Research On Non-contact Detection Technology Of Glass Bottle Defect

With the improvement of living standards,people’s requirements for product packaging have been gradually improved.As one of the most commonly used containers,the quality detection of glass bottles has become the top priority of industrial production.Traditional artificial glass bottle detection technology is no longer sufficient to meet the high requirements of mass production.The wide application of computer vision in quality detection technology is the general trend.At present,domestic glass bottle defect detection technology still needs to be broken through.It is particularly important to explore the combination of deep learning and glass bottle defect detection to achieve better detection results.Glass bottle defect detection technology faces the problems of single background,irregular object size and small objects.Therefore,this article applies the object detection technology based on deep convolutional neural networks to glass bottle defect detection.Firstly,the main architectures and common modules of the convolutional neural networks are deeply studied in this thesis.According to the characteristics of glass bottle defects,a collection system of glass bottle defects was built,a data set of glass bottle defects was created,and the glass bottle defect detection technology based on Faster R-CNN was studied.Aiming at the problem that VGG16 cannot meet better detection accuracy,ResNet101 is used as the feature extraction network,and the relevant parameters of the anchor are optimized to improve the low detection accuracy of small objects.Secondly,this thesis proposes a method that combines deformable convolution and FPN with Faster R-CNN to construct a symmetrical deformation feature pyramid network structure,which improves the effect of extracting multi-scale features and enhances the generalization ability of the model.After verification on the data set of glass bottle defect detection in PASCAL VOC2007 format,the mAP value of the algorithm in this thesis reached 86.13%.Finally,the algorithm model of this thesis is integrated into the interface designed by PyQt5,and a glass bottle defect detection software system is designed and implemented,which can complete the detection function,record the detection results,and improve the quality inspectors’ experience of using this algorithm for glass bottle defect detection.