Research On Pill Detection And Recognition Algorithm Based On Machine Vision

Taking pills and checking pills in large hospital pharmacies is an important task and has an important impact on hospital productivity.Traditionally,hospital pharmacies rely on manual labor to take pills and check pills.This method has the following problems: the wrong pills are taken due to visual fatigue;the pills are contaminated;the work efficiency is low.In response to this problem,the Automated Tablet Dispensing & Packaging System(ATDPS)has appeared in the society.The system consists of two main components: the Tablet Packing Machine System(TPMS)and the Tablets Check Machine System(TCMS).The TPMS technology is very well developed and the TCMS technology is in the development stage.Many hospitals only use TPMS technology,and the work of checking pills is still done manually.Therefore,the research on TCMS technology has important practical significance.This paper designs a pill detection and recognition algorithm based on machine vision.In the pill detection phase,the paper first extracts the region where the pills is located to obtain the foreground target,and then separates the adhesive pills to obtain a single region where the pills is located.In the stage of extracting the region where the pills is located,this paper considers four algorithms:otsu,one-dimension maximum entropy image segmentation,improved otsu,improved one-dimension maximum entropy image segmentation,the forth algorithm works best.In the stage of splitting the adhesive pills,the paper first performs concave points detection and then performs concave points matching.In the stage of concave points detection,this paper consider tree algorithms: curvature scale space corner detector with adaptive threshold and dynamic region of support,local region angle,improve local region angle,the third algorithm works best.In the stage of concave points matching,this paper considers K-M algorithm and improved KNN algorithm,the latter works well.In the pill recognition phase,this paper mainly uses three important features of area,shape and color to judge the type of pills.When measuring the area,the connected domain labeling method is used to count the number of pixels in the connected domain.When identifying shapes,shape matching based on shape context and a nine-layer convolutional neural network model are compared,the latter are superior to the former in accuracy and speed.In the stage of color classification,this paper designs a color classifier based on hue histogram and SVM.