Spectroscopic Data Sets Research Articles

Calibration-free PAT: Locating selective crystallization or precipitation sweet spot in screenings with multi-way PARAFAC models.

When developping selective crystallization or precipitation processes, biopharmaceutical modalities require empirical screenings and analytics tailored to the specific needs of the target molecule. The multi-way chemometric approach called parallel factor analysis (PARAFAC) coupled with ultraviolet visible light (UV/Vis) spectroscopy is able to predict specific concentrations and spectra from highly structured data sets without the need for calibration samples and reference analytics. These calculated models can provide exploratory information on pure species spectra and concentrations in all analyzed samples by representing one model component with one species. In this work, protein mixtures, monoclonal antibodies, and virus-like particles in chemically defined and complex solutions were investigated in three high-throughput crystallization or precipitation screenings with the aim to construct one PARAFAC model per case. Spectroscopic data sets of samples after the selective crystallization or precipitation, washing, and redissolution were recorded and arranged into a four-dimensional data set per case study. Different reference analytics and pure species spectra served as validation. Appropriate spectral preprocessing parameters were found for all case studies allowing even the application of this approach to the third case study in which quantitative concentration analytics are missing. Regardless of the modality or the number of species present in complex solutions, all models were able to estimate the specific concentration and find the optimal process condition regarding yield and product purity. It was shown that in complex solutions, species demonstrating similar phase behavior can be clustered as one component and described in the model. PARAFAC as a calibration-free approach coupled with UV/Vis spectroscopy provides a fast overview of species present in complex solution and of their concentration during selective crystallization or precipitation, washing, and redissolution.

A Review of Machine Learning for Near-Infrared Spectroscopy.

The analysis of infrared spectroscopy of substances is a non-invasive measurement technique that can be used in analytics. Although the main objective of this study is to provide a review of machine learning (ML) algorithms that have been reported for analyzing near-infrared (NIR) spectroscopy from traditional machine learning methods to deep network architectures, we also provide different NIR measurement modes, instruments, signal preprocessing methods, etc. Firstly, four different measurement modes available in NIR are reviewed, different types of NIR instruments are compared, and a summary of NIR data analysis methods is provided. Secondly, the public NIR spectroscopy datasets are briefly discussed, with links provided. Thirdly, the widely used data preprocessing and feature selection algorithms that have been reported for NIR spectroscopy are presented. Then, the majority of the traditional machine learning methods and deep network architectures that are commonly employed are covered. Finally, we conclude that developing the integration of a variety of machine learning algorithms in an efficient and lightweight manner is a significant future research direction.

New strong lensing modelling of SDSS J2222+2745 enhanced with VLT/MUSE spectroscopy

Context. SDSS J2222+2745, at z = 0.489, is one of the few currently known lens clusters with multiple images (six) of a background (z = 2.801) quasar with measured time delays between two image pairs (with a sub-percent relative error for the longer time delay). Systems of this kind can be exploited as alternative cosmological probes through high-precision and accurate strong lensing models. Aims. We present recent observations from the Multi Unit Spectroscopic Explorer (MUSE) on the Very Large Telescope (VLT) and new total mass models of the core of the galaxy cluster SDSS J2222+2745. Methods. We combine archival multi-band, high-resolution imaging from the Hubble Space Telescope (HST) with our VLT/MUSE spectroscopic data to securely identify 34 cluster members and 12 multiple images from 3 background sources. We also measure the stellar velocity dispersions of 13 cluster galaxies, down to HST F160W = 21 mag, enabling an independent estimate of the contribution of the sub-halo mass component to the lens total mass. By leveraging the new spectroscopic dataset, we build improved strong lensing models. Results. The projected total mass distribution of the lens cluster is best modelled with a single large-scale mass component, a galaxy-scale component, anchored by the VLT/MUSE kinematic information, and an external shear component. The best-fit strong lensing model yields a root mean square separation between the model-predicted and observed positions of the multiple images of 0″​​.29. When analysing the impact of systematic uncertainties, stemming from modelling assumptions and used observables, we find that the resulting projected total mass profile, the relative weight of the sub-halo mass component, and the critical lines are consistent, within the statistical uncertainties. The predicted magnification and time-delay values are, instead, more sensitive to the local details of the lens total mass distribution, and vary significantly among lens models that are similarly good at reproducing the observed multiple image positions. In particular, the model-predicted time delays can differ by a factor of up to ∼1.5. Conclusions. SDSS J2222+2745 is a promising lens cluster for cosmological applications. However, due to its complex morphology, the relatively low number of secure ‘point-like’ multiple images, and current model degeneracies, it becomes clear that additional information (from the observed surface brightness distribution of lensed sources and the measured time delays) needs to be included in the modelling for accurate and precise cosmological measurements. The full VLT/MUSE secure spectroscopic catalogue presented in this work is made publicly available.

