Research And Implementation Of Accurate Model Of Drug Configuration Based On Image Recognition

Pharmacy Intravenous Admixture Services(PIVAS)is a service in which hospital pharmacists mix multiple medications in a certain ratio and give them to patients by intravenous infusion according to medical orders and patient conditions.This service can improve the effectiveness and safety of drug treatment and is an essential part of modern medical care.However,there are some shortcomings in the working process of PIVAS: in the drug verification process,the drug verification can only be done by manual observation,and the visual fatigue and repetitive labor may make the medical staff misjudge the drugs;in the drug dispensing stage,the drug extraction may be affected by factors such as angle and light,resulting in the extraction amount not matching the expected dosage,and the manual verification may mistakenly believe that the extraction amount meets the expected dosage.In the drug dispensing stage,the extracted amount may be affected by factors such as angle and lighting,resulting in the extracted amount not matching the expected amount,and the manual checking may mistakenly think that the extracted amount is in line with the expected amount,thus leading to the wrong calculation of drug amount.As image recognition technology continues to advance,it has been widely used in the medical field.Therefore,it is feasible to explore how to use image recognition technology to identify drug types and dosages.The study aims to improve the accuracy and safety of drug allocation in PIVAS,and integrate the image recognition technology into the whole PIVAS process to study and analyze the identification of drug types and drug dosages.The main research contents are as follows.(1)Constructing an intravenous medication dataset.Due to the lack of publicly available datasets of drugs required for intravenous drug dispensing,the study chose to photograph the target drugs from different angles,different orientations and different placement situations in the actual environment of PIVAS to collect a large amount of data on intravenous drug use,and to label the collected drug targets for model training.For the constructed dataset,data enhancement is carried out to complete data augmentation,and random erasure processing as well as Gaussian noise processing is done to improve the accuracy of the model in recognizing drug types in complex scenes;(2)A model applicable to drug classification recognition is proposed.A modified YOLOv5 s model is used for drug classification recognition.Firstly,the Ghost module is introduced in the backbone feature extraction network,and for the problem of missed and false detection in drug target recognition,the coordinate attention mechanism is fused and the connection of the network is changed with the idea of Bi-FPN.Finally,CIoU Loss is adopted for the bounding box regression.experimental results show that the mAP of the improved model reaches 95.31%,which is better than all other models.(3)Study of a model for identifying the dosage of pharmaceuticals to be dispensed.The study identifies the position of the plunger in the syringe and thus calculates the actual dosage of the drug.The nature of the image target object is first analyzed and a pre-processing strategy is designed.Then a corner point detection model is designed for the processed images and corner point coordinates are extracted.Finally,the extracted corner point coordinates are correlated and calculated to obtain the actual dosage of drugs in the PIVAS dispensing session;...