Estimating Vegetation Coverage in Wheat Using Digital Images
نویسندگان
چکیده
No method exists to reliably predict percent vegetation coverage using indirect measures. This study was conducted to evaluate the use of digital image processing techniques applied to digital color, red-green-blue (RGB), images of crop canopies to estimate percent vegetation coverage and biomass. Two field experiments with winter wheat (Triticum aeslivum L.) "Tonkawa" were planted in October 1996 and 1997 at Perkins, OK on a Teller sandy loam (Udic Argiustoll) and at Tipton, OK on a Tipton silt loam (Pachic Argiustoll). Plot images from winter wheat canopies were taken using a Kodak DC40 Digital Camera(1995)' with an image resolution of 756 x 504 pixels. Spectral inadiance readings were taken from wheat canopies in red (67 1+6 nm) and near infrared (78W6 nm) wavelengths, and normalized difference vegetation index (NDVI) was calculated. Percent vegetation coverage was estimated using image-processing routines in Micrografx Picture Publisher8 version 7.0. The digital images wereconverted from &bit; RGB tagged image file format (TIFF) files, which were produced by processing the images from the camera with 'Mention of trade or company name is for information only and does not imply an endorsement by the authors or Oklahoma State University. Copyright O 1999 by Marcel Dekker, lnc 342 LUKINA, STONE, AND RAUN Photo Enhancer@, to binary pseudo-color images. Percent of pixels corresponding to the vegetation color was then calculated and used as the percent coverage for each plot. Binary pseudo-color images provided useful estimates of percent vegetation coverage that were highly correlated with wheat canopy NDVI measurements. Conventional methods of fertilizer application employ soil testing to determine appropriate rates. Soil testing is a good estimator of soil nutrient availability for immobile nuhents, phosphorus (P) andpotassium (K). Raunet a1. (1998) highlighted that in-season nitrogen (N) deficiencies can now be detected and treated using sensor-based methods. However, past soil testing for in-season treatment has been cumbersome largely because of the time lag required between testing and final fertilizk application. In the last 20 years, newer non-destmctive methods of measuring mobile and immobile nutrient availability have been developed. The normalized difference vegetation index (NDVI) has recently proven to be a al reliable estimator of N deficiency in winter wheat (Stone et al., 1996b). On-the-go NDVI measurements can be used for detecting N deficiency and for making in-season topdress N applications. Stone et al. (1997) found that this method could significantly increase nitrogen fertilizer use efficiency. Individual plants, their shadows, and soil background contribute to spectral measurements made in vegetation canopies. In early experiments with visible and infrared reflectance from wheat canopies, Stanhill et al. ( I 972) suggested that the difference in crop absorbtivity could be accounted for by the differences in biomass and degree of ground cover. Wanjura and Hatfield (1987) pointed out that vegetation indices were affected more by ground cover than by other variables such as fresh anddry biomass or leaf area index. They estimated groundcoverby measuring canopy shadow width in different crops. Considering ground cover as an important variable, Huete et al. ( 1985) measured percent green canopy cover by projecting a 35 mm slide onto a dot grid and counting the dots of light and shaded surface. Later, using the same technique of ground cover estimation, Heilman and Kress ( 1987) concluded that soil background reflectance had the greatest influence for 50 to 75% groundcover on soils withhighreflectance. Lowvegetationcovmge didnot affect soil irradiance significantly whereas soil reflectance was insignificant at high vegetation coverage. Vegetation density and amount of soil included in the sensor view can also affect spectral measurements. In order to evaluate the impact of vegetation coverage on sensor readings Lukina et al. (1997) evaluated percent vegetation coverage at different growth stages and row spacings. Their work demonstrated a high correlation (0.8-0.97) between percent vegetation coverage andNDVI measurements. The objective of this study was; to evaluate the use ofdigital image processing techniques applied to digital color, RGB, images of crop canopies to estimate percent coverage and biomass. VEGETATlON COVERAGE IN WHEAT USlNG DiGiTAL IMAGES 343 Picillre I. O~iginal image. Picbre 11. 