Development Of Wooland Cashmere Similar Fiber Identification Algorithms And Application

The identification of wool and cashmere similar fibers has always been one of the important research topics in the field of textile and clothing production industry.The traditional methods mainly include physical,chemical,biological and sensory identification which is complex,low efficiency,and easy to be affected by external factors.The image method is based on the objective evaluation and analysis,and has the advantages of batch operation,fast response,and simple experiment.Therefore,this paper proposes an image method based on digital image processing and multi feature fusion image analysis technology for automatic recognition of wool and cashmere similar fibers.The main contents of this paper are as follows: fiber image acquisition and processing system construction;fiber image preprocessing scheme design;fiber texture and morphological characteristics extraction and analysis;fiber type identification.Firstly,this paper summarizes the research progress of texture and morphological characteristics analysis of wool and cashmere fiber based on image processing technology at home and abroad in recent 30 years,summarizes the main process of image recognition system applied to fiber identification,and briefly introduces the research content,innovation and research thinking of this paper.Secondly,the hardware systems such as optical microscope and digital camera used in fiber image observation and acquisition are introduced in detail,as well as the software systems used in fiber image transmission,storage,processing and analysis.A method of sample preparation of segmented fiber is proposed to ensure that only one fiber is in the field of vision observed by the microscope.Thirdly,through the research of image enhancement,denoising,segmentation,morphological processing,edge detection,connected domain analysis and other digitalimage processing algorithms,six different fiber image preprocessing schemes are designed,three of which are applied to texture feature algorithm,three of which are applied to morphological feature algorithm.After discussing the advantages and disadvantages of the six schemes in detail,the third and sixth schemes are selected in this scheme to pre-process the collected fiber images to obtain different fiber result images for fiber texture and morphological analysis.In addition,in the case of fully investigating the feature analysis algorithm and its application in the field of fiber identification,three different texture feature analysis algorithms,i.e.gray level co-occurrence matrix,fractal analysis and Markov random field,are used to extract the surface texture features and model parameters of fiber respectively.At the same time,an interactive measurement method is proposed to analyze the fiber morphological characteristics,in order to achieve the measurement of fiber diameter,scale height,diameter height ratio and other characteristics.Finally,through the overview and analysis of machine learning algorithm,a new support vector machine classifier is constructed with the help of ten folds cross validation method and Gauss kernel function to realize the recognition of wool and cashmere fiber under different texture and morphological feature analysis methods.The experimental results show that the identification accuracy of four algorithms is 90.15,82.26%,94.49%,94.78% respectively.Because there are many kinds of fibers,and the texture or shape is easy to change during the growth process,a single feature analysis algorithm is more likely to have recognition errors.Therefore,on the basis of single feature recognition and analysis,this paper proposes a fiber image analysis technology based on texture morphology multi feature fusion.The fractal analysis algorithm with high recognition accuracy in texture feature analysis is combined with the interactive measurement algorithm in morphological feature analysis.At the same time,the texture and morphological features of fibers are collected and normalized to form a new five-dimensional array,and then the same support vector machine model is used for recognition.The experimental results show that the final accuracy of this method is 96.99% compared with single texture or single morphological feature analysis.