Research Of Visual System For Bullet Size And Defect

In the process of the production of bullets,it is inevitable to produce unqualified bullets,or the size inconformities,or the defects of the shell surface,such as the card injury and the bump,and so on.Therefore,the cartridge case should be detected in the production.At present,artificial detection is mainly relied on.Because the human eye identification is not only inefficient,it will also cause the instability of the test results because of fatigue.Therefore,an automatic detection method is needed to improve the efficiency and accuracy of the bullets detection.In this paper,a set of automatic visual inspection system for cartridge case is designed.(1)The research status of the visual measurement technology and the defect detection technology at home and abroad is analyzed.The defect detection system based on the computer vision is designed,including the measurement of bullets and the classification of the surface defects,and the experimental platform is built.It has the characteristics of large field of view and multi-position.(2)In dimension detection,according to the requirements of bullets measurement,the most suitable algorithm is selected by comparing the experimental results of image denoising,image enhancement,image segmentation and feature extraction.Through the accurate calibration of the test bench,the size results of bullets measurement are calculated,and the error analysis is carried out.The automatic bullet detection system is realized.(3)In the classification of surface defects,first of all,three kinds of defects on the surface of the shell are given: card injury,bump and contusion.Then the background noise is added as false damage,and the classification of defects is classified as 4categories.Then an improved local signal to noise ratio method is introduced to extract770 surface defects of cartridge case and represent each defect into 13 dimensional eigenvectors.By comparing 7 kinds of machine learning algorithms,namely support vector machine,normal Bias,KNN algorithm,Adaboost algorithm,decision tree,random forest and gc Forest algorithm.Through the experiment analysis,the final selection is gc Forest algorithm.In this paper,the research status of bullets detection at home and abroad is analyzed,and a bullets detection system based on computer vision is designed to finish the dimension measurement and the classification of surface defects.The accuracy of size measurement reaches the accuracy requirement of bullet case.For defect detection,13 dimension eigenvectors are used to characterize the defect,and the gc Forest algorithm is used to realize the damage classification with the correct rate of 96.1%.