Introduction
Wheat is cultivated all over the world and is grown on more land area than any other food crop. The trade market for wheat is greater than that of all other crops combined and world production of wheat is well over seven hundred million metric tons annually. It is an important source of carbohydrates and the leading source of vegetable protein in food. It can be consumed as a whole grain or milled into flour and used to make numerous types of food. Wheat straw is also used as an animal feed and for manufacture of many different types of products. Proper planting, harvesting, processing, storage, and transport are essential. Monitoring of moisture in wheat is essential for proper harvesting time, pest and disease management, and avoiding spoilage during storage and transport. Protein and carbohydrates are important nutritional components as well as total gluten, glutenin, and gliadin. Wheat grain hardness is classified into three major hardness classes: soft, hard hexaploid, and durum. These are generally related to endosperm texture and although extensively studied, no direct relationship between the genetic and physicochemical basis of endosperm texture has been established. However, protein, starch, and color differences do relate to hardness in grain. An increase in particle size increases the absorption of NIR light and particle size can be directly correlated to wheat hardness. Determining wheat hardness from NIR spectroscopy is a certified AACC method. The need for minimal gluten products for people with celiac disease and related ailments makes gluten content especially important. It is important to monitor wheat straw residue composition potential in wheat fields because the level of decomposition needed to keep the soil healthy varies based on rainfall levels in the region. Wheat straw is also an important precursor for biofuel production and there are parameters that must be monitored after the necessary pretreatment, such as weight loss, residual lignin content, and hydrolysable sugars. As is the case with many agricultural products, adulteration is a problem with wheat as the nutritional value and price can vary greatly in different products. A need exists to authenticate wheat on a large scale without expensive tests and the use of subjective sensory monitoring. While research and development of transgenic wheat strains does lag behind that of other mass produced agricultural products like rice, it is still prominent especially for developing strains that are resistant to herbicides and that are low in gluten content. There is a need to develop fast, non-invasive testing methods to meet the evolving challenges in producing quality wheat. One such method that has been examined is NIR Spectroscopy.
Analytes
- Moisture
- Protein
- Total gluten content
- Glutenin
- Gliadin
- Hardness
- Neutral Detergent Fiber (NDF)
- Acid Detergent Fiber (ADF)
- Acid Detergent Lignin (ADL)
- Cellulose
- Hemicellulose
- Carbon
- Nitrogen
- Weight loss
- Residual lignin content
- Hydrolysable sugars
- Authentication of origin
- Wild vs. transgenic discriminant analysis
Summary of Published Papers, Articles, and Reference Materials
Predicting Wheat Quality Characteristics and Functionality Using Near-Infrared Spectroscopy
NIR Spectroscopy was examined as a method for predicting a number of grain, milling, flour, dough, and breadmaking quality parameters in both red winter and red spring wheat and flour samples. One hundred of both Hard Red Winter and Hard Red Spring samples were provided for the study by the USDA Grain Inspection, Packers, and Stockyard Administration Federal Grain Inspection Service. Samples were specifically chosen for a range of protein content. HRW samples ranged from 9.2% to 15.8% protein and HRS samples ranged from 11.4% to 19.3% protein. All samples were scanned with four different NIR spectrometers to study the effects of different wavelength ranges and scanning technologies on the modeling and prediction results. Wavelengths ranges were 835 nm to 2502 nm, 850 nm to 1050 nm, 450 nm to 2498 nm, and 950 nm to 1650 nm. A portion of each sample was milled into flour and spectra were collected for each sample using both whole grains and flour. Various pre-processing methods were applied to the spectral data before chemometric modeling. Many of the parameters tested showed poor correlation for a number of reasons, including concentration below the threshold of detection, small range of values, or the parameter being a measurement that does not have an effect on the NIR spectra. However, good correlation was obtained for moisture, protein, gluten parameters, and mixograph absorption. The results for these parameters from all four instruments are shown below.
