| In the area of clinical detection,blood detection is the necessary program of blood transfusion and has pretty wide application.As the normal work in blood transfusion,blood grouping and cross matching are the primary presupposition and guarantee.Full automatic digital detector has been familiar to the public while many companies have started produce micro-column blood grouping card on a large scale.But the domestic automatic detectors matching with the cards are lack,we need to import relevant productions abroad.So it's extremely pressing to achieve and improve domestic recognition technology of automatic mass blood detection.The detail research content and results of this paper will be shown as follows.(1)It needs to make complete ready for extracting characteristic data of micro-column blood grouping card.It asks us to ready for three aspects: the first aspect is that making up image preprocessing operation flow,including grayscale,binarization,image enhancement techniques,morphological operations.In this aspect,we introduce its principle and effect of image processing,which lays the groundwork for extracting characteristic data.Secondly,we propose the way of using mask processing to clear pipe wall in order to avoid the micro-column blood grouping cards' pipe walls influencing extracting characteristic data and the next result analysis.We introduce the principle and program of mask processing in details,which attributes to the accuracy of extracting characteristic data.Thirdly,(2)This paper extracts data set and information attributing to pattern recognition from images,then get the description and expression of data set of images.We choose the micro-column blood grouping card as research object,make analysis and comparison to its characteristic description from many aspects,and deal the problem of choosing independent and the most dipartite feature vector among data sets.This paper makes analysis research of agglutination strength in agglutination reaction area of single micro-column pipe from image texture,shape description,gray integral projection curve trend description and projection curve peak point sample description.(3)In order to overcome the shortcomings of that 2+ type sample size is less,combined with the advantages of SVM classifier in small sample,non-linear and high-dimensional pattern classification,support vector machine classification is used to train and test the sample data and verify the recognition rate of feature vector.The method of data normalization and parameter optimization in this experiment is also introduced in detail.This paper describes the experiment procedure of pointing out the sort of the red blood cell agglutination strength.This paper compares the results of the same kernel function to different eigenvectors and the results of the same eigenvectors to different kernel function then select best eigenvector,kernel function and the optimal parameters. |
