Real-time Identification Of Capsule Defects Based On Machine Vision

With the rapid development of science and technology and economic construction,people’s health consciousness continues to increase, and they have higher requirements for thequality of medicines which is closely related to their health. Many pharmaceutical companiesrely on quality inspectors to visually observe the contour and color of medicines one by one,for judging the quality of medicines. As the cost of production and management is high, andhuman eyes have shortcomings such as instability and fatigue, this method has been unable tomeet the modern industrial production requirements. With the improvement of mechanization,informatization and intelligence, machine vision is gradually applied into the fast and accuratedetection of product quality. Machine vision is to use a computer to simulate the human visualfunction, of which a very important part is image processing. Currently, image-based defectdetection method has two categories: the first category is to judge the quality of productsbased on the intensity information of image, which simply uses a single threshold method tosplit products and defect information, but may loses some defect information; the secondcategory is to judge the quality of products based on the texture information of image, whichhas some deficiencies in the detection speed and the cluster of defect information.For these deficiencies, this paper designs a new capsule defect recognition method, aimedat improving the detection performance of existing equipment in pharmaceutical companies.It studies and designs the capsule defect recognition algorithm, to realize the qualitativeidentification of defect capsule, namely judge whether the image to be detected has defects ornot, and remove the unqualified capsules.In this paper, the main work is described as follows:(1) According to the features that the standard capsule image is divided into four distinctregions, the gray values within the region are relatively stable, and the gray values of differentregions have great differences, use eight connected method to achieve regional classificationand recognition of the enhanced capsule image, and detect the defects of facial contour. It is toovercome the problem of big error of the geometric property parameter in the instable regionof edge detection effect in the existing detection method. And it only involves addition andsubtraction in the calculation process of the area classification and recognition, therebyimproving the detection speed.(2) Based on the characteristics that the capsule linear defects concentrate in the capsuleedges and are in irregular polygonal scalloped shape, this paper proposes the improvedalgorithm of L-shaped corner point recognition on the basis of Harris algorithm to determinewhether there are linear defects in local areas. It overcomes the inaccuracy of calculating theminimum distance between the capsules to be recognized and standard capsules in theexisting detection method.(3) It proposes the improved angular value calculation algorithm based on the L-shapedcorner point to achieve the quantitative identification of linear defect angular values. Itovercomes the inaccuracy of judging whether there are linear defects relying solely onwhether there is L-shaped corner point. Through experimental verification, the capsule defect recognition method has higheraccuracy and real-time ability for facial contour defects and linear defects, meeting therequirements for efficient, automated testing in the actual production process.