Crop phenotyping gives access to vegetation characteristics called phenotypic traits. New phenotypic traits other than yield-related or morphological traits have emerged. They can be directly or indirectly related to growth, development, architecture and to resistance, resilience and tolerance to biotic and abiotic stresses [1
]. Quantifying a complex trait such as drought tolerance or adaptation to heat stress often requires a set of measurable indicators (secondary traits or indirect measurement) over the whole plant [2
]. These indicators of interest must be measured accurately to be valuable for plant-breeding [4
]. However, accessing this information directly to the field is cost- and time-consuming [7
]. For this reason, phenotyping for plant-breeding purpose is limited.
Phenotyping using an optical instrument is a mean to provide phenotypic information in an objective, rapid and non-destructive manner [8
]. Presently, many optical techniques are an integral part of phenotyping procedures such as multispectral imaging and NIR spectrometry [10
]. The most common approach is to propose indices based on a combination of a few spectral bands and related to biochemical variables, biomass or photosynthetic activity. Associated with a vector, pedestrian, tractor or drone, these technologies demonstrate the feasibility of relatively inexpensive tools for high-throughput phenotyping [8
]. As phenotypic traits have diversified, innovative low-cost optical instruments need to be developed to better describe new genotypes.
In recent years, the use of laser speckle has developed mainly in the biomedical field [12
] and very recently in the agricultural field [14
]. Laser speckle imaging is a low-cost, sensitive and noninvasive method. A typical setup for speckle measurement is very simple. It requires an expanded laser light, which might be a diode laser, a detector such as a CCD camera with a lens and a PC with a frame grabber that is able to record a set of images. These features make this method attractive to many applications, which require fast and non-destructive sampling. Combined with polarization, speckle image analysis techniques provide new parameters describing media properties (such as scattering or absorption) [16
] which makes this method very attractive for precision agriculture applications.
Leaf scattering and absorption parameters are related to its chemical composition and structure (cuticle, epidermis, mesophyll) [22
] which is the seat of the photosynthesis process [23
]. They are likely to vary according to the onset of stress or disease, resulting in leaf dryness or wilting for example.
As laser speckle is sensitive to scattering and absorption properties, its measurement is likely to identify structural variations between plants. Statistical parameters are computed from speckle images and provide information on the surveyed sample. These statistical parameters are then potential indicators usable for non-destructive plant phenotyping.
All the statistical parameters constitute a so-called multivariate observation. To analyze a set of speckle measurements associated with a design of experiments, methods of analysis of variance must be adapted to multivariate data. The method called Reduction of Repeatability Error-Analysis of Variance-Simultaneous Component Analysis (REP-ASCA) [24
] is suitable to study the influence of identified factors while reducing error due to the lack of repeatability of measurements.
In this paper, we propose an original method to assess the potential of laser speckle measurements analyzed with chemometrics as a non-destructive phenotyping tool. In particular, sensitivity to physiological criteria, such as inner structure, will be discussed. An optical system combining laser speckle and polarization at two wavelengths will be tested on a case study: the discrimination of sunflower genotypes under identical water conditions.
The objectives are (1) to characterize possible sources of error degrading the speckle measurements, (2) identify variables involved in genotype dissociation, (3) to establish a genotype discrimination model based on these variables.
In this paper, we investigated the potential of laser speckle measurements as a plant phenotyping tool. We were interested in the discrimination of sunflower genotypes.
The combination of laser speckle and polarization at two wavelengths has provided us a set of 20 parameters that can be used as multivariate observations. Using the analysis of variance method called REP-ASCA, we demonstrated that the genotype factor was strictly significant. Therefore, these parameters can be used to differentiate genotypes. We also identified the sources of error due to the lack of repeatability. Mostly, the analysis showed that the errors come from unwilling specular reflections on the leaf and from the variation of leaf chlorophyll content.
With the same method, studying the loadings of the genotype factor principal components pointed out the importance of variables measured at . These variables are mainly related to scattering properties within the leaf. Based on scores on these principal components, a DA can discriminate genotypes with an average error of 20% showing the ability of this new setup to distinguish genotypes.
Speckle measurements were performed at two selected wavelengths where different types of interactions with vegetation leaves occurred. Other wavelengths, specific to water, anthocyanins or carotenoids for example, could be added to improve the discrimination ability of the proposed approach. Moreover, for different genotypes, scattering and absorption properties are expected to vary heterogeneously in the presence of stress. We can make the assumption that the speckle pattern measured on leaves with the proposed setup will vary as well. Therefore, the next step will be to address the feasibility of using laser speckle to compare genotype responses to biotic or abiotic stresses.