Study Of Quality Detection Method Of Blood Type Card Filled Based On The Image

The production process of the blood type card,which is based on the glass bead agglutination method,includes the procedures of filling liquid(Antigen reagent),filling powder(glass bead)and sealing film.To obtain the accurate blood type detection result,the quality of the blood type card must be comply with the standard.The detection on filled quality of the blood type card mainly includes two aspects,one is to measure the amount of filling,the other is to detect whether there are bubbles in the blood type card after filling.At present,manual method is the domestic main method of the detection on filled quality of the blood type card.Therefore,method based on the image is proposed in this paper.The main research content and results will be shown as follows.(1)Blood card images are captured by the camera,First,the straight lines on the top edge of the blood card is fitted to judge whether the image is inclined and to compute the tilt angle.The tilt image is rotated and corrected.Then,the method of gray integral projection is employed to cut and segment the image,After useless image information is removed,six separate micro-column images are obtained in each blood type card.The horizontal edge position of the mixed solid and liquid phase is extracted respectively by Sobel edge detection method after image-enhancing the micro-column tube image.The corresponding height is calculated according to the correspondence between the pixel length in the image and the actual length in the practice.Comparing the measurement results obtained by algorithm in this paper with the manual measurement results,the error distribution can be made to statistics histogramsfit.According to the 3? principle,the height error of liquid phase is in the range of(-0.525,0.357),and the height error of solid-liquid mixed phase is in the range of(-0.273,0.225).(2)Aiming at the problem of whether bubbles are in the blood type card images,the image classification is carried out by the method of support vector machine.The HOG,Haar,LBP features are extracted for the single micro-column image respectively.The polynomial kernel function,RBF kernel function and Sigmoid kernel function were selected to train the classification model respectively.The best parameters of each model were obtained by cross-validation.Experimental results show that the proposed method achieves the best classification accuracy with the HOG extraction feature and the RBF kernel selection function.The accuracy is over 90% and the total time of height measurement and bubble detection is less than 1.5s,which can meet the real-time detection demand.(3)In order to determine where the bubbles are located,the method of convolutional neural networks is used for image classification.The residual network are selected as the basic model.According to the characteristics of a single micro-column image,el,the residual network is improved.The single micro-column image is divided into two regions of liquid and solid-liquid mixed phase.Three convolution kernels of 3×3?5×5 ?7 ×7 sizes are selected for each region to extract the features respectively.The feature fusion is performed by channel concatenation.The residual part of the network is designed as two residual groups with four residual units in each group.Experimental results show that the classification accuracy of the blood type card bubble image is as high as 98.63% by the convolutional neural.It can effectively detect whether the presence or absence of air bubbles in the blood card.