The detection and control of diseases constitute a primary objective of French viticultural research. In this paper, we present a bottom-up hierarchical approach for selecting spectral bands suitable for class discrimination of spectra acquired by Infrared spectroscopy. Our method entails evaluating neighboring bands using various similarity metrics, applying aggregation criteria, and ultimately identifying a limited number of the most relevant bands for the separation of classes. The bandwidths are limited within a range as is typically required for choosing existing optical filters or specifying colored filter arrays. Our approach facilitates the discovery of distinctive spectral bands associated with a disease of interest, enabling the customization of multispectral cameras to meet specific requirements. It was applied to spectra collected on vine leaves spanning a three-year period with the goal to identify the most discriminant bands for the detection of grapevine yellows. The results show that a limited number of bands are sufficient to identify this class of interest through a classifier based on Linear Discriminant Analysis.

The detection and control of diseases is a primary objective of French viticultural research. In this context, we have collected Near Infra-Red (NIR) spectra on vine leaves over different acquisition times, three years from 2021 to 2023, with the aim of selecting the discriminating spectral bands of yellowing of the vine compared to healthy plants, confounding symptoms, and other diseases of vines: rolling of leaves and esca. We also want these bands to be insensitive with respect to the acquisition time. To achieve this, we adapt the Correlation Explanation (CorEx) algorithm so that it can select suitable bands from NIR spectra under the constraint of insensitivity versus the acquisition times (CorEx-BS). Our two-steps method consists firstly in searching for a set of bands that best explain the correlations between the wavelengths, as measured by the multivariate mutual information, and secondly to bind these bands with the labels and the acquisition time to get the bands relevant to the labels. This approach facilitates the discovery of distinctive spectral bands associated with a class of interest, yellowing of the vine in our application, which are robust regarding the different acquisition times of spectra, the years in our case. The results in terms of Davies-Bouldin index (DB) and Calinski-Harabasz Index (CH) show that our method outperforms other classical bands clustering selection techniques.

The use of RGB cameras or multispectral imaging systems can provide a wide range of applications for crop monitoring, plant phenotyping and disease detection. Although several approaches have been proposed, they increasingly use convolutional neural network-based architectures, which have, however, become increasingly cumbersome for improving classification results and difficult to train with few labeled data. Other increasingly popular approaches consist of using an ensemble of convolutional neural networks, in which each model solves a different problem. Since the inference is time- and resource-consuming due to the execution of multiple models, recent works have focused on transferring knowledge from an ensemble of models to a compact model to obtain better performance. In this paper, we propose an original approach that improves both accuracy and speed by reusing feature maps extracted by heterogeneous models from different data. Linked to each model, a transformation block allows keeping the correct number of feature maps and changing their dimension if necessary. To generate the feature maps, we only need the first layers of the ensemble models, thus taking advantage of ensemble learning methods, while adding only a few layers of a second model dedicated to aggregation of features. This approach allows an ensemble of models to be combined with different architectures that can process different data, such as several representations of the same input image or multispectral images, while being fast enough at the inference stage. This approach is adapted to hierarchical classification tasks by re-exploiting the same feature maps with different transformation blocks, offering accuracy gains in tasks not handled by the ensemble model. The results are provided for the PlantVillage dataset, with RGB images converted to three different color spaces, and for a custom Grapevine Yellow dataset, with multispectral images acquired with two different multispectral cameras.

Convolutional neural networks (CNNs) are deep learning architectures used for image classification that have been improved in recent years to increase their accuracies and reduce their computation times. Hierarchical approaches are based on a step-by-step strategy and aim to optimize performance on difficult tasks by solving successive subtasks. The gain provided by these solutions must be relativized with the explosion in the number of parameters they imply, which makes their implementation on embedded systems difficult. New constraints also appear in the choice of the architectures of the branches when one seeks to have a global network providing predictions at different levels. We propose a strategy that allows the merging of heterogeneous CNNs by following a hierarchical approach in which the information extracted by first-level networks can be fed back at any location into second-level networks. Despite the differences in the number and size of the feature maps, such grafting can be done by using clustering, dimension reduction, and interpolation techniques. This strategy eliminates the computational redundancy induced by the recalculation of low-level features. The proposed grafting approach significantly reduces the inference time of the second network without impacting the accuracy. Tests performed on MNIST, CIFAR-10, and PlantVillage datasets with several CNNs illustrate the possibility of implementation in various situations. Our solution allows us to consider in an innovative way the implementation of hierarchical solutions on devices with limited capacities.

