease in kernel moisture at harvest from maturity class yellow 1 to orange and little change in moisture from orange to black. Loading 3 had significant intensity in the green region (512 nm). Linear correlation was computed between the MPB column number of each sample and the MSC of log(1/R) at each wavelength. The resulting correlation plot (fig. 1) shows a high positive correlation from 550 nm to 800 nm, where absorbance increased with maturity. Absorbance due to moisture shows a high negative correlation with maturity. First‐derivative data pretreatment, level of gap, and smoothing of visible and Vis/NIR spectra did not significantly reduce SECV. Therefore, only validation statistics from the calibrations developed with MSC plus smoothed data pretreatment will be discussed. Sample sets 2 and 5 were merged and predicted for MPB column number. Validation statistics for basal pod and dorsal saddle calibrations are given in table 3. Basal pod samples were predicted with an SEP of 1.7 and 1.6 columns with Vis and Vis/NIR spectra, respectively. Bias, coefficient of determination (r2), and linear regression of actual MPB column number versus predicted column number (slope) were similar. In contrast, the SEP was significantly higher and the r2 was lower for the dorsal saddle samples predicted with the dorsal saddle Vis model. However, the validation statistics for the dorsal saddle Vis/NIR model were lower and similar to both basal pod models. The validation statistics followed the same trend as the calibration statistics (tables 1 and 2). The lower SEP for the dorsal saddle Vis/NIR calibration was due to the presence of the 976 nm water band in the calibration model. Histograms were created with predicted column numbers for sample sets 2 and 5 and plotted to mimic the MPB. Days until digging were estimated, and the percentage maturity class distributions were calculated. The days until digging estimated by the National Peanut Research Laboratory technical staff for sample set 2 was 3 to 7 days, with brown and black pods representing 31% and 60%, respectively, of the whole sample (table 4). Estimates of the days until digging with both pod models were 7. The maturity class distributions resulting from the saddle models were 3 to 7days longer than the corresponding MPB estimates. All

494

models underestimated the percentage of pods in the black class and overestimated the percentage in the orange class. The models did, however, class brown and black pods between 67% and 76% of the total. The MPB days until digging for sample set 5 was estimated at 10 to 14 days, with brown and black pods representing 36% of the total sample (table 5). Estimates of days until digging from all spectroscopic models were identical to the MPB, and on the average brown and black pods represented 38% of the total samples. In both validation sample sets, the inclusion of the 976 nm water band in Vis/NIR models did not appear to help differentiate between the orange, brown, and black maturity classes due to similarities in the absorbance of water among the classes (fig. 1). Validation of sample set 5 with visible spectroscopic models did not follow the observed MPB classes for yellow 1 and yellow 2. Shahin et al. (2000) reported misclassification of yellow pods with a fuzzy logic model to predict pod maturity based on NMR signals of corresponding kernels. However, the Vis/NIR models did improve the separation between and within the yellow 1 and yellow 2 maturity classes due to the 976 nm water band in the models (figs. 3 and 4). Williams and Drexler (1981) reported that, along with color, the soft watery structure of the pod was important to help separate yellow 1 from yellow 2 due to the subtle differences in yellows. Table 4. Validation of days until digging and maturity class distributions for sample set 2. Maturity Class Days (%) Until Digging Yel 1 Yel 2 Org Brn Blk Calibration/Validation Maturity profile board Pod visible spectroscopy Saddle visible spectroscopy Pod Vis/NIR spectroscopy Saddle Vis/NIR spectroscopy

3‐7 7 10 7 10

0 0 0 0 0

0 4 9 5 8

9 21 25 19 17

31 28 23 26 30

60 48 44 49 45

Table 5. Validation of days until digging and maturity class distributions for sample set 5. Maturity Class Days (%) Until Digging Yel 1 Yel 2 Org Brn Blk Calibration/Validation Maturity profile board Pod visible spectroscopy Saddle visible spectroscopy Pod Vis/NIR spectroscopy Saddle Vis/NIR spectroscopy

10‐14 10‐14 10‐14 10‐14 10‐14

16 6 0 13 11

27 40 40 31 27

21 19 18 21 19

17 15 23 15 20

19 20 19 20 23

TRANSACTIONS OF THE ASABE

MPB maturity distribution Vis maturity distribution

18

Number of Pods

15 12

Projection line Slope

9 6 3 0

Yellow 1

Yellow 2

Orange BraBrbBrcBlaBlbBlcBldBleBlf 31 28 24 21 17 14 10 7 3 Days Until Digging

Figure 3. MPB‐determined maturity distribution for sample set 5 versus Vis‐predicted maturity distribution. MPB maturity distibution Vis/NIR maturity distribution

18

Number of Pods

15 12

Projection line

9

Slope

6 3 0

White Yellow 1

Yellow 2

Orange BraBrbBrcBlaBlbBlcBldBleBlf 313128282424 2121171714141010 77 33 Days Until Digging

Figure 4. MPB‐determined maturity distribution for sample set 5 versus Vis/NIR‐predicted maturity distribution.

