The Research On Near-infrared Test Of Internal Quality And Correlation With Appearance Of Tobacco

There are two main methods to evaluate the quality of tobacco leaves. One of the methods is based on the outside quality of tobacco leaves, which were classified through subjective experience according to appearances of them in traditional methods. The other is based on the internal quality of tobacco leaves, which were classified by chemical method and sampling analysis. When we come to the first method we need a lot of people and finance. The unity of judging standard can not be guaranteed some times. The advantage of the second method is that we could get data precisely, but on the other hand, this method is not suitable for general survey but only for sampling analysis. We need grind the tobacco leaves when we used the chemical method.This topic is going to research based tobacco leaf appearance quality factor database and classification using image processing on the quality and appearance quality of tobacco leaf. The content of this topic have three parts in the main. (1) Obtain near-infrared spectroscopy of tobacco leaf samples using Fourier near-infrared diffuse reflectance spectrometer. Then analyze the spectrum after de-noising with software OPUS and calculate the maximum value, minimum value and mean value of nicotine, reducing sugar and protein.(2) According to the processing above, values were divided into three classes of high, normal and low, and also into five classes of highest, higher, normal, lower and lowest. Analyze samples data with factor analysis. Analyze correlation between internal qualities: nicotine, reducing sugar and protein and appearance characteristics of tobacco leaf. By use of probabilistic neural networks, it was established the mathematical model of correlation between the three kinds of internal quality factors and appearance characteristics. Evaluation test effect with different value of SP. (3) predict contents of three internal factors according to appearance qualities of tobacco leaf use probabilistic neural networks and compare the results with different values of SP. Based on test error curves, it could conclude that regardless of the speed in training and the size of test error, the performance of probabilistic neural network is superior than the traditional BP neural network.The research results of topic can improve accuracy and reliability of classification through combined with method of classification based on appearance quality of tobacco leaf.