Grain Protein Content | R² 0.97 – 0.99 | SECV 0.18 – 0.29 14% mb |
Grain Moisture Content | R² 0.95 – 0.97 | SECV 0.16 – 0.19 % |
Single Kernel Moisture | R² 0.92 – 0.94 | SECV 0.22 – 0.26 % |
Flour Protein Content | R² 0.92 – 0.99 | SECV 0.29 – 0.45 14% mb |
Gluten Content | R² 0.88 – 0.93 | SECV 0.14 – 0.19 g/10 g of flour |
Soluble Glutenins | R² 0.75 – 0.77 | SECV 0.40 – 0.52 mg |
Soluble Gliadins | R² 0.85 – 0.89 | SECV 0.64 – 0.76 mg |
Insoluble Glutenins | R² 0.84 – 0.85 | SECV 0.64 – 0.67 mg |
Total Glutenins | R² 0.81 – 0.93 | SECV 0.59 – 1.02 mg |
Mixograph Absorption | R² 0.90 – 0.92 | SECV 0.67 – 0.76 % |
Results showed that moisture and protein could be predicted with a level of accuracy suitable for process control purposes while the gluten parameters can be predicted for quality control. A number of other parameters were modeled and showed predictions good enough for screening purposes, such as test weight, single kernel diameter, SDS sedimentation volume, color values, loaf volume, flour particle size, and the percentage of dark hard and vitreous kernels. Further analysis determined that many of these parameters were closely correlated to protein content. The influence from protein content was removed from the models and the results often got worse. The potential was shown to predict many grain quality and functionality traits from NIR spectroscopy, but many parameters are modeled based on their correlation to protein content.
https://www.ars.usda.gov/ARSUserFiles/30200525/368PredictingWheatQualityCharandFunctionality.pdf
Hardness Measurement of Bulk Wheat by Single-Kernel Visible and Near-Infrared Spectroscopy
Wheat grain hardness is the most important quality trait for milling properties and end use. There are three classifications of wheat hardness used in the United States: soft, hard hexaploid, and durum. Grain hardness is defined in more detail as endosperm texture and the various techniques that are used for grain hardness measurement are classified into diverse groups according to grinding, crushing, and abrasion. These methods include PSI, SKCS, pearling index, SDS-PAGE, and PCR markers as well as NIR spectroscopy. NIR is an AACC approved method for determining hardness as the reflectance signal and NIR absorption increase with increasing particle size. It is proven that hardness of grain increases with particle size. Thus, NIR spectroscopy can be used to determine particle size using a method much less labor-intensive and faster than other methods. In this study, samples of both hard and soft wheat single kernels were used to determine the feasibility of single kernel hardness analysis using NIR spectroscopy. Hard wheat samples were obtained from the National Institute of Standards and Technology (NIST). Soft wheat samples were obtained from the USDA Soft Wheat Quality Laboratory (SWQL). In total, thirty-five samples were used as a calibration set and one hundred single kernels were randomly selected from each of them. Likewise, one hundred single kernels from thirty separate sample sets were used as a validation set. Each one hundred kernel set was loaded into an automated hopper for single automated measurements of NIR spectra collection, single kernel hardness, and single kernel moisture. NIR spectra were collected from 400 nm to 1700 nm in reflectance mode. Eight spectra were collected per sample and averaged into a single spectrum. All kernels were first classified as soft, hard or mixed at various kernel amount averages ranging from one kernel to fifty kernels. The hardness measurement used was hardness index with a score greater than forty-six corresponding to hard and less than forty-six corresponding to soft. The thirty kernel average group was then chosen to create a PLS calibration model correlating the NIR spectra to hardness. Results are shown below.
Hardness Classification:
1 Kernel Average | R² = 0.49 |
5 Kernel Average | R² = 0.83 |
10 Kernel Average | R² = 0.86 |
20 Kernel Average | R² = 0.90 |
30 Kernel Average | R² = 0.91 |
50 Kernel Average | R² = 0.91 |
Correct Classification between Hard and Soft Kernels Compared to Reference Method: 97%
Hardness PLS Model | R² = 0.91 | SECV= 7.70 |
The results here confirmed the potential of using NIR spectroscopy and calibration models to both classify kernels based on grain hardness and to quantify hardness index. One factor must be noted as using another spectroscopic method as the reference method when building calibration models using NIR spectra can often introduce error, even when the spectroscopic method is a certified AACC method. Results would likely improve using a different hardness test for the reference values. Still, the correlation coefficient having a value higher than 0.9 indicates that the correlation is accurate. It is advised that more study be done before using this method in a real-time setting.
