Next Article in Journal
Influence of Seed Origin on Morphological Characteristics and Phytochemicals Levels in Brassica oleracea var. acephala
Next Article in Special Issue
Seed Yield and Lodging Assessment in Red Fescue (Festuca rubra L.) Sprayed with Trinexapac-Ethyl
Previous Article in Journal
Planting Density Induced Changes in Cotton Biomass Yield, Fiber Quality, and Phosphorus Distribution under Beta Growth Model
Previous Article in Special Issue
Estimating Biomass of Black Oat Using UAV-Based RGB Imaging
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Agronomy
Volume 9
Issue 9
10.3390/agronomy9090501
agronomy-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles thumb_up
...
Endorse textsms
...
Comment
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

An Object-Based Image Analysis Approach to Assess Persistence of Perennial Ryegrass (Lolium perenne L.) in Pasture Breeding

by
Chinthaka Jayasinghe
1,2,
Pieter Badenhorst
1,
Junping Wang
1,
Joe Jacobs
3,
German Spangenberg
4,5 and
Kevin Smith
1,2,*
1
Agriculture Victoria Research, Hamilton, Victoria 3300, Australia
2
School of Agriculture and Food, Faculty of Veterinary and Agricultural Sciences, The University of Melbourne, Victoria 3010, Australia
3
Agriculture Victoria Research, Ellinbank, Victoria 3821, Australia
4
Agriculture Victoria Research, AgriBio, Centre for AgriBioscience, Bundoora, Victoria 3083, Australia
5
School of Applied Systems Biology, La Trobe University, Bundoora, Victoria 3086, Australia
*
Author to whom correspondence should be addressed.
Agronomy 2019, 9(9), 501; https://doi.org/10.3390/agronomy9090501
Submission received: 26 July 2019 / Revised: 23 August 2019 / Accepted: 26 August 2019 / Published: 30 August 2019
(This article belongs to the Special Issue Remote Sensing Application for Monitoring Grassland and Forage Production)
Browse Figures
Versions Notes

Abstract

:
Perennial ryegrass (Lolium perenne L.) is one of the most important forage grass species in temperate regions of the world, but it is prone to having poor persistence due to the incidence of abiotic and biotic stresses. This creates a challenge for livestock producers to use their agricultural lands more productively and intensively within sustainable limits. Breeding perennial ryegrass cultivars that are both productive and persistent is a target of forage breeding programs and will allow farmers to select appropriate cultivars to deliver the highest profitability over the lifetime of a sward. Conventional methods for the estimation of pasture persistence depend on manual ground cover estimation or counting the number of surviving plants or tillers in a given area. Those methods are subjective, time-consuming and/or labour intensive. This study aimed to develop a phenomic method to evaluate the persistence of perennial ryegrass cultivars in field plots. Data acquisition was conducted three years after sowing to estimate the persistence of perennial ryegrass using high-resolution aerial-based multispectral and ground-based red, green and blue(RGB) sensors, and subsequent image analysis. There was a strong positive relationship between manual ground cover and sensor-based ground cover estimates (p < 0.001). Although the manual plant count was positively correlated with sensor-based ground cover (p < 0.001) intra-plot plant size variation influenced the strength of this relationship. We conclude that object-based ground cover estimation is most suitable for use in large-scale breeding programs due to its higher accuracy, efficiency and repeatability. With further development, this technique could be used to assess temporal changes of perennial ryegrass persistence in experimental studies and on a farm scale.

