Search

International audience; Spectroscopy is today and for two decades strongly used in many fields (pharmacy, agriculture, process, medicine…). This use in a very large number of applications is linked to the great spectral richness of the measurement and therefore to the large amount of accessible chemical information. For plant breeding, spectral reflectance in the visible and near-infrared range (VIS–NIR) embeds a lot of information about vegetation (pigments, structure, water, etc.). Discriminatory power between genotypes can be greatly improved by using high spectral resolution. NIR spectroscopy is still limited in the field for phenotyping compared to existing imaging solutions that are easier to implement.In this study, we will address the potential of high spectral resolution data by using NIR spectroscopy to describe phenotypic responses of maize genotypes to water stress. To that end, data acquired following an experimental design with water-deficient environment are processed using an analysis of variance method adapted to multivariate data called REP-ASCA. For each factor, this method gives its significance, the loadings describing the impacted spectral regions and the scores to classify observations. For a date with proven water stress, the treatment and genotype factors and the interaction term are significant with a p-value threshold at 0.05. Treatment term loadings highlight the spectral regions impacted by the change in irrigation while those of the genotype factor allows to group genotypes according to the yield potential regardless the irrigation. The interaction term loadings are used as a phenotyping trait related to water stress response. Based on this signature, tolerant genotypes are differentiated from sensitive genotypes according to a ranking based on final yield (R = 0.81). This spectral signature was then applied to another environment with a moderate water deficit. For most genotypes, we were able to recover the ranking previously established by the stressed environment (R = 0.60).

[Departement_IRSTEA]Ecotechnologies [TR1_IRSTEA]INSPIRE; Accurate estimation of leaf chlorophyll content (Cab) from remote sensing is of tremendous significance to monitor the physiological status of vegetation or to estimate primary production. Many vegetation indices (Vls) have been developed to retrieve Cab at the canopy level from meter- to decameter-scale reflectance observations. However, most of these Vls may be affected by the possible confounding influence of canopy structure. The objective of this study is to develop methods for Cab estimation using millimeter to centimeter spatial resolution reflectance imagery acquired at the field level. Hyperspectral images were acquired over sugar beet canopies from a ground-based platform in the 400-1000 nm range, concurrently to Cab, green fraction (GF), green area index (GAI) ground measurements. The original image spatial resolution was successively degraded from 1 mm to 35 cm, resulting in eleven sets of hyperspectral images. Vegetation and soil pixels were discriminated, and for each spatial resolution, measured Cab values were related to various Vls computed over four sets of reflectance spectra extracted from the images (soil and vegetation pixels, only vegetation pixels, 50% darkest and brightest vegetation pixels). The selected Vls included some classical Vls from the literature as well as optimal combinations of spectral bands, including simple ratio (SR), modified normalized difference (mND) and structure insensitive pigment index (SIPI). In the case of mND and SIPI, the use of a blue reference band instead of the classical near-infrared one was also investigated. For the eleven spatial resolutions, the four pixel selections and the five VI formats, similar band combinations are obtained when optimizing VI performances: the main bands of interest are generally located in the blue, red, red edge and near-infrared domains. Overall, mND(blue)[728,850] defined as (R-440-R-728)/(R-440+R-850) and computed over the brightest green pixels obtains the best correlations with C-ab for spatial resolutions finer than 8.8 cm with a root mean square error of prediction better than 2.6 mu g/cm(2). Conversely, mND(blue)[728,8501 poorly correlates with variations in GF and GAI, thus reducing the risk of deriving non-causal relationships with Cab that would actually be due to the covariance between C-ab and these canopy structure variables. As mND(blue)[728,8501 can be calculated from most current multispectral sensors, it is therefore a promising VI to retrieve C-ab from millimeter- to centimeter-scale reflectance imagery. (C) 2017 Elsevier Inc. All rights reserved.

AGAP : équipe GE2pop (Génomique évolutive et gestion des populations); Light propagation modeling in 3-dimensional virtual scenes has been successfully applied to many fields, including plant canopies. However, its application to detailed analyses on how multiple scattering affects spectral-based biochemistry assessments has never been proposed. In this article, a wheat canopy model has been built using simulation models included in the open source software platform Open-Alea. Adel-Wheat, a 3D dynamic model of the aerial growth of winter wheat, has been associated with spectra collected on wheat leaves with an ASD spectrometer, and then used as input of the Caribu light propagation model. Caribu calculates the proportion of direct and scattered light for all polygons of the 3D scene. Principal component analysis was first applied to analyze the distribution of resulting spectra in the spectral feature space. Then the influence of canopy structure on quantitative regression models has been considered. For this purpose, a typical agronomical problem, i.e. nitrogen content retrieval, was addressed, using a Partial Least Square regression model. This study exhibits some important results concerning the distribution of collected spectra in the spectral feature space due to multiple scattering, and underlines the physical interpretation of these results. In the short term, it shows that satisfactory nitrogen content prediction (error about 0.5% of dry matter) can be obtained at the plant level, when considering only the plant top leaves. Moreover, its paves the way for future researches to develop spectral analysis tools able to overcome such multiple scattering phenomena.

A method to reduce repeatability error in multivariate data for Analysis of variance-Simultaneous Component Analysis (REP-ASCA) has been developed. This method proposes to adapt the acquisition protocol by adding a set containing repeated measures for describing repeatability error. Then, an orthogonal projection is performed in the row-space to reduce the repeatability error of the original dataset. Finally, ASCA is performed on the orthogonalized dataset. This method was evaluated on NIR spectral data of coffee beans. This study shows that the repeatability error due to physical variations between measurements can alter results of the analysis of variance. These effects are predominant in factors analysis and can be seen on spectra as constant or non-constant baselines. By reducing repeatability error with REP-ASCA, baselines are removed and factor analysis provides more information about chemical content of the factors of interest. (C) 2019 Elsevier B.V. All rights reserved.

