Search

In the paper, the research present situation and development in the wavelet neural network model and a novel learning algorithm for wavelet neural network based on extended Kalman filter are discussed. Based on combining the exceptional property of localization of the wavelet transform and characteristics of self-learning of neural networks, the non-line time series model and network architecture model which combines affine transform with revolving transform is discussed. A novel learning algorithm for wavelet neural network based on extended Kalman filter is proposed to predict the deformation of structure. In comparison with the WNN algorithm, the EKF learning algorithm has improved convergence and can provide much more accuracy learning results.

Systematic errors, such as multipath and residual atmospheric errors in GPS double difference carrier phase observations, are the main error sources for high precision GPS applications. The complicated properties of systematic errors limit the modelling and mitigation of these errors when parametric models are used. In this paper, a new approach based on wavelet analysis is developed to tackle them. Wavelet transform is used as a tool to separate systematic errors from the signal. The signal here is double differential phase observation, which is an input for wavelet transform system. Through multi-step decomposition for original double differential observation, systematic errors identity is exhibited at a certain scale in the decomposition and the double differential observation is then reconstructed by a filtering to the signal with systematic errors eliminated. The newly formed observation is then used as the input for baseline adjustment unit to achieve desired positioning efficiency. Experiment and analysis is conducted to show the applicability of the proposed approach.

A pyramid approach for remote sensing image registration based on mutual information and grey relational analysis is presented. The methodology applies edge information of image extracted using grey relational analysis and combines with mutual information regarding as similarity measure. First, the image pyramid is constructed by wavelet transform, and then coarse-to-fine multi-resolution search approaches have been put forward to increase accuracy and efficiency. An exhaustive search algorithm is applied at the coarsest level of the image pyramid. Registration at higher levels can be performed with the result at the pervious level serving as the initial condition. The effective and exact matching results can be obtained by applying the algorithm to remote sensing image registration.

The characteristics of three GPS kinematical data processing models, Least Square, Kalman filtering, are discussed and their advantages and disadvantages are compared. With observational data and pertinent data processing software, the applicable condition, context and effect of the three models are experimented. Results show that when the mobile platform is in uniform motion, the accuracy of the three models are almost equal; when the mobile platform is in stochastic acceleration, the accuracy of new filtering model is superior to that of LS, and that of Kalman filtering is the worst.