Colour gradients of low-redshift galaxies in the DESI Legacy Imaging Survey

ABSTRACT Radial colour gradients within galaxies arise from gradients of stellar age, metallicity, and dust reddening. Large samples of colour gradients from wide-area imaging surveys can complement smaller integral-field spectroscopy data sets and can be used to constrain galaxy formation models. Here, we measure colour gradients for low-redshift galaxies (z < 0.1) using photometry from the DESI Legacy Imaging Survey DR9. Our sample comprises ∼93 000 galaxies with spectroscopic redshifts and ∼574 000 galaxies with photometric redshifts. We focus on gradients across a radial range 0.5Reff to Reff, which corresponds to the inner disc of typical late-type systems at low redshift. This region has been the focus of previous statistical studies of colour gradients and has recently been explored by spectroscopic surveys such as MaNGA. We find that the colour gradients of most galaxies in our sample are negative (redder towards the centre), consistent with the literature. We investigate empirical relationships between colour gradient, average g − r and r − z colour, Mr, M⋆, and sSFR. Trends of gradient strength with Mr (M⋆) show an inflection around Mr ∼ −21 ($\log _{10} \, M_\star /\mathrm{M_\odot }\sim 10.5$). Below this mass, colour gradients become steeper with increasing M⋆, whereas colour gradients in more massive galaxies become shallower. We find that positive gradients (bluer stars at smaller radii) are typical for galaxies of $M_{\star }\sim 10^{8}\, \mathrm{M_\odot }$. We compare our results to age and metallicity gradients in two data sets derived from fits of different stellar population libraries to MaNGA spectra, but find no clear consensus explanation for the trends we observe. Both MaNGA data sets seem to imply a significant contribution from dust reddening, in particular, to explain the flatness of colour gradients along the red sequence.

Application of elastic net in quantitative analysis of major elements using Martian laser-induced breakdown spectroscopy datasets

Multiple sets of laser-induced breakdown spectroscopy (LIBS) instruments, including ChemCam (Curiosity), SuperCam (Perseverance), and the Mars Surface Composition Detector (MarSCoDe, Zhurong), are working currently on the surface of Mars, a booming scene of LIBS technology application to planetary exploration. One of the primary challenges faced by those LIBS instruments is accurate and quantitative chemical composition determination of unexplored geological targets. Elastic net is a linear regression model that combines ridge regression and lasso model, which inherits the sparseness of lasso and the stability of ridge. In this work, we investigated the spectral features selected by elastic net model, the model performance and application to over 23,000 LIBS points. Selected features exhibit significant emission lines that are attributed to a certain single element and a few lines attributed to other related elements. These features enhance the interpretability of the model and weaken the matrix effect in LIBS quantitative analysis. In comparison, the results of elastic net are comparable and may be slightly better in some cases than other common linear regression models. For igneous calibration targets (norite, picrite, and shergottite), the predictions of elastic net are closer to the values measured in Earth laboratory than the final multivariate oxide composition values predicted by ChemCam team. Finally, we predicted the oxide abundances of 8 major elements from Mars LIBS spectra of over 23,000 points using elastic net. The predicted values of the elastic net models are highly correlated with those from the ChemCam team model. These results indicate that elastic net is viable for quantitative analysis Mars LIBS spectra. We propose that elastic net is an important candidate model for the quantitative analysis of the Zhurong MarSCoDe LIBS spectra.

Nitrate Classification Based on Optical Absorbance Data Using Machine Learning Algorithms for a Hydroponics System.

