Research On Image Recognition Method Of Cashmere And Wool Fiber Based On Back Propagation

In the field of fiber testing,the identification of cashmere and wool fibers has always been a challenge.At present,my country’s fiber testing departments usually use the fiber projection method to identify cashmere and wool,but this method has problems with low efficiency,poor stability,and subjective emotions of the inspectors.Therefore,it is imminent to develop an efficient and practical method for automatic identification of cashmere wool.Based on digital image processing technology,the image preprocessing method,feature parameter extraction method and BP neural network classification and recognition model in the identification of cashmere and wool fiber are studied in the article.The main research contents of the subject are followings:1.Determination of cashmere wool fiber image preprocessing scheme.Using the self-made fiber slicing machine to make cashmere wool fiber slices,and introducing a differential interference contrast microscope into the image acquisition,the fiber images with clear textures on the outer edges and the contours were obtained;By observing the cashmere wool fiber images under the microscope,the differences between the characteristics of cashmere and wool fibers were analyzed.And by using gray-scale,Laplace enhancement,median filtering,Canny detection,binarization,morphological processing,contour detection and other image processing methods,the target Fiber area binary map was finally got.2.Extraction of cashmere wool fiber morphology and texture features.The extraction of morphological features of diameter was obtained by using the segmented scanning method,the scale height was obtained by using the scale skeleton method,and the fiber scale perimeter and the scale area were obtained by using the pixel point statistics method.The ratio of diameter to height,the density of scale,the relative circumference of scale and the relative area of scale can be indirectly obtained by using these four intuitive indexes,which has 8 morphological characteristics.The extraction of texture features was based on the gray-level co-occurrence matrix to obtain the average values of energy,contrast,entropy,and correlation from the four directions of 0°,45°,90°,and 135°.These four average values are used to describe the texture.Statistical analysis was carried out on the extracted 12 cashmere and wool fiber characteristic parameter data,and they were selected as the basis for the classification and identification of cashmere and wool fibers.3.The establishment and identification results and analysis of the classification model of cashmere wool fibers.The principal component analysis method was used to reduce the dimension of the 12 cashmere wool fiber characteristic parameters obtained,and four characteristic parameter systems that could represent the comprehensive indicators of cashmere wool fiber were obtained.The data sample of cashmere wool fiber was divided by 50% cross-validation.Then,the structure of the BP Propagation is designed,and three training functions under different hidden layer nodes are selected to train the BP Propagation.The experiment proves that the classification effect of the model is good in the text.The comprehensive recognition rate of cashmere wool obtained by this method in this experiment is 94.89%.