A. M. Mouazen

Near Infrared Spectroscopy for Agricultural Materials: An Instrument Comparison:

The selection of a spectrophotometer for the measurement of constituents of agricultural materials with acceptable accuracy and cost effectiveness requires a comparative study of the performance of different spectrophotometers. Four commercially available spectrophotometers were evaluated, based on measurements performed on three agricultural materials. These spectrophotometers, differing mainly in wavelength range and measurement principles, comprised a diode array (DA) of 300–1700 nm, a combination of diode array and scanning monochromator (DASM) of 350–2500 nm, a Fourier transform (FT) of 750–2500 nm and a scanning monochromator (SM) of 400–2500 nm spectrophotometers. They were used to measure the moisture content of soil, the chemical constituents of hog manure and to detect bruising in apples. Three spectral pre-treatments were considered. Calibrations were developed using partial least squares (PLS) regression with the leave-one-out cross-validation technique for soil and manure and principal component analysis (PCA) for apple. The four instruments provided good predictions for soil moisture content, with the largest coefficient of determination (r2) values between 0.84–0.86 and with the largest ratio of prediction to deviation (RPD) of standard deviation (SD) to root mean square error of cross-validation (RMSECV) ranging from 2.53 to 2.75. The DASM and SM were comparable and slightly better than the DA and FT. For hog manure, total nitrogen was predicted more accurately with the four instruments (r2 = 0.83–0.89 and RPD = 2.43–3.01) than the phosphorus (r2 = 0.66–0.85 and RPD = 1.72–2.61) and potassium (r2 = 0.70–0.84 and RPD = 1.83–2.50). The order of accuracy of the four spectrophotometers for the measurement of total nitrogen was: FT–SM–DA–DASM. For phosphorus and potassium the order was: DA–DASM–FT–SM and SM–FT–DASM–DA, respectively. The DA performed better than the DASM and FT for discrimination of bi-colour and single-colour bruised apple, respectively. Therefore, selection of a spectrophotometer depends mainly on the type of material analysed and the constituent to be measured. Wavelengths above 1700 nm were found unnecessary for the applications considered and the DA spectrophotometer was of sufficient accuracy, as it is robust, significantly cheaper and can be used in the field for on-line measurements.

Classification of soil texture classes by using soil visual near infrared spectroscopy and factorial discriminant analysis techniques

Texture is one of the main properties affecting the accuracy of visible (vis) and near infrared (NIR) spectroscopy during on-the-go measurement of soil properties. Classification of soil spectra into predefined texture classes is expected to increase the accuracy of measurement of other soil properties using separate groups of calibration models, each developed for one texture class. A mobile, fibre-type, vis-NIR spectrophotometer (Zeiss Corona 1.7 vis-NIR fibre), with a light reflectance measurement range of 306.5–1710.9 nm was used to measure the light reflectance from fresh soil samples collected from many fields in Belgium and northern France. A total of 365 soil samples were classified into four different texture classes, namely, coarse sandy, fine sandy, loamy and clayey soils. The factorial discriminant analysis (FDA) was applied on the first five principal components obtained from the principal component analysis performed on the vis-NIR spectra in order to classify soils into the four assigned groups. Correct classification (CC) of 85.7% and 81.8% was observed for the calibration and validation data sets, respectively. However, validation of the vis-NIR-FDA technique on the validation set showed poor discrimination between the coarse sandy and fine sandy soil groups, with a great deal of overlapping. Therefore, the soil groups were reduced to three groups by combining the coarse sandy and fine sandy soil groups into one group and FDA was applied again. A better classification was obtained with CC of 89.9 and 85.1% for the calibration and validation data sets, respectively. However, the CC for the sand group in the validation set was rather small (46.7%), which was attributed to the small sample number and poor correlation between sand fraction and vis-NIR spectroscopy. It was concluded that vis-NIR-FDA is an efficient technique to classify soil into three main groups of sandy (light soils), loamy (medium soils) and clayey (heavy soils). Additional samples from the sandy and clayey groups should be included to improve the accuracy of the vis-NIR-FDA classification models to be used for an on-the-go vis-NIR measurement system of soil properties.