1. Introduction

Perennial ryegrass (Lolium perenne L.) is a major forage grass for livestock in temperate agriculture [1]. Compared to annual pasture species, perennials have an extended production period with a higher year-round ground cover, growing well under a wide range of environmental conditions, and show a greater ability to tolerate grazing pressure [2]. Longevity in perennial pastures reduces the annual cost for pasture renovation, however, many perennial species including perennial ryegrass can have poor persistence due to their low tolerance of biotic and abiotic stresses including water deficiency, high temperature, pests and diseases [3]. Persistence of perennial pasture species is the ability to maintain plant density and dry matter production throughout the life of the sward [4] which has a relationship with seasonal variation in dry matter production and pasture nutritive characteristics (pasture quality). The profitability of pasture-based farm systems is related to the utilization of forage from grazed pasture and the subsequent conversion into animal products [5]. Pasture species with low persistence reduce sward density rapidly over time, creating bare ground which provides space for invasion by opportunistic, but less productive species and weeds. Invasion of weeds and less productive species reduces herbage nutritive characteristics and dry matter production in grazed systems [6]. Therefore, pasture persistence has a positive relationship with farm profitability. Perennial ryegrass has a symbiotic relationship with a fungus (endophyte) that colonizes the base of perennial ryegrass tillers and migrates to the growing point during the reproductive phase [7]. Endophyte produces unique bio-active alkaloids, which may enhance the tolerance of the host plant to various abiotic and biotic stresses [8]. Due to the interaction between perennial ryegrass genotype, endophyte and environment, cultivars exhibit different levels of persistence within a sward. Estimation of pasture persistence may help farmers and pasture breeders select cultivars that are more productive over a more extended period for their farm system.
A major challenge in measuring persistence is that the assessment of pasture survival depends on manual ground cover estimation or counting the number of plants in a given area [9], making persistence slow and costly to measure. Plant ground cover is defined as the percentage or fractional area of soil covered by plants when viewed from a nadir position [10]. The patterns of ground cover changes are dynamic, they respond to stressors such as prolonged drought, extreme high or low-temperature events and pest attack. Plant ground cover measures indicate the pattern of plant distribution and the quantity of plant survival [11]. The assessment of ground cover is an interest for precision agriculture and which may provide information to understand the relationship of the plant resistance to abiotic and biotic stresses [12]. Visual estimation of plant ground cover is a widely used method because of its simplicity. As the human eye is often combined with subjective perception, there are a number of well-known disadvantages of visual ground cover estimation including time, repeatability, labour and user bias [11] which may limit the data collection to reference area in large scale breeding programs [13].
Technological developments over the last three decades have seen increased use of sensors for data collection [14], including plant canopy characteristics [15]. Sensor-based methods remove the inconsistency of conventional methods, offer large-scale data collection and provide a non-invasive, quick method for collecting vegetation cover data [16]. Use of high-resolution airborne images and image analysis software offer low-cost platforms to obtain ground-based information for ground cover estimation. The object-based technique is one of the most successful methods for the extraction of various features from images by removing redundant details found in higher spatial resolution images at the pixel level [17]. In the object-based technique, the image is segmented into small sized groups of pixel units that are neighbouring and are spectrally and spatially similar objects [18]. These homogenous units are known as image objects and are classified according to spectral characters to extract various kinds of canopy information such as size, shape, texture, pattern, shadow, site, and association characteristics [18].
High-resolution airborne images have been used for ground cover estimation in forests, and grasslands [19,20] and for plant disease monitoring in a range of crops [21]. They are also used to estimate the impact of natural disasters [22], forest vegetation classification [23] and to estimate grassland distribution [24]. The current trend in remote sensing studies of grasslands is to use vegetation indexes (VIs) derived from spectra collected by unmanned aerial vehicles (UAV) to determine an empirical relationship with ground truth sampling to measure plant canopy characteristics (leaf area index, ground cover or above ground biomass). The relationships developed between ground truth sampling and remotely sensed data may not be accurate for situations other than those that produced the data. For instance in wheat, the mathematical relationships developed between ground truth observations of plant canopy characteristics, and corresponding values of vegetation indices for five different geographical locations were qualitatively similar but differed in the specific values of the coefficients in the relationships among the sites [25]. In this study, we developed an object-based image analysis script to estimate perennial ryegrass ground cover comparing two different sensor-based approches and ground truth data. The estimation of temporal changes of pasture ground cover could provide canopy characteristics to explain the persistence of perennial pasture cultivars and cultivar ranking based on their persistence and dry matter yield.

2. Materials and Methods

2.1. The Study Site

The study site was located on the Agriculture Victoria research station at Hamilton in south-west Victoria, Australia (Coordinates: -37.819440S, 142.062171E). Hamilton has 660 mm long-term average rainfall with a pronounced maximum in July, August and September and average annual temperature of 27 °C, which facilitates ideal environmental conditions for pasture growth and selection. The field trial was planted on 06 May 2015 and consisted of a completely randomised design comprising 72 plots. The size of a single plot was 1.5 m long and 1.5 m wide and consisted of 100 plants with 15 cm inter-row spacing.

2.2. Ground-truth Sampling

Data collection occurred three years after establishment in May 2018. Before the plots were harvested, manual visual ground cover and manual plant count (the number of surviving plants) were recorded in each plot by two experienced pasture breeders and plot level ground cover was scored visually using a standard method which is described in the EverGraze [26]. Following assessment, the plots were mechanically harvested to a height of 5 cm using a 21" self-propelled lawn mower (Model: HRX217K5HYUA; Engine capacity—190 cc; Honda Motor Co., Ltd, Tokyo, Japan) and fresh weight of green plant material from each plot was taken using a Mettler Toledo GmbH compact scale (Model: ICS6x5-1; Mettler-Toledo Ltd., Toledo, OH, USA). A subsample (200–300 g) of fresh grass was separated from each bulk sample with initial fresh weight recorded. The oven-dry weight of subsamples was obtained after subsamples had been dried at 60 °C for 48 hours.

