The Learning Algorithm Of Serial And Parallel Neural Networks In The Application On B Ultrasonic Image Classification

The purpose of this thesis is to utilize the predominance of the neural networks in informational process for the recognition of B ultrasound image classification.This paper revolves around the central task of B ultrasound image classification.It is mainly about feature extraction of target images,the structure and learning algorithm of serial and paratactic neural networks and the identification technology of the neural networks.The obtained results make us further strengthen the understanding and widen the vision on the neural networks.As an important medical image diagnostic means,it is important and essential for ordinary people to recognize and classify B ultrasound image at an efficient and fast manner.Our research works are summarized as follows:At learning algorithm of neural networks,this paper afford Topological Structure of serial and paratical neural networks and two theorems of learning algorithm.The learning process on serial and paratical neural networks are described in detail.Serial and paratactic neural networks are novel artificial neural network models developed from the common neural networks according to the needs of practical problems.In serial and paratactic neural networks,a complex network input is decomposed into several simple sub-networks inputs,and each sub-network connectes with each other via Serial or paratactic way.Serial and paratactic neural networks have small size in each sub-network and have faster learning speed.At the process of feature extraction of B ultrasound image,we put the image into gray scale,remove the secret information and choose the interested region from the image firstly,then we extract the features of the image including moment invariants, spectrum feature,texture statistical moments and co-occurrence matrices texture feature.According to the data of this paper,the four type feature data have good and rotating invariants and conform to the classifical parameter of B ultrasound imageAt classification of B ultrasound image with serial and paratical neural networks, combining with computer technology,biomedicine technology and pattern recognition,we propose live B ultrasound classfication using serial and paratical neural networks.We use 120 samples(30 samples each)which are from the same doctot and the same ultrasound equipment to distinguish these images to interested regions of 200×200 pixels.And we extract the features from the interested regions. Considering the great difference of data forms,we unit all data in order to apply to class test with the standardization of feature data.The experiment results show that the two method achieve high classifical speeds and high identifiabilitiesAt last,we have used MATLAB7.4 to develop a B ultrasound imge classification based on serial and paratical neural networks system.It can identify B ultrasound image of normal live,hepatic adipose infiltration,schistosome live and live cancer.