International audience; Hyperspectral imaging is a versatile tool, with a promising prospective for the quick identification of foodborne microbial organisms. One of the main limitations for the application of Near Infrared (NIR) Spectral imaging for the identification of microorganisms in culture media is sample presentation. Polystyrene plastic petri dishes are one the most common sample holder used in microbiological studies. They are cheap, disposable, easy to handle and well standardized. Nevertheless, their use is generally proscribed for NIR spectral imaging, as the spectral signature of Polystyrene dishes can interfere with the spectral information of interest from the sample. The aim of this study is to characterize the effect different types of sample presentation, glass and polystyrene petri dishes, and to test orthogonal projection as a pretreatment to reduce the interference of polystyrene dishes for NIR imaging of Luria Bertani Agar. The combination of Savitzky Golay smoothing, SNV and orthogonal projection allowed to attenuate the spectral differences between agar samples presented on polystyrene and glass dishes. This same approach allowed to reduce the effect of the polystyrene dishes on samples with different agar thicknesses. Future research will test this approach as a method to facilitate the identification of bacteria on polystyrene petri dishes through NIR hyperspectral imaging.

International audience; Applying a calibration model onto hyperspectral (HS) images is of great interest because it produces images of chemical or physical properties. HS imaging is widely used in this way in food processing industries for monitoring product quality and process control. In this context, one of the main difficulties in the application of regression models to HS images is to evaluate the error of the obtained predictions, since in a proximal imaging set up, the size of the pixels is usually much smaller than the area required to obtain a wet chemical reference. Moreover, the selection of regression model parameters, such as the number of latent variables (LV) in a partial least squares (PLS) model, can modify the appearance of the prediction maps. The objective of this work is to propose an approach based on geostatistical indices to use spatial information of prediction maps for supporting the evaluation of regression models applied to HS images. This work stablishes a theoretical connection between linear regression model performance estimates and the spatial decomposition of variance in prediction maps, when the ground truth to estimate is spatially structured. This approach was tested in a simulated dataset and two real case studies. Geostatistical indices of the prediction maps were compared to model performance metrics for PLS models with increasing number of LV. The theoretical framework was proven by the results on the simulated dataset. In particular, the evolution of the nugget effect, C0; corresponded with the evolution of the random error of the model. Conversely, the error term of the model related with the slope of the model corresponded with the evolution of the structured variance observed in the prediction maps. On the real case studies, geostatistical indexes, extracted from the prediction maps, allowed to quantitatively evaluate the spatial structure of the estimations and complement the Root Mean Standard Error of Cross Validation (RMSECV) for the choice of optimal number of LV to consider in the model. The main advantage of this approach is that it does not require ground truth values. It could be used as a source of information for supporting the choice of optimum calibration parameters, such as the number of latent variables, or the choice of pre-treatments, complementing the traditional visual inspection of prediction maps with quantitative and objective metrics. (C) 2019 Elsevier B.V. All rights reserved.

This study aims to investigate the potential of an original polarized hyperspectral imaging (HSI) setup in the spectral domain of 400–1000 nm for sunflower leaves in real-world. Dataset 1 includes hypercubes of sunflower leaves in two varieties with different life growth stages, while Dataset 2 is comprised of healthy and contaminated sunflower leaves suffering from powdery mildew (PM) and/or septoria leaf spot (SLS). Cross polarised ( R ⊥ ), parallel polarised ( R | | ) reflectance signals, R BS ( R | | + R ⊥ ) and R SS ( R | | - R ⊥ ) spectra were obtained and used to develop partial least squares-discriminant analysis (PLS-DA) models. Surface information played an important role in separating two varieties of leaves due to the fact that the best model performance was achieved by using R SS mean spectra, while both surface and subsurface were equally important in classifying leaves between two major growth stages because model of R BS mean spectra outperformed other models. The best classification model for disease detection was achieved by using pixel R ⊥ spectra with the correct classification rate (CCR) of 0.963 for both cross validation and prediction, meaning that subsurface spectral features were the most important to detect infected leaves. The resulting classification maps were also displayed to visualize the distribution of the infected regions on the leaf samples. The overall results obtained in this research showed that the developed polarized-HSI system coupled with multivariate analysis has considerable promise in agricultural real-world applications.

This study was carried out to investigate the feasibility of an original polarized hyperspectral imaging setup in the spectral range of 400–1100 nm for enhancement of absorbance signal measurement on highly scattering samples. Spatial response and spectral calibration have been verified, indicating the consistency of this system and reliability of the acquired data. Model samples consisting of layered sand were prepared and used to uncover the hidden spectral information. In the model matrix, sand worked as scattering particle and dye E141 as absorbing material. Cross ([Formula: see text]) and parallel ([Formula: see text]) reflectance signals, along with the back-scattered reflectance, [Formula: see text]([Formula: see text]+[Formula: see text]) and the weakly scattered reflectance [Formula: see text] ([Formula: see text]−[Formula: see text]) spectra were computed and compared. Results demonstrated that cross-polarized images showed more subsurface information from the second layer due to the rejection of the superficial reflectance, while weakly scattered reflectance ([Formula: see text]) preserved only the surface information from the first layer. In addition, polarized light spectroscopy absorbance based on Dahm's equation in the frame of the representative layer theory and standard normal variate preprocessing [Formula: see text] spectra were also obtained from the prepared model matrix. The visual inspection of spectral curves revealed that [Formula: see text] and PoLiS absorbance showed two narrow peaks at 405 nm and 630 nm that were less impacted by multi-scattering effects. Partial least squares regression models were developed to predict dye concentration in the mixture sample. Consistent with the spectral profiles, [Formula: see text] and PoLiS absorbance presented the best model performances with determination of coefficients of prediction ([Formula: see text]) equal to 0.96 and 0.95, respectively. The resulting distribution maps of S1/S2 sand sample again confirmed the superior performance of [Formula: see text] and PoLiS absorbance, manifesting their better ability to reveal chemically related information. The overall results obtained in this research showed that the developed polarized-hyperspectral imaging system coupled with scattering correction methods has great potential for the analysis of powdered or turbid samples.