2.3. Sensor-Based Data Acquisition and Extraction

The multispectral images were taken two weeks after the final harvest using a 3DR Solo quadcopter (3D Robotics, Berkeley, CA, USA) equipped with a Parrot Sequoia multispectral sensor (Parrot Drones S.A.S, Paris, France). The sensor was integrated to the Global Navigation Satellite System (GNSS) to locate the camera when images were being collected. The Parrot Sequoia has synchronised 1.2 Mpx monochrome sensors with four narrow bands (Green—550 nm wavelength, 40 nm bandwidth; Red—660 nm wavelength, 40 nm bandwidth; Red Edge—735 nm wavelength, 10 nm bandwidth; and Near Infrared—790 nm wavelength, 40 nm bandwidth). The 3DR solo quadcopter is an autonomous flight model and powered by 5200 mAh Li-Po Solo battery which gives a flight time about 10–20 minutes.
Tower 4.0.1 Beta (Aero Hawk Technologies, USA; https://aero-hawk.com/) android application was used for designing, saving and loading an autonomous flight path into the quadcopter and used to automate the flight path during data acquisition. Turning points of the flight path were defined outside the area of interest to ensure that the whole experimental site was measured. The image overlap of the flight mission was set to 80% forward and 75% sideways which is recommended for the Parrot Sequoia sensor. The quadcopter was flown at an altitude of 20 m above ground level, with ground speed of the quadcopter set to 3 ms−1 (11 km/h) to enable capture of stable images. AIRINOV calibration targets (Paris, France) were utilised to calibrate the Parrot Sequoia sensor before the flight mission commenced and to convert the digital number to reflectance values in the orthomosaic workflow. The flight mission was operated under sunny conditions to avoid interruption from shadows during data acquisition. After multispectral image acquisition, individual RGB images from each plot were captured in a 12-megapixel Android mobile camera (Model: Samsung SM-G930F; F-stop: f/1.7) 1.5 m above the plot at nadir view.
A Digital Terrain Model (DTM) and orthomosaic image of each multispectral band weregenerated using Pix4D mapper 4.2.16 software (Pix4D SA, Switzerland; https://www.pix4d.com/) and, the average ground sampling distance (GSD) of orthomosaic images was 1.77 cm/0.70 in per pixel. Pix4D is a 3D computer graphics software that uses photogrammetry and computer vision algorithms to generate both RGB and multispectral 3D maps and models. The radiometric calibration of multispectral images was undertaken manually using AIRINOV reference reflectance values. Ground control points (GCPs) were located within the experimental area to obtain the georectification of the orthomosaic to within 2 cm in Pix4D workflow. Apart from NIR, green, red and red edge bands, the normalized difference vegetative index (NDVI) orthomosaic were generated and the default coordinate reference system (CRS) WGS 84 54S was used as an output coordinate system for all DTM and orthomosaic in Pix4D software workflow.
The plot overlay with the same CRS was created importing NDVI orthomosaic into QGIS 2.18.20 software (QGIS Development Team, 2017, Raleigh, NC, USA; https://qgis.org/). Multispectral image analysis was performed utilising eCognition developer 9.3.2 (Trimble Germany GmbH, Munich, Germany; http://www.ecognition.com/) rule sets. An algorithm “multi-threshold segmentation” was used to separate the soil and perennial ryegrass plants and created a new object layer (Figure 1). After image segmentation, the spectral distance between soil and perennial ryegrass was checked using the nearest neighbour analysis in eCogntion 9.3.2. software (Trimble Germany GmbH, Munich, Germany; http://www.ecognition.com/). The ground cover from each plot was extracted through QGIS vector calculator with WGS 84 UTM 54S coordinate reference system (CRS). The average plot level vegetation indices were estimated using the QGIS geospatial open source tool, HTP Geoprocessor 1.1. (GitHub, Inc., San Francisco, CA, USA).

2.4. Statistical Analysis

Two different sensor-based ground cover estimates (RGB sensor-based ground cover and multispectral sensor-based ground cover), five UAV-based vegetation indices and reflectance of the NIR band were compared with three different ground-based observations (manual ground cover, manual plant count and plot level dry matter) using the Pearson’s correlation coefficient (r). The perennial ryegrass ground cover predicted fitting a liner regression model between sensor-based data and ground truth data. Statistical analyses were performed in Minitab 17.2.1 (https://www.minitab.com).

3. Results

3.1. Validation and Calibration of Sensor-based Ground Cover

The Multispectral image-based ground cover estimates positively correlated with manual ground cover (Figure 2; r = 0.720, p < 0.001) and manual plant count (Figure 2; r = 0.637, p < 0.001).The RGB image-based ground cover also showed a significant positive correlation with visual ground cover (r = 0.746, p < 0.001) and manual plant count (r = 0.637, p < 0.001), and RGB sensor-based ground cover showed a positive correlation with plot level plant dry weight (r = 0.513, p < 0.001). Both sensor-based platforms provided accurate estimates of plot-level ground cover, and ground cover extracted from both types of sensors were—significantly correlated with each other (Table 1; p < 0.001). The calculated multispectral sensor-based vegetation indices (VIs) are presented in Table 2. Vegetation indices showed a moderate positive correlation with sensor-based plant ground cover estimates and ground truth data (Table 3; p < 0.05).

3.2. Prediction Manual Ground Cover

Multispectral sensor-based ground cover showed a positive linear relationship with manual ground cover (Figure 3; R2 = 0.6379, p < 0.001, standard error 8.783). The relationship between RGB sensor-based ground cover and manual ground cover data also showed a positive linear relationship (Figure 3; R2 = 0.5570, p < 0.001, standard error 9.7149) but the manual plant count showed a moderate relationship with multispectral sensor-based ground cover (Figure 3; R2 = 0.3938, p < 0.001, standard error 12.83), and RGB sensor-based ground cover (Figure 3; R2 = 0.4063, p < 0.001, standard error 12.69).