CONCLUSION Visible near‐infrared reflectance spectroscopy from 400 to 1100 nm was used to predict peanut pod maturity classes on a pod‐by‐pod basis, and the resulting class distributions were used to estimate the days until digging of the crop. Using basal pod and dorsal saddle spectra, PLSR was used to develop calibration models for MPB column number. Standard error of cross‐ validation was about 1.5 column numbers, which translate into an error of ±3 days until digging. Estimates of days until digging from all spectroscopic models were equal to and/or 3 to 7 days longer than the corresponding MPB estimates. Assessment of PLSR loadings suggested that prediction was dependent on the spectral variation of pod color at 472 nm, 512 nm, 640 nm, and moisture at 976 nm, which decreased as the pod matured. The results from this research present Vis/NIR models that can be used to successfully predict peanut pod maturity class distributions and estimate days until digging. However, these models should be further validated and the calibration population expanded to include other runner‐type peanuts grown in the southeastern U.S., and their utility tested over multiple growing seasons.

REFERENCES Branch, W. D. 1996. Registration of `Georgia Green' peanut. Crop Sci. 36(3): 806.

Vol. 53(2): 491-495

Carley, D. S., D. L. Jordan, L. C. Dharmasri, T. B. Sutton, R. L. Brandenburg, and M. G. Burton. 2008. Peanut response to planting date and potential of canopy reflectance as an indicator of pod maturity. Agron. J. 100(2): 376‐380. Cowe, I. A., and J. W. McNicol. 1985. The use of principal components in the analysis of near‐infrared spectra. Applied Spectroscopy 39(2): 257‐266. Ghate, S. R., M. D. Evans, C. K. Kvien, and K. S. Rucker. 1993. Maturity detection in peanuts (Arachis hypogaea L) using machine vision. Trans. ASAE 36(6): 1941‐1947. Isaksson, T., and T. Naes. 1988. The effect of multiplicative scatter correction (MSC) and linearity improvement in NIR spectroscopy. Applied Spectroscopy 42(7): 1273‐1284. Jordan, D. L. 2006. Peanut production practices. In Peanut Information, 15‐34. AG‐331. Raleigh, N.C.: North Carolina Cooperative Extension Service. Jordan, D. L., D. Johnson, J. Spears, B. Penny, B. Shew, R. Brandemburg, J. Faircloth, P. Phipps, A. Herbert, Jr., D. Coker, and J. Chapin. 2005. Determining peanut pod maturity and estimating the optimal digging date: Using pod mesocarp color for digging Virginia market type peanut. AG‐633. Raleigh, N.C.: North Carolina Cooperative Extension Service. Kim, N. K., and Y. C. Hung. 1991. Mechanical properties and chemical composition of peanuts as affected by harvest date and maturity. J. Food Sci. 56(5): 1378‐1382. Martens, H., and T. Naes. 1989. Chapter 4: Assessment, validation, and choice of calibration method. In Multivariate Calibration, 237‐266. H. Martens and T. Naes, eds. New York, N.Y.: John Wiley and Sons. Mozingo, R. W., T. A. Coffelt, and F. S. Wright. 1991. The influence of planting and digging dates on yield, and grade of four Virginia‐type peanut cultivars. Peanut Sci. 18: 55‐62. Naes, T., T. Isaksson, T. Fern, and T. Davis. 2002. Chapter 18: Qualitative analysis/classification. In Multivariate Calibration and Classification, 221‐259. T. Naes, T. Isaksson, T. Fern, and T. Davis, eds. Chichester, U.K.: NIR Publications. Rowland, D. L., R. B. Sorensen, C. L. Butts, and W. H. Faircloth. 2006. Determination of maturity and degree‐day indices and their success in predicting peanut maturity. Peanut Sci. 33(2): 125‐126. Sanders, T. H., and K. L. Bett. 1995. Effect of harvest date on maturity, maturity distribution, and flavor of florunner peanuts. Peanut Sci. 22: 124‐129. Sanders, T. H., A. M. Shubert, and H. E. Pattee. 1982. Maturity methodology and postharvest physiology. In Peanut Science and Technology, 624‐654. E. Pattee and C. T. Young, eds. Yoakum, Tex.: American Peanut Research and Education Society. Sanders, T. H., J. R. Vercellotti, K. L. Crippen, and G. V. Civille. 1989. Interaction of maturity and curing temperature on descriptive flavor of peanuts. J. Food Sci. 54(4): 1066‐1069. Shahin, M. A., B. P. Verma, and E. W. Tollner. 2000. Fuzzy logic model for predicting peanut maturity. Trans. ASAE 43(2): 483‐490. Tollner, E. W., V. Boudolf, R. W. McMlendon, and Y. C. Hung. 1998. Predicting peanut maturity with magnetic resonance. Trans. ASAE 41(4): 1199‐1205. Whitaker, T. B., H. E. Pattee, W. F. McClure, and J. W. Dickens. 1987. Measuring peanut maturity using near‐infrared reflectance. In Peanut Quality: Its Assurance and Maintenance from the Farm to End Product, 14‐28. E. M. Ahmend and H. E. Pattee, eds. Bulletin 874. Gainesville, Fla.: Florida Agricultural Experiment Station. Williams, E. J., and S. D. Adams. 1990. A 3‐D vision system for peanut pod maturity. Proc. SPIE 1379: 236‐245. Williams, E. J., and J. S. Drexler. 1981. A non‐destructive method for determining peanut pod maturity. Peanut Sci. 8: 134‐141. Williams, P. C., and K. H. Norris. 1987. Chapter 15: Qualitative applications of near‐infrared reflectance spectroscopy. In Near‐Infrared Technology in the Agricultural and Food Industries, 241‐246, P. C. Williams and K. H. Norris, eds. St. Paul, Minn.: American Association of Cereal Chemist.

495

496

TRANSACTIONS OF THE ASABE