International audience; The paper proposes a methodology based on near-infrared (NIR) spectrometry for studying stratigraphy and depth profiles in archaeological excavations. The NIR spectra can be used to describe and complement the wet chemical analysis. Soil samples were collected from a 0.8 m deep stratigraphy of a Neolithic site that were analyzed by three different NIR instrumentations. Phosphate- and magnetic susceptibility and inductively-coupled plasma mass spectrometry measurements were also conducted as reference analysis. Principal component analysis on the data from three different NIR instrumentations gave useful score plots that allowed grouping of the samples. The results from the lab spectrometer were most useful, although the hyperspectral NIR camera was the fastest method to obtain spectra of many samples from one image. The paper shows how the NIR spectral data can be used for multivariate analysis to get meaningful conclusions on archaeological soils and sediments, especially in terms of understanding site development/phases and soil formation.

The detection of plant diseases, including fungi, is a major challenge for reducing yield gaps of crops across the world. We explored the potential of the PROCOSINE radiative transfer model to assess the effect of the fungus Pseudocercospora fijiensis on leaf tissues using laboratory-acquired submillimetre-scale hyperspectral images in the visible and near-infrared spectral range. The objectives were (i) to assess the dynamics of leaf biochemical and biophysical parameters estimated using PROCOSINE inversion as a function of the disease stages, and (ii) to discriminate the disease stages by using a Linear Discriminant Analysis model built from the inversion results. The inversion results show that most of the parameter dynamics are consistent with expectations: for example, the chlorophyll content progressively decreased as the disease spreads, and the brown pigments content increased. An overall accuracy of 78.7% was obtained for the discrimination of the six disease stages, with errors mainly occurring between asymptomatic samples and first visible disease stages. PROCOSINE inversion provides relevant ecophysiological information to better understand how P. fijiensis affects the leaf at each disease stage. More particularly, the results suggest that monitoring anthocyanins may be critical for the early detection of this disease.

Most methods for retrieving foliar content from hyperspectral data are well adapted either to remote-sensing scale, for which each spectral measurement has a spatial resolution ranging from a few dozen centimeters to a few hundred meters, or to leaf scale, for which an integrating sphere is required to collect the spectral data. In this study, we present a method for estimating leaf optical properties from hyperspectral images having a spatial resolution of a few millimeters or centimeters. In presence of a single light source assumed to be directional, it is shown that leaf hyperspectral measurements can be related to the directional hemispherical reflectance simulated by the PROSPECT radiative transfer model using two other parameters. The first one is a multiplicative term that is related to local leaf angle and illumination zenith angle. The second parameter is an additive specular-related term that models BRDF effects. Our model was tested on visible and near infrared hyperspectral images of leaves of various species, that were acquired under laboratory conditions. Introducing these two additional parameters into the inversion scheme leads to improved estimation results of PROSPECT parameters when compared to original PROSPECT. In particular, the RMSE for local chlorophyll content estimation was reduced by 21% (resp. 32%) when tested on leaves placed in horizontal (resp. sloping) position. Furthermore, inverting this model provides interesting information on local leaf angle, which is a crucial parameter in classical remote-sensing.

International audience; This paper proposes a novel approach to classify hyperspectral (HS) images using both spectral and spatial information. It first consists of a supervised spectral dimension reduction step that transforms the HS image into a score image that has fewer channels. These channels are chosen so as to enhance distances between classes to be discriminated and to reduce background variability, thus leading to edges that correspond to actual class borders. In the second step, applying an edge-preserving spatial regularization on this score image leads to a lowered background variability. Therefore, in the third step, the pixel-wise classification of the regularized score image is greatly improved. We implement this approach using the partial least squares (PLS) method for spectral dimension reduction and the anisotropic diffusion for spatial regularization. We then compare linear discriminant analysis (LDA), K-nearest neighbors (KNN), and support vector machine (SVM) for the class decision. The effectiveness of our method was evaluated with three remotely sensed HS images. Its robustness was also assessed for different training sets, since the latter has a crucial influence on classification performance. On average, our method gave better results in terms of classification accuracy and was more robust than other classification methods tested with the same images.

International audience; This dataset presents two series of hyperspectral images of healthy and infected apple tree leaves acquired daily, from two days after inoculation until an advanced stage of infection (11 days after inoculation). The hyperspectral images were calibrated by reflection correction and registered to match the geometry of one reference image. On the last experiment day, scab positions are provided.

