Font Size: a A A

Study Of Data Mending Based On BP Neural Network

Posted on:2004-11-18Degree:MasterType:Thesis
Country:ChinaCandidate:Y J WanFull Text:PDF
GTID:2168360092981324Subject:Measuring and Testing Technology and Instruments
Abstract/Summary:PDF Full Text Request
During the procedure of digitalizing geometric features of the curved surface, exact data mending of the insufficient measuring data is the precondition of reliable Vreconstructing curved surface, so it is especially noticed by researching personnel, corporation manufacturer and engineering technician.BP(Back - propagation) neural network is the outcome of artificial intelligence technology rapidly developing and extensively applying in many engineering fields, and is especially attached importance by the searcher in many disciplines with the powerful nonlinear mapping capability and original strongpoint solving the high nonlinear and uncertainty problem in nature. Now, it is widely used in nonlinear modeling, function approaching, mode classifying, and is the most widely applied neural network model, work well, and easily be trained.In this paper, BP neural network is used to mending the insufficient data during the measuring procedure of digitalizing geometric feature of curved surface. At first, the principle of this method, which is using the nonlinear mapping capability of BP neural network to study and memory the geometric feature of the object surface, and to mend the insufficient data for curved surface, is described. Then, the network structure and algorithm are designed. After designing the hidden-layer structure of network, the neural network data -mending program is developed with Matlab, and with the contrast of results gained from single and double hidden-layer the number of hidden-layer is selected. After the training algorithm is studied, using the visual application-designingtool, Visual Basic, the application adopting the designed algorithm is developed. With plenty of experimental study, the node number of the hidden-layer is decided and the rational algorithm is designed, which can advance the speed of training network, and decrease the probability of running into the local teeny value.The experimental results of mending the insufficient data of several curved surface indicate that the data mending system has the performance of good university, rapid speed and high precision , and then has high practical value.
Keywords/Search Tags:insufficient data, nonlinear mapping, data mending, BP neural network
PDF Full Text Request
Related items