Design Of Machine Vision Algorithm For Magnetic Powder Detection And System Development

With the rapid development of high-speed railway and the increasing number of running trains in China,the wheel,an important part of the train which plays a crucial role in the safe driving.At present,traditional fluorescent magnetic powder detection method is still utilized to guarantee the quality in the manufacture of the wheels.However,this kind of detection method has the problems of low accuracy,subject to human error,poor working environment and the expense of manpower.The developments of machine vision measurement technology,image processing technology and artificial intelligence technology provide efficient methods for solving the above problems.Implementing these new technologies to construct a machine vision measurement system is of great substantial significance and challenge.The main contents are as follows:(1)Based on the analysis and comparison of the existing image pre-processing and image segmentation algorithms,the guided filter and maximum entropy segmentation has been selected to identify whether there are defects or not.Where the segmented image still has noise,an improved guided filter algorithm was proposed,taking the median filter image as the guided graph and the local entropy value as the weighted parameter.The experimental results showed that the improved algorithm is not only superior to the original algorithm in the structure similarity index and peak signal to noise ratio values,but also the final segmented image can meet the requirement of subsequent applications.(2)To identify the defect types,the traditional method is used to extract gray feature,geometric feature and texture feature of the defect area.The most representative features were imported into softmax classifier and the average recognition rate is 80%.In view of the traditional method which has the problem of long training time and low classification recognition rate,a new method using the hidden layer of sparse coding to extract the feature from image was given.Finally,the average recognition rate is 91.4%and it require less training time,which also confirms the effectiveness of the new method.(3)The interface of the machine vision measurement system was adjustable to meet the actual demands.The function of data acquisition,data processing and data storage have been achieved by using Qt software.The software system demonstrated the reliability,stability and accuracy by detecting the actual wheel sample and simulate defect samples,laying a solid foundation for further research and practical application.