Research On Classification Of Raw Materials Of Pulp Combined With Near-infrared Spectroscopy Technology And Pattern Recognition

As important raw materials of pulps, the variaty of vegetable fiber is closely associated with pulp's quality. Discrimination of pulp's raw materials is to confirm the categories of paper fiber and the categories of paper fiber made by different processing, which is greatly useful to the quality inspection of paper products, research of paper characteristic and reproduction of products. The traditional identification methods are time-consuming, expensive and tedious, which can't meet the need of rapid and numerous inspecting in online monitoring of production and production process. Near-infrared spectroscopy (NIRS) is a fast, low cost and non-destructive real-time analytical method, and which is used in original sample analysis and overcomes the shortcomings of traditional methods.Taking the NIRS of the hand-made paper sheet as research object, this thesis researched the identification of paper fiber in pattern recognition with near-infrared spectroscopy techniques to meet the need of rapid identification. This research analyzed the error factor, made contrast through the experiments analysis, and optimize the strategy of NIRS measuring, finished the collection of six kinds pulp as well as the sample preparation and the NIRS measurement of hand paper sheet. The spectroscopy of pulp is pretreated by the techniques of filtering and the first derivative. The clustering analysis based on principle components analysis (PCA) was employed for sample identification. The results show that the method provides the rapid and accurate classification for raw materials of pulp fibers. The analysis of the influencing factor get the results as follows:bleaching technology has significant effect on the classification of raw materials of pulp, every kind of pulp could be successfully classified without the affection of substance weight and additive chemicals.Two methods, BP-ANN and SIMCA, were employed to establish identification model based on NIRS of raw materials of paper fiber, and the key techniques were introduced during the SIMCA's application. The results show that every kind of pulp could be successfully classified by the established model. Due to the deficiencies of traditional feature extrtraction tecniques, a method by integrating wavelet transform with fractal dimension was proposed. The wavelet and fractal parameters computed by NIRS of the hand-made paper sheet were employed for pulp sample identification and its applying effect was discussed preliminarily. key words:Near infrared spectroscopy; clustering analysis; raw materials of paper fiber; Soft independent modeling of class analogy (SIMCA); pattern recognition...