Much faster cross‐validation in PLSR‐modelling by avoiding redundant calculations

Carbohydrate levels are important regulators of the growth and yield of tree crops. Current methods for measuring foliar carbohydrate concentrations are time consuming and laborious, but rapid imaging technologies have emerged with the potential to improve the effectiveness of tree nutrient management. Carbohydrate concentrations were predicted using hyperspectral imaging (400–1000 nm) of leaves of the evergreen tree crops, avocado, and macadamia. Models were developed using partial least squares regression (PLSR) and artificial neural network (ANN) algorithms to predict carbohydrate concentrations. PLSR models had R2 values of 0.51, 0.82, 0.86, and 0.85, and ANN models had R2 values of 0.83, 0.83, 0.78, and 0.86, in predicting starch, sucrose, glucose, and fructose concentrations, respectively, in avocado leaves. PLSR models had R2 values of 0.60, 0.64, 0.91, and 0.95, and ANN models had R2 values of 0.67, 0.82, 0.98, and 0.98, in predicting the same concentrations, respectively, in macadamia leaves. ANN only outperformed PLSR when predicting starch concentrations in avocado leaves and sucrose concentrations in macadamia leaves. Performance differences were possibly associated with nonlinear relationships between carbohydrate concentrations and reflectance values. This study demonstrates that PLSR and ANN models perform well in predicting carbohydrate concentrations in evergreen tree-crop leaves.

The objective of this study was to compare the predictions of the protein contents and hardness values by partial least squares regression (PLSR) and principal components regression (PCR) models for bulk samples of Canadian wheat, which were obtained from different locations and crop years. Wheat samples of Canada Western Red Spring (CWRS), Canada Western Hard White Spring (CWHWS), Canada Western Soft White Spring (CWSWS), and Canada Prairie Spring Red (CPSR) classes were obtained from nearby agricultural farms in the main wheat growing locations in the Provinces of Alberta, Saskatchewan, and Manitoba from 2007, 2008, and 2009 crop years. Wheat samples were conditioned to moisture levels of 13, 16, and 19 % (wet basis) and pooled together for developing the regression models. A database of the near-infrared (NIR) hyperspectral image cubes of bulk samples of wheat classes was created in the wavelength region of 960–1,700 nm with 10 nm intervals. Reference protein contents and hardness values were determined using the Dumatherm method and single kernel characterization system (SKCS), respectively. A tenfold cross-validation was used for the ten-factor partial least squares regression (PLSR) and principal components regression (PCR) models for prediction purposes. Prediction performances of regression models were assessed by calculating the estimated mean square errors of prediction (MSEP), standard error of cross-validation (SECV), and correlation coefficient (r). Using the full data set in the protein prediction study, the ten-component PLSR model gave 1.76, 1.33, and 0.68 for the estimated MSEP, SECV, and r, respectively, which were better than the results for the ten-component PCR model (2.02, 1.42, and 0.62, respectively). For the hardness prediction, the estimated MSEP, SECV, and r values were 147.7, 12.15, and 0.82, respectively, for the ten-component PLSR model using the full data set. The PLSR models prediction performances outperformed the PCR models for predicting protein contents and hardness of wheat.

This study aimed to compare stepwise multiple linear regression (SMLR), partial least squares regression (PLSR) and support vector machine regression (SVMR) for estimating soil total nitrogen (TN) contents with laboratory visible/near-infrared reflectance (Vis/NIR) of selected coarse and heterogeneous soils. Moreover, the effects of the first (1st) vs. second (2nd) derivative of spectral reflectance and the importance wavelengths were explored. The TN contents and the Vis/NIR were measured in the laboratory. Several methods were employed for Vis/NIR data pre-processing. The SMLR, PLSR and SVMR models were calibrated and validated using independent datasets. Results showed that the SVMR and the PLSR models had similar performances, and better performances than the SMLR. The spectral bands near 1450, 1850, 2250, 2330 and 2430 nm in the PLSR model were important wavelengths. In addition, the 1st derivative was more appropriate than the 2nd derivative for spectral data pre-processing. PLSR was the most suitable method for estimating TN contents in this study. SVMR may be a promising technique, and its potential needs to be further explored. Moreover, the future studies using outdoor and airborne/satellite hyperspectral data for estimating TN content are necessary for testing the findings.

Motivated by recent research on two-phase flow void fraction measurement, this paper presents a study on void fraction measurements of a lubricant air-oil two-phase flow using a near infrared (NIR) optical-fiber spectrometer associated with partial least squares (PLS) regression models. To overcome the measurement inaccuracy due to interface scattering, this study used PLS regression models to analyse the spectrum of air-oil two-phase flow. First, an NIR optical-fiber spectrometer experimental system was developed. Second, a flow regime analysis was conducted for the bubble flow transition inside the experimental system. Finally, the PLS regression model validations, the void fraction measurement accuracies, and the uncertainties of the experimental system were evaluated and discussed, and the different PLS regression models were compared in detail. The results indicate that the NIR optical-fiber spectrometer combined with the PLS regression model can achieve void fraction measurements of lubricants in air-oil two-phase flow, with a maximum squared correlation coefficient (R 2) of 0.981, a minimum Type A uncertainty of 0.039% and a minimum expanded uncertainty of 0.077%.

