The Research Of Online Visual Recognition And Classification For Visible Contaminating Particles In Transfusion Liquid

With the example of transfusion liquid of medicinal solution, a research of visualforeign substances recognition and classification is conducted. Machine vision is usedto imitate the human eye to make sure that the recognition and classificationtechnology of visible foreign substances must be effective and accurate. Theapplication of the recognition and classification algorithm into the real productionsystem has improved the detection accuracy and dectetion speed, and brings lots ofeconomic and social benefit.Firstly, the principles and applications of machine vision technology aresummarized in this paper, and the paper elaborates the application status of visualtechnology utilized in current medical environment at home and abroad. Based on thefeature of real detection objects, the difficulty of foreign substances recognition andclassification is analyzed.Secondly, the structure of visual system is briefly introduced and the paperelaborates the design of the foreign substances recognition and classification systemin transfusion liquid, including the construction of optical lighting systems,mechanical systems and electrical control system.Thirdly, the recognition algorithm is focused in the paper. After analyzed therecognition object, the detection difficulty, the status of key technology and the entirevisual recognition program is presented. After the original image is dealt with anadaptive weighted mean filter, the region of interest is got by the normalizedself-adaptive projection method to abandon parts of the interference of the bottom,sidewall of the bottle and then it is enhanced by composite Laplace. This paper hasanalyzed the characteristics of the target and the difficulties in recognition. Based onthe image sequences, the accumulated difference image is calculated and theinter-frame difference method and background difference method are combined tosegment the foreign substance. The method of mathematical morphology and theKalman filter are used to track the segmented target objects to judge the foreignsubstances and the bubble interference correctly. The experimental results are givento verify the reliability and validity of the recognition algorithm.Subsequently, after extracting target features on the basis of tracking the foreignsubstances, BP neural network and support vector are used to classify the foreign objects to determine their categories. For the problem of slow convergence rate of thestandard BP algorithm, a kind of improved BP algorithm is presented and theimprovement in the performance is proved by the experiments on comparing the twokinds of BP algorithm on classifying the Iris data. When support vector is used to be aclassifier, the one-versus-rest algorithm is used to solve multi-classification problem.At the end of the chapter, the experimental results of BP neural network and supportvector classifier are given and compared to verify the effectiveness and practically ofthe classification algorithm.Finally, the software system development and implementation of theexperimental platform in this paper is presented. After introducing an overview of theentire software system formation, a brief description of the user login module, thesystem control module, the motion control module and database module are given andthen the function and design of the camera parameter setting module and visualrecognition and classification module are presented in detail.