4. Discussion

In this paper, we describe a remote sensing methodology to assess perennial ryegrass ground cover which could replace traditional visual estimations of perennial ryegrass persistence in pasture breeding programs. The technique integrates relatively low-cost sensor-based techniques and image processing software. The study indicated estimates of perennial ryegrass persistence could be derived from proximal RGB image and UAV-based multispectral image analysis. Both airborne and ground based RGB sensors are relatively lower cost devices for precision agriculture. In this study the proximal, RGB image-based technique provided a low-cost approach for data acquisition to estimate perennial ryegrass ground cover. However, we consider RGB image-based ground cover estimate not suitable for large scale breeding programs for several reasons. While pre-training and preparation time are relatively low for RGB image acquisition and RGB image analysis is combined with open source software which reduces the cost of digital image analysis, it took considerably time required to identify the precise threshold HSB colour profile of individual RGB images for the pixel-based supervised image analysis, making the RGB imaging technique unsuitable for large scale plant ground cover estimation.. However, this study suggests that the pixel-based supervised classification technique in ImageJ offers accurate plant ground cover estimation to assess perennial ryegrass persistence, particularly where perennial ryegrass ground cover is relatively low [16]. The multispectral sensor-based ground cover estimate based on eCognition software which converts homogenous pixels of digital images into objects that are classified into user-defined classes and this user-defined classification is one of the specific functions of eCognition that most of the currently available remote sensing software does not possess [32]. This is an object-based image classification that provides a closer classification to human visual interpretation [33]. In the object-based image classification, spectral distance between class “soil” and class “perennial ryegrass” was 95% in the nearest neighbour analysis which was strong enough to successfully distinguish between perennial ryegrass ground cover from soil in each plot. An object-based image analysis has been used in previous studies to estimate ground cover and distinguish vegetation types of plots using proximal RGB imaging in different agricultural crops [11]. In this study, we applied an efficient image analysis algorithm to multispectral images to obtain a clear separation of pixels representing soil and green perennial ryegrass plants so UAV-based multispectral images provide a comprehensive and intensive sense of pasture persistence in a relatively short period over large areas which could be used to rank novel perennial ryegrass cultivars in terms of their long-term persistence. Airborne multispectral images have been successfully used for crop ground cover estimation in different field crop species and showed a strong correlation between the ground cover and manual plant count [34]. However, the correlation between the sensor-based ground cover and manual plant count of perennial ryegrass was only moderate, most probably due to plant size variation within plots. Perennial ryegrass is a diploid species (2n) with a two-locus self-incompatibility system [35], and the outbreeding habit of perennial ryegrass leads to a high degree of genetic variation in natural populations [36]. Due to this genetic variation, the resistance of individual plants to abiotic and biotic stresses in the same population may vary, which creates intra-population plant size variations resulting in different numbers of surviving tillers per plant. Counting the number of surviving plants within a given area may be less biased than manually estimating of ground cover but, it may not be an accurate variable to assess perennial ryegrass persistence of a sward or breeding plots. The relationship between plot level ground cover estimates and dry matter production was moderate due to plant height variation among individual plants within each cultivar. This resulted contrasting dry matter production in the plants with the similar size of canopy cover. Therefore, in relation to persistence assessment, sensor-based ground cover may not an accurate approach to estimate perennial ryegrass dry matter production. In a sward, perennial ryegrass propagates by producing more tillers using a short underground rhizome [37]. Perennial ryegrass plants in breeding programs may also produce seeds which may germinate and grow within the breeding plots. The in-depth visual examination of the base of seedling plants may differentiate unsown plants in breeding plots, but proximal RGB or UAV-based multispectral sensors may not have adequate sensitivity to identify these plants as both sown and unsown plants show a similar spectral signature in the visible and NIR regions.
Optical sensors are more suitable for pasture monitoring compared to other sensors as they have a higher sensitivity to identify plant canopy characteristics. For instance, light detection and ranging (LiDAR) imaging can be scattered on the surface of small objects like pasture leaves which are prone to create false images. The biggest challenge to the use of optical imagery is atmospheric noise [24]. The initial development of optical sensing techniques for pasture monitoring was based on a 35 mm RGB camera which required a long tripod or a hydraulic boom for image acquisition that was more disruptive than current RGB sensor-based methods [11]. Previous optical sensing techniques for pasture were useful for estimating presence, absence, or approximate estimation of biomass, but these techniques did not have the ability to distinguish different vegetation types or accurately estimate biomass [38]. With increased resolution of optical sensors, several authors have applied manual, proximal image-based techniques to estimate plant ground cover at plot scale [39], and direct visual estimation has been used as a reference method to calibrate the sensor-based techniques because visual ground cover estimation is the simplest and quickest method with no required special equipment or training. The most common image-based ground cover estimation in pasture studies relies on proximal three-band RGB imagery [10] and application of these techniques for low accessibility areas and large-scale breeding trials may not be appropriate [11,12]. By flying UAVs over the area of interest more frequently at lower altitudes, offers an alternate remote sensing tool at low-cost to assess persistence of perennial ryegrass, therefore UAV-based techniques are likely to be more useful than the other imaging platforms in terms of precision and efficiency.
Many studies have found that vegetation indices are accurate for estimation of grassland ground cover and other grass traits in arid and semi-arid areas but, these indices have a limited value for ground cover estimation of plots/swards in temperate pastures. The major limitation of using vegetation indices for ground cover estimation is that red and NIR portion of the electromagnetic spectrum approach a saturation level after a certain biomass density or ground cover [19]. In addition, these indices are sensitive to soil background effect and atmospheric induced variation when fractional ground cover is low. Due to these limitations, the strength of correlation between five indices (NDVI, SAVI, RVI, NDGI and VIN) and manual ground-based observation is moderate, compared to a correlation between sensor-based ground cover and manual ground-based observation.
Changing the distance between ground surface and low altitude airborne sensor directly impacted on consistencey of ground sampling distance (GSD) which leads to the acquisition of poor quality image data [40] therefore, the geometrical complexity of breeding trials may create a great challenge for breeders to use this technique for accurate ground cover estimation. However, application of terrain correction into the flight path to change camera angle and flying height can reduce spatial variation of GSD, this allows to use this technique for acurate ground cover estimation in uneven terrain and hillsides. Proximal sensing platforms acquire images near to ground level and these have a relatively low field of view. Therefore, topographical changes have a minimal impact on proximal sensor-based data acquiscition. Thus, RGB sensor-based technique can be implemented for breeding trial in uneven and complex geographical area.
Data acquisition from passive sensors depends on direct or indirect solar radiation, therefore, data acquisition for relatively large area will be affected by differing image spectral signatures caused by changing light conditions, scattering processes in the atmosphere, the surrounding environment and adjacent objects. The image sensors record the intensity of the electromagnetic radiation of each pixel as a digital number which includes reflected radiation from the target object, and radiation scattered and emitted by the atmosphere. Corrections for the atmospheric and illumination conditions must be applied to obtain the actual surface reflectance value of the target objective. In this study we used AIRINOV calibration targets before multispectral image acquisition and GCPs points, and the radiometric calibration were used manually in the orthomosaic software to convert the digital numbers to actual reflectance values to keep the ground level accuracy below 2 cm to estimate the ground cover at individual plant level.