334AACChardness.pdf (usda.gov)
Modeling Research on Wheat Protein Content Measurement Using Near-Infrared Reflectance Spectroscopy and Optimized Radial Basis Function Neural Network
NIR spectroscopy was examined as a method for determining protein content in wheat. One hundred forty wheat samples from a dozen different wheat producing areas were provided by the Institute of Agricultural Quality and Safety in China. Protein content in the samples ranged from 10.85% to 18.31% and were chosen from areas representative of the actual wheat growing conditions in China. Samples were scanned from 850 nm to 1050 nm at 2 nm intervals. Ten scans were collected per reading and averaged into one spectrum. Reference values for protein were determined by the semimicro-Kjeldahl method, which is time-consuming and impractical for implementing on a large scale. Instead of the traditional Partial Least Squares (PLS) algorithm, the artificial neural network algorithm known as Radial Basis Function (RBF) was used to correlate the spectral data to protein value. RBF is favored by many researchers for its ease of use, high fitting, and high nonlinear approximation. The Particle Swarm Algorithm (PSA) was used to optimize the number of cluster centers in the hidden layers of the RBF network. One hundred samples were used to create the RBF model and the remaining samples were used as a validation set for independent predictions.
Protein | R² = 0.975 | RMSEP = 0.266% |
The results of this study were excellent and proved the feasibility of using NIR spectra, protein values, and the RBF algorithm to predict protein in wheat. The independent prediction confirmed the validity of the model. NIR spectroscopy can be used to replace traditional time-consuming and expensive methods for determining protein content in wheat.
Rapid Estimation of Wheat Straw Decomposition Constituents Using Near-Infrared Spectroscopy
In areas where wheat in grown and rainfall amounts can vary from region to region, crop and soil management practices must be adjusted to account for high and low rainfall. One example of this is using no-till production systems for wheat crops, which is an excellent technique for reducing soil erosion. It is important for wheat straw residue to decompose rapidly in winter months in high rainfall regions to avoid planting complications in the spring. Likewise, in low rainfall regions, wheat straw residue needs to decompose slowly to cover the soil during the entirety of the fallow season. Fallowing is an old term for soil management defined as allowing soil time to rest and recover. A need exists for determining wheat straw decomposition parameters that is effective, fast, non-destructive, and requires minimal labor. NIR spectroscopy was examined for this purpose. Straw from a panel of four hundred eighty soft winter wheat cultivars from the Pacific Northwest were provided for the study. This region is particularly known for variance in annual rainfall totals and the samples were from two separate regions, one known for high annual rainfall and the other for low annual rainfall. Reference tests were performed for the following parameters: Neutral Detergent Fiber (NDF), Acid Detergent Fiber (ADF), Acid Detergent Lignin (ADL), Cellulose, Hemicellulose, Carbon, and Nitrogen. Samples were scanned from 400 nm to 2498 nm at 2 nm intervals in a sampling cup. Each sample was rotated 90 degrees after the first scan and scanned again, with the two spectra for each sample then averaged into one spectrum. Various pre-processing methods were applied to the spectral data before chemometric modeling. Partial Least Squares (PLS) calibration models were created correlating the NIR spectra to the parameters of interest. Results are shown below.
NDF | R² = 0.87 | SECV 1.52% |
ADF | R² = 0.89 | SECV 1.38% |
ADL | R² = 0.68 | SECV 0.56% |
Cellulose | R² = 0.91 | SECV 1.10% |
Hemicellulose | R² = 0.45 | SECV 1.10% |
Carbon | R² = 0.76 | SECV 1.12% |
Nitrogen | R² = 0.75 | SECV 0.05% |
Cross-validation was used to pull out representative samples from the calibration models and perform independent predictions. Prediction results were successful in predicting NDF, ADF, and cellulose with an accuracy suitable for screening purposes. Accuracy was lower for the other parameters and not suitable to be used in any real-time setting. Results may improve if more samples were incorporated into the calibration set. For real-time use, classifying wheat straw samples into a high or low category for decomposition potential by predicting NDF, ADF, and cellulose along with subsequent tests for carbon and nitrogen using another reference method would work to provide a good estimate of fast or slow decomposition potential.