Nutrient solution plays an essential role in providing macronutrients to hydroponic plants. Determining nitrogen in the form of nitrate is crucial, as either a deficient or excessive supply of nitrate ions may reduce the plant yield or lead to environmental pollution. This work aims to evaluate the performance of feature reduction techniques and conventional machine learning (ML) algorithms in determining nitrate concentration levels. Two features reduction techniques, linear discriminant analysis (LDA) and principal component analysis (PCA), and seven ML algorithms, for example, k-nearest neighbors (KNN), support vector machine, decision trees, naïve bayes, random forest (RF), gradient boosting, and extreme gradient boosting, were evaluated using a high-dimensional spectroscopic dataset containing measured nitrate-nitrite mixed solution absorbance data. Despite the limited and uneven number of samples per class, this study demonstrated that PCA outperformed LDA on the high-dimensional spectroscopic dataset. The classification accuracy of ML algorithms combined with PCA ranged from 92.7% to 99.8%, whereas the classification accuracy of ML algorithms combined with LDA ranged from 80.7% to 87.6%. The PCA with the RF algorithm exhibited the best performance with 99.8% accuracy.

Prognostic potential and pathological validation of a diagnostic application using Raman spectroscopy in the characterization of degenerative changes in the cartilage of the humeral head.

Raman spectroscopy is a well-established analytical method in the fields of chemistry, industry, biology, pharmaceutics, and medicine. Previous studies have investigated optical imaging and Raman spectroscopy for osteoarthritis (OA) diagnosis in weight-bearing joints such as hip and knee joints. However, to realize early diagnosis or a curable treatment, it is still challenging to understand the correlations with intrinsic factors or patients’ background. To elucidate the correlation between the Raman spectral features and pathological variations of human shoulder joint cartilage. Osteoarthritic cartilage specimens excised from the humeral heads of 14 patients who underwent shoulder arthroplasty were assessed by a confocal Raman microscope and histological staining. The Raman spectroscopic dataset of degenerative cartilage was further analyzed by principal component analysis and hierarchical cluster analysis. Multivariate association of the Raman spectral data generated three major clusters. The first cluster of patients shows a relatively high Raman intensity of collagen. The second cluster displays relatively low Raman intensities of proteoglycans (PGs) and glycosaminoglycans (GAGs), whereas the third cluster shows relatively high Raman intensities of PGs and GAGs. The reduced PGs and GAGs are typical changes in OA cartilage, which have been confirmed by safranin–O staining. In contrast, the increased Raman intensities of collagen, PGs, and GAGs may reflect the instability of the cartilage matrix structure in OA patients. The results obtained confirm the correlation between the Raman spectral features and pathological variations of human shoulder joint cartilage. Unsupervised machine learning methods successfully yielded a clinically meaningful classification between the shoulder OA patients. This approach not only has potential to confirm severity of cartilage defects but also to determine the origin of an individual’s OA by evaluating the cartilage quality.

Whose line is it anyway? A self-training spectral line identification code for plasma physics experiments

The process of associating spectral peaks in emission radiation data with particular charge states of specific elements is a common task in the field of plasma diagnostics in both laboratory and astrophysical settings. Existing techniques for this purpose are often highly manual or can rely heavily on theoretical models and assumptions of plasma parameters. We present a numerical approach to largely automate this process. The approach combines statistics from experimental data with theoretical predictions of transition strengths and observed emission intensity data in order to accomplish the task of spectral line identification in a rigorous, quantitative way, reporting confidence levels in its own predictions for each wavelength. Weighted by this confidence, the method identifies sources of 31 test emission lines in the C-2W device with 99.99% accuracy (compared to manual identification). Similar performance is demonstrated on synthetic datasets and spectroscopic observations of the planetary nebula NGC 6543, with accuracies of between 95% and 100%. The approach is scalable, portable to a wide variety of spectroscopic datasets and significantly faster and more rigorous than manual methods.

Unsupervised Feature Selection by a Genetic Algorithm for Mid-Infrared Spectral Data.

Dimensional reduction of highly multidimensional datasets such as those acquired by Fourier transform infrared spectroscopy (FTIR) is a critical step in the data analysis workflow. To achieve this goal, numerous feature selection methods have been developed and applied in a supervised context, i.e., using a priori knowledge about data usually in the form of labels for classification or quantitative values for regression. For this, genetic algorithms have been largely exploited due to their flexibility and global optimization principle. However, few applications in an unsupervised context have been reported in infrared spectroscopy. The aim of this article is to propose a new unsupervised feature selection method based on a genetic algorithm using a validity index computed from KMeans partitions as a fitness function. Evaluated on a simulated dataset and validated and tested on three real-world infrared spectroscopic datasets, our developed algorithm is able to find the spectral descriptors improving clustering accuracy and simplifying the spectral interpretation of results.

