Research On Surface Defects Detection And Recognition Method Of Automobile Electroplated Parts

In the production process,the surface defects of automotive electroplated parts will affect the surface beauty,corrosion resistance and user experience.Therefore,it is greatly significant to carry out the detection research on the surface defects of automobile electroplated parts,find and identify the types of defects in time,and improve the process.The surface quality inspection of automobile electroplated parts is often tested by artificial vision,this method relies on the experience and subjectivity of the tester,which is inefficient and difficult to meet the production needs of modern enterprises.As a non-contact detection method,machine vision has the characteristics of high efficiency and high accuracy.Therefore,this thesis has carried out the research on the detection and recognition method of surface defects of electroplated parts based on machine vision to provide reference for surface defects detection of electroplated parts.The thesis obtains the surface defect images of the electroplated parts of the automobile by the image acquisition system,and reduces the noise of the particle noise in the collected images.Due to the irregular geometric structure of automobile electroplated parts,there is a problem that reflected light-spot in the collected defect images affect the detection rate of defects.A method for removing reflected light-spot is proposed.The area distribution is determined according to the two conditions of separation of the defect and the light-spot or partial coverage of the defect image.Repair the light-spot interference area,retain the flawed area,and then achieve the purpose of eliminating the light spot interference.Then,the image defects information are enhanced with piecewise linear function.In view of the defect detection caused by the non-maximum suppression algorithm,the generalized overlap degree replaces the traditional overlap degree calculation method,and combined with the Gaussian function to attenuate candidate frames with higher confidence scores to improve the accuracy of defective target recognition,And realize the effective detection and recognition of three kinds of defects,including scratches,pits and patches on the surface of automobile electroplating parts.The main research contents and conclusions of this thesis are as follows:(1)On the basis of analyzing the types and characteristics of surface defects of automobile electroplated parts,the research status of surface defect detection methods based on machine vision and the research status of surface defect detection based on convolutional neural networks are introduced.(2)According to the technical index requirements for the surface defects detection of automobile electroplated parts,the overall scheme of the inspection system was designed,the image acquisition device in the inspection system was selected,and the image acquisition system was built to obtain the original image of the performance defects of the automobile electroplated parts.(3)The filtering effect of the three filtering methods on the original image is compared and analyzed,and the bilateral filtering method is determined as the method of removing image particle noise.The optical model of machine vision inspection was established to analyze the cause of reflected light-spot interference,and the thesis proposes a reflected light-spot removal method.This method extracts the light component of the image by convolution and weighting the gray value of the image with Gaussian functions of different scales,obtains the distribution law of the light gray,establishes the clustering loss function,and iteratively calculates the membership degree and cluster center,and make the membership degree and cluster center meet the convergence condition,through the characteristic law of the gray value change of the image and the gray value change gradient of the adjacent background area,determine the light-spot transition area and the defects area,repair the light-spot area according to the determined area distribution,retain the detects area,and then achieve the elimination the light-spot interference,finally,the defects information are enhanced by piecewise linear function.(4)The Faster R-CNN network model is used to detect and identify the surface defects of the electroplated parts of the automobile,and the defects information are extracted through the VGG16 network.The Region Proposal Network is used to generate the proposal area from the feature map,and then the proposal area and the feature map are transported to the Region Of Interest pooling layer,and finally the defects are detected and recognized by the fully connected layer and the classifier,and remove the redundant detection box through the non-maximum suppression algorithm.Considering that the traditional non-maximum suppression algorithm is sensitive to the defects position and will cause the missed detection of the defects,an improved nonmaximum suppression algorithm is proposed,which replaces the traditional overlap calculation method with a generalized overlap degree,and the candidate boxes with higher confidence score are attenuated by combining with Gaussian function,to achieve the purpose of improving the detection accuracy.(5)The VOC2007 data set was established,and the training samples were marked.The training set was trained according to the specific parameter settings,and the test set was used to detect and analyze the improved non-maximum suppression algorithm.Experimental results show that the improved algorithm can effectively detect and identify scratches,pits and patches on the surface of automotive electroplated parts,and mark the exact location of the defects.The detection accuracy of scratches,pits and patches reached 92.5%,90.6% and 92.1%,respectively.Compared with the original Faster R-CNN algorithm,the average accuracy of defects recognition improved by 2.7%.