Research On Tablet Defect Detection Technology Based On Machine Vision

Product quality inspection is a key link in automated production,in order to ensure product quality,domestic and foreign pharmaceutical companies have invested a lot of resources in testing technology research.At present,foreign drug defect detection technology and defect detection equipment based on machine vision are gradually mature.Due to the late start in China,the self-developed equipment is difficult to meet the needs of real-time detection of enterprises,which hinders the transformation of the production mode of domestic small and medium-sized enterprises.In order to form an independent localized defect detection system as soon as possible,it is key to increase the research of defect detection technology.Based on machine vision technology,this paper studies the defect detection technology of tablets.The specific research content is as follows:First,a drug defect detection system was designed and an image acquisition device was built.In order to reduce the influence of external factors on the image,the common light source and lighting method are selected based on the characteristics of the tablet.Then the performance of industrial cameras was compared,and Hikvision’s CCD area scan camera was selected;Then the lens is selected according to the camera parameters,and the working distance of the lens is determined by calculation;After the selection is completed,complete the construction of the image acquisition system;Finally,the types of defects and the number of defects to be collected in the experiment are analyzed,and the image acquisition is completed.Secondly,the data processing method of defective drugs was studied,and the image preprocessing and data set production were completed.Aiming at the problems existing in the image acquisition process,a set of image preprocessing scheme is proposed.Firstly,compared with the common grayscale methods,weighted average treatment was selected.Secondly,in order to reduce the interference of noise signals,compared with different filtering methods,adaptive median filtering with high flexibility is selected.In order to further improve the image quality and highlight the defect characteristics,the image is enhanced by HE and piecewise linear transformation.Finally,the dataset is enriched by mirroring,clipping,etc.,and the annotation of the dataset is completed.Finally,the defective drug detection method is studied,based on the YOLOv5 m model,an improved detection model YOLOv5＿G is proposed,and a visual interface is designed.Firstly,the network structure of YOLO model is introduced,and then in view of the high real-time requirements in automated production,the CSP structure,Focus structure and CBL structure of YOLOv5 m are pruned,and the number of residual components and convolution kernels in the model is reduced without changing the model structure,the lightweight of the model is realized,and a new lightweight detection model YOLOv5＿G is proposed,and then the parameter optimization and model comparison experiments and result analysis are carried out.Finally,a visual interface for drug defect detection is designed.Experiments show that the improved model has significant advantages in detection accuracy and detection speed.