[Departement_IRSTEA]Ecotechnologies [TR1_IRSTEA]INSPIRE [ADD1_IRSTEA]Équiper l'agriculture; International audience; The use of digital soil mapping, with the help of spectroscopic data, provides a non-destructive and cost-efficient alternative to soil property laboratory measurements. Visible, near-infrared and short wave infrared (VNIR/ SWIR, 400–2500 nm) hyperspectral imaging is one of the most promising tools for topsoil property mapping. The aim of this study was to test the sensitivity of soil property prediction results to coarsening image spectral resolution. This may offer an analysis of the potential of forthcoming hyperspectral satellite sensors, e.g., HYPerspectral X IMagery (HYPXIM) or Environmental Mapping and Analysis Program (EnMAP), and existing multispectral sensors, e.g., SENTINEL-2 Multispectral Sensor Instrument (MSI) or LANDSAT-8 Operational Land Imager (OLI), for soil properties mapping. This study used VNIR/SWIR hyperspectral airborne data acquired by the AISA-DUAL sensor (initial spectral and spatial resolutions of approximately 5 nm and 5 m, respectively) over a 300 km2 Mediterranean rural region. Ten spectral configurations were built and divided in the following two groups: i) six spectral configurations corresponding to simulated sensors with regular spectral resolution from 5 nm to 200 nm (i.e., the Full Width at Half Maximum (FWHM) remains constant throughout the considered spectral domain; this includes the simulation of the forthcoming HYPXIM and EnMAP hyperspectral satellites) and ii) four spectral configurations corresponding to existing multispectral sensors with irregular spectral resolution (i.e., the FWHM differs from spectral sampling interval; Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER), SENTINEL-2 MSI, LANDSAT-7 Enhanced Thematic Mapper (ETM+) and LANDSAT-8 OLI). The soil property studied in this paper is the clay content, defined as the percentage of granulometric fraction finer than 2 μm by weight of the soil, which will be estimated using the partial least squares regression method. Our results showed that i) spectral configurations with regular spectral resolutions from 5 to 100 nm provided similar and good clay content prediction performances (R2val > 0.7 and RPIQ > 3) and allowed clay mapping with correct short-scale variations, ii) the spectral configuration with a regular spectral resolution of 200 nm provided unsatisfactory clay content prediction performance (R2val ≃ 0.01 and RPIQ ≃ 1.65) and iii) the ASTER sensor was the only existing multispectral sensor that provided both correct performance of clay content estimation (R2val ≃ 0.8 and RPIQ ≃ 3.72) and correct clay mapping. Therefore, clay mapping by the ASTER multispectral data should be pursued while awaiting the launch of forthcoming hyperspectral satellite sensors (e.g., HYPXIM and EnMAP), which will be good candidates for future large clay mapping campaigns over bare soils.

We modeled spectral variability obtained with wheat and weed reflectance spectra.We examined pre-treatment influence on discrimination results using PLS-LDA and SVM.We discussed the bias and class separability obtained with PLS-LDA scores.Results prove that smoothing and logarithm are the best pre-treatments for this study. The optimal processing of spectral data often requires specific pre-treatments. In the context of spectral discrimination, results can be greatly improved using the relevant pre-treatment. Most importantly, the pre-treatment must be suited to the nuisance variability that has to be removed. This study focuses on discrimination of weed and wheat using spectra acquired in outdoor conditions with uncontrolled lighting and leaf orientations. Both vegetation spectra are highly similar, and due to the context of acquisition, a lot of spectral variability is present. This nuisance variability is modeled using an additive, multiplicative and noise term, each of which affects the measured spectra. Several pre-treatments were therefore evaluated according to their potential to deal with this variability and their effects were described in the feature space. Finally, results obtained with these pre-treatments combined with two discrimination methods (PLS-LDA and Gaussian SVM) are compared and discussed. Results showed that, thanks to their ability to remove nuisance variability, most pre-treatments are effective in terms of classification accuracy. Gaussian SVM classification results are less influenced by pre-treatments than those of PLS-LDA, since the former compensates the pre-treatment effect by using a different non-linear kernel. For this data-set, the best discrimination result was obtained using the combination logarithm and PLS-LDA. Logarithm actually transforms the multiplicative effect into and additive one, which is then effectively dealt with by PLS-LDA.

International audience; Laboratory Visible-Near Infrared (Vis-NIR) spectroscopy is a good alternative to costly physical and chemical soil analysis to estimate a wide range of soil properties. Various statistical methods relate soil Vis-NIR spectra to soil properties including partial least-squares regression (PLSR), the most common multivariate statistical technique in soil science. Most efforts are generally dedicated to the comparison of methodologies and their optimization for the estimation of soil properties. Instead, the focus of this paper is to assess the prediction of soil properties from laboratory Vis-NIR spectroscopy data in regards to spectral degradation. Consecutively, both spectra quality and PLSR models quality are analyzed across the definition of different spectral configurations, each one characterized by three parameters: the number of spectral bands, the spectral resolution and the spectral sampling interval. The originality of this work is to perform this study on four soil properties with different spectral absorption features due to their various physico-chemical interactions with soil substrate, namely: clay, free iron oxides, calcium carbonate (CaCO3) and pH. The initial database is composed of 1961 spectral bands, spectral resolutions of 3 and 10 nm in the 400'1000 nm and 1000'2500 nm ranges, respectively, with a resampled spectral interval of 1 nm. Seven degraded spectral configurations were built from this reference database with a number of spectral bands decreasing from 328 to 10, a spectral resolution decreasing from 3 nm to 200 nm, and a spectral sampling interval equaling the spectral resolution (i.e., uniform interval sampling). All of these databases were composed of 148 soil samples collected at a Mediterranean site. PLSR predicted the four selected soil properties, and the results were as follows: (1) the prediction performances of the PLSR models were accurate and globally stable with a spectral resolution between 3 and 60 nm regardless of the soil properties (R2 decreased from 0.8 to 0.77 for clay, from 0.88 to 0.84 for CaCO3, from 0.66 to 0.58 for pH and remained constant at 0.78 for iron), (2) the prediction performances decreased, but remained acceptable for clay, iron oxides and CaCO3 at spectral resolutions between 60 and 200 nm (R2 > 0.7), (3) the sensitivity of a given soil property to spectral configurations depended on its spectral features and correlations with other soil properties.

