Research On Defect Detection Algorithm For Quality Inspection Of Semi-finished Zippers

After entering the 21st century,with the rapid development of science and technology,all links of manufacturing products are becoming more intelligent and automated.In 2015,the State Council proposed "intelligence" as the main direction for the future development of the manufacturing industry.As one of China's representative manufacturing industries,the transition of the zipper industry from labor-intensive to technology-intensive is imperative.In the industrialized zipper production line,zipper quality inspection is a crucial link,which is generally divided into semi-finished zipper sampling inspection and finished zipper full inspection.For semi-finished zippers sampling inspection is still relying on manual inspection methods,there will be unstable accuracy and high cost,Low efficiency and other shortcomings.Therefore,this article applies digital image technology and machine vision technology to the quality inspection of semi-finished zippers to realize the "intelligence" of quality inspection of semi-finished zippers.The main content of this article is divided into two stages: image preprocessing of semi-finished zipper and defect detection of semifinished zipper.1)The pre-processing stage of the semi-finished zipper image provides the cut and separated tape image and sprocket image for the semi-finished zipper defect detection stage.Aiming at the image quality problems and the inclination of the zipper position in current zipper images,this article adopts the following image preprocessing process.First,the median filter and Gaussian filter are used to filter the impulse noise and Gaussian noise of the image;secondly,the histogram specification algorithm is used to enhance the contrast of the image;then,the local binary mode based on the rotation invariant uniform mode is calculated The angle at which the zipper is tilted in the image,and the position of the zipper in the image is corrected by a rotation operation;finally,the boundary between the upper and lower tape and the background in the zipper image is determined based on the Ostu threshold method,and the tape part and the sprocket part image are cut out accordingly,Provide high-quality image data sets for the defect detection stage of semi-finished zipper.2)In the semi-finished zipper defect detection stage,in order to achieve a balance of detection accuracy and efficiency,the Zipper Net algorithm proposed in this thesis uses the Res Ne Xt network for feature extraction.By introducing grouped convolution,the feature extraction network becomes structured,which reduces the manual adjustment of hyperparameters.Without significantly increasing the detection time,the accuracy of semifinished zipper defect detection is improved;in order to further improve the detection accuracy,the Zipper Net algorithm proposed in this thesis adds a feature enhancement module based on cavity convolution to the feature pyramid network,and through different expansion rates Hole convolution to obtain feature information of different receptive fields in the original image,enhance the extraction of semantic features of the fusion feature map,and further improve the accuracy of semi-finished zipper defect detection.The data set used in this experiment is the semi-finished zipper images collected on the actual production line.The data set contains 5 types of tape defects and 6 types of chain tooth defects studied.The experimental results show that in the image preprocessing stage,the tape image and the sprocket image can be cut and separated quickly and accurately.In the zipper defect detection stage,the Zipper Net algorithm proposed in this thesis can complete the detection of tape defects in 325 ms,with an average accuracy of 90% and m AP of 91%;it can complete the detection of sprocket defects in 405 ms,with an average accuracy Up to91%,m AP can reach 90.7%.The Zipper Net algorithm designed in this thesis takes much less time to detect a semi-finished zipper image than a skilled quality inspector.It has a good detection effect on various typical defects,and can achieve automated quality for the zipper production line.Inspection brings a driving effect.