Euclid: Cosmological forecasts from the void size function

The Euclid mission – with its spectroscopic galaxy survey covering a sky area over 15 000 deg2 in the redshift range 0.9 < z < 1.8 – will provide a sample of tens of thousands of cosmic voids. This paper thoroughly explores for the first time the constraining power of the void size function on the properties of dark energy (DE) from a survey mock catalogue, the official Euclid Flagship simulation. We identified voids in the Flagship light-cone, which closely matches the features of the upcoming Euclid spectroscopic data set. We modelled the void size function considering a state-of-the art methodology: we relied on the volume-conserving (Vdn) model, a modification of the popular Sheth & van de Weygaert model for void number counts, extended by means of a linear function of the large-scale galaxy bias. We found an excellent agreement between model predictions and measured mock void number counts. We computed updated forecasts for the Euclid mission on DE from the void size function and provided reliable void number estimates to serve as a basis for further forecasts of cosmological applications using voids. We analysed two different cosmological models for DE: the first described by a constant DE equation of state parameter, w, and the second by a dynamic equation of state with coefficients w0 and wa. We forecast 1σ errors on w lower than 10% and we estimated an expected figure of merit (FoM) for the dynamical DE scenario FoMw0, wa = 17 when considering only the neutrino mass as additional free parameter of the model. The analysis is based on conservative assumptions to ensure full robustness, and is a pathfinder for future enhancements of the technique. Our results showcase the impressive constraining power of the void size function from the Euclid spectroscopic sample, both as a stand-alone probe, and to be combined with other Euclid cosmological probes.

Short GRB Host Galaxies. I. Photometric and Spectroscopic Catalogs, Host Associations, and Galactocentric Offsets

We present a comprehensive optical and near-infrared census of the fields of 90 short gamma-ray bursts (GRBs) discovered in 2005–2021, constituting all short GRBs for which host galaxy associations are feasible (≈60% of the total Swift short GRB population). We contribute 274 new multi-band imaging observations across 58 distinct GRBs and 26 spectra of their host galaxies. Supplemented by literature and archival survey data, the catalog contains 542 photometric and 42 spectroscopic data sets. The photometric catalog reaches 3σ depths of ≳24–27 mag and ≳23–26 mag for the optical and near-infrared bands, respectively. We identify host galaxies for 84 bursts, in which the most robust associations make up 56% (50/90) of events, while only a small fraction, 6.7%, have inconclusive host associations. Based on new spectroscopy, we determine 18 host spectroscopic redshifts with a range of z ≈ 0.15–1.5 and find that ≈23%–41% of Swift short GRBs originate from z > 1. We also present the galactocentric offset catalog for 84 short GRBs. Taking into account the large range of individual measurement uncertainties, we find a median of projected offset of ≈7.7 kpc, for which the bursts with the most robust associations have a smaller median of ≈4.8 kpc. Our catalog captures more high-redshift and low-luminosity hosts, and more highly offset bursts than previously found, thereby diversifying the population of known short GRB hosts and properties. In terms of locations and host luminosities, the populations of short GRBs with and without detectable extended emission are statistically indistinguishable. This suggests that they arise from the same progenitors, or from multiple progenitors, which form and evolve in similar environments. All of the data products are available on the Broadband Repository for Investigating Gamma-Ray Burst Host Traits website.

Single-model multi-tasks deep learning network for recognition and quantitation of surface-enhanced Raman spectroscopy.

