Spectral Pre-processing Methods Research Articles

Determination of pH and acidity in green coffee using near-infrared spectroscopy and multivariate regression.

Coffee is a raw material of global interest. Due to its relevance, this work evaluated the performance of calibration models constructed from spectral data obtained using near-infrared spectroscopy (FT-NIR) to determine the pH values and acidity in coffee beans in a practical and non-destructive way. Partial least squares regression was used during the calibration and the cross-validation to optimize the number of latent variables. The predictive capacity of the spectral pre-processing methods was also accessed. The results obtained showed that the best methods of pre-processing were the first derivative for the pH variable and the standard normal variate for the acidity, which produced models with correlations of 0.78 and 0.92, ratios of prediction to deviation of 2.061 and 2.966 and biases of -0.00011 and -0.152 to test set validation, respectively. The average errors between predicted and experimental values were lower than 7%. FT-NIR was successfully applied to predict properties related to the quality of coffee. The method was demonstrated to be a fast and non-destructive tool which allows the rapid inline evaluation of samples facilitating industrial and commercial processing. © 2020 Society of Chemical Industry.

Predicting macronutrients and energy content of snack products using FT-NIR analysis and chemometric techniques

Nowadays, the consumption of snack products is permanently increasing. Because of the growing trend of snack consumption, it is more and more difficult to guarantee the quality and safety of the products. Near infrared spectroscopy (NIRS) method, combined with chemometric techniques provide outstanding solutions, due to its rapidity and simple sample preparation. The objective of this study was to investigate the possibilities of using NIRS to predict fat, protein, carbohydrate, sugar and salt content of all in all 155 commercially available snack products from 25 countries. The prediction models were performed using partial least squares regression (PLSR) with different spectral pre-processing methods. Different pre-processing methods proved to be the best to predict the five macronutrients, however, the final models showed good accuracy |R2/Q2 > 0.94/0.82|. The energy content of the samples was calculated from the measured parameters and interval PLS regression was accomplished to improve prediction parameters. The methods developed are suitable for analyzing snacks made from single or mixed raw materials.

Prediction of soluble solid content of Agaricus bisporus during ultrasound-assisted osmotic dehydration based on hyperspectral imaging

Soluble solid content (SSC) is a critical index to evaluate the nutrition and flavor quality of food products. This study presents a novel strategy to predict the SSC in Agaricus bisporus slices during ultrasound-assisted osmotic dehydration (UOD). The spectral signatures of Agaricus bisporus were captured via a hyperspectral imaging (HSI) system and different spectral preprocessing methods and models were used to fit and evaluate the SSC behaviour of samples during UOD. The results showed that the support vector machine (SVM) preprocessed with orthogonal signal correction (OSC) provided the best fit for the full-band spectra of samples, with a higher correlation coefficient of prediction (R2 P, 0.883) and residual predictive deviation (RPD, 3.04). Moreover, the competitive adaptive reweighted sampling (CARS) algorithm can screen 67 key wavelengths from the complex original full-band wavelengths, and the OSC-CARS-SVM model showed the best predicted performance of SSC for the simplified spectra. In addition, the distribution of SSC in different UOD periods of the samples were demonstrated in a pseudo-colour map, which further revealed the SSC distribution of samples during UOD. The overall results showed the great potential of HSI to detect and predict the SSC of Agaricus bisporus rapidly, accurately, and non-destructively.

The Effect of Light Intensity, Sensor Height, and Spectral Pre-Processing Methods when using NIR Spectroscopy to Identify Different Allergen-Containing Powdered Foods.

