Research And Improvement Of HU Moment And SIFT Algorithm On PCB Board Identification

With the fast development in computer technology,machine vision technology has been widely applied to face recognition,object detection,and other fields ranging remote sensing.And image recognition technology is one of the important technology of machine vision.First of all,this article introduced the characteristics and significance of machine vision,and also introduced the importance of PCB board in life.According to the PCB board structure characteristics,in the PCB board inspection production line,the machine vision can identify the matching way to the PCB.The structural characteristics and shape features of the board are identified and matched,and the accuracy and efficiency of matching can be improved compared to the traditional manual recognition and matching.Secondly,this paper introduces the domestic research of image recognition and matching algorithm,including image preprocessing,gray-scale transformation,image blur,and edge extraction.The Harris corner method,SIFT algorithm.PCA-SIFT algorithm,SURF algorithm are compared.Hu Invariant Moments The advantages and disadvantages of these image feature matching algorithms,according to the characteristics of the PCB board with obvious texture features and more edge information,the feature point recognition algorithm SIFT and the shape recognition algorithm Hu invariant moment are selected.Then,after introducing the SIFT algorithm and the Hu moment invariant algorithm respectively,it is analyzed that the SIFT algorithm is sensitive to noise when extracting the edge feature points,it is easy to find the wrong feature points,and the reason why the wrong feature points match occurs;The reason why Hu invariant moments are inefficient in shape recognition is analyzed for scaled simple contour images.The SIFT algorithm is improved by using RANSAC algorithm,improved image distance formula and Canny operator;the recognition rate of the scaled contour image shape is improved by correcting the central normalized operator of Hu moment invariant.Based on Visual Studio 2015,the improvement of the algorithm was verified through experiments in the OPENCV2.4.13 configuration environment.It was proved that the improved SIFT algorithm reduced the number of erroneous feature points in the extraction of edge feature points and improved the correct matching rate.The shape recognition rate of the moment is significantly improved after the image is scaled.Finally,the paper uses the improved SIFT algorithm and Hu invariant moments,and based on the same experimental platform,the image matching recognition experiment of PCB is carried out.Experiments show that the improved algorithm can effectively reduce the number of erroneous feature points in PCB image matching and improve the matching accuracy of feature points.For the PCB image after rotation,translation or scaling,the shape matching degree is improved.