Research And Implementation Of Biochemical Detection Image Recognition Method

At present,the blood cell test and joint inspection card diagnosis are two of the most commonly used methods in biochemical detection.Blood cell test,which can count the number of white blood cells(WBC),red blood cells(RBC)and platelets in blood cells,determines whether the patient has a disease.The joint inspection card diagnosis,which can judges the result of the reaction between secretion sample and chemical reagents in the reaction area of the joint inspection card,identifies the health status of the patient.However,these two detection methods both require excellent image processing algorithms to ensure accuracy.In response to this problem,this thesis designs and implements automatic recognition algorithms for biochemical detection images.The main work is as follows.1)Designed and implemented blood cell test method.Firstly,to improve YOLOv4 algorithm for raising the recognition precision of the algorithm for small targets,changing the single-channel convolution feature extraction operation in the original residual module to three-channel feature extraction,distinguishing three kinds of blood cells,and improving detection precision.Secondly,the adhesion cells in the image are processed to eliminate the influence of the adhesion cells.The thesis uses the mark-based watershed algorithm to segment the adhesion cells to solve the phenomenon of the adhesion cells,and the over-segmentation problem is successfully solved after testing.Through these improvements,the average precision of YOLOv4 algorithm for the three kinds of blood cells reached 94.7%,87.3%and 88.5%respectively,and achieved better results than the original network(84.7%).2)Designed and implemented joint inspection card detection method.Firstly,taking the colorimetric card as the comparison standard,process the image of the colorimetric card,reduce noise,extract the edge of the image,and then extract the contour of the edge image to obtain 7 sub-colorimetric cards,and then,use the K-means clustering algorithm to complete color vector clustering,which can remove irrelevant colors and just retain useful color blocks for comparison to generate a new colorimetric card.Secondly,the joint inspection card is processed.Using hough circle transform to detect the reaction holes and identify the reaction holes in the joint inspection card.However,due to the influence of light,there is a phenomenon of local reflections appeared in the image,which has a serious affect on the very important step of color recognition of reaction hole.In order to eliminate the phenomenon of local reflections,the L of each reaction hole is calculated by using Lab color space to distinguish the normal area from the reflective area,so as to remove the reflective area of the image.Finally,the color of the processed joint inspection card and the colorimetric card are matched,the detection accuracy of the algorithm on the joint inspection card reaches 95.1%.In this thesis,the methods of recognizing blood cell image detection and joint inspection card image detection are improved.The accuracy of blood cell image detection(90.2%)and the accuracy of joint inspection card image detection algorithm(95.1%)are better than other methods.It realizes the accurate recognition of the two kinds of biochemical detection images and assist in treatment.