Monitoring of crops during their growth period is essential to increase the profitability of production and reduce costs. Unmanned aerial vehicles (UAVs) have been introduced as an appropriate tool in precision farming, due to their affordability and high-spatial resolution data. The potential of plant height and vegetation indices derived from UAV-based RGB imagery at 12-leaf growth stage (V12) is investigated for prediction of corn forage yield. Using the structure from motion techniques, the crop height model (CHM) was created by subtracting crop surface model from digital elevation model. Next, by evaluating the CHM using ground measurements of plant height, it was shown that UAV imagery provides high capability in modeling the plant height ( R 2 = 0.85). Pearson’s correlation coefficient ( R ) between variables showed that the plant height had the highest correlation with forage yield ( R = 0.84). Furthermore, among the vegetation indices, visible atmospherically resistant index with R = 0.77, normalized difference index with R = 0.75, and excess red with R = 0.74 were better than other indices. Finally, the partial least squares regression (PLSR) model was constructed using plant height and vegetation indices to predict forage yield. The beta regression coefficients of the PLSR model were used to identify the most relevant predictors, and the final PLSR model was rebuilt with relevant predictors. The PLSR analysis confirmed that the plant height was the most important predictor, although integrating the vegetation indices in the PLSR model has slightly improved the prediction model ( R 2 = 0.74, RMSE = 0.316 kg / m2, and relative RMSE = 12.39 % ), compared to the model that used only plant height ( R 2 = 0.702, RMSE = 0.339 kg / m2, and relative RMSE = 13.29 % ). Overall, these results indicate that the UAV-based images are efficient and cost-effective data to predict corn forage yield as well as plant height modeling.

<p>Soil properties could be assessed with reflectance spectroscopy (soil spectroscopy, SS) in vis-NIR region (400-2500 nm) through absorption features found in soil spectra. A high spectral resolution (up to 1 nm) drives to high dimensional and multicollinear data. This issue is usually addressed prior to modelling with feature extraction methods such as Principal Component Analysis, or embedded methods such as Partial Least Squares Regression (PLSR). Feature Selection (FS) wrapper methods are promising dimensionality reduction approaches barely used in SS. The objective of this study was two-fold: i) evaluate the performance of FS wrapper methods built from Random Forest (RF) algorithm to predict soil organic matter (SOM), clay and carbonates using laboratory spectroscopy, ii) test the performance of FS methods for dimensionality reduction in SS. The reflectance of 100 soil samples from Sierra de las Nieves National Park (Spain), was measured under laboratory conditions using an ASD FieldSpec Pro JR. A spectral preprocessing method, Continumm Removal (CR), was applied to raw spectra. The RF wrapper considered two different feature searching approaches: Sequential Forward Selection (SFS) and Sequential Flotant Forward Selection (SFFS). The performance of RF with FS (RF-FS) was compared to that of Partial Least Squares Regression (PLSR) and RF (without FS). Models were evaluated with R-squared, root mean squared error (RMSE) and ratio of prediction to deviation (RPD).</p><p>RF-FS models outperformed PLSR and RF models for the three SAP. RF-FS best models had a RPD of 2.19 for SOM, 1.64 for carbonates and 1.52 for clay, whereas PLSR models had RPD values of 1.59, 1.22 and 1.3, and RF 1.38, 1.23 and 1.23 for SOM, carbonates, and clay, respectively. Therefore, FS was useful in obtaining models with improved accuracy by reducing redundant features and avoiding multicollinearity (Hughes effect). The application of FS wrapper methods reduced the number of features in the RF-FS models to less than 1% of the starting features. Features were selected across all spectra from SOM and clay, and around 900, 1900 and 2350 nm for carbonates. This research, thus, shows an alternative to different feature extraction approaches for modelling soil properties based on FS methods and machine learning.</p>

Estimating economic benefit of sugar beet based on three-dimensional computer vision: a case study in Inner Mongolia, China

Fractional-order Savitzky–Golay filter for pre-treatment of on-line vis–NIR spectra to predict phosphorus in soil

Quantification and classification of vegetable oils in extra virgin olive oil samples using a portable near-infrared spectrometer associated with chemometrics

Prediction of leaf traits of lianas and trees via the integration of wavelet spectra in the visible-near infrared and thermal infrared domains

Knowledge of environmental factors controlling soil organic carbon (SOC) stocks can help predict spatial distribution SOC stocks. So, this study was carried out to select the best environmental factors to model and estimate the spatial distribution of SOC stocks in northwestern Iran. Soil sampling was performed at 210 points by multiple conditioned Latin Hypercube method (cLHm) and SOC stocks were measured. Also, environmental factors, including terrain attributes, moisture index, and normalized difference vegetation index (NDVI), were calculated. SOC stocks were modeled using random forest (RF) and partial least squares regression (PLSR) models. Modeling SOC stocks by RF model showed that the efficient factors for estimating the SOC stocks were slope height (slph), terrain surface texture (texture), standardized height (standh), elevation, relative slope position (rsp), and normalized height (normalh). Also, the PLSR model selected standardized height (standh), relative slope position (rsp), slope, and channel network base level (chnl base) to model SOC stocks. In both RF and PLSR methods, the standh and rsp factors were suitable parameters for estimating the SOC stocks. Predicting the spatial distribution of SOC stocks using environmental factors showed that the R2 values for RF and PLSR models were 0.81 and 0.40, respectively. The result of this study showed that in areas with complex land features, terrain attributes can be good predictors for estimating SOC stocks. These predictors allow more accurate estimates of SOC stocks and contribute considerably to the effective application of land management strategies in arid and semiarid area.