Infrared spectroscopy provides useful information on the molecular compositions of biological systems related to molecular vibrations, overtones, and combinations of fundamental vibrations. Mid-infrared (MIR) spectroscopy is sensitive to organic and mineral components and has attracted growing interest in the development of biomarkers related to intrinsic characteristics of lignocellulose biomass. However, not all spectral information is valuable for biomarker construction or for applying analysis methods such as classification. Better processing and interpretation can be achieved by identifying discriminating wavenumbers. The selection of wavenumbers has been addressed through several variable- or feature-selection methods. Some of them have not been adapted for use in large data sets or are difficult to tune, and others require additional information, such as concentrations. This paper proposes a new approach by combining a naïve Bayesian classifier with a genetic algorithm to identify discriminating spectral wavenumbers. The genetic algorithm uses a linear combination of an a posteriori probability and the Bayes error rate as the fitness function for optimization. Such a function allows the improvement of both the compactness and the separation of classes. This approach was tested to classify a small set of maize roots in soil according to their biodegradation process based on their MIR spectra. The results show that this optimization method allows better discrimination of the biodegradation process, compared with using the information of the entire MIR spectrum, the use of the spectral information at wavenumbers selected by a genetic algorithm based on a classical validity index or the use of the spectral information selected by combining a genetic algorithm with other methods, such as Linear Discriminant Analysis. The proposed method selects wavenumbers that correspond to principal vibrations of chemical functional groups of compounds that undergo degradation/conversion during the biodegradation of lignocellulosic biomass.

The analysis of lignocellulosic materials is crucial to optimizing the conversion efficiencies in biorefineries and to studying crop residue input to soil nutrient cycles. Mid-infrared (MIR) and near-infrared (NIR) spectroscopies are rapid, simple, and nondestructive methods for the determination of biomass compositions. However, the analysis of a small set of plant biomass is not generally possible with conventional methods of data processing, such as partial least squares. Additionally, IR spectra do not distribute spherically in the data space. To overcome these problems, we propose a weighted-covariance factor fuzzy C-means clustering method combined with bootstrapping. The algorithm can classify spherical and nonspherical clusters, in contrast to classic fuzzy C-means, which is only adapted to spherical clusters. Bootstrapping enables resampling of the available spectra to generate several datasets on which the classification is performed. This unsupervised clustering methodology was tested to classify a small set of maize roots in soil according to genotype or period of their biodegradation process based on their NIR and MIR spectra. This methodology is applied to determine the optimal pretreatment of IR spectra, to study the contribution of the combination of MIR and NIR spectra and to compare the results on spectral and chemical data. The results show that the best methods of pretreatment are the first-order Savitzky-Golay derivative followed by standard normal variate. The MIR spectra produce a better result than NIR spectra for the initial characterization and for dynamic samples, while MIR spectra acquired on raw samples, without soluble extraction, provided better classification than wet chemistry.

Monitoring the health of ancient artworks requires adequate prudence because of the sensitive nature of these materials. Classical techniques for identifying the development of faults rely on acoustic testing. These techniques, being invasive, may result in causing permanent damage to the material, especially if the material is inspected periodically. Non destructive testing has been carried out for different materials since long. In this regard, non-invasive systems were developed based on infrared thermometry principle to identify the faults in artworks. The test artwork is heated and the thermal response of the different layers is captured with the help of a thermal infrared camera. However, prolonged heating risks overheating and thus causing damage to artworks and an alternate approach is to use pseudo-random binary sequence excitations. The faults in the artwork, though, cannot be detected on the captured images, especially if their strength is weak. The weaker faults are either masked by the stronger ones, by the pictorial layer of the artwork or by the non-uniform heating. This work addresses the detection and localization of the faults through a wavelet based subspace decomposition scheme. The proposed scheme, on one hand, allows to remove the background while, on the other hand, removes the undesired high frequency noise. It is shown that the detection parameter is proportional to the diameter and the depth of the fault. A criterion is proposed to select the optimal wavelet basis along with suitable level selection for wavelet decomposition and reconstruction. The proposed approach is tested on a laboratory developed test sample with known fault locations and dimensions as well as real artworks. A comparison with a previously reported method demonstrates the efficacy of the proposed approach for fault detection in artworks.

