Cover Defects Detection Based On Machine Vision

Machine vision technology is a new subject,mainly used to identify objects through the processing of images directly or indirectly like our eyes,which often used to actual measuring or detecting.It improves the working efficiency and degree of automation in industrial production.With time goes by,the application of highly developed technology spreads fast in almost every range both at home and abroad in recent years.We have explored and researched the solution on online detecting for defects of the metal lids used in industrial production,referring to a great deal of professional literature,to improve the shortcomings,such as high cost,low precision,low efficiency coursed by the old detecting method rely on hand-actuated work.The main research work in this project listed as,(1)A defect detection algorithm based on the bottom of the tank cover is proposed.Firstly,we get the contour image by filtering and edge contour extraction,then find the center of the contour with different algorithms,and divided the image into three main areas: outer circle of regional,glue injection area,inner circle area etc.Then we integrated a series of image processing algorithms to detect these three regions.(2)We raise an algorithm for roundness error measurement based on Genetic Algorithm and nonlinear programming,and study circle detection algorithm.Traditional way using Hough transform has a limited speed,and using least squares cannot discard interference points which may cause errors.Genetic Algorithm has strong ability to do universal searching,while weak in local searching,which means we often get suboptimal solutions.But nonlinear programming is strong in local searching.Combining Genetic Algorithm and nonlinear programming,we raise an algorithm for roundness error measurement,which has a high precision and fast speed of convergence,working well measuring roundness errors.(3)A method of defect detection based on extreme learning machine is proposed.Firstly,we can extract 13 morphological characteristics,10 frequency domain texture features and 32 spatial texture features with the knowledge of Image Processing.Secondly,we establish a defect classification model using the ELM.Lastly,we make the comparisons with two classification results using BP neural network and LVQ neural network separately.Experiments show that this algorithm can distinguish the defective lid with a higher accuracy rate up to 97%,which enables the production line lids effective classification,and thatt the defect detection rate is up to 10 per second,providing a good support for lids' defect detection.(4)We develop flaw detection software.The software is developed in C++ based on demand analysis.With graphic user interface provided to simplify users' operation and maintenances,the software can obtain images.And with image algorithms integrated,the software implements the function to process images of can tops and to output the images.This system has been tested by a large number of samples,it can accurately detect the image of the metal tank cover at high speed,and detect the defects such as cover deformation,outer ring gap,glue too much or too little,cover surface pollution,scratch,funnel and other defects.high speed,high accuracy,no pollution,has a strong practical significance.The system has the advantages of high speed,high accuracy,no pollution and so on,has a strong practical significance.