International audience; Remote sensing has gained much attention for agronomic applications such as crop management or yield estimation. Crop phenotyping under field conditions has recently become another important application that requires specific needs: the considered remote-sensing method must be (1) as accurate as possible so that slight differences in phenotype can be detected and related to genotype, and (2) robust so that thousands of cultivars potentially quite different in terms of plant architecture can be characterized with a similar accuracy over different years and soil and weather conditions. In this study, the potential of nadir and off-nadir ground-based spectro-radiometric measurements to remotely sense five plant traits relevant for field phenotyping, namely, the leaf area index (LAI), leaf chlorophyll and nitrogen contents, and canopy chlorophyll and nitrogen contents, was evaluated over fourteen sugar beet (Beta vulgaris L.) cultivars, two years and three study sites. Among the diversity of existing remote-sensing methods, two popular approaches based on various selected Vegetation Indices (VI) and PROSAIL inversion were compared, especially in the perspective of using them for phenotyping applications. Overall, both approaches are promising to remotely estimate LAI and canopy chlorophyll content (RMSE'10%). In addition, VIs show a great potential to retrieve canopy nitrogen content (RMSE=10%). On the other hand, the estimation of leaf-level quantities is less accurate, the best accuracy being obtained for leaf chlorophyll content estimation based on VIs (RMSE=17%). As expected when observing the relationship between leaf chlorophyll and nitrogen contents, poor correlations are found between VIs and mass-based or area-based leaf nitrogen content. Importantly, the estimation accuracy is strongly dependent on sun-sensor geometry, the structural and biochemical plant traits being generally better estimated based on nadir and off-nadir observations, respectively. Ultimately, a preliminary comparison tends to indicate that, providing that enough samples are included in the calibration set, (1) VIs provide slightly more accurate performances than PROSAIL inversion, (2) VIs and PROSAIL inversion do not show significant differences in robustness across the different cultivars and years. Even if more data are still necessary to draw definitive conclusions, the results obtained with VIs are promising in the perspective of high-throughput phenotyping using UAV-embedded multispectral cameras, with which only a few wavebands are available.

Visible Near-Infrared (Vis-NIR, 400-2500 nm) hyperspectral imaging has proven to be a valuable tool for mapping soil properties over bare soils. To date, most predictive models presented in literature, have been built from calibration databases made up of both Vis-NIR imaging spectra (predictor variables) and soil properties (response variables). Nevertheless, the constitution of such calibration databases is costly both in terms of time and money, as this implies soil sampling over hundreds of pixels of bare soils and physico-chemical soil analysis of these samples. We propose to test the transfer of predictive models built from laboratory database, to Vis-NIR hyperspectral image. Four transfer methods were tested (model updating, Repfile, Transfer by Orthogonal Projection TOP and Piecewise Direct Standardization PDS). Their respective performances were evaluated and compared to those obtained using two other predictive models. The first predictive model was obtained by directly applying a partial least square regression (PLSR) calibrated in laboratory, onto the hyperspectral image. The second model was obtained by applying a PLSR built from Vis-NIR spectra extracted from the hyperspectral image, onto the hyperspectral image. The transferred models are based on the use of standards. Standards were selected in order to (i) take into account or (ii) ignore field soil stratification knowledge. Moreover, the impact of the number of standards used in the transferred models was studied. This approach was tested for clay content prediction over the La Peyne Valley area in Southern France (24 km(2), in a Mediterranean context) covered by HyMap Vis-NIR airborne data. The results show that 1) applying PLSR model from laboratory spectra to hyperspectral image without calibration transfer, provides poor prediction performances (R-val(2) = 0.41, RMSEP = 103 g/kg), 2) transferring this model from laboratory spectra to hyperspectral image offered good prediction performances (R-test(2) median above 0.5 and RMSEP median below 70 g/kg), whatever the transfer method used and from only 15 used standards, and 3) taking into account field soil stratification knowledge in the standards selection improves prediction performances (R-test(2) median above 0.6 and RMSEP median below 60 g/kg) whatever the transfer method and from only 10 used standards. Finally, transferring models from the laboratory to a hyperspectral image gave better mapping performances than those obtained using PLSR models built from airborne spectra.

International audience; Raman chemical imaging provides both spectral and spatial information on a pharmaceutical drug product. Even if the main objective of chemical imaging is to obtain distribution maps of each formulation compound, identification of pure signals in a mixture dataset remains of huge interest. In this work, an iterative approach is proposed to identify the compounds in a pharmaceutical drug product, assuming that the chemical composition of the product is not known by the analyst and that a low dose compound can be present in the studied medicine. The proposed approach uses a spectral library, spectral distances and orthogonal projections to iteratively detect pure compounds of a tablet. Since the proposed method is not based on variance decomposition, it should be well adapted for a drug product which contains a low dose product, interpreted as a compound located in few pixels and with low spectral contributions. The method is tested on a tablet specifically manufactured for this study with one active pharmaceutical ingredient and five excipients. A spectral library, constituted of 24 pure pharmaceutical compounds, is used as a reference spectral database. Pure spectra of active and excipients, including a modification of the crystalline form and a low dose compound, are iteratively detected. Once the pure spectra are identified, multivariate curve resolution-alternating least squares process is performed on the data to provide distribution maps of each compound in the studied sample. Distributions of the two crystalline forms of active and the five excipients were in accordance with the theoretical formulation.