Classification of Soils into Different Moisture Content Levels based on VIS-NIR Spectra

Soil moisture content (MC) affects the accuracy of the visible (VIS) and near infrared (NIR) spectroscopic measurement of other soil properties e.g. carbon, nitrogen, etc. This study was conducted to subtract the MC contribution to VIS-NIR spectra by classifying soil spectra into different MC groups. A mobile, fiber-type, VIS-NIR spectrophotometer (Zeiss Corona 1.7 visnir fiber), with a measurement range of 306.5 – 1710.9 nm was used to measure the light reflectance of two sample sets; one (275 samples) collected from a single field and the other (360 samples) collected from multiple fields in Belgium and northern France. The Partial Least Squares (PLS) regression analysis and Factorial Discriminant Analysis (FDA) were applied on the VIS-NIR spectra in order to quantify MC and classify spectra into different MC groups, respectively. The PLS for the single-field sample set provided better estimation of MC (R2=0.98) than for the multiple-field sample set (R2=0.88). For the single-field sample set, spectra were successfully classified into six MC groups with correct classification (CC) of 94.1% and 95.6% for the calibration and validation data sets, respectively. Due to large variability of the multiple-field sample set, soils were successfully classified into 3 MC groups only. The CC obtained were 88.1% and 79.7% for the calibration and validation sets, respectively. These results suggested that the FDA can be successfully used to classify soil VIS-NIR spectra into different MC levels, particularly when soil variability is minimal.

Vis-NIR Spectroscopy Coupled with Multivariate Statistical Analyses as a Tool for the Classification of Soil Texture

Soil texture is an important property for the assessment of soil quality and the sustainability of agricultural management practices. The composition of sand, silt and clay affects soil–water retention characteristics, leaching and erosion potential, plant nutrient storage, organic matter dynamics and carbon sequestration capability. Soil texture analysis like other laboratory conventional analyses of soil properties can be costly, labour intensive and time consuming. Spectroscopic methods are being increasingly considered as possible alternatives to substituting the conventional laboratory methods for determining soil properties. There is a correlation between the shape of visible (vis) and near infrared (NIR) spectra and soil texture. Therefore, some researchers utilised this fact to measure the clay content using NIR spectroscopy. Others experienced a decline in prediction accuracy of other soil components due to the interfering effect of texture. To exclude the effect of texture from soil spectra, researcher provided quantitative calculations of texture or spectra transformation that can alter the original shape of the spectra, leading to loss of important information of other components to be quantified. Therefore, it is essential to exclude the effect of texture without losing important information from soil spectra. This can be ensured by classification of soil spectra using proper multivariate statistical methods. When a successful classification of spectra is reached, quantitative analyses are expected to provide more accurate prediction of other soil physical and chemical components using calibration models developed separately for each class of texture.

A Comparison and Joint Use of VIS-NIR, MIR and Fluorescence Spectroscopic Methods for Differentiating Between the Manufacturing Process and Sampling Zones of Ripened Soft Cheese

Ten traditional M1 (n=5) and M2 (n=5), and five stabilised M3 (n=5) retail soft cheeses, with different manufacturing process were studied using visible (VIS) and near infrared (NIR), mid infrared (MIR) and front face fluorescence spectroscopies. VIS-NIR, MIR, tryptophan, riboflavin) and vitamin A spectra were recorded for two sampling zones (external (E) and central (C)) of the investigated cheeses. When the factorial discriminant analysis (FDA) was applied to the MIR spectral region, the classification was not satisfactory. A slightly better classification was obtained from the VIS-NIR spectra. Better classifications were obtained using vitamin A fluorescence spectra, since 91.8% and 80.6% of the calibration and validation spectra, respectively, were correctly classified. The first 5 principal components (PCs) of the principal component analysis (PCA) extracted from each data (VIS-NIR, MIR, tryptophan, riboflavin and vitamin A fluorescence spectra, and physicochemical data) were pooled into a single matrix and analysed by FDA. The classification was considerably improved, obtaining CC of 100% of the calibration and 88.9% of the validation spectra. The discrimination of the investigated cheeses according to their manufacturing processes and their sampling zones was excellent. It was concluded that concatenation of the physico-chemical and spectroscopic information is an efficient technique for the identification of soft cheese varieties.