| In recent years, the mathematic theory and application based on the wavelet analysis caused a great attention in many scientific fields. It has been applied in many scientific fields, such as image processing, pattern recognition, computer vision and signal analysis, with that a great deal of productions, which have a significant scientific meaning and applied value, were obtained. This is mainly attributed that wavelet analysis has many eximious characteristics that are absent by other signal processing methods, such as information extraction, multi-distinguish and direction selection. Application of wavelet analysis and its combination with neural network can improve the sensitivity and distinguish ratio in analytical signal processing and accelerate the development of analytical chemistry. The thesis is divided into three chapters, and the author's contributions are mainly focus on the follows:1 . Based on the different characteristics of wavelet coefficients of signal and noise after wavelet transform, three denoising methods are put forward and the comparison of these methods is studied in principle and application.(1) A dyadic wavelet transform modulus maximum de-noising method, which is based on the fact that the characteristics of wavelet transform modulus maximum of real signals are distinctively different from those of noise, is founded. The method is applied in capillary electrophoresis signals, and the results show that this method can pick up real signal and de-noise effectively.(2) A self-adaptive filter method, which is based on the different characteristics of signal and noise in different wavelet scales, is founded. The method is directly comparing the unified correlative value of wavelet coefficients in the neighboring scales and wavelet coefficients to determine whether the data point should be kept. And the standard deviation 6 m of noise at each scale is employed to stop the iterating. The method is used to the de-noising processing of capillary electrophoresis signals. And the results show that it is a better adaptive filter method and can be usedIto eliminate the noise effectively.(3) Based on the wavelet analysis, a novel de-noising method - threshold filtering based on the region relativity of wavelet coefficients is founded and used to the de-noising processing of the simulated signal and chromatogram. The experiment results shown that the noise contained in the original signal can be cleanly eliminated, and the characteristics of the signal can be preserved completely, furthermore the peak shape of the de-noised signal is ideal as well as the smooth baseline can be obtained.(4) Based on the principle of wavelet analysis, the principle of the threshold methods, wavelet modulus maximum method and the threshold filtering method based on the region relativity, are described, and then the de-noised results of simulated signals of these methods are compared. In conclusion, the applicable range of these five methods is put forward, this work can offer a reference for how to use the denoising methods based on the wavelet analysis accurately, rapidly and effectively.2. Combination of wavelet transform with neural network can be divided into two modes ?assistant combination and nested combination. In the first combination method, the wavelet transform is firstly used in the extraction of useful information and data compression, and then the compressed data is used as the input of the neural network. The method can enhance the prediction accuracy and convergence rate of neural network effectively. In the nested combination method, the wavelet basis function is used as the activation function of the neural node and a new method -wavelet neural network is formed. The method possesses doughty capacity of approximate and robust. Two methods are both used in the oscilligraphic determination and the experimental results show that the assistant combination method needs less training time and has lower prediction error, while the wavelet neural network has a higher automatiza...
|
PDF Full Text Request