Research On Defect Detection Technology Of Glass Wine Bottle Packaging

With the development and progress of society,the level of science and technology has become more and more developed,especially the rise and rapid development of automation technology,my country’s industrial development is facing a transformation from traditional manufacturing to intelligent manufacturing.At present,in the production process of wine bottles in my country,there are generally 3 to 5 links in a production line that produce different types of defects.At present,it mainly relies on manual inspection,which has problems such as high labor cost and low detection accuracy.Using machines to replace manual quality inspection can improve detection accuracy and detection efficiency,and create economic benefits.In recent years,as the development of computer technology has reached a new level,computer vision and deep algorithm theories have been continuously improved.In particular,detection methods based on deep convolutional neural networks are suitable for automatic feature extraction and high detection accuracy.More and more complex target detection.This paper studies bottle defects and designs a set of detection methods for bottle defects.The main research contents of the thesis are as follows:First of all,this article summarizes the current research status of bottle defect detection at home and abroad,and analyzes the advantages and disadvantages of existing detection methods.And it focuses on the basic principle and structure of the two-stage target detection algorithm Faster R-CNN based on convolutional neural network,and the effectiveness of the algorithm is verified through experiments.Secondly,in response to the problem of low detection accuracy encountered in experimental detection,combined with the characteristics of the wine bottle data set,the method of data enhancement is adopted,so that the model can be trained with a small amount of information,and the defect capture ability of the model is improved,thereby improving the model The generalization ability.Subsequently,in view of the long detection time and memory consumption caused by too many ResNet network layers,this paper optimizes the network model,compares and analyzes the network models of different depths,and selects the network model that is most suitable for the data set of this paper..Finally,this paper improves the algorithm for the characteristics of unbalanced target categories,diverse scales,and different shapes in the wine bottle data set,as well as the error detection and missing detection problems in the experiment.Because Faster R-CNN is limited by its inherent structure,it can only use regular grid points to sample the fixed positions of the feature map,which results in the hindrance of geometric transformation operations during model training,and the detection effect of multi-scale and diverse defects is poor.The deformable convolutional network can avoid these problems because of its irregular sampling method and larger receptive field.At the same time,the feature pyramid network can integrate high-level feature information due to its unique construction method,thereby improving the detection accuracy.Through experimental verification,this method can improve the detection accuracy.Compared with the existing algorithms,the algorithm in this paper has higher detection accuracy and has obvious advantages.