'Pure-color' emect applied. Picture III. 'Threshold' effect Pirmre N. Binary pseudo-color applied. image. FIGURE I . Image processing procedure, steps I-IV Winter wheat (Triticum aesrivum L.) was planted in October 1996 and 1997 at P e r h , OK on a Teller sandy loam (fine-loamy, mixed thermic Udic Argiustoll) and atTipbq OK on aTlptonsilt loam (fme-loamy, mixed, thermic, Pach'ic Argiustoll). Four N rates, 0, 56, 1 12, 168 kg ha-' as ammonium nitrate were broadcast and incorporatedpreplant. Seeding rates were 99,80,59, and49 kg ha-', at row spacings of 15.2,19.0,25.4, and 30.5 cm,respecnvely. Eachplot was 2.6 mx 6.1 m. Canopy irradiance measurements were taken from wheat in-situ using red (671+6 nm) and near-infrared (78016 nm) wavelengths at Feekes growth stages 4 and 5 (Large, 344 LUKlNA, STONE, AND RAW TABLE 1. Method to convert digital images ofwheat in vegetative stages to percent vegetation cover. Gum ~ u & ~ r a m & < I m a g M k t Y EnaOopi0. bthclulc\EUa~dira(ny < l o u m ~ \ chrma C d a r S m n h option Calm Adjud uds C a b r A d j u r m h m u Calm Smustion> In thc Cabr Ssbmtiom bo. (m thc right) m the &W; lo pure (-lo), thca click APMY In Uls Irnmlc EKaOdimc!my A b i n o w of each band (R G. and 8) is wmputsd Soil and plant dawd pixels a-emnld. Shadow 1 VEGETATlON COVERAGE IN WHEAT USING DIGITAL IMAGES 345 ! 1954). Plotimages were fakenusing aKodakDC40 Digital Camera (Kodak, 1995) I with a resolution of 756 x504 pixels at both locations (Tipton and Perkins), and at I the same time and places as spectral irradiance measurements. Percent vegetation I coverage was estimated using image-processing routines in Micrografx Picture Publishem version 7.0 (Micrografx, 1997). The digital images were converted from 8-bit RGB TIFF files, which were produced by processing the images from the camera with Photo Enhancem (Kodak, 1995), to binary pseudo-color images. Percent of pixels corresponding to the vegetation color was then calculated and used as the percent of coverage for each plot as illustrated in Figure I. Table 1 summarizes the procedure used for processing the images. It should be noted that some adjustments of contrast and color balance are requlred for images taken under different light and soil reflectance conditions. The color of the Udic Argiustoll soil (dark brown 1OYR 413) at Perkins was lighter than that of the Pachic Argiustoll (very dark grayish brown IOYR 312) at Tipton. The Perkins images, taken at Feekes growth stage 5 in 1997, were adjusted for conmast and color balance to improve color separation. Before applying the procedure in ! Table 1, the contrast (Map\Color Balance ->Joystick\Contrast) was increased (by 5%) and the balance (Map\Color Balance ->Joystick\ Balance) of the red channel was shifted towards red (by -10%). Color saturation was thenadjusted to maximum (pure): steps I. 1l.a 1l.c inTable 1. The images were then color thresholded in red, i oreen, and blue at a 2496 level in the same way as described in steps 1l.d 1l.f in a Table 1. A 'chroma mask' was then generated for the combination of red and black portions of the image, and filled with red color. These parts corresponded to soil. The mask was then invertedand filled with a black color and processed as described in steps 1ll.g through I V b inTable I. Extremely bright images, taken on a very bright sunny day at Tipton, Feekes growth stage 4, 1998 were treated differently. Soil color at this location was darker than that of Perkins. First images were 'smoothed' (Image-Effects-PhotographicSmooth) by 2 units. Thenall steps were executed as described in the Table 1 with the only difference in step Il.b, where 'chroma mask' was applied not only to red, but also to white and purple colors. 'Smoothing' can assist with separating bright spots of leaves with that of soil. I However, this procedurecan misclassify some soil-related pixels as plant pixels. It was noticed that images taken on a cloudy day were converted with better i 1 precision; largely due to the absence of glare. Thus, future Images were either bken on acloudy day or under a shadow, createdby blackposter boards (81xl02cm). I I Images taken under the shadow were processed with slight differences. For images obtainedat Tipton, at Feekes growth stage 5, 1998, thecontrast was increased (by I 5D:) and the balance of the red channel was shifted towards red (by 5%). The 'chroma mask' was genented for red and purple colors (Table 1, step 1l.b) and then I 1 filled with red. The rest of the procedure was unchanged. Images taken at Perkins, Feekes growth stages 4 and 5, 1998 were adjusted in contrast and balance(by 5 and-5 ' X I , respectively). Colors were thresholded in red, 348 LUKINA, STONE, AND RAUN FlGURE 4. Correlation between NDVI and dry matter, N concentration, and total N uptake at Perk~ns at Feekes 4, 1997. ! In the second year of the experiment, the range of vegetation coverage was ! 