Research On Monitoring Method Of Image Data Based On RTC

With the wide applications of machine vision technology in industrial manufacturing,the process monitoring of image data has become an important research direction.In this thesis,images are taken as the research object.And the study focuses on the monitoring high-dimensional image data when the shape and location of the image defects are unknown.Firstly,considering the unknown size,shape and location of an image defect,image difference and image enhancement are used to highlight an image defect.And image data are constructed based on the pixel mean of image regions.A monitoring method based on the idea of Real-time Contrast is proposed for high-dimensional image data and transforms the monitoring process into a classification problem.And a kind of under-sampling method is used to solve the problem of sample imbalance.Performance of the proposed method is compared with that of an altermative method proposed by Megaled et al.under both single and multiple defects scenarios via simulations.The results show that the proposed method can effectively identify the occurrence of shifts in different situations.Secondly,on the basis of Real-time Contrast method,in order to improve the delay problem caused by real-time moving windows,a weighted SMOTE method is proposed to generate new real-time image samples,increase the proportion of potential defective images in the real-time image set,and improve the sensitivity of classification accuracy at the sample level.At the same time,the statistics are further improved,and a generalized likelihood ratio control chart based on Real-time Contrast is proposed to monitor image data and improve monitoring sensitivity at statistical level.The simulation results show that comparing with the monitoring method based on Real-time Contrast,the proposed method can detect shifts relatively quickly.Finally,the method of image data monitoring based on Real-time Contrast is applied to a real case,and steps of applying the proposed control chart are explained in detail.The results show that the proposed method can identify defects quickly and determine the location of the defect by comparing regional statistics with that of in-control images,which can provide help for practitioners to recovery the process as soon as possible.