International audience; Visible and near-infrared (Vis-NIR) spectra are generated by the combination of numerous low resolution features. Spectral variables are thus highly correlated, which can cause problems for selecting the most appropriate ones for a given application. Some decomposition bases such as Fourier or wavelet generally help highlighting spectral features that are important, but are by nature constraint to have both positive and negative components. Thus, in addition to complicating the selected features interpretability, it impedes their use for application-dedicated sensors.In this paper we have proposed a new method for feature selection: Application-Dedicated Selection of Filters (ADSF). This method relaxes the shape constraint by enabling the selection of any type of user defined custom features. By considering only relevant features, based on the underlying nature of the data, high regularization of the final model can be obtained, even in the small sample size context often encountered in spectroscopic applications. For larger scale deployment of application-dedicated sensors, these predefined feature constraints can lead to application specific optical filters, e.g., lowpass, highpass, bandpass or bandstop filters with positive only coefficients.In a similar fashion to Partial Least Squares, ADSF successively selects features using covariance maximization and deflates their influences using orthogonal projection in order to optimally tune the selection to the data with limited redundancy. ADSF is well suited for spectroscopic data as it can deal with large numbers of highly correlated variables in supervised learning, even with many correlated responses. © 2016 Elsevier B.V.

International audience; Radiative transfer models have long been used to characterize the foliar content at the leaf and canopy levels. However, they still do not apply well to close-range imaging spectroscopy, especially because directional effects are usually not taken into account. For this purpose, we introduce a physical approach to describe and simulate the variation in leaf reflectance observed at this scale. Two parameters are thus introduced to represent (1) specular reflection at the leaf surface and (2) local leaf orientation. The model, called COSINE (ClOse-range Spectral ImagiNg of lEaves), can be coupled with a directional-hemispherical reflectance model of leaf optical properties to relate the measured reflectance to the foliar content. In this study, we show that, when combining COSINE with the PROSPECT model, the overall PROCOSINE model allows for a robust sub-millimeter retrieval of foliar content based on numerical inversion and pseudo bidirectional reflectance factor hyperspectral measurements.The relevance of the added parameters is first shown through a sensitivity analysis performed in the visible and near-infrared (VNIR) and shortwave infrared (SWIR) ranges. PROCOSINE is then validated based on VNIR and SWIR hyperspectral images of various leaf species exhibiting different surface properties. Introducing these two parameters within the inversion allows us to obtain accurate maps of PROSPECT parameters, e.g., the chlorophyll content in the VNIR range, and the equivalent water thickness and leaf mass per area in the SWIR range. Through the estimation of light incident angle, the PROCOSINE inversion also provides information on leaf orientation, which is a critical parameter in vegetation remote sensing.

International audience; In order to be independent from light source and atmospheric conditions, radiance values extracted from a remote hyperspectral image have to be converted into reflectance values before data processing. Several methods have been proposed in the literature but they require that the lighting/and or atmospheric conditions to be estimated. In the framework of supervised classification, we propose an approach to deal with such lighting and atmospheric temporal fluctuations without reference measurement. Assuming that materials to the surface objects to be discriminated is Lambertian, we show that the difference in lighting conditions after a log-transformation of both reflectance and radiance signals can be expressed as an additive effect. This effect remains additive after the use of a linear dimension reduction method and can be efficiently estimated in a low dimensional feature space. In the feature space, this difference in ligthing can be estimated and thus corrected by finding the translation for which the class densities obtained for each image best overlaps (using cross correlation). This novel approach was applied on a remote sensing data set over the Quiberon peninsula France. For the tested images, classification results obtained with this approach were comparable to those obtained using a classical reflectance correction technique.

In current industrial environments there is an increasing need for practical and inexpensive quality control systems to detect the foreign food materials in powder food processing lines. This demand is especially important for the detection of product adulteration with traces of highly allergenic products, such as peanuts and tree nuts. Manufacturing industries dealing with the processing of multiple powder food products present a substantial risk for the contamination of powder foods with traces of tree nuts and other adulterants, which might result in unintentional ingestion of nuts by the sensitised population. Hence, the need for an in-line system to detect nut traces at the early stages of food manufacturing is of crucial importance. In this present work, a feasibility study of a spectral index for revealing adulteration of tree nut and peanut traces in wheat flour samples with hyperspectral images is reported. The main nuts responsible for allergenic reactions considered in this work were peanut, hazelnut and walnut. Enhanced contrast between nuts and wheat flour was obtained after the application of the index. Furthermore, the segmentation of these images by selecting different thresholds for different nut and flour mixtures allowed the identification of nut traces in the samples. Pixels identified as nuts were counted and compared with the actual percentage of peanut adulteration. As a result, the multispectral system was able to detect and provide good visualisation of tree nut and peanut trace levels down to 0.01% by weight. In this context, multispectral imaging could operate in conjuction with chemical procedures, such as Real Time Polymerase Chain Reaction and Enzyme-Linked Immunosorbent Assay to save time, money and skilled labour on product quality control. This approach could enable not only a few selected samples to be assessed but also to extensively incorporate quality control surveyance on product processing lines.

International audience; Raman chemical imaging provides chemical and spatial information about pharmaceutical drug product. By using resolution methods on acquired spectra, the objective is to calculate pure spectra and distribution maps of image compounds. With multivariate curve resolution-alternating least squares, constraints are used to improve the performance of the resolution and to decrease the ambiguity linked to the final solution. Non negativity and spatial local rank constraints have been identified as the most powerful constraints to be used. In this work, an alternative method to set local rank constraints is proposed. The method is based on orthogonal projections pretreatment. For each drug product compound, raw Raman spectra are orthogonally projected to a basis including all the variability from the formulation compounds other than the product of interest. Presence or absence of the compound of interest is obtained by observing the correlations between the orthogonal projected spectra and a pure spectrum orthogonally projected to the same basis. By selecting an appropriate threshold, maps of presence/absence of compounds can be set up for all the product compounds. This method appears as a powerful approach to identify a low dose compound within a pharmaceutical drug product. The maps of presence/absence of compounds can be used as local rank constraints in resolution methods, such as multivariate curve resolution-alternating least squares process in order to improve the resolution of the system. The method proposed is particularly suited for pharmaceutical systems, where the identity of all compounds in the formulations is known and, therefore, the space of interferences can be well defined.