5. Conclusions

Sensor-based methods and manual ground-based techniques showed similar trends in perennial ryegrass persistence estimates (Figure 4). The proximal RGB imaging and pixel-based supervised image analysis offer a low-cost technique to assess perennial ryegrass persistence but is only applicable to small-scale breeding programs. The use of high-resolution multispectral sensors with low-cost UAVs and object-based image analysis software offer more powerful analysis, which provides a more accurate perennial ryegrass persistence estimation in large scale. The monitoring of ground cover with regular images may provide a dynamic assessment of persistence of perennial ryegrass and further development may allow us to use this technique for ground cover estimation of other pasture species and agricultural crops.

Author Contributions

Data collection and analysis, writing-original draft preparation C.J.; review and editing, J.W., P.B., J.J. and G.S.; supervision, review and editing, K.S.; read and approved the final manuscript, C.J., P.B., J.W., J.J., G.S. and K.S.

Funding

This research was funded by Dairy Australia, Gardiner Dairy Foundation and Agriculture Victoria Research.

Acknowledgments

We thank Carly Elliot, Andy Phelan, Darren Picket and Russell Elton for their technical assistance.

Conflicts of Interest

The authors declare no conflict of interest.

References

  1. Reed, K.F.M.; Nie, Z.N.; Miller, S.; Hackney, B.F.; Boschma, S.P.; Mitchell, M.L.; Albertsen, T.O.; Moore, G.A.; Clark, S.G.; Craig, A.D.; et al. Field evaluation of perennial grasses and herbs in southern Australia. 1. Establishment and herbage production. Aust. J. Exp. Agric. 2008, 48, 409–423. [Google Scholar] [CrossRef]
  2. Pontes, L.; Carrère, P.; Andueza, D.; Louault, F.; Soussana, F.J. Seasonal productivity and nutritive value of temperate grasses found in semi-natural pastures in Europe: Responses to cutting frequency and N supply. Grass Forage Sci. 2007, 62, 485–496. [Google Scholar] [CrossRef]
  3. Culvenor, R.A.; Simpson, R.J. Persistence traits in perennial pasture grasses: The case of phalaris (Phalaris aquatica L.). Crop Pasture Sci. 2014, 65, 1165–1176. [Google Scholar] [CrossRef]
  4. Waller, R.A.; Sale, P.W.G. Persistence and productivity of perennial ryegrass in sheep pastures in south-western Victoria: A review. Aust. J. Exp. Agric. 2001, 41, 117–144. [Google Scholar] [CrossRef]
  5. García, S.C.; Islam, M.R.; Clark, C.E.F.; Martin, P.M. Kikuyu-based pasture for dairy production: A review. Crop Pasture Sci. 2014, 65, 787–797. [Google Scholar] [CrossRef]
  6. Doyle, P.T.; Rowe, J.B.; Warren, B. Nutritive value of pasture species. In Proceedings of the 5th Australian Agronomy Conference, Perth, Western Australia, 24–29 September 1989. [Google Scholar]
  7. Hume, D.E.; Sewell, J.C. Agronomic advantages conferred by endophyte infection of perennial ryegrass (Lolium perenne L.) and tall fescue (Festuca arundinacea Schreb.) in Australia. Crop Pasture Sci. 2014, 65, 747–757. [Google Scholar] [CrossRef]
  8. Malinowski, D.P.; Belesky, D.P. Adaptations of Endophyte-Infected Cool-Season Grasses to Environmental Stresses. Crop. Sci. 2000, 40, 923. [Google Scholar] [CrossRef]
  9. Wang, J.; Pembleton, L.W.; Cogan, N.O.I.; Forster, J.W. Evidence for Heterosis in Italian Ryegrass (Lolium multiflorum Lam.) Based on Inbreeding Depression in F2 Generation Offspring from Biparental Crosses. Agronomy 2016, 6, 49. [Google Scholar] [CrossRef]
  10. Calera, A.; Martínez, C.; Melia, J. A procedure for obtaining green plant cover: Relation to NDVI in a case study for barley. Int. J. Remote Sens. 2001, 22, 3357–3362. [Google Scholar] [CrossRef]
  11. Luscier, J.D.; Thompson, W.L.; Wilson, J.M.; Gorham, B.E.; Dragut, L.D. Using digital photographs and object-based image analysis to estimate percent ground cover in vegetation plots. Front. Ecol. Environ. 2006, 4, 408–413. [Google Scholar] [CrossRef] [Green Version]
  12. Lynch, T.; Barth, S.; Dix, P.J.; Grogan, D.; Grant, J.; Grant, O.M. Ground Cover Assessment of Perennial Ryegrass Using Digital Imaging. Agron. J. 2015, 107, 2347–2352. [Google Scholar] [CrossRef]
  13. Smith, A.M.; Hill, M.J.; Zhang, Y. Estimating Ground Cover in the Mixed Prairie Grassland of Southern Alberta Using Vegetation Indices Related to Physiological Function. Can. J. Remote Sens. 2015, 41, 51–66. [Google Scholar] [CrossRef]
