Research On Recognition Of Cotton And Ramie In Blending Textile Based On Support Vector Machine

In fiber trade, the blending ratio is a major factor affecting the intrinsic fabric quality and value, so usually need to indicate its composition and content. At present, artificial recognition is the widely used method in the quality inspection departments, the method is time-consuming, high cost and low efficiency. Therefore, fiber detection departments need an objective and accurate method for automatic identification of cotton and ramie fiber.In this paper, Fiber automatic cutting instrument is first used to produce cotton and ramie fiber section. Then using the differential interference contrast microscopy, the spiral phase contrast imaging method is introduced into the identification of cotton cotton and ramie fiber.By this method, the contrast ratio of fiber local feature is enhanced.and an images which can representation cotton and ramie fiber texture feature distinctly is acquired. Finally, the texture feature of the fiber is extracted by the method of image recognition,combined with morphological characteristics, and based on support vector machine(SVM), automatic identification of cotton and ramie fiber. Experiment shows that this algorithm can effectively identify target fiber,and the recognition accuracy can reach 94.4%.Papers specific contents are as follow:1.Cotton and ramie fiber image acquisition.Using fiber automatic slicing instrument making cotton and ramie fiber section. Then by the differential interference contrast microscopy, images of cotton and ramie fiber which is high image contrast and texture features detailed are getted. By observing the fiber image, the scheme of image preprocessing is proposed,and the recognition method based on texture features and morphological features.2.Segmentation of single fiber image.Using the knowledge of image processing and algorithm,through gray scale processing,self-adapt local thresh,binary morphology processing,improved region growing method and region of interest extraction, a single fiber is divided from the whole image successfully.Through this pretreatment algorithm,single fiber image can be obtained without noise,and the target area is complete.3.The extraction of longitudinal feature data for cotton and ramie fiber.It mainly includes two aspects: the longitudinal morphological characteristics and the texture feature.Morphological feature extraction is the fiber diameter and its irregularity,through the axes method.While texture feature extraction is to transform the image into 4 different directions of the gray level co-occurrence matrix(GLCM),and 4 characteristic values corresponding to each direction can be obtained.Finally, the mean value and the standard deviation of the same characteristic value in different directions are calculated,that a total of 8 characteristic values can be getted.Combined with the morphological characteristics, the paper finally can get 10 characteristic values for automatic recognition.4.Automatic identification of cotton fiber.The key problem of automatic recognition is the design of classifier.On the basis of actual situation, support vector machine(SVM) model is decided.Before using the classifier to automatically identify and classification,it is required to determine its kernel function and its corresponding optimal parameter pair. Through contrasting and analyzing,this paper finally adopted is radial basis function(RBF) kernel function,and the optimum parameters is(0.0625,1024).By the SVM,the accuracy of cotton and ramie fiber automatic identification is 94.4%.