66-90% and 67-93% at Tipton. atFeekes growth stages 4 and5, respectively, and 1 26.34% at Perkins, Feekes growth stage 5. The Pearson correlation coefficient between percent vegetation coverage and NDVI was higher than 0.92 at both locations and growth stages (Figure 2) . The Pearson correlation coefficienb beween NDVI and biomass ranged from 0.35 to 0.80, and correlation coefficients between NDVI andN concentration in dry biomass ranged from 0.12 to 0.45 for data obtained in 1997. Higher r-values were observed behveenNDV1 and totalNuptake, ranging i between 0.47 and 0.83. Stone et al. (1996a) reported that NDVI measurements 1 depend on two factors, N concentration and biomass. Assuming that a change in N concentration or biomass affects NDVI, total N uptakc should to be a better predictor ofNDVI since it takes into account variations in both (N concentration and biomass) factors. This is also consistent with results illustrated in Figure 4 where correlation was improved between NDVI and total N uptake as compared to NDVI versus N concentration andlor biomass alone. The Pearson correlation coefficients between percent vegetation coverage and biomass ranged from 0.33 to 0.81. As was expected, high r-values were observed betweenperceni vegetation coverage and total N uptake, which ranged from 0.42 to 0.82.
منابع مشابه
Non-destructive Method for Estimating Biomass of Plants Using Digital Camera Images
Abstract Plant growth and biomass assessments are required in production and research. Such assessments are followed by major decisions (e.g., harvest timing) that channel resources and influence outcomes. In research, resources required to assess crop status affect other aspects of experimentation and, therefore, discovery. Destructive harvests are important because they influence treatment s...
متن کاملApplication of Satellite Data and Data Mining Algorithms in Estimating Coverage Percent (Case study: Nadoushan Rangelands, Ardakan Plain, Yazd, Iran)
Assessing and monitoring rangelands in arid regions are important and essential tasks in order to manage the desired regions. Nowadays, satellite images are used as an approximately economical and fast way to study the vegetation in a variety of scales. This research aims to estimate the coverage percent using the digital data given by ETM+ Landsat satellite. In late May and early Ju...
متن کاملHyperspectral Field Spectrometry for Estimating Greenbug (homoptera: Aphididae) Damage in Wheat
Remote sensing techniques have the potential to provide information about vegetation characteristics. Therefore, this study was designed to determine the ability of a hyperspectral field spectrometer along with a digital camera to estimate greenbug (Schizaphis graminum Rondani) damage to wheat (Triticum aestivum L.) in field condition. Percentage reflectance data and digital images were collect...
متن کاملPresentation of Suitable Model to Estimate Vegetation Fraction Using Satellite Images in Arid Region (Case Study: Sadough- Yazd, Iran)
One of the influential tools concerning the rangeland and vegetation sciences isthe technology of remote sensing and satellite data. Satellite data have played essentialroles in preparing the needed information for studying the vegetation. Vegetation has beenwidely recognized as one of the best indicators for determining the land conditions. Usingvegetation indices is one of the techniques of r...
متن کاملDevelopment of a Wireless Computer Vision Instrument to Detect Biotic Stress in Wheat
Knowledge of crop abiotic and biotic stress is important for optimal irrigation management. While spectral reflectance and infrared thermometry provide a means to quantify crop stress remotely, these measurements can be cumbersome. Computer vision offers an inexpensive way to remotely detect crop stress independent of vegetation cover. This paper presents a technique using computer vision to de...
متن کاملذخیره در منابع من
با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید
عنوان ژورنال:
دوره شماره
صفحات -
تاریخ انتشار 2007