Study On Intelligent Visual Inspection Machine Based On Machine Vision

With wide clinical use, infusion fluid is one of the five major types of preparations in China. Due to production process or environmental reasons, scrap glass or scrap aluminium might be contained in some infusion fluid products, which could cause great damage to human bodies. The national Pharmacopoeia stipulates that the examination of visible foreign materials and diaphaneity against injections must be conducted on an ampul-by-ampul and bottle-by-bottle basis. Currently, visual inspection is the most popular method among domestic manufacturers. Subjectivity, low efficiency and high emission factor of this manual process are the bottleneck of the automatic injection production line.My focus is the machine-vision-based intelligent visual inspection machine. First the series images of the tested liquor are collected using machine vision technique, then proper image processing algorithm is used to identify visible foreign materials in the injections, and to automatically determine if the liquor meets the standards. The objection is to improve the speed and accuracy of visual inspection on liquor, and fully automate the production of medical infusion fluid. The economic and social benefits of this approach is enormous.When machine vision technique is used to identify visible foreign materials in injections, the precision depends on the quality of the acquired images and the validity of the image processing algorithm. In this paper, I conduct experimental research on two key techniques, namely, the online collecting of injection images, and the foreign material identification algorithm. The content is as follows:(1) I design an automatic visual inspection system based on machine vision technique and the algorithm which I research, and research is conducted on the key techniques involved. I develop special light source and lighting system to ensure the high quality of the collected images. Proper vidicons, vidicon lens and image data acquisition card are chosen in order to meet the real-timeness and high precision of the system. (2) The research on image preprocessing method includes:removing the noise in acquired images with median filtering, applying mathematical morphology operator to enhance image contrast, employing instant frame difference method to restrain background, All of the above ensures the good quality of the acquired binary image.(3) Research is performed on the fluid detecting algorithm that's based on feature matching and track association. First the features of the foreign objects are extracted, and then moving-object-detection method based on multi-frame image sequence is used to determine the existence of small particles based on the continuity of the moving tracks and the randomness of the interfering noise moving trails. I verify this algorithm against the acquired liquid images. The results show that the effect of this algorithm on large particles (according to the definition in national pharmacopoeia) is good, and can satisfy the requirements for foreign material detection on infusion fluid.(4) I research on foreign material detection algorithm for fluid based on artificial immune algorithm. Firstly, I construct the possible flight track trees for small objects, and optimize all the trails to ensure that all the small-sized foreign materials in the image are searchable, thus avoiding omission. I develop a trail encoding schema, define fitness function according to the requirements in detecting foreign particles, and on top of that, propose a trail optimization algorithm, which is based on artificial immune algorithm. Finally, I verify the algorithm against the real fluid images. The results prove that this algorithm has good effect on not only large visible particles in fluid, but also small particles (according to the definition in the national pharmacopoeia.) Therefore, it is concluded that this algorithm can meet the requirements for the examination of the diaphaneity and visible foreign matters in infusion fluid.