Food allergens present a significant health risk to the human population, so their presence must be monitored and controlled within food production environments. This is especially important for powdered food, which can contain nearly all known food allergens. Manufacturing is experiencing the fourth industrial revolution (Industry 4.0), which is the use of digital technologies, such as sensors, Internet of Things (IoT), artificial intelligence, and cloud computing, to improve the productivity, efficiency, and safety of manufacturing processes. This work studied the potential of small low-cost sensors and machine learning to identify different powdered foods which naturally contain allergens. The research utilised a near-infrared (NIR) sensor and measurements were performed on over 50 different powdered food materials. This work focussed on several measurement and data processing parameters, which must be determined when using these sensors. These included sensor light intensity, height between sensor and food sample, and the most suitable spectra pre-processing method. It was found that the K-nearest neighbour and linear discriminant analysis machine learning methods had the highest classification prediction accuracy for identifying samples containing allergens of all methods studied. The height between the sensor and the sample had a greater effect than the sensor light intensity and the classification models performed much better when the sensor was positioned closer to the sample with the highest light intensity. The spectra pre-processing methods, which had the largest positive impact on the classification prediction accuracy, were the standard normal variate (SNV) and multiplicative scattering correction (MSC) methods. It was found that with the optimal combination of sensor height, light intensity, and spectra pre-processing, a classification prediction accuracy of 100% could be achieved, making the technique suitable for use within production environments.

Detection of aflatoxin B1 on corn kernel surfaces using visible-near infrared spectroscopy

In this study, visible-near infrared spectroscopy over the spectral range of 400–2500 nm was utilized to detect surface contamination of corn kernels with aflatoxin B1. The artificially contaminated samples were prepared by dropping known amounts of aflatoxin B1 standard dissolved in 50:50 ( v/ v) methanol/water solution, onto corn kernel surface to achieve different contamination levels of 10, 20, 50, 100, 500, and 1000 ppb. Both endosperm and germ sides of corn kernels were used for artificial contamination, and a total of 210 contaminated and control kernels were scanned with the visible-near infrared spectroscopy in reflectance mode. Spectral preprocessing methods including standard normal variate, first derivative, second derivative, first derivative + standard normal variate, and second derivative + standard normal variate were applied on the original absorbance spectra. Using the original and preprocessed spectra, the 3-class and 7-class discriminant models were established by the chemometric methods of principal component analysis-linear discriminant analysis and partial least squares discriminant analysis separately. The results showed that in discriminating the aflatoxin B1 contamination levels, the spectral range II (1120–2470 nm) generally performed better than using the corresponding spectra type over range I (410–1070 nm). Compared to using the original spectra, the first derivative and second derivative spectra generally improved the performance of the classification models. For classification thresholds of 20 and 100 ppb in aflatoxin B1, the best 3-class models achieved the same overall accuracy of 98.6% for prediction over both ranges I and II. For the 7-class discriminant models, the best overall accuracies obtained over ranges I and II are 91.4 and 97.1% for prediction.

An integrated strategy for rapid discovery and prediction of nucleobases, nucleosides and amino acids as quality markers in different flowering stages of Flos Chrysanthemi using UPLC–MS/MS and FT-NIR coupled with multivariate statistical analysis

Nucleobases, nucleosides, nucleotides and amino acids are important nutritional compounds, and some of them are positively correlated with the umami taste in scented tea. Taking these nutrients of Flos Chrysanthemi from three different flowering stages as the research object, an integrated strategy of ultra high-performance liquid chromatography coupled with triple-quadrupole linear ion-trap tandem mass spectrometry (UPLC–MS/MS), Fourier transform near infrared spectroscopy (FT-NIR) and multivariate statistical analysis was proposed to set up a systematic quality evaluation system for Flos Chrysanthemi from discovery to the rapid prediction of quality markers. First, a rapid, sensitive and accurate UPLC–MS/MS method was established to simultaneously determine 40 nucleobases, nucleosides, nucleotides and amino acids in three different flowering stages of Flos Chrysanthemi. The results indicated that Flos Chrysanthemi from three different flowering stages are rich in 40 nutritional compounds detected above and could be developed as functional products possessed nutraceutical and bioactive properties. Next, ten potential quality markers for three flowering stages of Flos Chrysanthemi were screened using partial least squares discriminant analysis (PLS-DA) and the plot of variable importance for projection (VIP), and the related metabolic pathways were identified for the first time. Furthermore, three of the ten potential quality markers, namely, glutamate, asparagine and aspartate were selected as key quality markers. Finally, three best PLS regression models constructed with FT-NIR were developed to predict the contents of the three key quality markers using the optimal spectral preprocessing methods and spectral interval combinations screened by synergy interval-PLS (Si-PLS), which yielded excellent prediction performance: 0.9766 ≤ R2p (coefficient of determination for prediction) ≤ 0.9858 and 6.3059 ≤ RPD (ratio of prediction to deviation) ≤ 7.3613. To sum up, the integrated quality marker discovery and FT-NIR quantitation strategy was suitable for fast quality evaluation of Flos Chrysanthemi.