Surface-enhanced Raman scattering (SERS) spectroscopy analysis has long been the central task of nanoscience and nanotechnology to realize the ultrasensitive recognition/quantitation applications. Recently, the blooming of artificial intelligence algorithms provides an edge tool to revolutionize the spectroscopy analysis, especially for multiple substances analysis and large-scale data handling. In this study, a single-model multi-tasks deep learning network is proposed to simultaneously achieve the qualitative recognition and quantitative analysis of SERS spectroscopy. The SERS spectra of two kinds of hypoglycemic drugs (PHE, ROS) and the corresponding mixtures are collected, respectively, with the concentration grade from 10-4 M to 10-8 M. Based on the SERS spectroscopy dataset, the loss functions and hyperparameters of the multi-tasks classifications model are optimized, and the recognition accuracies are tested by simulation experiments. It is demonstrated that the accuracy rates of qualitative and quantitative analysis can reach up to 99.0% and 98.4%, respectively. Moreover, the practical feasibility of this multi-tasks model is demonstrated by using it to achieve qualitative and quantitative analysis of PHE and ROS in complex serum matrix. Overall, this single-model multi-tasks deep learning network shows significant potential for the recognition and quantitation of SERS spectroscopy, which provides the algorithmic and experimental basis for large-scale and multiple substances SERS spectra analysis.

Why it is Unfortunate that Linear Machine Learning "Works" so well in Electromechanical Switching of Ferroelectric Thin Films.

Machine learning (ML)is relied on for materials spectroscopy. It is challenging to make ML models fail because statistical correlations can mimic the physics without causality. Here, using a benchmark band-excitation piezoresponse force microscopy polarization spectroscopy (BEPS) dataset the pitfalls of the so-called "better", "faster", and "less-biased" ML of electromechanical switching are demonstrated and overcome. Using a toy and real experimental dataset, it is demonstrated how linear nontemporal ML methods result in physically reasonable embedding (eigenvalues) while producing nonsensical eigenvectors and generated spectra, promoting misleading interpretations. A new method of unsupervised multimodal hyperspectral analysis of BEPS is demonstrated using long-short-term memory (LSTM) β-variational autoencoders (β-VAEs) . By including LSTM neurons, the ordinal nature of ferroelectric switching is considered. To improve the interpretability of the latent space, a variational Kullback-Leibler-divergency regularization is imposed . Finally, regularization scheduling of β as a disentanglement metric is leveraged to reduce user bias. Combining these experiment-inspired modifications enables the automated detection of ferroelectric switching mechanisms, including a complex two-step, three-state one. Ultimately, this work provides a robust ML method for the rapid discovery of electromechanical switching mechanisms in ferroelectrics and is applicable to other multimodal hyperspectral materials spectroscopies.

Accretion and Host-Galaxy Properties of 14 New “Changing-Look” Active Galactic Nuclei Identified from the SDSS-V Survey

Abstract The widely accepted active galactic nucleus (AGN) paradigm has been recently challenged by the discovery of the so-called ``changing-look'' (CL) phenomenon characterized by spectral-type transitions. By comparing the SDSS-V and SDSS DR16 spectroscopic datasets, here we report the identification of 14 new CL-AGNs (redshift $z<0.5$) exhibiting spectral-type changes on a timescale of $\sim 10$\,yr. Follow-up spectroscopy was conducted with the Lick Shane 3\,m and Keck 10\,m telescopes for three of the objects. Detailed analysis of these spectra enables us to arrive at the following two main results. (1) By compiling a sample of 65 CL-AGNs with good measurements, we reinforce the previous claim that CL-AGNs tend to be biased against both a high Eddington ratio ($\lesssim 0.1$) and a high bolometric luminosity ($\lesssim 10^{46}\,\mathrm{erg\,s^{-1}}$). This bias suggests that the disk-wind broad-line-region model is a plausible explanation of the CL phenomenon. (2) The host galaxies of CL-AGNs tend to be dominated by intermediate stellar populations, which motivates us to propose that CL-AGNs are probably particular AGNs at a special evolutionary stage, such as a transition stage from ``feast'' to ``famine'' fueling of the supermassive black hole. In addition, with our spectra, we identify SDSS\,J025951.22+003744.2 as a new repeat CL narrow-line Seyfert 1 galaxy with a rapid ``turn-on'' timescale of $\sim 1$\,yr.

