Study On Key Technology And System Realization For Detection Of Glass Defects Online

With the continuous improvement of glass production technology,the detection technology online and real-time of glass defects based on machine vision has gradually replaced the manual detection.This new detection technology not only saves human resources,but also has the characteristics like high accuracy,fast speed,more real-time.And it is able to communicate with other production processes in real time,achieving automated production of glass production line.At present,the research about the detection technology online of glass defects is lag in our country.Both the research of detection algorithm and system application are at a low level.Most automated inspection systems rely on imports.For these problems,based on the discussing about the characteristics of glass defects and the existing testing theory,the overall scheme of the detection online of glass defects is designed in this paper;The algorithm of glass defect detection is studied deeply,and a variety of glass image preprocessing techniques are compared and analyzed;Based on the simulation analysis of the feature extraction and selection of glass defective image,a U-ReliefF feature selection algorithm model is proposed;The well results were obtained by simulating the glass defect samples collected in the field.The main contents are represented as follows:Firstly,according to the requirement of glass production engineering,a real-time detecting system online is constructed based on the analysis of the characteristics of glass image data.The system consists of image acquisition device,light source system,image processing system,real-time display interface and software.Secondly,the existing image preprocessing technology are analyzed and compared.Based on the research of image filtering,threshold segmentation and edge detection algorithm,a quadratic median filter preprocessing model is proposed aiming at the problems of glare and hole in glass defective image.The actual glass defect samples are simulated by using the Matlab software,comparing the results.Thirdly,based on the study of the characteristics of glass defect image,theinvariant moment characteristics and geometric features of the image are discussed and extracted;In order to obtain more accurate and strong correlation features,this paper presents a U-Relief F feature selection algorithm based on the study of the Relief algorithm in the training session of defective image,and gives the algorithm flow chart.Finally,the U-Relief F algorithm is used to select the features of of UCI database and the real glass defect samples,and also compared with the Relief F algorithm.The experimental results show that the accuracy of classification and recognition is improved by 13.73% through the feature selection process,and all the technical requirements of the production line are satisfied.