Species- and moisture-based sorting of green timber mix with near infrared spectroscopy

Methods suitable for the determination and classification of green timber mix (western hemlock and amabilis fir), with respect to species and moisture content, were developed and tested using near infrared spectroscopy and chemometrics. One thousand two hundred samples were distributed into a calibration set (720 samples) and a prediction set (480 samples). Partial least squares (PLS) and least squares-support vector machines (LS-SVM) for both regression (PLSR and LS-SVR) and classification (PLS-DA and LS-SVC) with different spectral preprocessing methods were implemented. LS-SVM outperformed PLS models for both regression and classification. The coefficient of determination (R2p) and root mean square error (RMSEP) of prediction for the best LS-SVR model with spectra pretreated by smooth and first derivative were 0.9824 and 8.7%, respectively, for wood moisture content prediction in the range of 30% to 253%. The best classification model was LS-SVC with spectra pretreated by smooth and second derivative, with overall accuracies of 99.8% in the prediction set, when the samples were divided into four classes. NIRS combined with LS-SVM can be used as a rapid alternative method for qualitative and quantitative analysis of green hem-fir mix before kiln drying. The results could be helpful for sorting green hem-fir mixes with an on-line application.

Rapid and nondestructive determination of sesamin and sesamolin in Chinese sesames by near-infrared spectroscopy coupling with chemometric method.

Sesame was one of the most important crops in Africa and east Asia. The sesamin and sesamolin in sesames have shown various pharmacological, biological and physiologic activities. In this study, a rapid and nondestructive method for determination of sesamin and sesamolin in Chinese sesames by near-infrared spectroscopy coupled with chemometric method was proposed. The near infrared spectra of sesame samples from three different Chinese areas were collected and the partial least squares (PLS) was used to construct the quantitative models. The spectral preprocessing and variable selection methods were adopted to improve the predictability and stability of the model. Reasonable quantitative results can be obtained when the samples used for model construction and prediction were harvested in same years. For sesamin and sesamolin, the correlation coefficient (R) and root mean square error prediction (RMSEP) were 0.9754, 0.9636 and 151.2951, 39.7720, respectively. The optimized models seem less effective when they were used to predict the samples harvested in other years or countries. However, acceptable results can still be obtained.

Rapid Classification of Wheat Grain Varieties Using Hyperspectral Imaging and Chemometrics

The classification of wheat grain varieties is of great value because its high purity is the yield and quality guarantee. In this study, hyperspectral imaging combined with the chemometric methods was applied to explore and implement the varieties classification of wheat seeds. The hyperspectral images of all the samples covering 874–1734 nm bands were collected. Exploratory analysis was first carried out while using principal component analysis (PCA) and linear discrimination analysis (LDA). Spectral preprocessing methods including standard normal variate (SNV), multiplicative scatter correction (MSC), and wavelet transform (WT) were introduced, and their effects on discriminant models were studied to eliminate the interference of instrumental and environmental factors. PCA loading, successive projections algorithm (SPA), and random frog (RF) were applied to extract feature wavelengths for redundancy elimination owing to the possibility of existing redundant spectral information. Classification models were developed based on full wavelengths and feature wavelengths using LDA, support vector machine (SVM), and extreme learning machine (ELM). This optimal model was finally utilized to generate visualization map to observe the classification performance intuitively. When comparing with other models, ELM based on full wavelengths achieved the best accuracy up to 91.3%. The overall results suggested that hyperspectral imaging was a potential tool for the rapid and accurate identification of wheat varieties, which could be conducted in large-scale seeds classification and quality detection in modern seed industry.