This paper is focused on the optimization of the pulsed thermography inspection of coating thickness. To this aim, an analysis method was developed for evaluating variations of coating thickness using active infrared thermography. This method, based on a linear decomposition of thermographic surface response evolution, uses a Partial Least Squares Regression algorithm to identify different phenomena affecting the overall thermal responses. Results show that the proposed technique allows to overcome the experimental conditions (positions, intensity) of heating sources and to enhance the robustness of the pulsed thermography technique for coating thickness heterogeneity evaluation.

In this paper, a pulsed Infrared thermography technique using a homogeneous heat provided by a laser source is used for the non-destructive evaluation of paint coating thickness variations. Firstly, numerical simulations of the thermal response of a paint coated sample are performed. By analyzing the thermal responses as a function of thermal properties and thickness of both coating and substrate layers, optimal excitation parameters of the heating source are determined. Two characteristic parameters were studied with respect to the paint coating layer thickness variations. Results obtained using an experimental test bench based on the pulsed Infrared thermography laser technique are compared with those given by a classical Eddy current technique for paint coating variations from 5 to 130μm. These results demonstrate the efficiency of this approach and suggest that the pulsed Infrared thermography technique presents good perspectives to characterize the heterogeneity of paint coating on large scale samples with other heating sources.

Mid-infrared (MIR) and near-infrared (NIR) spectroscopy provide useful information on the molecular composition of biological systems. Because they are sensitive to organic and mineral components, there is a growing interest in these techniques for the development of biomarkers that reflect intrinsic characteristics of plants and their mode of degradation. Due to their complexity and complementary nature, an important challenge is the combining of MIR and NIR information to identify discriminating wavenumbers in each wavenumber region, with the ultimate goal of assessing the biodegradation process of a lignocellulosic biomass at different time scales. This work investigates the potential of using the outer product to combine MIR and NIR spectra to highlight the connections between fundamental molecular vibrations and their combinations and bonds. Because this operation yields high-dimensional spectra, we propose to use a genetic algorithm to select the most discriminant wavenumbers within the degradation process. The results from two lignocellulosic biomasses with different biodegradation kinetics, miscanthus aerial parts and maize roots, confirm that the outer product combination of MIR and NIR spectral information allows a better discrimination of the biodegradation kinetic compared with the simple concatenation of MIR and NIR spectra or with the use of MIR or MIR spectral information separately. We show that the genetic algorithm selects wavenumbers that correspond to principal vibrations of chemical functional groups of compounds that undergo degradation/conversion during the biodegradation of the lignocellulosic biomass.

In this paper, we propose a factor weighted fuzzy c-means clustering algorithm. Based on the inverse of a covariance factor, which assesses the collinearity between the centers and samples, this factor takes also into account the compactness of the samples within clusters. The proposed clustering algorithm allows to classify spherical and non-spherical structural clusters, contrary to classical fuzzy c-means algorithm that is only adapted for spherical structural clusters. Compared with other algorithms designed for non-spherical structural clusters, such as Gustafson-Kessel, Gath-Geva or adaptive Mahalanobis distance-based fuzzy c-means clustering algorithms, the proposed algorithm gives better numerical results on artificial and real well known data sets. Moreover, this algorithm can be used for high dimensional data, contrary to other algorithms that require the computation of determinants of large matrices. Application on Mid-Infrared spectra acquired on maize root and aerial parts of Miscanthus for the classification of vegetal biomass shows that this algorithm can successfully be applied on high dimensional data.

