Research On Independent Function Element For Single Channel Mixed Signal Separation And Signal Characterization

In this dissertation, a research has been made on an new signal statistical represent-ation for solving the problem that is single channel mixed signal separation and signal characterization.The single channel mixed signal exists widely in actual scientific research and engineering applications due to some constraints, whose processing and analysis methods usually are that separate the mixed signal and extract interest sources, then characterize them. The conventional signal separation is achieved by means of prior knowledge, but as the prior is limited or even no, the traditional methods can not solve the problem of signal separation effectively. In addition, since the feature vector obtained from signal especially non-stationary signal does not meet the statistical independence, there is a certain degree of redundancy in the signal characterization, so it is necessary to design complex classifier for classification in application. Sometimes it can not get satisfactory results even wrong judgment. Therefore, this paper takes these two issues as a starting point, and combining the fundamental theory of statistical signal processing and features of classical linear transformation technology, proposes a new signal decomposition and reconstruction model, that is independent function element transformation, further presents the method for single channel mixed signal separation and signal characterization on the basis of independent function element, which focuses on the application methods of independent function element in different projects, especially in the research field including under-determined blind source separation, feature extraction and pattern recognition.The main works of this dissertation are summarized as follows:Firstly, as inheriting the advantages of traditional linear transformation in time and frequency domain such as Wavelet transform and Hilbert-Huang Transform, the signal decomposition and reconstruction method in the statistical domain is put forward based on independent function element. The signal components obtained by independent function element transformation can mostly reduce correlation and meet the statistical indepen-dence. The independent function element model, definition as well as the selection of independence criterion are mainly discussed, furthermore, the nature and application area of independent function element are analyzed, especially the detailed theoretical analysis involves dimension expansion of single channel signal and signal characterization are carried out in the process of signal analysis.Secondly, the specific acquisition method of independent function element is investigated based on theory model above mentioned, the mechanism of signal hierarchy, the decision criteria of hierarchical number, the conditions for independent function element transformation, and determination of weighting coefficients are deeply studied. The hierarchical technique and selection of independent component analysis are mainly discussed in the process of independent function element transformation, then some experiments are conducted for comparative analysis, in which the equal length hierarchical technique by using improved cyclic convolution decomposition is proposed, and a new algorithm for determining the optimal hierarchical number is presented which does not rely on any prior knowledge.Thirdly, a novel under-determined blind source separation algorithm for single channel mixed signal is proposed, which does not require any prior knowledge. And the effectiveness and feasibility are validated by some experiments; From the experiment, we can see the two separation results are similar, but the under-determined blind separation method based on independent function element is on the condition of independence, which is easily met comparing with sparsity assumption. As a useful complement, the under-determined blind separation algorithm based on independent function element can enhance the applicability in certain extent.Fourthly, then on this basis, the application of independent function element in physiological signal processing is studied. Take heart sound signal for example, obtain independent function element of heart sound signal(HSIFE), and achieve feature extraction as well as classification; Finally, apply the HSIFE in vehicle active safety for monitoring the driver’s physiological status. In the specific implementation process, on the basis of the characteristics between heart sound signal and background sound in car, a PCG acquisition devices is designed and heart sound signal model in automotive environment is presented, and a kind of heart sound signal classification method is given based on HSIFE. In which how to take the HSIFE as a new physiological feature for characterizing the heart sound is mainly discussed.Fifthly, this paper studies the application of independent function element in the field of two-dimensional image understanding. According to the law of visual cognition and the theory of independent function element transformation, expand the method of independent function element transformation to the field of image processing. By analyzing the three function elements of the wild scene images:ground, vertical objects, sky, give the equations for outdoor scene image, then propose rapid image classification method to realize the wild scene image segmentation based on independent function element transformation,Lastly, further the method of the water body identification in the ground is presented. In which the water illumination model in outdoor scene is mainly discussed, the relevant physical characteristics of the wild scene is analyzed and some new concepts are defined such as water quality parameters, environmental dyeing parameters, independent image element and so on, finally present the wild scene analysis method and water body identification technology by using data fusion algorithm including location, texture, geological factors and environmental dyeing parameters and other parameters.This article puts emphasis on the separation method of single channel mixed signal and a new form of signal characterization based on independent function element, and in-depth research is conducted on its experimental application such as under-determined blind source separation, driver’s heart health monitoring as well as wild scene image understanding and get some positive results. For the promotion of signal processing method, this work has important engineering applications and can be widely used in blind source separation, artificial intelligence and biometric identification technology and other areas.