Process quality control of the manufacturing of Chinese Materia Medica by process analysis technology

Process quality control is essential for the manufacturing of Chinese Materia Medica. Non-destructive monitoring is still a challenge for the quality control of Chinese Materia Medica manufacturing. As the commonly used non-destructive process analysis technology, near infrared spectroscopy, near infrared chemical imaging, and laser-induced breakdown spectroscopy have been applied to the quality control of Chinese Materia Medica manufacturing. The characteristic near infrared bands of 29 natural chemical components have been assigned. Ten spectral pre-processing methods and five variable selection methods have been optimized, respectively. Given the interrelationship among modeling parameters, a system modeling concept was put forward to establish a global model. Accuracy Profile was further proposed as the validation method of models. In addition, the homogeneity of chlorpheniramine maleate tablets from six brands was successfully visualized by near infrared-chemical imaging. As and Hg variation in An-Gong-Niu-Huang Wan have been rapidly monitored by laser-induced breakdown spectroscopy. A systematic modeling method (model establishment, evaluation, and validation) and the non-destructive technology of homogeneity visualization and toxic element detection have been developed for quality control of Chinese Materia Medica manufacturing, which is shown in Figure 1 of the article. These achievements have greatly promoted the research and application of process analysis technology in Chinese Materia Medica manufacturing, laying a solid foundation for the development of intelligent Chinese pharmaceutical industry.

Moisture content detection of maize seed based on visible/near‐infrared and near‐infrared hyperspectral imaging technology

SummaryTo realise accurate and nondestructive detection on moisture content of maize seed based on visible/near‐infrared (Vis/NIR) and near‐infrared (NIR) hyperspectral imaging technology, the hyperspectral images on two sides (embryo and endosperm sides) of each maize seed of four varieties were collected. The effects of average spectra extraction regions, that is centroid region and whole seed region, and different spectral preprocessing methods, were investigated. Uninformative variable elimination (UVE) was used to extract the feature wavelengths, and the partial least squares regression (PLSR) prediction models were established. The results showed that extracting the average spectra from the centroid region did better than from the whole seed region, and S‐G smoothing was prior to other preprocessing methods. The PLSR models established with NIR spectra had better performance than that with Vis/NIR spectra. The model developed for a single variety was superior to that for all varieties together.

Rapid Identification of Rainbow Trout Adulteration in Atlantic Salmon by Raman Spectroscopy Combined with Machine Learning.

This study intends to evaluate the utilization potential of the combined Raman spectroscopy and machine learning approach to quickly identify the rainbow trout adulteration in Atlantic salmon. The adulterated samples contained various concentrations (0–100% w/w at 10% intervals) of rainbow trout mixed into Atlantic salmon. Spectral preprocessing methods, such as first derivative, second derivative, multiple scattering correction (MSC), and standard normal variate, were employed. Unsupervised algorithms, such as recursive feature elimination, genetic algorithm (GA), and simulated annealing, and supervised K-means clustering (KM) algorithm were used for selecting important spectral bands to reduce the spectral complexity and improve the model stability. Finally, the performances of various machine learning models, including linear regression, nonlinear regression, regression tree, and rule-based models, were verified and compared. The results denoted that the developed GA–KM–Cubist machine learning model achieved satisfactory results based on MSC preprocessing. The determination coefficient (R2) and root mean square error of prediction sets (RMSEP) in the test sets were 0.87 and 10.93, respectively. These results indicate that Raman spectroscopy can be used as an effective Atlantic salmon adulteration identification method; further, the developed model can be used for quantitatively analyzing the rainbow trout adulteration in Atlantic salmon.