Recycled polypropylene (rPP) often contains a small amount of polyethylene (PE). Since polypropylene (PP) and PE are incompatible, the presence of PE compromises the performance of rPP materials and needs to be closely monitored. In our previous work, Raman and near-infrared (NIR) spectrometries were evaluated to monitor PE content in rPP with partial least square regression (PLSR) modeling. The NIR spectrometry exhibited a wider application range, but the accuracy of the prediction models might be further improved. In the current work, a different modeling method, principal component regression (PCR) was employed to analyze PE content in rPP with NIR spectrometry. Spectrum pretreatment methods, including multivariate scatter correction (MSC), standard normal variate transformation (SNV), smoothing, and first derivative, were investigated to improve the NIR spectrum quality. Forward and backward interval methods were used to optimize spectral range selection. The outcomes were compared with our previous PLSR modeling results. The highest accuracy in independent validation was achieved by a PCR model with an R 2 of 0.9991 and a root-mean-square error of prediction (RMSEP) of 0.1596 PE%. On the other hand, a PLSR model achieved the lowest RMSEP of 0.9712 PE% for a non-colored post-consumer rPP sample. The PCR models might be sensitive to interference and more suitable for post-industrial materials, which have a simpler chemical composition. The PLSR models might have better stability and be more suitable for complicated post-consumer samples. Both the PCR and PLSR models were successfully applied to a gray commercial rPP sample.

Angelicae Sinensis Radix is a widely used traditional Chinese medicine and spice in China. The purpose of this study was to develop a methodology for geographical classification of Angelicae Sinensis Radix and determine the contents of ferulic acid and Z-ligustilide in the samples using near-infrared spectroscopy. A qualitative model was established to identify the geographical origin of Angelicae Sinensis Radix using Fourier transform near-infrared (FT-NIR) spectroscopy. Support vector machine (SVM) algorithms were used for the establishment of a qualitative model. The optimum SVM model had a recognition rate of 100% for the calibration set and 83.72% for the prediction set. In addition, a quantitative model was established to predict the content of ferulic acid and Z-ligustilide using FT-NIR. Partial least squares regression (PLSR) algorithms were used for the establishment of a quantitative model. Synergy interval-PLS (Si-PLS) was used to screen the characteristic spectral interval to obtain the best PLSR model. The coefficient of determination for calibration (R2C) for the best PLSR models established with the optimal spectral preprocessing method and selected important spectral regions for the quantitative determination of ferulic acid and Z-ligustilide was 0.9659 and 0.9611, respectively, while the coefficient of determination for prediction (R2P) was 0.9118 and 0.9206, respectively. The values of the ratio of prediction to deviation (RPD) of the two final optimized PLSR models were greater than 2. The results suggested that NIR spectroscopy combined with SVM and PLSR algorithms could be exploited in the discrimination of Angelicae Sinensis Radix from different geographical locations for quality assurance and monitoring. This study might serve as a reference for quality evaluation of agricultural, pharmaceutical, and food products.

Increasing the accuracy of atmospheric ammonia concentrations calculated from Open-Path Fourier transform infrared spectra using partial least squares model by scanning and removing interference spectral data

A comparative quantitative structure–retention relationship (QSRR) study was carried out to predict the retention time of polycyclic aromatic hydrocarbons (PAHs) using molecular descriptors. The molecular descriptors were generated by the software Dragon and employed to build QSRR models. The effect of chromatographic parameters, such as flow rate, temperature, and gradient time, was also considered. An artificial neural network (ANN) and Partial Least Squares Regression (PLS-R) were used to investigate the correlation between the retention time, taken as the response, and the predictors. Six descriptors were selected by the genetic algorithm for the development of the ANN model: the molecular weight (MW); ring descriptor types nCIR and nR10; radial distribution functions RDF090u and RDF030m; and the 3D-MoRSE descriptor Mor07u. The most significant descriptors in the PLS-R model were MW, RDF110u, Mor20u, Mor26u, and Mor30u; edge adjacency indice SM09_AEA (dm); 3D matrix-based descriptor SpPosA_RG; and the GETAWAY descriptor H7u. The built models were used to predict the retention of three analytes not included in the calibration set. Taking into account the statistical parameter RMSE for the prediction set (0.433 and 0.077 for the PLS-R and ANN models, respectively), the study confirmed that QSRR models, associated with chromatographic parameters, are better described by nonlinear methods.