International audience; In this work, Raman hyperspectral images and Multivariate Curve Resolution Alternating Least Squares (MCR-ALS) are used to study the distribution of actives and excipients within a pharmaceutical drug product. This article is mainly focused on the distribution of a low dose constituent. Different approaches are compared, using initially filtered or non-filtered data, or using a column-wise augmented dataset before starting the MCR-ALS iterative process including appended information on the low dose component. In the studied formulation, magnesium stearate is used as a lubricant to improve powder flowability. With a theoretical concentration of 0.5% w/w in the drug product, the spectral variance contained in the data is weak. By using a Principal Component Analysis (PCA) filtered dataset as a first step of the MCR-ALS approach, the lubricant information is lost in the non-explained variance and its associated distribution in the tablet cannot be highlighted. A sufficient number of components to generate the PCA noise-filtered matrix has to be used in order to keep the lubricant variability within the data set analyzed or, otherwise, work with the raw non-filtered data. Different models are built using an increasing number of components to perform the PCA reduction. It is shown that the magnesium stearate information can be extracted from a PCA model using a minimum of 20 components. In the last part, a column-wise augmented matrix, including a reference spectrum of the lubricant, is used before starting MCR-ALS process. PCA reduction is performed on the augmented matrix, so the magnesium stearate contribution is included within the MCR-ALS calculations. By using an appropriate PCA reduction, with a sufficient number of components, or by using an augmented dataset including appended information on the low dose component, the distribution of the two actives, the two main excipients and the low dose lubricant are correctly recovered.

The use of a common environment for processing different powder foods in the industry has increased the risk of finding peanut traces in powder foods. The analytical methods commonly used for detection of peanut such as enzyme-linked immunosorbent assay (ELISA) and real-time polymerase chain reaction (RT-PCR) represent high specificity and sensitivity but are destructive and time-consuming, and require highly skilled experimenters. The feasibility of NIR hyperspectral imaging (HSI) is studied for the detection of peanut traces down to 0.01% by weight. A principal-component analysis (PCA) was carried out on a dataset of peanut and flour spectra. The obtained loadings were applied to the HSI images of adulterated wheat flour samples with peanut traces. As a result, HSI images were reduced to score images with enhanced contrast between peanut and flour particles. Finally, a threshold was fixed in score images to obtain a binary classification image, and the percentage of peanut adulteration was compared with the percentage of pixels identified as peanut particles. This study allowed the detection of traces of peanut down to 0.01% and quantification of peanut adulteration from 10% to 0.1% with a coefficient of determination ( r2) of 0.946. These results show the feasibility of using HSI systems for the detection of peanut traces in conjunction with chemical procedures, such as RT-PCR and ELISA to facilitate enhanced quality-control surveillance on food-product processing lines.

Independent component analysis (ICA) was used as a blind source separation method on a Raman image of a pharmaceutical tablet. Calculations were performed without a priori knowledge concerning the formulation. The aim was to extract the pure signals from the initial data set in order to examine the distribution of actives and major excipients within the tablet. As a method based on the decomposition of a matrix of mixtures of several components, the number of independent component to choose is a critical step of the analysis. The ICA_by_blocks method, based on the calculation of several models using an increasing number of independent components on initial matrix blocks, was used. The calculated ICA signals were compared with the pure spectra of the formulation compounds. High correlations between the two active principal ingredient spectra and their corresponding calculated signals were observed giving a good overview of the distributions of these compounds within the tablet. Information from the major excipients (lactose and avicel) was found in several independent components but the ICA approach provides high level of information concerning their distribution within the tablet. However, the results could vary considerably by changing the number of independent components or the preprocessing method. Indeed, it was shown that under-decomposition of the matrix could lead to better signal quality (compared to the pure spectra) but in that case the contributions due to minor components or effects were not correctly identified and extracted. On the contrary, over-decomposition of the original dataset could provide information about low concentration compounds at the expense of some loss of signal interpretability for the other compounds.

In the context of precision agriculture, hyperspectral image classification has already proved its potential to classify highly similar vegetation that could not be achieved with standard RGB camera. However, the in-field conditions do not allow a perfect classification while using conventional methods that only deal with spectral information. Several study already showed the potential of methods taking into account the spatial information embedded within hyperspectral images. In this paper, the spectral-spatial information is enhanced using a 3D multi-resolution discrete wavelet decomposition of the hyperspectral image. The smoothed image is obtained by zeroing noisy coefficients in the appropriate decomposed data cube before wavelet reconstruction. Then, Partial-Least-Squares Linear-Discriminant-Analysis is applied for classification. Experimental results obtained with this method outperformed standard spectral pre-processing and also non-linear classification method such as SVM.