Quantitative Analysis of Total Nitrogen Content in Monoammonium Phosphate Fertilizer Using Visible-Near Infrared Spectroscopy and Least Squares Support Vector Machine

А quantitative analysis method to determine the total nitrogen content in monoammonium phosphate (MAP) fertilizer using visible-near infrared (Vis-NIR) spectroscopy and least squares support vector machine (LS-SVM) is proposed. Sample set partitioning based on the joint x–y distance (SPXY) was used to select the calibration set. Fourteen spectral pre-processing methods were then employed to deal with the spectral data including Savitzky–Golay (SG) smoothing, fi rst derivative (D1) and second derivative (D2) with SG smoothing, multiplicative scatter correction (MSC), standard normal variate (SNV), wavelet, and combination thereof. Next, the LS-SVM model with radial basis function kernel was established with the best pre-processing method, and its performance was compared with that of partial least squares (PLS) model. The results revealed LS-SVM calibration with the discrete wavelet transform provided the best prediction for total nitrogen content in MAP fertilizer, yielding R2, root mean square error of prediction (RMSEP), and ratio of performance to deviation (RPD) values of 0.91, 0.101, and 3.34, respectively.

Convolutional neural network for simultaneous prediction of several soil properties using visible/near-infrared, mid-infrared, and their combined spectra

No single instrument can characterize all soil properties because soil is a complex material. With the advancement of technology, laboratories have become equipped with various spectrometers. By fusing output from different spectrometers, better prediction outcomes are expected than using any single spectrometer alone. In this study, model performance from a single spectrometer (visible-near-infrared spectroscopy, vis-NIR or mid-infrared spectroscopy, MIR) was compared to the combined spectrometers (vis-NIR and MIR). We selected a total of 14,594 samples from the Kellogg Soil Survey Laboratory (KSSL) database that had both vis-NIR and MIR spectra along with measurements of sand, clay, total C (TC) content, organic C (OC) content, cation exchange capacity (CEC), and pH. The dataset was randomly split into 75% training (n = 10,946) and the remaining (n = 3,648) as a test set. Prediction models were constructed with partial least squares regression (PLSR) and Cubist tree model. Additionally, we explored the use of a deep learning model, the convolutional neural network (CNN). We investigated various ways to feed spectral data to the CNN, either as one-dimensional (1D) data (as a spectrum) or as two-dimensional (2D) data (as a spectrogram). Compared to the PLSR model, we found that the CNN model provides an average improvement prediction of 33–42% using vis-NIR and 30–43% using MIR spectral data input. The relative accuracy improvement of CNN, when compared to the Cubist regression tree model, ranged between 22 and 36% with vis-NIR and 16–27% with MIR spectral data input. Various methods to fuse the vis-NIR and MIR spectral data were explored. We compared the performance of spectral concatenation (for PLSR and Cubist model), two-channel input method, and outer product analysis (OPA) method (for CNN model). We found that the performance of two-channel 1D CNN model was the best (R2 = 0.95–0.98) followed closely by the OPA with CNN (R2 = 0.93–0.98), Cubist model with spectral concatenation (R2 = 0.91–0.97), two-channel 2D CNN model (R2 = 0.90–0.95) and PLSR with spectral concatenation (R2 = 0.87–0.95). Chemometric analysis of spectroscopy data relied on spectral pre-processing methods: such as spectral trimming, baseline correction, smoothing, and normalization before being fed into the model. CNN achieved higher performance than the PLSR and Cubist model without utilizing the pre-processed spectral data. We also found that the predictions using the CNN model retained similar correlations to the actual values in comparison to other models. By doing sensitivity analysis, we identified the important spectral wavelengths variables used by the CNN model to predict various soil properties. CNN is an effective model for modelling soil properties from a large spectral library.

