基于视觉词典与多尺度特征融合的药品外包装图像检索系统设计与实践OA

Objective To develop a drug outer packaging retrieval model and integrate it into the drug dispensing system of the hospital outpatient pharmacy,in order to optimize the outpatient drug dispensing business process and improve the work efficiency and safety of the drug dispensing link.Methods A Feature extraction method based on the fusion of scale-invariant feature transform(SIFT)and speeded up robust features(SURF)was proposed.The drug outer packaging image retrieval algorithm combining vector of locally aggregated descriptors(VLAD)feature coding was used to train the collected outer packaging images of commonly used outpatient drugs.The running time of key steps,the prediction probability vector and the prediction result were taken as the evaluation indicators of this model.An interactive visual control was designed using Windows Form,and the algorithm was directly invoked by pharmacists through the operator interface.Results The visual dictionary model based on feature fusion took less time in generating the visual dictionary,VLAD encoding and the total training time compared with the visual dictionary model based on SIFT.Moreover,it has better prediction performance,with an accuracy of up to 99.02%and an average probability value of 94.77%for a single sample output.The outpatient pharmacy dispensing system optimized by the algorithm in this study has significantly improved in terms of waiting time for dispensing and the number of dispensing errors compared with that before optimization,and the differences were statistically significant(P<0.05).Conclusion Integrating the algorithm of feature fusion of SIFT and SURF and feature coding with VLAD into the dispensing system of the outpatient pharmacy can accurately retrieve the names of drug outer packaging,play an auxiliary role in the dispensing process of pharmacists,improve work efficiency,and reduce a series of medical disputes caused by dispensing errors.