| Summary: | 博士 === 國立高雄第一科技大學 === 工學院工程科技博士班 === 103 === Electrocardiogram (ECG) signal analysis is a noninvasive modality widely used for heart disease diagnosis. With high sensitivity, real-time reporting the use of a multi-lead ECG signal is a necessary measure in prevention-oriented healthcare and elderly care. An efficient option in this regard would be equipping patients with a wearable and wireless body sensor network (WBSN). However, WBSN has the problem of short stand-by. There are two methods commonly used for solving this problem: 1) emerging compressed sensing signal, and 2) transmitting transformed encoded and quantized data. Both methods involve transform-based ECG data compression. The former directly plugs the quantization scale into transform coefficients. Since quantization scales cannot be adjusted, this method does not satisfy the requirement of guaranteeing reconstruction quantity. The latter chooses the best quantization scale by a quality guarantee mecha-nism (QGM), and can maintain reconstruction quality stability, but at a time-consuming cost and complex hardware. In the transform domain, the wavelet-based compression method has the best compression performance
The transform-based quantization process is part of the frequency domain, rendering a coefficient to each sub-band, and the reconstruction quality is processed in the time domain. In regard to this method, we proposed a non-linear quantization algorithm by adjusting the single quantization factor (QF) based on genetic algorithm and curve fit-ting. The goal of this algorithm is to maintain a linear relation between the distortion and compression ratio (CR). This linear relationship is not only based on improving the com-pression performance of the whole system, but also building a linear distortion predic-tion model. In this dissertation, a fuzzy decision-making method is used for minimizing the data dependency effect; for this reason, the training data should be as diverse as possible. To this end, four training datasets (TD#1, TD#2, TD#3 and TD#4) were built. A fuzzy decision-making method under multiple criteria consideration is needed to in-tegrate various linguistic assessments and weights to evaluate and determine the best selection. With minimum dynamic range (MDR), we proposed a reversible round-off non-recursive wavelet transform in the transform section. Finally, a modified set parti-tioning in hierarchical trees (MSPIHT) algorithm is proposed in the lossless coding. MSPIHT uses the bit-plane and is a suitable method to implement on hardware. The performance evaluation is based on the percentage root-mean-square difference (PRD) and visual inspection of the reconstructed signals. The experimental results show that our proposed method can obtain superior compression performance compared to SPIHT.
|
|---|