SPECTRAL PREPROCESSING FOR HYPERSPECTRAL REMOTE SENSING OF HEAVY METALS IN WATER

Abstract. This study aims to investigate the feasibility of using hyperspectral remote sensing technique by visible-near infrared spectroradiometer (VNIR, FieldSpec HandHeld 2) for rapid water monitoring of heavy metal, followed by comparison of different spectral preprocessing methods for the development of quantitative predictive model. The water samples evaluated in this study were prepared in our laboratory by dilution of stock solutions. Heavy metals of lead (Pb), zinc (Zn) and copper (Cu) in the range of concentration between 100 to 2000 mg/L were selected as the target samples in this study. The sensitive bands for the target metals were characterized in the range from 800 nm to 1075 nm, based on the reflectance spectral data. Spectral data for developing of the quantitative predictive model was first preprocessed with first derivative and logarithm transformation, followed by establishing of the prediction model using multivariate linear regression (MLR). It was observed that increase in the number of sensitive bands for the MLR can significantly improve the adjusted R2 for the model. The prediction model for Cu was found to have the highest adjusted R2 of 0.92 and least normalized root mean square error (NRMSE) of 0.065, while using the reflectance values of 7 sensitive bands. This result could be attributed to the blue color characteristic of the solution, whereas the others remain clear. Additionally, the first derivative transformation was determined as the best method for predicting Pb, whereas the logarithm transformation provided the best outcomes for predicting Cu and Zn.

Mapping changes in sarcoplasmatic and myofibrillar proteins in boiled pork using hyperspectral imaging with spectral processing methods

Hyperspectral imaging technology was investigated to examine the changes of sarcoplasmatic and myofibrillar proteins contents in boiled pork. Different spectral pre-processing methods were employed to improve the partial least squares regression (PLSR) models. Seven optimal wavelengths in the region of 400–2000 nm without spectral pre-processing were used to build the best PLSR model for predicting the sarcoplasmatic proteins contents with R2p of 0.9185 and RMSEP of 2.50 mg/g. Based on multiplicative scatter correction (MSC) method, eight wavelengths in the region of 400–1000 nm were also selected to construct the best predictive PLSR model for myofibrillar protein contents with R2p of 0.9161 and RMSEP of 2.71 mg/g. The changes in sarcoplasmatic and myofibrillar proteins contents of boiled pork during different cooking periods were successfully visualized.

Identification of mildew degrees in honeysuckle using hyperspectral imaging combined with variable selection

Mildew is one of the main reasons for the quality degradation of honeysuckle, which can lead to economic loss and threaten human safety. In order to detect different mildew degrees of honeysuckle, a method based on hyperspectral imaging technology was investigated. Different spectral pre-processing methods including Savizky–Golay filter (SG), standard normalized variable (SNV), multiple scatter correct (MSC), SG–MSC (a combination of SG and MSC) and SG–SNV were conducted for raw spectral data. A comparison was made among different pre-processing methods based on partial least squares regression models, of which the best method was SG–SNV. Then the uninformative variable elimination (UVE), competitive adaptive reweighted sampling (CARS) and successive projections algorithm (SPA) were applied to extract effective variables (wavelengths) from the preprocessed data by SG–SNV. The extreme learning machine (ELM) models, which the ‘Sigmoidal’ was chosen as the incentive function and the number of neurons in the hidden layer was 40, were developed for identifying honeysuckle with different mildew degrees using full-spectrum data and the selected variables obtained by UVE, CARS, SPA, UVE–CARS, UVE–SPA, CARS–SPA and UVE–CARS–SPA. The classification results showed that the UVE–SPA–ELM model performed the highest accuracy of 100% for both training set and testing set and the proposed UVE–SPA method was optimal and powerful for the variable selection. The results of this study indicated that hyperspectral imaging technology could be a rapid and non-destructive analytical tool for identifying different mildew degrees of honeysuckle.

Photoacoustic Spectrometric Evaluation of Soil Heavy Metal Contaminants

Heavy metal pollution in soil has severe impact on the human health and global environment. There is an urgent need for cost-effective devices capable of recognizing and detecting heavy metallic matters in soils. A broadband photoacoustic spectrometric (PAS) system was established for the detection of heavy metal contaminants in soil. The heavy metal element lead (Pb) was selected as a preferred detection target. The near infrared photoacoustic spectra of contaminated soil samples with various concentrations of Pb were collected and the spectroscopic relationship between the soil absorption peaks and Pb concentrations was analyzed. Four advanced spectral preprocessing methods were explored to improve the robustness of the prediction model. Based on the maximal correlation coefficient and minimal root mean square error criteria of prediction, a two-layer feed-forward network with a continuum removing preprocessing method with a correlation coefficient as 0.96 was tested as the most appropriate method for Pb concentration prediction. This fact verifies that the broadband PAS evaluation methodology, as a nondestructive testing method, can be potentially used as an alternative quantification detection method of heavy metal contaminants in soil without the involvement of complicated sample pretreatment.

