| Popis: |
A new system of peak component recognition and measurement in digitized waveforms is detailed. Two input parameters identify waveform context (scale and noise content), and a third specifies baseline tolerance (if applicable). The input waveform is preprocessed by a discrete linear piecewise approximation algorithm yielding a segmentation in endpoint/slope/constant format. Slope values are encoded as symbols of a string which is parsed by a syntax-directed finite-state automaton. One of three different machines and sets of semantic routines is chosen depending upon whether the waveform is baseline-free, unipolar, or bipolar. In the last case, the endpoint values relative to the baseline are encoded as symbols of a second string which modifies the action of the machine. An electrocardiogram is employed as an example waveform to demonstrate the bipolar algorithm, which runs with sufficient speed to allow real-time processing. A proposed on-line implementation of the system is outlined. |