  14. Li, L.; Zhang, Q.; Huang, D. A Review of Imaging Techniques for Plant Phenotyping. Sensors 2014, 14, 20078–20111. [Google Scholar] [CrossRef] [PubMed]
  15. Rajan, N.; Maas, J.S. Mapping crop ground cover using airborne multispectral digital imagery. Precis. Agric. 2009, 10, 304–318. [Google Scholar] [CrossRef]
  16. Lei, G.; Li, A.; Bian, J.; Zhang, Z.; Jin, H.; Nan, X.; Zhao, W.; Wang, J.; Cao, X.; Tan, J.; et al. Land Cover Mapping in Southwestern China Using the HC-MMK Approach. Remote Sens. 2016, 8, 305. [Google Scholar] [CrossRef]
  17. Moskal, L.; Styers, D.; Halabisky, M. Monitoring Urban Tree Cover Using Object-Based Image Analysis and Public Domain Remotely Sensed Data. Remote Sens. 2011, 3, 2243–2262. [Google Scholar] [CrossRef] [Green Version]
  18. Lizarazo, I.; Elsner, P. Fuzzy segmentation for object-based image classification. Int. J. Remote Sens. 2009, 30, 1643–1649. [Google Scholar] [CrossRef]
  19. Lu, D. The potential and challenge of remote sensing-based biomass estimation. Int. J. Remote Sens. 2006, 27, 1297–1328. [Google Scholar] [CrossRef]
  20. Zhao, F.; Xu, B.; Yang, X.; Jin, Y.; Li, J.; Xia, L.; Chen, S.; Ma, H. Remote Sensing Estimates of Grassland Aboveground Biomass Based on MODIS Net Primary Productivity (NPP): A Case Study in the Xilingol Grassland of Northern China. Remote Sens. 2014, 6, 5368–5386. [Google Scholar] [CrossRef] [Green Version]
  21. Shanmugam, L.; Adline, A.L.A.; Aishwarya, N.; Krithika, G. Disease detection in crops using remote sensing images. In Proceedings of the 2017 IEEE Technological Innovations in ICT for Agriculture and Rural Development (TIAR), Chennai, India, 7–8 April 2017; pp. 112–115. [Google Scholar]
  22. Westen, C.J. Remote Sensing for Natural Disaster Management. Available online: https://www.isprs.org/proceedings/XXXIII/congress/part7/1609_XXXIII-part7.pdf (accessed on 30 August 2019).
  23. Jing, X.; Wang, J.; Huang, W.; Liu, L.; Wang, J. Study on Forest Vegetation Classification Based on Multitemporal Remote Sensing Images; Springer: Boston, MA, USA, 2008; pp. 115–123. [Google Scholar]
  24. Cawkwell, F.; Ali, I.; Green, S.; Dwyer, E.; Barrett, B. Satellite remote sensing of grasslands: From observation to management. J. Plant Ecol. 2016, 9, 649–671. [Google Scholar]
  25. Wiegand, C.L.; Maas, S.J.; Aase, J.K.; Hatfield, J.L.; Pinter, P.J.; Jackson, R.D.; Kanemasu, E.T.; Lapitan, R.L. Multisite analyses of spectral-biophysical data for wheat. Remote Sens. Environ. 1992, 42, 1–21. [Google Scholar] [CrossRef]
  26. McCormick, L.H.; Lodge, G.M. A field kit for producers to assess pasture health in the paddock. In Proceedings of the 10th Australian Agronomy Conference, Hobart, Tasmania, 28 January–1 February 2001. [Google Scholar]
  27. Rouse, J.W.; Haas, R.W.; Schell, J.A.; Deering, D.W.; Harlan, J.C. Monitoring the Vernal Advancement and Retrogradation (Greenwave Effect) of Natural Vegetation; NASA: Greenbelt, MD, USA, 1974.
  28. Huete, A.R. A soil-adjusted vegetation index (SAVI). Remote Sens. Environ. 1988, 25, 295–309. [Google Scholar] [CrossRef]
  29. Pearson, R.L.; Miller, L.D. Remote mapping of standing crop biomass for estimation of the productivity of the shortgrass prairie, Pawnee National Grasslands, Colorado. In Remote Sensing of Environment; Dept. of Watershed Sciences, College of Forestry and Natural Resources, Colorado State University: Fort Collins, CO, USA, 1972. [Google Scholar]
  30. Nedkov, R. Normalized differential greenness index for vegetation dynamics assessment. C. R. l’Acad. Bulg. Sci. 2017, 70, 1143. [Google Scholar]
  31. Collins, T.J. ImageJ for microscopy. BioTechniques 2007, 43, S25–S30. [Google Scholar] [CrossRef] [PubMed]
  32. Congalton, R.G. Accuracy assessment and validation of remotely sensed and other spatial information. Int. J. Wildland Fire 2001, 10, 321–328. [Google Scholar] [CrossRef] [Green Version]
  33. Wong, T.H.; Mansor, S.B.; Mispan, M.R.; Ahmad, N.; Sulaiman, W.N.A. Features extrcation based on object oriented analysis. In Proceedings of the ATC 2003 Conference, Selangor, Malaysia, 20–25 February 2003. [Google Scholar]
