Research On Defect Detection Method Of Copper Sealing Cap Based On Deep Learning

In recent years,with the increasingly fierce international competition,the development of science and technology in China has become more and more important.As a basic industrial component,copper sealing cap provides important support for industrial equipment.The copper sealing cap is a semi open workpiece with regular thread lines inside,which provides sealing and fastening when coupled with other parts.Its output is huge.Therefore,it is necessary to design a set of software system for copper sealing cap defect detection to speed up the quality inspection efficiency of enterprises and liberate workers from dry repetitive work.However,the internal thread of the copper sealing cap is small,and the defect characteristics are easy to be hidden by the thread boundary contour,which is not easy to detect;Secondly,the appearance of the copper sealing cap is a cylinder,and the image acquisition inside it has higher requirements for the camera lens,and the image distortion correction is required,so the modeling process is more complex.The copper cap defect detection algorithm based on deep learning proposed in this paper,combined with computer vision technology and the third-party open source vision library opencv,detects the defects of copper cap image samples.The accuracy of tooth damage detection is as high as 100%,and the accuracy of comprehensive defect detection is 96.3%.Firstly,the original copper sealing cap image is collected through the industrial endoscope,the median filter is used to reduce the noise,find the contour,and adjust the pose of the workpiece by using the minimum circumscribed rectangle.Manually set the mask to extract the region of interest of the internal thread,and then establish the two-dimensional and three-dimensional equations of the thread line according to the control position of the workpiece and the camera and the standard parameters of the internal thread of the copper sealing cap.By fitting the equation with the boundary of the region of interest of the internal thread,the workpiece with tooth damage can be divided.Two sets of training sample sets are made,one of which is prepared by geometric transformation and polarization transformation,and the other is the original data set.The rapid convergence of polarization data set and the accuracy of the model are verified by cross comparison test.Using the polarization data set,the loss function can converge stably in the 600 theory iterative training,while the original training set needs 900 rounds to complete the convergence under the same conditions.In terms of deep learning model,the mainstream detection model yolov3 network is adopted,and the squeeze exception module is added in its prediction head to improve the decision-making ability of the model feature channel.Through comparative experiments with the original yolov3 model,faster CNN and SSD network,the improved yolov3 network model has better detection effect for the defect detection of copper sealing cap.Compared with the original model,the accuracy of the improved yolov3 model in polarization data set is improved from 91.1% to96.3%.