Visual Detection Of Blood Glucose Monitoring Data Based On Image Processing

For patients with diabetes at home,it is a commonly used telemedicine method to send the blood glucose curve displayed on the CGM screen to the doctor who remotely manages the blood sugar.Considering that doctors need more accurate data for diagnosis,it is necessary to consider CGM blood glucose image data extraction through visual detection.In this paper,the visual detection algorithm of data extraction is designed for the screen image of CGM blood glucose meter,and the image data detection system of CGM blood glucose meter is developed.The main contents of this paper are as followsDue to the need to design different visual detection algorithms for different types of CGM blood glucose meter,and it is difficult to obtain a large number of CGM blood glucose meter screen images at the same time,the task of CGM blood glucose meter screen image classification in small sample scenes is very difficult.Because the commonly used hog features mainly consider the edge information,an improved hog feature is proposed.After calculating the gray distribution of the image,it is spliced with the hog feature.The improved feature has not only the edge information but also the gray distribution information of the image.After the SVM classification experiment on the data,the classification results of the original hog features and the results obtained are improved.Before data extraction of CGM blood glucose monitor screen image,image correction is designed.The first step of correction is to segment the screen image,and an improved grabcut segmentation algorithm is proposed.The foreground preselection box is calculated by region growing algorithm,which makes up for the problem that grabcut algorithm needs to set preselection box manually.For the segmented screen,the projection transformation is used for further correction,the DP algorithm is used for quadrilateral fitting of the frame,and the projection matrix is calculated by the fitting vertex position to get the corrected screen image.For screen images,the data to be extracted include blood glucose data and text data.The first step of blood glucose data detection is to detect the key points of the coordinate axis of blood glucose curve.Through the further correction of the coordinate axis,the curve image with the direction calibrated is obtained.After the threshold segmentation of the calibrated curve image,the blood glucose curve part is obtained,and then the image is refined.The most obtained image data is interpolated to calculate the corresponding relationship between time and blood glucose value.For text data detection,including text region detection and text segmentation.The text region detection is realized by semantic segmentation.A semantic segmentation network is designed to detect the text region,and the output of the network is processed to get the text region.Text recognition uses crnn network,including CNN and LSTM.For the problem that the output is not aligned with the actual label,CTC loss function is used.The text area is put into the trained crnn network as the input to get the recognized text.Finally,after designing the visual detection algorithm,the CGM blood glucose meter image data detection system on the web is developed.The system uses Django framework,including user module,detection module and management module.Patients or doctors can upload image data by opening the web page.The system will call the detection module to automatically detect,generate excel structure and save records.The system can assist doctors in telemedicine.