  34. Chapman, D.F.; Beca, D.; Hill, J.; Tharmaraj, J.; Jacobs, J.L.; Cullen, B.R. Increasing home-grown forage consumption and profit in non-irrigated dairy systems. 4. Economic performance. Anim. Prod. Sci. 2014, 54, 256–262. [Google Scholar] [CrossRef]
  35. Cornish, M.A.; Hayward, M.D.; Lawrence, M.J. Self-incompatibility in ryegrass. Heredity 1979, 43, 95. [Google Scholar] [CrossRef]
  36. Guthridge, K.M.; Dupal, M.P.; Kölliker, R.; Jones, E.S.; Smith, K.F.; Forster, J.W. AFLP analysis of genetic diversity within and between populations of perennial ryegrass (Lolium perenne L.). Euphytica 2001, 122, 191–201. [Google Scholar] [CrossRef]
  37. Ozkose, A.; Ahmet, T. Morphological and Agronomic Characteristics of Perennial Ryegrass (Lolium perenne L.) Genotypes. Turk. J. Field Crops Res. 2014, 192, 231–237. [Google Scholar] [CrossRef]
  38. Masubuchi, I.; Kajiwara, K.; Honda, Y. Biomass Estimation by The Stereophonic Image Analysis. Available online: https://www.geospatialworld.net/article/biomass-estimation-by-the-stereophonic-image-analysis/ (accessed on 30 August 2019).
  39. Vanha-Majamaa, I.; Salemaa, M.; Tuominen, S.; Mikkola, K. Digitized photographs in vegetation analysis—A comparison of cover estimates. Appl. Veg. Sci. 2000, 3, 89–94. [Google Scholar] [CrossRef]
  40. Topan, H.; Maktav, D.; Jacobsen, K.; Buyuksalih, G. Information content of optical satellite images for topographic mapping. Int. J. Remote Sens. 2009, 30, 1819–1827. [Google Scholar] [CrossRef]
Agronomy 09 00501 g001 550
Figure 1. (a) Multispectral orthomosaic of the experimental site where Green, NIR, Red and Red edge multispectral bands converted to false colours composite to differentiate individual plots; (b) the ground cover classification map in eCognition software package where green represents perennial ryegrass ground cover; orange represents soil and weed cover within the experimental plot boundaries.
Figure 1. (a) Multispectral orthomosaic of the experimental site where Green, NIR, Red and Red edge multispectral bands converted to false colours composite to differentiate individual plots; (b) the ground cover classification map in eCognition software package where green represents perennial ryegrass ground cover; orange represents soil and weed cover within the experimental plot boundaries.
Agronomy 09 00501 g001
Agronomy 09 00501 g002 550
Figure 2. The pairwise comparison of estimated sensor-base plant ground cover and ground truth data. Where MPC is manual plant count, MGC is manual ground cover, MSGC is multispectral sensor-based ground cover, RGBGC is RGB sensor-based ground cover. The diagram also shows the median (the horizontal middle line between upper and lower box boundaries) and mean (₊) observation for a particular method. Data falling outside the lower quartile (Q1) - quartile upper (Q3) range is plotted as outliers (*).
Figure 2. The pairwise comparison of estimated sensor-base plant ground cover and ground truth data. Where MPC is manual plant count, MGC is manual ground cover, MSGC is multispectral sensor-based ground cover, RGBGC is RGB sensor-based ground cover. The diagram also shows the median (the horizontal middle line between upper and lower box boundaries) and mean (₊) observation for a particular method. Data falling outside the lower quartile (Q1) - quartile upper (Q3) range is plotted as outliers (*).
Agronomy 09 00501 g002
Agronomy 09 00501 g003 550
Figure 3. Linear regression n = 72 (a) multispectral sensor-based ground cover (MSGC) vs. manual ground cover (MGC); (b) multispectral sensor-based ground cover vs. manual plant count (MPC); (c) RGB sensor-based ground cover (RGBGC) vs. manual ground cover; (d) RGB sensor-based ground cover vs. manual plant count (MPC).
Figure 3. Linear regression n = 72 (a) multispectral sensor-based ground cover (MSGC) vs. manual ground cover (MGC); (b) multispectral sensor-based ground cover vs. manual plant count (MPC); (c) RGB sensor-based ground cover (RGBGC) vs. manual ground cover; (d) RGB sensor-based ground cover vs. manual plant count (MPC).
Agronomy 09 00501 g003
Agronomy 09 00501 g004 550
