Research On Surface Defect Detection Technology Of Bullet Casing Based On Machine Vision

As an important part of the bullet,the cartridge case will inevitably have unqualified products during the production process.Therefore,it is necessary to perform surface inspection on the cartridge case and timely find and remove the defective cartridge case.Traditional methods rely on human eyes to observe and make judgments.The detection results are affected by the skill level and working status of the workers,and the detection stability is poor.Machine vision-based surface defect detection technology is widely used in the surface quality inspection of industrial products due to its characteristics of non-contact detection,as well as high detection accuracy,good stability,and fast speed.Based on the shell of a certain type of long sporting cartridge,this paper detects the surface defects of the shell by the technology of image processing that is based on machine vision.The main research contents are as follows:(1)Aiming at the problem that the area array camera can't collect the image of the complete shell wall at one shot,a prism group image acquisition scheme is proposed.The key components of the image acquisition system are selected Build an image acquisition module to collect highquality images on the surface of the cartridge case.(2)Aiming at the problem that traditional filtering algorithms cannot balance the noise filtering and protect the image details,a gradient-based median filtering algorithm is proposed.Designing a filtering window according to the edge gradient information for edge pixels polluted by noise to realize targeted filtering of the edge points.Finally,the feasibility of the algorithm is verified by experiments.(3)Aiming at the problem that the surface defect target of the cartridge case is inconsistent with the background and difficult to segment,an object segmentation algorithm based on pixel search and a multi-threshold segmentation algorithm based on improved gravitational search are proposed.The pixel search algorithm is based on the idea of blocking and uses the similarity and connectivity of pixels in the same target area.Determine the area to which the current pixel belongs one by one.For a single threshold that cannot accurately segment the target area,a multi-threshold Otsu segmentation method is used,and an improved gravitational search algorithm is used to optimize the multi-threshold parameters.The validity of the segmentation algorithm is verified by experiments.(4)Due to the existence of pseudo-target and noise in the segmented image,morphological transformation is used for post-processing and a mask image is generated to extract the grayscale image of the defect area.Support Vector Machine-Recursive Feature Elimination(SVM-RFE)is used to select the extracted defect feature set to obtain the optimal feature set of the defect sample.Aiming at the small amount of sample data on shell surface defects,the gc Forest algorithm was used to classify the surface defects of the shell case,and 10-fold crossvalidation was used to evaluate the classification accuracy of the algorithm.The experiment verify that the classification ability of the gc Forest algorithm was superior to common machine learning Algorithm,classification accuracy can reach 94.5%.