Research On Defect Detection Technology Of Aluminum-plastic Blister Medicine

The research object of this paper is the Aluminum Plastic Blister.In the production process of medicine,there are some appearance defects.If some defects like piercing,hairy,stains,pitting or chipped edge on the medicine surface,not only influence business value,but also cause serious consequences to human physical and mental health.At present,manufactures mainly adopt the traditional manual visual inspection method to detect the surface defects of the products,which has low efficiency,high cost and high false inspection rate.To solve these problems,this paper proposes a series of algorithms to achieve the detection positioning and classification of five defects based on machine vision.Firstly,the paper introduces the hardware part of the defect dectection system to collect images.The paper use a method of combination of coaxial light and bowl light to finish the image acquisition.Then,this paper proposed a medicine plate region and medicine granules region extraction algorithm.It has been proved by experiment,the algorithm extracted the medicine plate images and medicine granules images successfully.Secondly,in the image preprocessing,through cutting the medicine plate images to reduce the impact of uneven grayscale on defect detection.Then this paper proposed a Retinex image enhancement algorithm based on pyramid reconstruction.Compared with the traditional method,this algorithm got better experimental results.Then,based on the cascade detection method,a sub-image gray-scale difference algorithm is proposed to locate defects roughly.The algorithm greatly reduces the amount of calculation and lays the foundation for precise defect positioning in the later stage.To avoid the background texture interference when detecting defects,the paper proposed an improved texture suppression algorithm based on Gabor filtering.Compared with bilateral filtering and total difference model algorithm,this algorithm got a better experimental result.The algorithm preserved defect information while filtering out background texture and lays the foundation for the successful detection of defects.Then,this paper studied the defect detection algorithms of Aluminum Plastic Blister.To detect the defects on the medicine plate,a set of combination algorithms based on frequency domain is proposed in this article.To detect the defects of medicine granules,this paper proposed a partition threshold algorithm based on density,background modeling algorithm based on sparse and low rank matrix factorization.Then,author used a series of algorithms.For example,connecting and merging regions based on distance to achieve defect detection and positioning.Through a few tablets image test by MATLAB,the accuracy is up to 96.1%,the missed detection rate is 2.8%and the false detection rate is 1.1%.Finally,this paper studied the defect classification algorithm of aluminum plastic blister.By comparing the advantages and disadvantages of image classification algorithm,the paper finally used the random forest and deep convolutional network to identify the defect types.According to the experimental analysis,in this article,because of small number of samples,obvious characteristics of the image of the medicine particles which the defect feature is easy to extract,compared with the deep convolutional network,the random forest classification model obtains better experimental results,and is more suitable for the classification of the drug particle defects.The classification accuracy rate is 95.63%.