Feature Extraction Of Workpiece's Defect In The Complex Background

The feature extraction of workpiece defect is an important content of the workpiece quality testing. We propose a suitable and efficient detection algorithm on the basis of machine vision technology applied to industrial production,dealing with the problems of low speed, inefficiency and high missed detectable rate of workpiece. The following aspects have been studied and discussed:Firstly, propose the design of overall system and build an experimental platform.Secondly, study of image processing algorithms of the workpiece thoroughly.Choose filter image noise with the median algorithm after analysising and comparing various image processing algorithms,having advantages of simple design and better ratio of signal to noise.Combine the improved two-dimensional OTSU and mathematical morphology to detect image edge with ability of strong noise immunity.Thirdly, identify defect by BP neural network based on genetic algorithms, having advantages of high-accuracy, fast and reliability and overcoming the traditional BP neural network's shortcomings which is vulnerable to falling into local minimum and weaken slowly.Lastly, analyse the deviation of detection system and propose an effective method for reducing deviation.