Identification Of Cashmere And Wool Based On SEM Images

Cashmere and wool all belongs to animal fiber,and they have high similarity in appearance,chemical structure and physical properties.Therefore,it has always been a great challenge to identify them.At the same time,due to the low yield and the excellent performance in wearability of cashmere,It's more expensive than wool fiber.These facts lead to a constant adulteration in the market.In order to protect the interests of consumers,it's necessary to find a quick and effective identification method.For this problem,scholars have proposed many methods for the identification of wool and cashmere,which can be divided into four main types: optical microscopy method,electron microscopy method,near-infrared spectroscopy method and DNA identification method.Because cashmere and wool fundamentally belong to different kinds of fibers and their internal genetic makeup is different,so the DNA identification method can accurately identify the fiber types.But it also has a disadvantage of being an expensive method,and in fact it is commonly used in large testing institutions.Nowadays,the common used method in the textile industry is optical microscopy method.Professional testers can observe the morphology of the fibers under the optical microscope and identify them based on their experience.Optical microscopy method is simple to operate and can take pictures quickly,but compared to electron microscopy,it can not get the true outline of the surface of fiber and have lower resolution.With the development of computer application technology,the application of feature extraction in image recognition is more and more broad.Therefore,this paper combine the electron microscope and image recognition technology to identify cashmere and wool fiber.In this paper,an identification method based on fiber skeleton was firstly studied.The fiber skeleton include the two sides of fiber plate and the contour of the scale.This approach assumes that wool and cashmere fiber have certain difference on the scale shape and the scale pattern.Then some representative characteristic parameter was acquired to distinguish between the two.Such as,the average fineness of fiber,the variance of fineness,the density of scales,the average area of scales,the variance of scale area,the perimeter of scales,the variance of scale perimeter and the circular degree of scale.The specific steps include fiber image processing,skeleton feature extraction and the classification of feature parameters.The corresponding solutions were proposed to solve the problem of edge contour loss in image processing.In this paper 1000 scanning electron microscope(SEM)fiber images were used to do the experiments,70% of them as the training sample and the remaining 30% as test samples.After testing 3different kernel functions and tuning the related parameters,it was found that the average recognition rate was stable at 87% when using the combination of c-svc and RBF Kernel.Another method was based on the characteristics of Speed-up robust features(SURF).First of all,in order to rule out the effect of noise in the image background,image pre-processing was proposed to remove the background and enhance their characteristics.After that,SURF features were obtained and then these large number of features were clustered and translated into the form of histogram vector.In the end,support vector machine(SVM)classifier was used to classify the labeled vector.In the experiment part,fiber blend ratio and the sample were changed to test the performance of algorithm.It was found that the results were relatively stable in the situation of different mixing proportion and the recognition rate can reach above 90% when the sample size was greater than 300.By comparing the two methods,it can be found that the identification method based on SURF feature has a better classification performance,and it has less demanding on the image processing which means it has a great advantage in speed.