A Study Of B-Ultrasonic Fetal Image Classification System

With the development of the application in B-ultrasonic machine, the phenomena that detecting the fetal sexual organ by B-ultrasonic machine is more and more popular. How to detecting B-ultrasonic fetal image and overlapping B-ultrasonic fetal sexual organ is more and more important. The aim of B-Ultrasonic fetal Image classification system is finding out the position of B-ultrasonic fetal sexual organ. It will avoid the lawless use of B-ultrasonic image to identify the gender of fetus. Therefore, it is very important for society and technology development. Based on statistical learning theory, this dissertation employed several feature extraction and dimension data reduction techniques to the sexual organ detection in a B-ultrasonic fetal image. Experiments demonstrated that they can greatly reduce detection error rate while increase the detection speed efficiently.According to the characteristic of B-ultrasonic fetal image and the requirement of application ,this dissertation do the research in preprocessing of images, dimension reduction arithmetic and the classification of image. The main achievements of this dissertation are as fellows:(1)This dissertation combines LLE with SVM, and then applies LLE/SVM into the classification system of B-ultrasonic fetal image. The experiment indicates that the system completely meets the demand of application.(2)This dissertation studies the application of LLE/SVM's every different core function in the classification system of B-ultrasonic fetal image.(3)This dissertation not only studies TwoClass-SVM, but also studies OneClass-SVm and MultiClass-SVM in the classification system of B-ultrasonic fetal image.(4)This dissertation does the further research in the compare between LLE/SVM and the combination of traditional dimension reduction method with SVM. The experiment indicates that LLE/SVM completely can greatly reduce detection error rate while increase the detection speed efficiently.