Characterization of Key Parameters for Biotechnological Lignocellulose Conversion Assessed by FT-NIR Spectroscopy
Wheat straw and oat straw are lignocellulosic materials that contain around 30% to 35% cellulose, 20% to 25% hemicelluloses, and 17% to 20% lignin. The carbohydrates in these materials can be hydrolyzed to fermentable sugars, which are precursor substances for biotechnological conversion to biofuels or building blocks for chemical syntheses. However, pretreatment for delignification is required to open up the lignocellulose structure and to increase the accessibility to microbial enzymes. Traditional methods for determining key parameters like weight loss, residual lignin content, and hydrolysable sugars entail expensive and time-consuming wet chemistry methods which are impractical to implement for large scale testing. NIR spectroscopy was examined as a method for determining key parameters for biotechnological lignocellulose conversion in wheat straw and oat straw. Eighty wheat straw samples and fifty-three oat straw samples were procured for the study. Initial wet chemistry analysis showed the following composition: 63% polysaccaharides, 21.5% lignin, 11.4% extractives, and 4.2% ash for wheat straw and 51% polysaccaharides, 19.6% lignin, 20.5% extractives, and 8.9% ash for oat straw. Samples were chopped to 1 cm length and 20 g per sample were treated at varying concentrations of acid, acid/H2O2, alkali, and alkali/H2O2. Samples were then rinsed and dried before wet chemistry reference testing to determine conversion parameter values. An FT-NIR spectrometer was used to scan samples from 10000 cm-1 to 4000 cm-1 at 8 cm-1 resolution. One hundred scans were collected per reading and averaged into one spectrum. This process was repeated four times for each sample and the four collected spectra were then averaged into one single spectrum per sample. Various pre-processing methods were applied to the spectral data before chemometric modeling. Partial Least Squares (PLS) models were created correlating the NIR spectra to the parameters of interest. Results are shown below.
Wheat Straw:
Weight Loss Range: 4.0% to 33.5% | R² = 0.85 | RMSEP= 3.5% |
Residual Lignins Range: 7.9% to 20.7% | R² = 0.95 | RMSEP= 0.9% |
Reducing Sugars Range: 128 mg/g – 1000 mg/g | R² = 0.94 | RMSEP= 83 mg/g |
Oat Straw:
Weight Loss Range: 5.0% to 44.0% | R² = 0.96 | RMSEP= 3.4% |
Residual Lignins Range: 8.3% to 18.5% | R² = 0.99 | RMSEP= 0.8% |
Reducing Sugars Range: 131 mg/g – 812 mg/g | R² = 0.96 | RMSEP= 64 mg/g |
The results here show the potential to use NIR Spectroscopy as a method for determining parameters essential for biotechnological lignocellulose conversion of both wheat straw and oat straw. Further calibration work was done to determine the feasibility of measuring parameters for anaerobic conversion of wheat straw to biogas: biogas production, total solids, and volatile solids content. While the results were not good enough for quantitative measurement of these parameters from NIR spectra, they were considered decent enough for estimation of values. Further study and more samples may improve these results. Overall, NIR spectroscopy shows promise as a fast, non-invasive, and non-destructive method for determining important precursor parameters of biofuel production in wheat straw and oat straw.
