| 摘要 |
PROBLEM TO BE SOLVED: To improve the accuracy of removal when removing noise in a waveform using a probability model.SOLUTION: Adopted as a probability model in a noise removal device which, when estimating the waveform at each time of day in accordance with a probability model using time-series data comprising data representing the waveform at each time of day, removes a waveform component whose occurrence probability is low in the estimation from the time-series data as noise, is a model that includes a visible element corresponding to the waveform at each time of day and a hidden element that makes a pair with the visible element, and also includes recurrence elements in forward and reverse directions that realize a transition between visible elements that are in a row in the time-series direction, a transition between hidden elements that are in a row in the time-series direction, or a transition between visible and hidden elements at adjacent times of day. The data at each time of day is correlated to the visible elements at each time of day, and the waveform at each time of day is estimated from the past visible and hidden elements earlier than the time of day and the future visible and hidden elements ahead of the time of day in accordance with the transition indicated by the recurrence element.SELECTED DRAWING: Figure 1 |
