Research Of PCA-based Paper Double-side Defect Recognition System

Paper production is an indispensable part of our national economy.With the the level of industrial automation improving rapidly,the width of the produced paper web is continuously widening,and the speed of the paper machine is gradually increasing.In the process of production,due to the poor quality of raw materials or changes in process parameters,the probability of various defects on the positive and negative surface of the web increases.It seriously affects the appearance quality of the paper.At present,the paper defect detection system based on machine vision has become the general trend and achieved preliminary results,but there are still several bottlenecks.Among them,the most prominent aspect of the detection algorithm is the problem of determining the characteristic amount of paper defects.In order to improve the identification accuracy of approximate defects,the intrinsic law of paper surface defects is often approached by increasing the dimension of feature quantities,which leads to the coupling between feature quantities and the expansion of computational quantities,but makes the system unable to achieve the ideal recognition effect,and prolongs The system's running time is made difficult to meet the real-time and identification accuracy of the system.In view of the above problems,based on analyzing the limitations of the existing paper defect recognition system,this paper proposes a PCA-based paper double-sided recognition algorithm,which uses the PCA algorithm to identify the type of double-sided defects and the paper on which they belong.The pre-extracted high-dimensional feature quantity matrix is reduced in dimension and decoupling while retaining the main feature information,so as to shorten the system operation time and improve the identification accuracy.This method is of practical significance to the improvement of papennaking production efficiency and product quality.The main work of this article is as follows:(1)Design and construction of system hardware structureThe hardware structure of the paper double-sided identification system is mainly divided into four parts:a light source system,an image acquisition module,a motion control module,and a computer responsible for data processing.Based on the system's function and performance requirements,the four types of hardware equipment are analyzed for type and model selection,and formed into a hardware system that meets the design requirements.(2)Research on PCA-based paper double-sided defect recognition algorithmThe algorithm research in this paper mainly includes three steps:image preprocessing(image denoising,double-sided defect segmentation,etc.),feature extraction and PCA feature extraction,and recognition and classification.First,use the Otsu method image segmentation algorithm to separate the defect areas existing in the front and back paper images;then perform preliminary extraction of the gray and morphological features of the defect area;then use the PCA feature extraction algorithm to initially extract the high-dimensional features The vector matrix performs dimensionality reduction and decoupling operation;finally,the new features extracted are input into the SVM classifier,and the SVM completes the judgment of the position and defect type of the double-sided defects on the paper.(3)Design and development of system software systemsThe software system of this paper uses C++language and MFC framework to develop under the environment of VS2013.By designing and developing the system's image acquisition module,algorithm processing module,interface display and data management module,the image acquisition of double-sided defects on paper is realized.,Detection and identification,data management and information storage and display functions.Finally,the system's operation effect was tested and analyzed.After testing,the system can accurately identify the surface and type of paper disease,and the recognition rate is 98.49%;compared with the traditional recognition method,the system running time is shortened by 53.3%.The research goals were fully achieved.