Research On Pill Recognition Algorithm Based On Deep Learning

With the development of modern medicine,medicine plays an increasingly important role in human life.Pills are one of the most commonly used forms of medicine in everyday life.However,due to the wide variety of tablets and their similar appearance,it is easy for human identification errors to occur.Therefore,the development of an accurate and efficient pill recognition algorithm is of great significance for the drug industry and the medical industry to provide fast and accurate drug identification and management services.Based on deep learning technology,this paper designs a recognition algorithm that takes pill shape as the key information for pill recognition.The details are as follows:First,this paper classifies and recognizes pills in six types of dosage forms commonly used in hospitals,pharmacies and families,and obtains required pill samples and organizes pill images by means of web crawler,manual shooting and Baidu search.In order to ensure the quantity and diversity of samples,data enhancement processing was carried out for some categories of pill images,and labelimg tool was used to label the images and produce pill data set Secondly,the backbone network and feature extraction network of the original YOLOv5s algorithm are improved.It is proposed to replace the backbone network of the original YOLOv5s algorithm with a more lightweight GhostNet model for deployment in mobile devices.The BIFPN weighted bidirectional pyramid structure is added to the feature extraction network to improve the detection effect of different scales.At the same time,CBAM attention mechanism is integrated to improve the network’s ability to effectively pay attention to useful feature information.Finally,multiple rounds of training were conducted on the model,and the pill recognition results before and after improvement were compared to determine whether the improved algorithm could effectively improve the pill recognition performance,and the improved YOLOv5s algorithm in this paper was displayed visually.The experimental results show that the pill recognition algorithm proposed in this paper can realize rapid and accurate pill recognition.The experimental results of this paper show that the pill recognition algorithm based on deep learning can realize rapid and accurate pill recognition.The average accuracy of the improved YOLOv5s model reaches 99.1%,and GFLOPs of the model size and computational complexity are significantly reduced.Pill picture recognition and video detection function can be used to identify pills in real time.The introduction of artificial intelligence in this study can identify a large number of pills in real time with high accuracy,thereby reducing the possibility of medical errors and helping medical staff focus on higher-level tasks by simplifying time-consuming lower-level tasks.