Design And Realization Of Ampule Bottle Surface Defect Inspection System Based On Deep Learning

Different types of defects such as stains and gas lines can be generated on the surface of medicinal ampoule glass bottle during the production process.Identifying and classifying the defects on the bottle surface will help improve the quality of the ampoule bottle and improve the production technology.Facing the monitoring need of ampoule bottle production,this thesis summarizes and analyzes the state-of-the-art about glass bottle defect inspection system and surface defect inspection based on deep learning,explores the algorithm of surface defect inspection for ampoule bottle based on object detection,and designs and realizes of the ampule bottle surface defect inspection system based on cloud platform.1.Surface defect inspection of ampoule bottle combined with YOLOv4 and LBP coding(LBP-YOLOv4).First,the edge and texture features of surface defect image for ampoule bottle are extracted.The LBP coding image of gradient amplitude image and gradient amplitude image is combined with the original gray image to obtain a three-channel image with rich edge and texture information.By means of three effective feature layers are obtained based on the model of YOLOv4.Based on K-means clustering,multi-scale prior frame can be estimated,and used to improve the prediction accuracy of the bounding box.In the end,the recognition of the six types of ampoule bottle surface defects and the location of the defects are realized based on the bounding box regression.Nine kinds of comparison experiments about the typical function modules in the proposed algorithm(LBP-YOLOv4)show that LBP-YOLOv4 algorithm has better inspection effect for small objects.For those defect object difficult to distinguish from background area,LBP-YOLOv4 algorithm can achieve more accurate object location.Compared with other three typical detection algorithms,the LBP-YOLOv4 algorithm has more obvious advantages for smaller defect object inspection.2.Design and realization of the ampule bottle surface defect inspection system based on cloud platform.In this thesis,the ampule bottle surface defect inspection system based on cloud platform is designed and realized,and the proposed LBP-YOLOv4 algorithm is also applied in the system.The aforementioned system contains surface image collection of ampoule bottles,ampoule bottle surface defect inspection based on terminal inspection machines,model optimization based on cloud platform,storage and viewing of ampoule bottle surface image and defect inspection results.Some modules of the system have been functionally tested,and good results have been achieved.