InfraRed spectroscopy (IR) provides useful information of the molecular composition of biological systems. Mid-InfraRed (MIR) spectroscopy reflects fundamental molecular vibrations whereas Near-InfraRed (NIR) spectroscopy exhibits the overtones and combinations of fundamental vibrations and bonds. In most applications, the samples are mixed with potassium bromide (KBr) powder, or simply unmixed. Two technics are investigated: IR absorption on mixed samples and Diffuse Reflectance IR Fourier Transform (DRIFT) on unmixed samples. IR spectra are collected in either MIR or NIR regions. However, the preprocessing of IR spectra, the choice of the spectral band and the combination of MIR-NIR information are important factors that could substantially influence analyses. This study investigates these factors while attempting to retrieve three different genotypes of maize roots via a Fuzzy C-Mean (FCM) classification of IR spectra. A bootstrapping procedure is used as the number of samples is limited. Results show that KBr spectroscopy is better than DRIFT spectroscopy for MIR region; MIR provides equivalent information as NIR for DRIFT spectroscopy; combination of MIR-NIR information gives preprocessing independent results. Several distances are tested in FCM classification. The city bloc distance gives optimal results compared with Euclidean, Chebyshev, correlation and diagonal distance.

Infrared thermography is a non-destructive method that is becoming attractive due to its ability to inspect noninvasively large areas in short times.This work demonstrates that the analysis of the thermographic response of coating layer can be greatly improved with a multiscale decomposition.Based on 2D discrete wavelets transform, effective scales in the thermographic responses of samples can be selected.Then, the dynamic thermal response was analyzed using cross-correlation measure.Results show that thermal transient images obtained with this decomposition, during both excitation and cooling step, provide important information for detecting the presence of the sol-gel coating.

The bearings of rotating machines are comprised of rolling elements arranged between two coaxial rings. Rolling elements transfer the external load from one raceway to the other. This load's distribution on the rolling elements is uneven and depends on the internal geometry of the bearing and on the amplitude of the external load. In order to measure the transferred charge by a rolling element, a capacitive probe is inserted into the fixed ring of the bearing. This forms with the raceway a capacitor with variable gap that depends on the transmitted load by the rolling element. A numerical model of this capacitor's capacitance as a function of transmitted load by the rolling element has been established. This numerical model shows that there is a linear relation between the capacitance of the probe and the transmitted load by the rolling element. An experimental prototype has been established in order to precisely measure the probe's capacitance; the numerical model facilitates the access to the load's value. By placing a capacitive probe in front of each rolling element, the load transmitted by each rolling element can be measured. Therefore, the reconstruction of the shaft's external load on the rotary machine's bearing can be easily done.

In this paper, we studied the association of random infrared thermography, Singular Value Decomposition (SVD) and Higher Order Statistics (HOS) to help to restore works of art. We present first, the principle of theses signal processing methods. We show then, that this association allows detection of inclusions of plastazote located in an academic fresco. We show in a third stage that this photothermal method permits the location of defect located in a real work of art. It is the painting on canvas entitled “Saint Martin” located in the church of “Bonnet”. Finally, we show that this association permits in one hand the use of a lower power density. On the other hand, it permits to reduce the influence of radiative properties of the pictorial layer and of the inhomogeneous energy deposition on the photothermal signal.

Active infrared thermography is a nondestructive method for evaluating defects in artworks. A conventional excitation radiation heats the sample and the photothermal response is recorded by an infrared (IR) camera. Classical pulsed excitation has shown the feasibility of such a detection system, but the energy deposition for a long period of time can alter samples. Random excitations can prevent such problem, but signal processing methods should be implemented to extract the useful information. We propose a processing method that combines Singular Value Decomposition (SVD) and Higher Order Statistics (HOS). The former decomposes the dataset in several subspaces, allowing to remove the influence of the acquisition environment and system. The latter is used to build up from the useful information one or two images for diagnostic. We show on a mural-type laboratory and on a in situ artwork that this method allows good identification of defects, providing a complementary detector to classical analysis.