The Research Of PRRS Diagnosis By Image Recognition Technology

Aimed at diagnosis of PRRS,we use the image recognition technology to analyze the content level of lymph node cells,which can assit the the diagnosis of PRRS. The main contents are as follows:(1)In this paper,we have survey the domestic and international research status of PRRS,as well as the image recognition technology in the field of medical diagnostics.(2) Image pre-processing research. Due to the complexity of medical images and experimental environment,we can not use the original medical images for image segmentation directly. In this paper, in view of problems such as uneven illumination,we have introduced the HSI color space model,which lift the relationship between color information and light intensity. We have also investigated commonly used smoothing filter methods、edge detection methods and the binary morphology.(3) Researchof image segmentation based on watershed algorithm. a) Smoothing filter:we select bilateral filter as the smoothing algorithm before the watershed algorithm by experiment.b) marker extraction:we introduced the artificial heuristic informationto marker extraction,and designed the extraction method of the mark based on the HSI color image segmentation and morphological operations, thus obtained meaningful cell markers.c) Region merging:aiming at the high time consumption problem for the traditional region merging method, we introduced the marker information,and designed a new region merging method based on region growing. When considering the design of the distance between regions,we take full advantage of the abundant information of the color image, put a region merging criteria from three aspects of color, texture, and public edge, and achieved good results. For the adhesive cells in the image, we also adopt the adhesive cell separation method based on searching the concave point.(4) Research of systemicly feature extraction and reduction. We carried the feature extraction systemicly from three aspects of shape, texture and color. In this paper, We combined the advantages of PCA feature transform and ReliefF feature selection, and proposed a the feature reduction method based on PCA and ReliefF, the method reduced the feature dimension effectively, eliminating the correlation between the features, and removing features which contributes little to the classification.(5) Research of classification based on integrated learning as well as sample denoising and reduction method. The complexity of medical images such as the lymph nodes,the recognition result by single classifier is not satisfactory. Integrated learning can significantly improve the performance of the classifiers, so we carried out in-depth study on integrated learning, described the classic Boosting and Bagging algorithm. For noise-sensitive characteristics of Adaboost, we proposed denoising method based on clustering of samples, and further research on large-scale sample set reduction methods based on clustering.Starting with improving the difference of the base classifiers, we introduce random perturbations to further enhance the learning generalization ability in Adaboost and Bagging method. At last, we use weighted voting to design the integrated approach of the two level classifiers, and proposed a complete set of pattern recognition algorithm framework based on modular design and pipe ideology.(6)Research of application of image recognition technology in medical diagnostics. We analyze the cell concentration levels obtained by experiments,and point out the diagnostic indicators of disease, provides a basis for PRRS diagnosis.