Application Of Machine Vision Detection In Medical Liquid Impurity Identification

At present, traditional domestic pharmaceutical companies still use artificial contrast detection method for detecting visible foreign matter in medicine. This method is simple but without unified standard, high precision, perfect repeatability. Because it is easily influenced by emotions of operators, false and miss detections are not rare. In addition, when the size of the foreign body is less than 50 microns, human eye is unable to identify them, we must rely on high resolution camera pictures to tell.Because artificial light inspection method cannot meet industrial requirements, this paper adopts machine vision technology for automatic identification of medicine liquid impurities, what’s more, impurity identification method proposed in this essay not only address the problem of impurity identification but also tell impurity kinds. Medical liquid impurity detection structure is mainly comprised of system mechanical structure、inspection electrical structure、software structure based on MFC and OPENCV where mechanical structure includes’rotate-halt-capture’mechanical equipment, electrical structure includes optical system、industrial camera、PLC and industrial PC, software structure includes open source software environment for the medical liquid impurity detection platform.Based on aforementioned hardware equipment, this paper explicitly describes algorithm process of impurity identification.Firstly, due to mechanical vibration, temporal sequential images captured need to be registered and aligned, this paper completely show core algorithm of scale invariant feature transform (SIFT) and its improvement for achieving high precision requirement, then temporal sequential images registration and alignment is done based on improved SIFT, experiments demonstrate the improvement realizes high-accuracy demands.Secondly, for registered and aligned temporal sequential images comprise much low psnr noise which easily affects impurity inspection, consequently segmentation of impurity and low psnr noise is needed. Segmentation model in this paper is based on Pulse Coupled Neural Network (PCNN). Two kinds of evolved PCNN segmentation model are proposed combining with the minimum cross entropy and region growing segmentation algorithm. Canny operator segmentation algorithm is compared with two kinds of algorithm.Lastly, identification of impurity is after the work of segmentation of impurity and low psnr noise, The introduction of the medical liquid impurities and their general categories is given in this chapter. Then tracking algorithm like Hidden Markov Model (HMM) and Unscented Kalman Filter (UKF) is harnessed, based on the temporal sequential images alignment and correction as well as the separation of background and impurities, to track impurities, thereby detecting and classifying impurities...