Research On The Key Technology Of Fetal Movement Detection Model Building

Fetal movement signal is an important fetal health indicator in the uterus. Ultrasound and maternal perception are mainly used in the hospital and home, but the expense and inaccuracy impacts the user's experience. Researchers and institudes have tried some methods to detect fetal movements, but most of them do not achieve the desired results. With the development of the Micro-Electro-Mechanical System, accelerometers become smaller, more sensitive, lower cost and comsumption, thus, more and more researchers focus on the fetal movement detection based on accelerometers.Now, some teams have already achieved some conclusions on the fetal movement dectection, but still have some problems need to be solved, such as how to build a platform to acquire fetal movement signal quickly and comfortable, how to preprocess the raw signal with lots of noisy and maternal movements, how to build a model to recognize the fetal movement signal, etc. To solve these problems, the subjects in this thesis can be concluded as follows:1) A wearable platform is built to acquire fetal movement signal with acclerometers sewed on the pregnancy belt. A MCU samples and uploads the signal to the smart devices with wireless technology. Thus, the acquisition can collect the fetal movement signal quickly and conveniently.2) Noisy and maternal movement signal features are studied in this thesis. A method is proposed to filter the raw signal with band-pass filter based on window functions. Wavelet denoising is also used to filter the signal to increase the signal noisy rate with the right wavelet basis, decomposition layers and thresholds.3) Some quadratic time-frequency distributions are studied in this thesis, a wigner-ville distribution with modified B distribution kernel function is used to analyse the fetal movement and maternal movement signal in the time frequency domain. Two methods with template and signal sparse decomposition are proposed to extract features in the time frequency domain. The simulation results show that these two algorithms can both effectively extract the features of the fetal movement in the time frequency domain, but they have different performance in recoginition and computing time.4) Besides time frequency features, some time features are studied. Signal energy, standard deviations, maxinum offset and kurtosis are used to recognize the fetal movement signal. A model is built with these features. Accoring to the observation and clinical experience, a series of thresholds are set for the model. The simulation results show that the fetal movement detection method based on accelerometers can achieve a effective recognition. Compared with the traditional methods, BP neural network, the model has a similiar recognition, but increase a lot in accuracy and reduces the dependency on trainning samples.