Research Identification Principle Based On Feature And Quick Identification Algorithm For Cashmere

In textile industry, the identification for cashmere and fine wool has been a challenge. For similar pattern and performance of physics and chemistry, it is very difficult to classify the two kind of fiber with higher accuracy. So far, though many methods are studied by researcher, the most used method is still microscopy. Following the development of computer technology, it has become a trend to combine computer technology with microscopy to identify cashmere and fine wool. Although these new technology advance quickly, there are still many problem needed to be solved.In this paper, the identification principle and quick identification algorithm for cashmere are studied. Following are the main work and contribution.(1) The surface pattern is observed under microscope with higher magnification. The ten parameter indexes describing the scale pattern of fiber are fixed according to the difference existed in the surface of fiber. A new concept Gene code of scale pattern was developed and what is the gene code are discussed. A method how to determine optimal gene group is provided. The concept of independent index is established and the relation between the identification accuracy rate and independent index of optimal gene group. The significance is discussed about the concept of theoretical accuracy rate and error. The method determining the theoretical error is provided and the limited value of accuracy is also determined.(2) The statistic characters of all scale parameters are studied. All the nine parameters except scale rectangle factor are following logarithm normal distribution. The theoretical accuracy is discussed based on each parameter and results show that the total errors for relative scale height and relative scale area are less than 25%. The sequence for parameters is determined according the total errors and the feasible combinations for parameters are determined. The accuracy rate is discussed based different gene group and several optimal gene groups with two, three, four and five parameters are determined. The highest accuracy rate is 98.5% for cashmere.(3) The testing errors provided by lab are discussed and the concept of feature picture is developed. The criterions of three type of feature picture are elected based on the scale pattern that we collect. These feature pictures are supplement for national criterion and can be a reference for fiber identification. The feasibility using integration value to identify fiber is also studied. The whole sense of fiber is described by the whole geometry characteristic, which can be divided into three parts including the characteristic in the cross section, in the longitudinal direction and in the region, respective. For the each geometry characteristic, its value is the average of its measuring parameter. This is a new method for fiber identification.(4) The nonlinear models are also studied. The highest accuracy rate is 99.8% for cashmere and 98.785% for wool when a Bayes model with ten parameters is used. So the theoretical error are determined that Cet is 0.2% and Wet 1.125%. The highest accuracy rate is 96.6% for cashmere and 96.1% for wool when a BP network model with ten parameters is used. So the theoretical accuracy rate are determined that Ct is 96.6% and Wt is 96.1%. The highest identification accuracy rate of 81.6% for cashmere is get when a fuzzy cluster model driven by data is used. A N-FCLASS model combining neural network and fuzzy system is also used to classify cashmere and fine wool fiber. The highest identification accuracy of 99.9% is obtained, which means the theoretical error Cet will approach 0 on the condition that more data are added in model or a batter model is developed.In this paper, the conclusions will be a reference to the related apparatus developed. Addition, the conclusion of this paper show that at present condition, if the identification model is proper, the identification accuracy for cashmere can achieve the limited value 1—Cet.