Application Of Deep Learning Technique In Intelligent Infusion System

As a common treatment,intravenous infusion is widely used in clinical emergency,etc.Furthermore,it is an inevitable part of medical work to supple nutrition and balance the regulation of body fluid through intravenous infusion.However,the current monitoring of the infusion process mainly relies on the irregular inspections of medical staff.The unintelligent approach not only aggravates the workload of medical staff,but also easily occurs the liquid tube empty or blood returning because the needle is not pulled out in time.When the situation is serious,even complications such as air embolism may occur,causing great harm to the patient and aggravating the contradiction between the doctor and the patient.Therefore,it is imperative to use the related technology to transform the existing infusion drip system.At present,many researchers have used photoelectric sensors or gravity sensors to monitor the infusion process intelligently.In the case of infusion monitoring of photoelectric sensors,it is easily affected by ambient light in actual use,thereby influencing the monitoring availability.For the infusion monitoring of gravity sensors,although the ambient light does not affect it,its main drawback is the measured weight containing the residual liquid and the infusion bottle.The generated errors by the gravity sensors will seriously influence the effect of the infusion monitoring and even cause false alarms when the residual liquid is fewer.Therefore,this paper will study the current defects of the infusion monitoring system based on gravity sensor,and use the computer vision technology based on deep learning to intelligently identify the infusion bottle material,specifications and manufacturers by extracting the information of the infusion bottle images.Thus,the errors caused by the weight of infusion bottles can be solved,which makes the more intelligent and accurate infusion and the less working pressure of medical staff.And the infusion monitoring system based on gravity sensor can be improved effectively,which promotes doctor-patient relationship and the establishment of intelligent medicine.For the material identification of infusion bottle body,we first collect the data in the practical application scenario and construct the data set with a reasonable method,so as to make the data set more consistent with the scenario.Then we adopt appropriate neural network model to learn the constructed data set and accomplish the classification of infusion bottles with different materials according to the learned characteristics.To verify the effectiveness of the proposed scheme,we organize the contrast experiments and the experimental results show that the scheme used in this paper has relatively high recognition accuracy.For the extraction of infusion bottle label information,it is necessary to distinguish the text area and background and locate the text area in the image of infusion bottle label firstly,and the accuracy of the positioning of the text area directly influences the effect of the text recognition.In this paper,three target detection models with excellent performance are used for text area detection experiments.And we analyze the detection results of the three schemes and choose the most suitable model for the application scenario.Then the proposed scheme extracts the detected text area by line and identifies the text of the extracted line.In order to improve the accuracy of recognition,convolutional neural network and cyclic neural network are combined to refer to the context information and to optimize the identified characters according to the semantics model.Finally,the experiments are organized to verify the proposed scheme,which show that the proposed scheme is more reasonable and effective.