Analytical Detection Of Intravenous Infusion Based On Image Processing

Intravenous infusion is an indispensable treatment measure in medical care,but during the infusion process,when the infusion is over and the medical staff cannot handle it in time,there will be blood return and other conditions that cause people to suffer,and also affect the doctor-patient relationship.Therefore,it is also very meaningful to detect the infusion level.The existing liquid level detection methods have some shortcomings: mechanical detection methods have the problems of complex structure and low accuracy;non-contact measurement methods like ultrasound are expensive.Therefore,this paper proposes an approach that uses image processing and deep learning.The position of the infusion bottle is first targeted and then the level of the infusion bottle is detected.In view of the problem of infusion bottle object detection,two methods are used to detect the infusion bottle.In order to have a certain understanding of object detection for infusion bottles,SURF(Speeded Up Robust Features)and FLANN(Fast Approximate Nearest Neighbor Search Library)based on traditional image processing are used.algorithm.Firstly,the SURF is used to extract feature points and calculate feature vectors.Then the FLANN algorithm is used to match the feature vector,filter out matching results that meet the conditions.Finally,the target box containing the infusion bottle is drawn on the image to be detected.In order to solve the problems of detection accuracy and detection speed of object detection,the SSD(Single Shot Multi Box Detector)algorithm based on deep learning is adopted,and feature extraction is performed based on the VGG16 network,and the multi-scale frame system is used for local feature learning.NonMaximum Suppression(NMS)filters out the default box,and finally outputs the object box position and category label.Through experimental comparison,the object detection method based on SSD was selected to detect the infusion bottle.In view of the problem of infusion bottle liquid level detection,two methods are used to detect the infusion bottle liquid level.In order to accurately detect the liquid level line,a projection method based on image processing is used.Firstly,the image is preprocessed,then Canny edge detection is performed;finally,the liquid level line is located by the projection method,and the threshold of the liquid level line is further judged,below Threshold value will give warning prompt.In order to quickly detect the liquid level,a method based on Convolutional Neural Networks(CNN)is adopted.In the first step,the sample set is put into the CNN model for training,continuous feature learning,then the full connection layer outputs all feature vectors in one-dimensional mode.Finally,Soft Max is used to classify and output category labels.In the second step,the images after object detection and region cropping are sent to the trained CNN model for liquid level classification and recognition,and the category labels are output.Through experimental comparison,a projection method based on image processing was selected to detect the liquid level of the infusion bottle.Through the above object detection method based on deep learning SSD and the projection method based on image processing,this paper achieves the functions of infusion bottle target detection and infusion bottle liquid level detection.The experiment proves the feasibility of this method with an accuracy of about 0.8,which solves the problem of detecting the liquid level during intravenous infusion.