Research On The Defect Detection Technology Of The Appearance Decoration Firing Of Liquor Bottle

Liquor has a long history and tradition in China.With the continuous increase of liquor production,liquor producers are increasing their requirements for product quality in order to better meet the market competition.In the process of wine bottle quality detection,the traditional manual detection mode is easy to be affected by visual fatigue and other factors,and the efficiency is low,so it can not meet the requirements of industrial production and application.Therefore,the research of machine vision-based liquor bottle detection system,instead of manual detection method,so as to effectively meet the requirements of liquor product quality detection,improve the detection accuracy,has a very wide range of application prospects and economic value.Based on machine vision technology,this paper introduces the related theory and technology of liquor bottle defect detection,designs a set of liquor bottle decoration firing defect detection system,and emphatically discusses and analyzes the Detection Algorithm of liquor bottle decoration firing defect.First of all,the background and significance of the subject are briefly discussed,the research situation and development status in the field of liquor bottle defect detection are analyzed,and the types,characteristics and causes of common defects of liquor bottle decoration firing are summarized,then,the technology and method related to the defect detection of liquor bottle decoration firing are introduced.Secondly,according to the characteristics and requirements of the object,an image acquisition system for liquor bottle are designed,this system uses a manipulator grab camera to capture the complete image of liquor bottle,it is composed of vision illumination module,manipulator grasp and control module,image acquisition module.Then,the acquired image of wine bottle decoration firing are preprocessed by debloom processing,image noise reduction and image segmentation.In order to detect the surface trademark defects of liquor bottles,the methods of image registration based on SURF operator,image difference and Blob analysis are used,the experimental results show that the average detection rate is 95.40%.However,the non-trademark defects are extracted from their 14-dimensional features and made into a data set of the defects of decoration firing in liquor bottles,which is used as the input vector of the XGBoost classifier in this paper,and the parameters of this classifier are optimized by using Bayesian optimization algorithm,finally,the cross-validation and result analysis of the classifier are carried out by using the self-built data set of liquor bottle decoration firing defect.Finally,the multi-dimensional evaluation and analysis of the defect detection model and eight other common machine learning models are carried out.,the experimental results show that the average detection rate of the model is 98.58%,and it has higher detection rate and stability than other models.