Chemometric and Machine Learning Analysis of Lithium Concentration and Solvation Behavior in Li-Ion Battery Electrolytes

The demand for batteries is rapidly growing across a range of technologies. The increasingly diverse use cases for batteries require various capabilities, particularly requirements for high energy densities, that are currently unmet by traditional Li-ion batteries. Electrolyte stability proves to be a bottleneck for battery advancement towards energy dense chemistries beyond Li-ion, including metal anodes. In situ spectroscopy tools, such as Fourier transform infrared (FTIR) spectroscopy, Raman spectroscopy, and X-ray spectroscopy, have provided insight into critical molecular-level interactions in batteries during cycling. These in situ tools have yielded continuous improvement of electrolyte properties. Spectroscopy datasets, however, contain many nuances that challenge meaningful human understanding. Artificial intelligence and chemometric tools can be coupled with in situ spectroscopy to find relevant interpretations of spectral datasets and elucidate complex molecular phenomena. In this research, an analysis was performed on FTIR spectroscopy data from an electrolyte composed of LiPF6 in ethylene carbonate (EC) and ethyl methyl carbonate (EMC) to discern solvation behavior using principal component analysis (PCA) and a convolutional neural network (CNN). PCA pinpointed exact band locations of solvation shifting behavior in the IR spectra and improved understanding of the relationship between lithium concentrations and peak changes. The CNN was trained with spectral datasets of electrolytes with known lithium concentrations and then could predict lithium concentrations from spectral datasets with extraordinarily high accuracy. Additionally, the CNN interpreted FTIR spectral datasets from a graphite half-cell with EC/EMC/LiPF6 electrolyte and accurately determined the lithium concentration in the bulk electrolyte. The CNN also observed lithium depletion events (up to 10% lithium depletion) in the graphite anode during fast-charging cycles of the galvanostatic intermittent titration technique. This research breaks new ground on using advanced computational tools for in situ spectroscopic analysis of battery electrolytes and demonstrates an improved understanding of complex molecular-level phenomena in electrolytes.

Prediction of malignant transformation in oral epithelial dysplasia using machine learning

A machine learning algorithm (MLA) has been applied to a Fourier transform infrared spectroscopy (FTIR) dataset previously analysed with a principal component analysis (PCA) linear discriminant analysis (LDA) model. This comparison has confirmed the robustness of FTIR as a prognostic tool for oral epithelial dysplasia (OED). The MLA is able to predict malignancy with a sensitivity of 84 ± 3% and a specificity of 79 ± 3%. It provides key wavenumbers that will be important for the development of devices that can be used for improved prognosis of OED.

Employing Auto-Machine Learning Algorithms for Predicting the Cold Filter Plugging and Kinematic Viscosity at 40 ºC in Biodiesel Blends using Vibrational Spectroscopy Data

This work aims to develop an auto-machine learning method using Mid-Infrared (MIR) spectroscopy data to determine the cold filter plugging point (CFPP) and kinematic viscosity at 40 ºC of biodiesel, diesel, and mixtures samples. The biodiesel was obtained by the transesterification reaction and later purified. The first dataset was composed of 108 blends (biodiesel obtained from different biomass such as soy, corn, sunflower, and canola) with binary, ternary and quaternary mixtures. The second dataset was composed of 227 blends of diesel-biodiesel and diesel-biodiesel-ethanol, respectively. The physical properties of the samples were obtained according to ABNT NBR 14747 and ABNT NBR 10441, respectively. The MIR Spectroscopy data were acquired from 7,800 to 450 cm-1, with a 4 cm-1 resolution and 20 scans. The spectra' baseline alignment was carried out using the asymmetric least squares method. A Savitzky–Golay filter was applied to a set of digital data points to smooth the data. This work used a first-order polynomial and a zero derivative function to smooth the spectra. The dataset was split into training and test sets using the function CreateDataPartition from the caret package. It was adopted 70% for training and 30% for test sets. In this work, the model training process was carried out using the open-source Python library LazyPredict. The LazyPredict returns the trained models and their performance metrics. The kinematic viscosity at 40 ºC of the biodiesel samples and their blends could be modeled using the MIR Spectroscopy dataset using different auto-machine learning algorithms. The RMSEP (Root Mean Square Error of Prediction) (≤ 0.02 mm2 s-1) was similar to the experimental error obtained after log transformation. The CFPP of the biodiesel samples and their blends could be modeled using the MIR Spectroscopy dataset by different auto-machine learning algorithms with an RMSEP (≤ 1.6 ºC) similar to the experimental error obtained by traditional methodology. Based on the lower computational time and the same performance observed by the RMSEP and R2 (coefficient of determination) values from different algorithms, it is recommended to use Ridge or Ridge Cross-Validation Regression models for both physical properties using MIR Spectroscopy data.

