Near Infrared Spectroscopy In The Tobacco Chemical Indicators Of Quantitative Analysis

Near-infrared (NIR) Spectroscopy, a new high-efficiency analytical technique, has received considerable attention due to the speed of analysis, minimum sample preparation, and low cost. Recently, the technique has been utilized to foodstuff, soybeans, juices as well as wine for the rapid determination of their major components. Here, the feasibility of using near-infrared spectroscopy to determine total sugar, reducing sugar, protein, moisture, potassium, chloride, total petroleum ether extracts, total nitrogen, total volatile bases, total volatile acids, alkaloid, sulfate radical, pH value, neutral extracts with petroleum ether, ash, and water-soluble ash in tobacco was examined.125 flue-cured tobacco samples, collected respectively from Shandong, Henan, and Heilongjiang of China and picked in 2004 and 2005, were ground and their spectra were recorded between 1100 and 2500 nm with 8 cm^-1 of resolution, moreover, the average of spectral data of 50 times scan was obtained. The original spectrums were pretreated with db5 wavelet for denoising, Standard Normal Variate, and first derivative, and then 6 outlying samples and 17 redundant samples were removed using PCA-M distance method. The quantitative analysis was performed based on 102 samples. The following parameters, including total sugar, reducing sugar, protein, water ratio, potassium, chloride, total petroleum ether extracts, total nitrogen, total volatile bases, total volatile acids, alkaloid, sulfate radical, pH value, neutral extracts with petroleum ether, ash, and water-soluble ash were determined according to traditional method.20 samples were picked out according to the distribution of their chemistry values and used as a calibration set (n = 20), and the remaining samples as a validation set (n = 82). Partial least-squares regression (PLSR) and BP neural network were used to create calibration models by relating chemical reference values to spectral data. Among them, PLS models revealed good prediction ability to total sugar, alkaloid, ash, and moisture in tobacco, with the values of root mean square error of prediction (RMSEP) which was 1.1965, 0.1783, 0.4881, and 0.1791, respectively. whereas, BP neural network models could predicted reducing sugar, total nitrogen, protein, chloride, potassium, sulfate radical, total petroleum ether extracts, pH value, total volatile acids in tobacco, with the values of root mean square error of prediction (RMSEP) which was 1.8605, 0.1121, 0.403, 0.0465, 0.2365, 0.1417, 0.4162, 0.1024, and 0.0215, respectively. The results indicated that NIR spectroscopy can be used for rapid determination of total sugar, reducing sugar, alkaloid, total nitrogen, protein, total petroleum ether extracts, pH value, ash, and moisture in tobacco, due to the forecasting error of these indexes was less than 10 % for 86 % of samples, and can so offer reference data, for the prediction of sulfate radical, chloride, potassium and total volatile acids, with errors less than 20% for 80% of samples. Nevertheless, the prediction ability for neutral extracts with petroleum ether, total volatile bases, water-soluble ash bases was unreliable, partly because the routine method of determination for these parameters should be improved.