Research On Defect Detection Technology Of Tablets In Aluminum Plastic Foamed Mask Package Based On Image Processing

Aluminum plastic foamed mask package which occupies the majority share in drugpackaging is popular for pharmaceutical companies and consumers. Because theproduction of aluminum plastic foamed medicinal package needs a series of complexpipeline operation, defective tablets will be inevitably produced. If defective productcirculates to the market, it will affect people’s lives and health. This paper adopts aseries of image processing algorithms to complete the defect detection and theclassification of defect tablets. The detection method based on machine vision has takenplace of traditional detection method, which improves production efficiency andproduct qualification rate.This paper mainly studies image segmentation algorithm, feature extraction anddefect detection method. Firstly, it uses a weighted type of color image filtering methodto pretreat the image, which not only removes the noise, but also protects the imagedetail and edges well. Secondly, as the effect of common image segmentation methodsis far from satisfactory, according to the characteristics information of tablets, this paperfirst classifies the tablets into two kinds: color ones and white ones. Then it adopts asignificant target segmentation method based on visual significance analysis torespectively segment two kinds of tablets image from the background. For color tablets,image pixels’ significant value is obtained by the histogram contrast to generate asaliency map, then the saliency map is segmented by OTSU threshold segmentationmethod; While white tablet images are first processed by Gabor filter to generate asaliency map, which is also segmented into a binary image by OTSU thresholdsegmentation, and morphology erode algorithm is used to process the binary image.Using the center of each region in the resulting image as seed points, region growingalgorithm is adopted to segment the original white tablets image. Eventually, colortablets and white tablets are all separated from the background successfully. Then, theedges of all tablets are obtained by Canny detection operator. As for defect detecting,defective tablets mainly include absence, scratch and so on. We first make statistics ofthe number of tablets by the connected component labeling algorithm to judge whether there exits missing tablets. Then we extract features of tablets and study two kinds ofdefects detection method that one is based on BP neural network while the other isbased on support vector machine (SVM). The experiment results show that theclassification accuracy of SVM algorithm is relatively higher, and it has lesstime-consuming and higher stability.