Data Feature Awareness Of FECG Extraction Algorithm

Fetal Electrocardiogram (FECG) monitoring is very important to ensure pregnant woman and fetal health. The traditional approach of FECG extraction is often operated by invasive method, which not only waste much time and energy but also is harmful to pregnant woman and fetal health. Recently, people usually extract FECG by abdominal electrodes which is a non-invasive and friendly method. However, due to the complex environment, the electrode is interfaced by expectant mother’s electrocardiogram (ECG) and other bio-electricity signal. So the extracted signal often is a mixed signal by FECG, expectant mother’s electrocardiogram and other noise. Thus signal processing is needed to get clear FECG.How to extract FECG is a hot point in biological signal processing field. Recently years, scholars proposed many methods to extract FECG, such as blind source separation (BSS), adaptive noise cancellation (ANC), singular value decomposition (SVD) and learning machine (LM).Support Vector Machine (SVM) is a new kind of learning machine based on statistical learning theory, its kernel content is based on structural risk minimization principle. SVM can be used in solving small sample problem, nonlinear problem and high dimensionality problem efficiently. By the way, SVM overcomes the over-fitting and under-fitting problem in machine learning algorithm.In this thesis, we combined the previous project background and the research problem and have done the following research based on the predecessor’s work:1) Project background of FECG extraction problem is presented, the significance of FECG is proposed, and we indicated the characteristic of FECG, noise source analysis is included; 2) We analyzed statistical learning theory and the principle of SVM which including three important concepts:the generalized optimal classification plane、kernel function feature space and support vector regression machine; 3) Aiming at solving the kernel function parameter optimization of least square support vector regression machine(LSSVM), we indicated a modified optimization algorithm based on glowworm swarm optimization(GSO) to replace the original grid research method(GSM). A mixed kernel function is proposed to replace the single global kernels or local kernels; 4) Using the modified LSSVM to solve the FECG extraction problem. We analyzed the mathematical model of this problem and got the solving steps. We do the extraction experiment on clinic data and compared the effect with the unimproved LSSVM method and ANC method to verify effectiveness of the modified LSSVM; 5) Finally pointed out the research deficiency and the future research direction.