Major ergot alkaloids in naturally contaminated cool-season barley grain grown under a cold climate condition in western Canada, explored with near-infrared (NIR) and fourier transform mid-infrared (ATR-FT/MIR) spectroscopy

Globally, mycotoxins in food and feed are serious threats to the health of human and animals. Ergot alkaloids (EAs) are frequently detected in barley as well as other cereals. The purpose of this study was to examine the potential of applying NIR and ATR-FT/MIR spectroscopy for the determination of major EAs in cool-season barley samples grown under cold climate condition in western Canada. Sixty-seven cool-season barley samples were collected from Western Canada and analyzed for six major EAs by LC-MS/MS technique. Forty-nine barley samples were positive for EAs. The mean concentrations of ergocornine, ergocristine, ergocryptine, ergometrine, ergosine, ergotamine, and the sum-total EAs in positive barley samples were 120.9, 554.8, 235.6, 411.0, 127.4, 351.2, and 1150.5 μg/kg, respectively. The NIR (680–2500 nm) and MIR (4000-700 cm−1) spectra of cool-season barley samples were collected and calibrated with EAs reference values using PLS regression technique based on different spectral preprocessing methods and selected wavelengths. Results showed that most of the constructed models obtained weak calibration and cross-validation parameters and none of the models was able to accurately predict external samples. The relatively low levels of EAs in the contaminated barley samples might be lower than the detection limit of the conventional NIR and ATR-FT/MIR spectroscopy. More efforts are needed to study the determination limit of molecular spectroscopic techniques in detecting EAs in cereal grain grown under the cold climate condition.

On‐the‐Go Sensor Fusion for Prediction of Clay and Organic Carbon Using Pre‐processing Survey, Different Validation Methods, and Variable Selection

Core Ideas Four different validation methods were investigated. Extensive pre‐processing and variable selection were used. Fusing electrical conductivity to spectra adds no value to models. Clay and organic carbon predicted with high accuracy on‐the‐go. Independent field prediction is not yet reliable. Using on‐the‐go visible and near‐infrared (Vis–NIR) spectroscopy singularly or in combination with other sensor data has been found to produce promising cross‐validation (CV) and prediction results. In this study, besides from testing two CV methods, we aim to predict clay and organic carbon (OC) not only in new samples but also on samples from new fields that were not included in either of the calibration or validation processes. Sensor data was collected from eight fields in different parts of Denmark using a mobile multi‐sensor platform with a Vis‐NIR spectrometer and electrical conductivity and temperature sensors. Fifteen samples were collected from each field using fuzzy c‐means clustering of the field spectra, and clay and OC were measured in the laboratory using traditional methods. Different combinations of spectral pre‐processing methods were tested and Partial Least Squares Regression (PLSR) was calibrated on the data and validated using four methods: (A) Venetian blinds CV; (B) one‐field‐out CV; (C) model on 70% of the data used to predict the remaining 30%; and (D) model on the data from six fields used to predict the remaining two fields. To simplify models and improve performance, interval PLS (iPLS) variable selection was used. Results indicate no or negligible improvement when spectral data was fused with other sensor data. In general, the results look very promising for Validations A, B and C for both clay and OC. However, in Validation D, models did not produce high accuracy results. Our findings show that good processing of Vis‐NIR spectral data can produce the best models, making sensor fusion for clay and OC determination pointless. These results have important implications for future sensor design and sensor measurements in soil science and call for further investigations into on‐the‐go site‐independent field prediction.