Tracing the Geographical Origin of Durum Wheat by FT-NIR Spectroscopy
Durum wheat is a cereal crop that is mainly cropped in the Mediterranean basin and is used to manufacture a wide range of products. Characteristics include large kernel size, hardness, bright yellow color, high protein content, and gluten strength. It is especially popular for making pasta and Italy is the country with both the highest production and consumption of durum wheat. Consumption of durum wheat is so high in Italy that despite being the largest producer of it, imports are required to meet demand. As is the case with many natural products, durum wheat can vary greatly in nutritional quality and market price based on origin. While labeling origin is a requirement of the EU, this is difficult to enforce in practice as current methods for determining origin require destructive and expensive methods such as isotopic, compositional, and elemental analysis that are impractical to implement for large scale testing. NIR spectroscopy was examined as a method for determining geographical origin of durum wheat. Fifty-nine durum wheat samples from eleven different regions coming from three separate geographical areas of Italy were procured for the study. Twenty-nine samples from eight different foreign countries were obtained as well. Samples were ground before scanning using an FT-NIR spectrometer from 10000 cm-1 to 4000 cm-1 at 8 cm-1 resolution. Thirty-two scans were collected per reading and averaged into one spectrum. For some samples, multiple separate portions of each sample were scanned as well and in total, one hundred eighty-one spectra were collected for the Italian samples. Likewise, a total of seventy-five spectra were collected for the non-Italian samples. Different pre-processing methods were applied to the NIR spectra before classification analysis. Two separate Principle Component Linear Discriminant Analysis (PC-LDA) classification models were created from the NIR spectra: one classifying the Italian samples based on the Northern, Central, and Southern geographical origins and the other classifying the Italian samples from non-Italian samples.
Italian Samples: Overall Discrimination Rate (OD) | 96.7% |
Non-Italian Samples: Overall Discrimination Rate (OD) | 100% |
Both classification models showed excellent results and were validated by using NIR spectra of separate samples from those used to create the calibration models. Some misclassification occurred between samples from the Northern and Central regions of Italy while the Southern region was almost always classified correctly. This likely occurred because growing conditions in Northern Italy are humid and cold while conditions in Southern Italy are warmer and dryer, leading to differences in the chemical composition in the wheat. A perfect discrimination rate was obtained for the Italian samples and samples from other countries. This study demonstrated the potential of NIR spectroscopy for use as a fast and non-invasive method for classifying durum wheat based on geographical origin.
Effective Identification of Low-Gliadin Wheat Lines by Near Infrared Spectroscopy (NIRS): Implications for the Development and Analysis of Foodstuffs Suitable for Celiac Patients
Gluten proteins account for 80% to 85% of total grain protein in wheat, with about 30% being gliadins and 50% being glutenins. These proteins are associated with celiac disease, which may affect up to 7% of the world’s population, and non-celiac gluten sensitivity (NCGS), which is estimated to be prevalent in up to 6% of the United States population. For people suffering from these ailments, it is recommended to follow a completely gluten-free diet. In practice, this is difficult to do as wheat is such a large part of food products and additives. One promising approach for reducing gluten toxicity for those affected with these disorders is the down regulation of immunodominant gluten peptides by RNA interference (RNAi), resulting in low-gliadin wheat lines. Research and testing has demonstrated the potential of this technology in bread wheat to develop food products that can be tolerated by those suffering from celiac disease and NCGS. There is a need to develop a system capable of distinguishing normal wheat lines from transgenic low-gliadin wheat lines and NIR spectroscopy was examined for this purpose. Two sets of samples were obtained for the study: Four hundred and nine wild and one hundred twenty-six transgenic whole grain samples and four hundred and fourteen wild and one hundred fifty-six transgenic flour samples. All samples were scanned using an NIR spectrometer from 400 nm to 2500 nm at a 2 nm scanning interval. Each sample was scanned twice and the results were averaged into a single spectrum. Various pre-processing methods were applied to the spectral data before chemometric analysis. Different Partial Least Squares-Discriminant Analysis (PLS-DA) classification models were created using different wavelength ranges and pre-processing methods on the spectral data. The best results obtained are shown below.
Whole Grain Classification | 96% correct |
Flour Classification | 99% correct |
PLS-DA models use an arbitrary number for two classification groups and a number is chosen based on the NIR spectra to classify samples. An independent validation set of samples was used for both whole grain and flour to perform predictions and the results were excellent. Further validating the results is that the validation set for both groups was used from two separate harvesting years, indicating that any classification is not based on different harvests. This study proved the feasibility of using NIR spectra and classification models to successfully classify wild and transgenic whole grain wheat and wheat flour.