Identification Of Wood And Cashmere Fiber Based On Support Vector Machine

At present,the inspection of wool cashmere in our fiber testing department mainly adopts the microprojection method,namely,the morphological characteristics of the fibers are observed by visual observation,so as to realize the detection of their classification and content.The method of testing is susceptible to the long hours of the examiners and the visual fatigue and subjective emotion,which is less stable,less accurate,and time-consuming.Therefore,it is particularly important to study an effective,practical and accurate detection method for wool and cashmere fibers.In this paper,the image of wool and cashmere fibers with enhanced shape and texture details is captured by differential interference phase contrast microscope system.The gray level co-occurrence moments were used to characterize the texture features of wool and cashmere fibers,such as scale warping,scale uniformity and surface roughness,The recognition and classification of wool cashmere fiber is realized by combining morphological and texture features and SVM classifier.according to the characteristics of wool and cashmere fiber images,the following processing is carried out: firstly,the effective separation of target fiber and image background is realized through local adaptive threshold binarization and morphological processing.Secondly,the fiber profile was used to segment the fiber,among which Harris detection and Hough transform detection were used for cross case discrimination and cross region detection.Finally,the single fiber is obtained by removing the intersecting and overlapping areas.The diameter,scale density,scale area and relative area of wool and cashmere fibers were taken as the morphological characteristic parameters.After obtaining single fiber,the characteristic parameters of wool and cashmere fiber were extracted.fiber characteristics were extracted by the following methods: first,The fiber diameter was extracted by the middle axis method,and the gray scale of rotating fiber was drawn to the horizontal and sobel treatment was performed.The purpose of this method was to enhance the edge of the scale to obtain the density of the scale,and the size and relative area of the scales were obtained by the pixel point cumulation.Secondly,local binarization patterns(LBP)was used to highlight the fiber texture and simplify the gray level information,and the fiber grayscale map is transformed into a gray level symbiotic matrix(GLCM),and the difference of the surface texture of wool cashmere fiber is characterized by the different characteristics of the generated gray level symbiotic moment,and the texture parameters are extracted by data.Combined texture features and morphological features,can get 12 used to automatically identify the characteristics of the value.Finally,support vector machine(SVM)is used as classifier to classify wool cashmere fiber.Taking the extracted wool and cashmere fiber feature data as training samples,the optimal parameters corresponding to the optimal classifier are determined by cross-validation and mesh search,and the fiber recognition and classification model is obtained by training.The results showed that the recognition rate of wool cashmere fiber was 93.5%.