Figure 4. Comparison of perennial ryegrass ground cover classification in ImageJ and eCognition developer software; a, b, c and d images represent plot no 3. (A) RGB image (12 megapixels) from nadir position; (B) converted RGB image using HSB colour space settings in ImageJ where red represents perennial ryegrass ground cover and white c represents weed and soil; (C) NDVI reflectance raster image; (D) object-based ground cover classification in eCognition developer where green represents perennial ryegrass ground cover, and orange represents soil and weed.
Figure 4. Comparison of perennial ryegrass ground cover classification in ImageJ and eCognition developer software; a, b, c and d images represent plot no 3. (A) RGB image (12 megapixels) from nadir position; (B) converted RGB image using HSB colour space settings in ImageJ where red represents perennial ryegrass ground cover and white c represents weed and soil; (C) NDVI reflectance raster image; (D) object-based ground cover classification in eCognition developer where green represents perennial ryegrass ground cover, and orange represents soil and weed.
Agronomy 09 00501 g004
Table
Table 1. The Pearson’s correlation between manual parameters and sensor-based observations. p value < 0.001 for all correlation.
Table 1. The Pearson’s correlation between manual parameters and sensor-based observations. p value < 0.001 for all correlation.
ParameterMPCMGCMSGCRGBGCDW
MPC
MGC0.720
MSGC0.6280.799
RGBGC0.6370.7460.892
DW0.6390.5330.4620.513
MPC is manual plant count, MGC is manual ground cover, MSGC is multispectral sensor-based ground cover, RGBGC is RGB sensor-based ground cover, and DW is dry weight.
Table
Table 2. Vegetation indices used for the perennial ryegrass ground cover estimation.
Table 2. Vegetation indices used for the perennial ryegrass ground cover estimation.
Vegetation IndexEquationRef.
Normalized Vegetation Index (NDVI)NDVI = (NIR−RED)/(NIR + RED)[27]
Soil-Adjusted Vegetation Index (SAVI)SAVI = (NIR−RED)/(NIR + RED + L) × (1 + L)[28]
Ratio Vegetation Index (RVI)RVI = RED/NIR[29]
Normalized Difference Greenness Index (NDGI)NDGI = (GREEN−RED)/(GREEN + RED)[30]
Vegetation Index Number (VIN)VIN = NIR/RED[29]
RGB images collected from each plot were analysed in ImageJ software. ImageJ is an open source Java-based image analysing software which has been developed at the National Institutes of Health and the Laboratory, the University of Wisconsin [31]. Different thresholds values of Hue, Saturation and Brightness (HSB) were applied to differentiate the green area of each plot to estimate the ground cover from RGB images (pixel-based supervised image analysis).
Table
Table 3. The Pearson’s correlations between plot level mean vegetation indices extracted from multispectral images and ground-based observations.
Table 3. The Pearson’s correlations between plot level mean vegetation indices extracted from multispectral images and ground-based observations.
ParameterNDVISAVIRVINDGIVINNIR
MPC0.1270.215−0.2090.2090.1650.240
0.2880.0700.0790.0790.1650.042 *
MGC0.2190.335−0.2420.2420.2450.375
0.0650.004 **0.040 *0.040 *0.038 *0.001 **
MSGC0.2380.406−0.2350.2350.2640.474
0.044 *<0.001 ***0.047 *0.047 *0.025 *<0.001 ***
RGBGC0.0280.456−0.2530.2530.3230.522
0.016 *<0.001 ***0.032 *0.032 *0.006 **<0.001 ***
MPC is manual plant count, MGC is manual ground cover, MSGC is multispectral sensor-based ground cover and RGBGC is RGB sensor-based ground cover. In each cell, the top value represents the correlation coefficient and the bottom value is p value. *, p < 0.05; **, p < 0.01; ***, p < 0.001.

Share and Cite

MDPI and ACS Style

Jayasinghe, C.; Badenhorst, P.; Wang, J.; Jacobs, J.; Spangenberg, G.; Smith, K. An Object-Based Image Analysis Approach to Assess Persistence of Perennial Ryegrass (Lolium perenne L.) in Pasture Breeding. Agronomy 2019, 9, 501. https://doi.org/10.3390/agronomy9090501

AMA Style

Jayasinghe C, Badenhorst P, Wang J, Jacobs J, Spangenberg G, Smith K. An Object-Based Image Analysis Approach to Assess Persistence of Perennial Ryegrass (Lolium perenne L.) in Pasture Breeding. Agronomy. 2019; 9(9):501. https://doi.org/10.3390/agronomy9090501

Chicago/Turabian Style

Jayasinghe, Chinthaka, Pieter Badenhorst, Junping Wang, Joe Jacobs, German Spangenberg, and Kevin Smith. 2019. "An Object-Based Image Analysis Approach to Assess Persistence of Perennial Ryegrass (Lolium perenne L.) in Pasture Breeding" Agronomy 9, no. 9: 501. https://doi.org/10.3390/agronomy9090501

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop