Wavelength Screening And Application Of Near-infrared Spectroscopy Based On Absorbance Optimization

With the advantage of simplicity, rapidness, non-destructive and real-time analysis,modern infrared spectroscopic technique has been widely used in many fields, such asagriculture, food, environment, and bio-medicine. As one of the key technologies,spectrophotometric wavelength screening method with high SNR is of significance forestablishing a high-precision analysis model, reducing the complexity of the model anddesigning dedicated spectrometer.In this paper, we overcame the problem that spectral noise raises while spectralabsorbance rate is too high and the spectral information misses while spectral absorbancerate is too low. A spectrophotometric wavelength screening method based on absorbanceoptimization was proposed which known as absorbance optimization partial least squarespreferential (AO-PLS). The absorbance from the whole spectrum was sorted in ascend andsubdivided into several parts with certain step. Two nodes were randomly selected from thewhole spectrum, and the values of them were used as the minimum and maximumabsorbance rate, respectively. The difference between minimum and maximum absorbancerate was absorbance rate range. With the relationship between the wavelength andabsorbance rate, the wavelength combination of corresponding absorbance rate wasdetermined. By using the mode of moving window, the PLS model of calibration andprediction process was established in each wavelength combination from the correspondingabsorbance rate. According to the prediction effect, the optimal absorbance rate can beselected, and then the optimal wavelength combination can be obtained, which achieveeddimensionality reduction and information extraction.Serum cholesterol is one of the most important clinical biochemistry test indicators forHyperlipidemia, existing measurement methods usually need chemical reagents and Highprofessional skills. Therefore, a rapid, reagent-free analysis of Serum cholesterol using nearinfrared (NIR) spectroscopy has significance application value for large population bloodfat screening. In this paper, the NIR spectroscopic analysis of serum cholesterol was used asan example to verify the proposed AO-PLS method and to compare with the direct PLS method, moving window PLS (MWPLS) method.PLS: Comparing the prediction effects of the whole spectrum, short-NIR region andlong-NIR region, we have found out that the prediction effect of long-NIR region was thebest of all. Its waveband was1100-2498nm. The validation root mean square error,validation correlation coefficients and relative validation root mean square error ofpredition were0.835mmol L-1,0.667,16.01%, respectively.MWPLS：The optimal wavebands were1556nm-1852nm. The optimal PLS factorwas14. The validation root mean square error, validation correlation coefficients andrelative validation root mean square error of predition were0.189mmol L-1,0.987,3.81%,respectively.AO-PLS: The optimal wavebands were1374nm-1392nm and1544nm–1852nm.The validation root mean square error, validation correlation coefficients and relativevalidation root mean square error of predition were0.181mmol L-1,0.988,3.61%,respectively.Experimental results showed that: The result based on the optimal wavebands screenedby AO-PLS is better than that of the whole spectrum and shortwave region, and the numberof wavelengths has been greatly reduced; the prediction results of AO-PLS and MWPLSwas close, and the former was slightly better. However, a number of high signal-bandscould be selected based on absorbance rate while using AO-PLS method, which can beused flexible. The model will greatly reduce the complexity of the spectroscopic system,which provides a valuable reference for the design of professional spectrometer.