Research And Application Of Wine Bottle Defect Detection System Based On Machine Vision

With the Made in China 2025 Initiative has been effectively carried out,automated assembly line equipment has been gradually taking place of traditional labor-intensive industries in the context of the Industry 4.0 Era with intelligent manufacturing as its core.In order to adapt to operating speed of automated production line,most enterprise used to invest a lot of human costs in the quality inspection department,during the production of bottle wine.However,with the improvement of filling technology,the way,manual visual inspection,featuring higher human costs and lower can no longer meet the actual production needs.This paper aims to study flaw detection system of wine bottle based on machine vision with flaw in wine bottle and its cap surface in the production of wine bottle.The main research contents are as follows:(1)In this study,we can determine the overall design architecture of the machine vision system according to the inspection objects and its requirements,ascertain the model selection of the image acquisition module by operation of the production line,build the hardware system construction of the flaw detection system of wine bottle and ultimately train and deploy the detection model after build software algorithm module based on the laboratory server.(2)The focus is on the wine bottle defect detection algorithm based on deep learning.First,the two commonly used target detection algorithms Faster R-CNN and YOLO v3 are applied to the self-made wine bottle defect data set.The experimental results show that the two The target detection algorithm can meet the requirements of the assembly line in terms of execution time,but in the industry that requires high accuracy,such as the detection of bottle defects,neither of the two algorithms can meet the requirements of detection accuracy.(3)In the case that the two target detection algorithms cannot meet the accuracy requirements,an improved multi-stage Cascade R-CNN algorithm is proposed.The innovative idea is to adopt the Anchor generation strategy of the clustering algorithm and use the multi-scale prediction backbone network as the backbone network.Feature extraction,using the alignment layer of interest to replace the original pooling layer of interest.The experimental results show that the improved model proposed is slightly inferior to other models in terms of detection speed,but the accuracy of model detection reaches 92.6%,which is much higher than other models.(4)In order to facilitate users to better use the wine bottle defect detection system,a multi-threaded design idea is used to design a supporting human-computer interaction interface for the system,and the production line products,model results and production data are displayed in real time through the interface.