Interesting features finder (IFF): Another way to explore spectroscopic imaging data sets giving minor compounds and traces a chance to express themselves

Today, we acquire larger and larger spectroscopic imaging data sets on complex samples. Even before proceeding with a spectroscopic analysis, we often have information concerning for example the presence of major compounds. However, we are not in the same position regarding the presence and location of minor compounds and traces in the sample, while this is generally a more interesting piece of information. Modern spectroscopic imaging uses chemometric tools that allow the simultaneous exploration of the entire spectral range available. These well-tested tools, such as principal component analysis, often exploit the variance contained in the data set to extract maximum chemical information. In general, we can say that these approaches are quite efficient but they are not completely adapted to the characteristics of the data acquired during a spectral imaging experiment. Indeed, such data sets have generally a limited signal to noise ratio and minor compounds and traces are often present on a small number of pixels compared to the total number in the considered data set. It is then quite possible that these compounds are present in a sample but not detected by the chosen multivariate tool. The goal of this work is to introduce an approach called Interesting Features Finder (IFF) able to retrieve minor compounds and traces independently of the variance they may express in the spectral data set. Like other state of the art methods, it uses the notion of convex hull but with the great advantage of working directly on the raw data, without prior projection to reduce their dimensionality. By means of a synthetic spectral data set and a micro-LIBS (laser-induced breakdown spectroscopy) imaging one acquired on a sample taken in Antarctica, we will show the potential of this approach in terms of sensitivity of detection of compounds but also of robustness against noise.

A new approach to feature-based asteroid taxonomy in 3D color space

The taxonomic classification of asteroids has been mostly based on spectroscopic observations with wavelengths spanning from the visible (VIS) to the near-infrared (NIR). VIS-NIR spectra of ~2500 asteroids have been obtained since the 1970s; the Sloan Digital Sky Survey (SDSS) Moving Object Catalog 4 (MOC 4) was released with ~4 × 105 measurements of asteroid positions and colors in the early 2000s. A number of works then devised methods to classify these data within the framework of existing taxonomic systems. Some of these works, however, used 2D parameter space (e.g., gri slope vs. z-i color) that displayed a continuous distribution of clouds of data points resulting in boundaries that were artificially defined. We introduce here a more advanced method to classify asteroids based on existing systems. This approach is simply represented by a triplet of SDSS colors. The distributions and memberships of each taxonomic type are determined by machine learning methods in the form of both unsupervised and semi-supervised learning. We apply our scheme to MOC 4 calibrated with VIS-NIR reflectance spectra. We successfully separate seven different taxonomy classifications (C, D, K, L, S, V, and X) with which we have a sufficient number of spectroscopic datasets. We found the overlapping regions of taxonomic types in a 2D plane were separated with relatively clear boundaries in the 3D space newly defined in this work. Our scheme explicitly discriminates between different taxonomic types (e.g., K and X types), which is an improvement over existing systems. This new method for taxonomic classification has a great deal of scalability for asteroid research, such as space weathering in the S-complex, and the origin and evolution of asteroid families. We present the structure of the asteroid belt, and describe the orbital distribution based on our newly assigned taxonomic classifications. It is also possible to extend the methods presented here to other photometric systems, such as the Johnson-Cousins and LSST filter systems.

Rapid Quantitation of Adulterants in Premium Marine Oils by Raman and IR Spectroscopy: A Data Fusion Approach.

This study uses Raman and IR spectroscopic methods for the detection of adulterants in marine oils. These techniques are used individually and as low-level fused spectroscopic data sets. We used cod liver oil (CLO) and salmon oil (SO) as the valuable marine oils mixed with common adulterants, such as palm oil (PO), omega-3 concentrates in ethyl ester form (O3C), and generic fish oil (FO). We showed that support vector machines (SVM) can classify the adulterant present in both CLO and SO samples. Furthermore, partial least squares regression (PLSR) may be used to quantify the adulterants present. For example, PO and O3C adulterated samples could be detected with a RMSEP value less than 4%. However, the FO adulterant was more difficult to quantify because of its compositional similarity to CLO and SO. In general, data fusion improved the RMSEP for PO and O3C detection. This shows that Raman and IR spectroscopy can be used in concert to provide a useful analytical test for common adulterants in CLO and SO.