Support vector machine regression on selected wavelength regions for quantitative analysis of caffeine in tea leaves by near infrared spectroscopy

Abstract

Caffeine is an important component that determines the quality of tea, and its rapid estimation is very much needed for the industry. In this pursuit, a near‐infrared (NIR) spectroscopy‐based technique for the estimation of caffeine is developed and presented in this paper. On the basis of responses of the different bonds present in caffeine, four specific wavelength windows—(a) 1075 to 1239.5 nm (C―H stretch second overtone); (b) 1339.25 to 1440.75 nm (C―H stretch and C―H deformation); (c) 1640.25 to 1700 nm (C―H stretch first overtone, ═CH & amp; ―CH3 asymmetric); and (d) 900 to 1700 nm (whole range of the spectrometer)—were analyzed in details for model development and to obtain the effective wavelength (EW). Five different preprocessing techniques followed by two regression techniques—(a) the partial least‐squares (PLS) and (b) the support vector regression (SVR) were implemented on raw data for analysis. Comparing all the models, the wavelength band of 1075 to 1239.5 nm and 1339.25 to 1440.75 nm were found to produce satisfactory results. The best discrimination result was obtained using the combination of standard normal variate (SNV) preprocessing with SVR at the 1075 to 1239.5 nm wavelength region. The SVR regression with 105 samples in the training set and 15 samples in the testing set resulted in the performance parameters as RMSECV = 0.134, RMSEP = 0.069, rcv2 = 0.869, rp2 = 0.65, and RPD = 5.626 at 1075 to 1239.5 nm, whereas the PLS model produced the best RMSECV = 0.287, RMSEP = 0.077, rcv2 = 0.637, rp2 = 0.675, and RPD = 5.218 at 1339.25 to 1440.75‐nm wavelength band.

A Machine Learning Framework to Predict Nutrient Content in Valencia-Orange Leaf Hyperspectral Measurements

“…For K, a three-band combination index predicted the nutrient with an R 2 equal to 0.74 [37]. In nutrients like Mg, S, P, Ca, and others, the predictions (R 2 ) variated between lower values of 0.27 up to 0.98, depending on the method applied and the plant evaluated [24,26,30,32,38,39,50].…”

“…Machine learning algorithms have the advantage of modeling data in a non-linear and a non-parametric manner. Unlike many traditional statistical methods, these algorithms are built with the advantage of dealing with noisy, complex, and heterogeneous data [16,23,[50][51][52]. These characteristics proved to be an advantage for this study, as the data used had higher variance, was not-normal (Table 3), and, while statistically significant, low-correlated in a pairwise manner (Figure 4).…”

“…They are also capable to perform numerous combinations and calculations in a matter of seconds, achieving relative success in remote sensing applications regarding plant analysis [48,49]. Concerning hyperspectral measurements, among the applications evaluated, these algorithms were able to return state-of-the-art performances for many situations [5,16,23,[50][51][52]. Even though, to date, no study evaluated the performance of machine learning algorithms in inferring both plant macro-and micro-nutritional content with only leaf hyperspectral measurements.…”

Prediction of Soil-Available Potassium Content with Visible Near-Infrared Ray Spectroscopy of Different Pretreatment Transformations by the Boosting Algorithms

“…Therefore, the feature of the VIS-NIR follows the normal distribution. The methods of PLS [9,10,29,30] and SVR [17,18] are widely used for the calibration of VIS-NIR spectra, but boosting regression algorithms are almost never used. GBRT and AdaBoost, which are boosting algorithms, can be effectively calibrated to predict the soil-available potassium content.…”

“…In chemometrics, partial least square (PLS) regression and support vector regression (SVR) are commonly used to build calibration models [17][18][19]. PLS analysis was utilized as a method to extract the latent variables (LVs) of the spectrum.…”