International audience; Visible-near-infrared hyperspectral imaging was tested for its suitability for monitoring the moisture content (MC) of wood samples during natural drying. Partial least-squares regression (PLSR) prediction of MC was performed on the basis of average reflectance spectra obtained from hyperspectral images. The validation showed high prediction accuracy. The results were compared concerning the PLSR prediction of MC mapping from raw spectra and standard normal variate (SNV) treatment. SNV pretreatment leads to the best results for visualizing the MC distribution in wood. Hyperspectral imaging has a high potential for monitoring the water distribution of wood.; Cet article s’intéresse à l’imagerie hyperspectrale visible-proche infrarouge en tant qu’outil de suivi de la teneur en eau d’échantillons de bois durant la phase de séchage naturel. Des prédictions par régression PLS ont été conduites à partir de spectres moyens de réflectance issus d’images hyperspectrales. La validation met en évidence une bonne précision de prédiction. Les résultats obtenus à partir de spectres bruts ou prétraités par SNV ont été comparés. Le prétraitement SNV conduit a un meilleur résultat. Il permet de visualiser la distribution spatiale de la teneur en eau dans l’échantillon, confirmant l’intérêt de l’imagerie hyperspectrale pour ce type de suivi.

Le concept d’agriculture de précision consiste à gérer spatialement les pratiques culturales en fonction des variabilités intra-parcellaires. La plupart des outils associés utilisent des capteurs optiques ou imageurs. L’imagerie aérienne est notamment souvent utilisée pour compléter ou remplacer des capteurs embarqués sur machine agricole, mais elle présente des inconvénients comparée aux capteurs optiques de " proxi-détection " : résolution, accessibilité, coût, temporalité, etc. Nous proposons dans ce chapitre d’explorer le domaine de la proxi-détection en s’intéressant d’abord au développement de systèmes d’acquisition d’image robustes, puis aux méthodes de traitement d’image à travers divers exemples d’applications : caractérisation de cultures de blé d’une part (détection/comptage d’épis, détection d’adventices), évolution de la germination de graines (chicorée) d’autre part. / The concept of precision agriculture consists to spatially manage crop management practices according to in-field variability. Most of the tools used for Precision Farming utilize optical and/or imaging sensors. Remote sensing imaging devices are often used to complete or replace embedded sensors onboard the agricultural machinery, however they get some drawbacks compared to “proxy-detection” optical sensors: resolution, accessibility, cost, temporality, etc. We propose in this chapter to explore the proxi-detection domain by focusing first on the development of robust image acquisition systems, and secondly on the use of image processing through different application examples dealing with wheat crop characterization (detection and counting of wheat ears, weed detection), and evaluation of seed development/germination performance of chicory achenes.

International audience; Calibrating consists in predicting Y, a quantitative variable of interest, using P explaining variables. PLSR-Projection to Latent Structures Regression (Wold,[1]) is the most popular method, very powerfull when a calibration dataset is available. Other methods don't need a calibration dataset, they are called direct calibrations. Two of them have been proposed previously: DC-Direct Calibration (as described in Martens and Naes, [2]) and SBC-Science Based Calibration (Marbach([3])). New method called IDC-Improved Direct Calibration is proposed. As for DC, this approach is based on an orthogonal projection. IDC projector is obtained by merging DC projector (consisting only in pure spectra of chemical compounds), and vectors characterising physical influence factors (consisting in PCA loadings onto a design dataset). Indeed, hyperspectral image analysis is a case where calibration data are not available. Thus, it's interesting to use direct calibration methods instead of PLSR. With the prior knowledge of a few reference spectra, and modelling some noise from the hyperspectral image itself, it's possible to identify objects of interest from the background. This method is also simple and understandable, very quick and easy to compute.

International audience; Few methods are proposed in the litterature for coupling the spectral and the spatial dimension available on hyperspectral images. This paper proposes a generic segmentation scheme named butterfly based on an iterative process and a cross analysis of spectral and spatial information. Indeed, spatial and spatial structures are extracted in spatial and spectral space respectively both taking into account the other one. To apply this layout on hyperspectral imgages, we focus particulary on spatial and spectral structures i.e. topologic concepts and latent variable for the spatial and the spectral space respectively. Moreover, a cooperation scheme with these structures is proposed. Finally, results obtained on real hyperspectral images using this specific implementation of the butterfly approach are presented and discussed.

International audience; In this paper, we describe a methodology based on imaging system shifting and Fourier transform to recover the spatial distribution of the sensivity of a hyperspectral imaging system. The methodology mainly adresses a hyperspectral imaging system based on a CCD sensor for proximity image acquisition. The principle is to look several times at the same scene by moving the camera slightly (a few millimetres) at each acquisition. Thus, it is possible to separate what moves (scene) from what remains fixed (response of the system) using properties of Fourier transform. Tests on synthetic images have reinforced theorical results on contraints shifts and given good results with more than ten translations. Tests on real and in-laboratory images have shown the need for accurate determination of translation to avoid some disruptive effects (pattern multiplication). Nevertheless, the results are promising and have shown the potential of the methodology to correct images from spatial non-uniformity due to the imaging system (radiometric aberration due to the sensor and optic). We notice that such a methodology remains valid for any imaging system based on a charged-coupled device (CCD) sensor.

International audience; Vitreousness is an important grading factor for durum wheat kernel that is associated with protein content. The European Union (EU) regulations stipulate the use of a visual method to determine the vitreousness rate. However, some authors have been interested in the development of automatic and non-destructive methods based on near infrared spectroscopy or digital imaging technology. In this paper, we propose to couple spectroscopy and digital imaging technology and, thus, to analyse the potential of a near infrared hyperspectral imaging system to classify wheat kernels by their vitreousness. Hyperspectral reflectance images of wheat kernel at different vitreousness rates have been achieved (wavelength range: 650-1100 nm). A discrimination method using latent variables from partial least squares has enabled satisfactory discrimination results to be obtained: perfect separation between vitreous and non-vitreous kernels and a classification rate reaching up to 94% for the discrimination of total and partial starchy kernel classes