Coded data generation or conversion – Digital code to digital code converters
Reexamination Certificate
2000-02-01
2001-08-28
Young, Brian (Department: 2819)
Coded data generation or conversion
Digital code to digital code converters
C367S099000, C702S191000, C702S035000
Reexamination Certificate
active
06281814
ABSTRACT:
TECHNICAL FIELD
The present invention relates to a data conversion method and apparatus which are used to extract characteristic portions from a time-series signal obtained for analysis, and a program storage medium.
BACKGROUND ART
When a wavelet transform is performed for a time-series signal representing vibrations, sounds, process data, or the like by using a complex type wavelet function, information about a time-frequency domain can be obtained. When intensities (absolute values) of this time-frequency domain information are computed, changes in various frequency characteristics contained in the time-series signal over time can be analyzed.
There is a technique (reference
1
: Japanese Patent Laid-Open No. 7-271763) of differentiating time-frequency information as a wavelet transform result in units of scales, extracting extreme values as feature amounts, and using them for analysis and diagnosis.
There is a technique (reference Japanese
2
: Patent Laid-Open No. 8-83265) of extracting feature amounts representing the periodicity of a signal, in units of scales, from time-frequency information as a wavelet transform result serving as a target signal and using them for analysis and diagnosis.
There is a technique (reference
3
: Japanese Patent Laid-Open No. 8-219955) of computing statistical amounts such as averages and variances from a wavelet transform result serving as a target signal in units of scales, extracting them as features, and using them for diagnosis.
There is a technique (reference
4
: Japanese Patent Laid-Open No. 8-177530) of comparing a wavelet transform result serving as a target signal with a predetermined threshold in units of scales, extracting values exceeding the threshold as features, and using them for diagnosis.
There is a technique (reference
5
: Japanese Patent Laid-Open No. 8-329046) of computing variances in units of scales from a wavelet transform result serving as a target signal, extracting the peaks of the obtained variance distributions as features, and using them for analysis.
DISCLOSURE OF INVENTION
[Problem to be Solved by the Invention]
In the conventional methods of extracting feature amounts from a time-series signal which appear in a wavelet transform result, a person visually determines feature amounts or detects them by using a threshold (reference
4
). According to these methods, however, it is difficult to discriminate such feature amounts from unnecessary feature amounts. In some case, therefore, desired feature amounts cannot be extracted.
When extreme values are extracted as feature amounts in units of scales from a wavelet transform result (reference
1
), desired features may not be extracted.
If the feature that can be obtained from a wavelet transform result is limited to periodicity (reference
2
), other useful features cannot be extracted.
If a statistical amount is used as a feature (references
3
and
5
), an overall given domain having a temporal width is expressed by one feature amount. Hence, a detailed feature per unit time cannot be extracted.
Assume that vibration components are to be extracted from the time-series signal shown in FIG.
10
. In this case, if a wavelet transform of the time-series signal is performed by using a complex type wavelet function to compute the intensity of the signal, the intensity signal shown in
FIG. 11
can be obtained.
FIG. 11
shows the wavelet intensity signal obtained by expressing the intensity of each signal component with luminance such that portions having higher intensities become brighter, and portions having lower intensities become darker. With a wavelet intensity signal like the one shown in
FIG. 11
, a stepped waveform and vibration waveform can be analyzed at the two axes, i.e., the frequency axis (ordinate) and time axis (abscissa).
Even if, a signal (data) is converted in this manner, it is not always easy to perform quantitative discrimination because of a large information amount as compared with the original signal, limitations in terms of resolution, and the like.
According to the technique in reference
4
, if values exceeding a predetermined threshold are extracted as features from the wavelet intensity signal in
FIG. 11
, the extraction result shown in
FIG. 12
can be obtained. With this technique, however, vibration components are difficult to extract.
More specifically, the method of setting a threshold, which is disclosed in reference
4
, is effective when it is found that only desired features exceed the threshold with fail. However, such a case is likely to occur less frequently. If desired features do not exceed the threshold or other features exceed the threshold owing to various factors, it is difficult to perform discrimination. In addition, it is difficult to set a threshold itself. As described above, according to the result shown in
FIG. 12
which is extracted with a threshold, the step portion of the original time-series signal cannot be satisfactorily discriminated from the vibration portion.
According to the technique disclosed in reference
2
, peaks (extreme values) of changes over time are detected in units of scales from a wavelet transform result. If peaks periodically appear, the distances between the peaks are extracted as the periodicity of the original signal. According to the technique in reference
2
, the extraction result shown in
FIG. 13
can be obtained from the wavelet intensity signal shown in FIG.
11
.
According to the extraction result in
FIG. 13
, however, no vibration component is extracted. That is, the technique in reference
2
cannot extract vibration components.
According to the technique in reference
1
, a wavelet transform result serving as a target signal is differentiated in units of scales to extract the extreme values of changes over time as feature amounts in units of scales. Since extreme values include maximum and minimum values, according to the technique in reference
1
, minimum values are extracted in units of scales from the wavelet intensity signal in
FIG. 11
, in addition to the maximum values extracted in units of scales as shown in FIG.
13
.
Since the minimum values extracted from the wavelet intensity signal in
FIG. 11
in units of scales are present in the dark portions in
FIG. 11
, i.e., the portions with low intensities, the extracted features include meaningless features. This makes it difficult to extract vibration components.
As disclosed in references
3
and
5
, when a statistical amount is extracted as a feature, the features of a domain having a temporal width are expressed by a typical value termed a statistical amount. As a consequence, the temporal feature of the domain is lost. For example, according to the technique in reference
5
, variance values are extracted as feature amounts in units of scales from a wavelet transform result. In this technique in reference
5
, the result shown in
FIG. 14
can be obtained from the wavelet intensity signal shown in FIG.
11
. According to the result shown in
FIG. 14
, although the scale of vibration components, i.e., the frequency, can be specified, temporal information is lost. For this reason, information about the occurrence time, duration, and the like of vibration components cannot be obtained.
As described above, according to the conventional techniques, step components cannot be discriminated from vibration components, or temporal information is lost, in particular, although the existence of vibration components can be checked, because the feature amount of vibration components can be extracted only partly. For this reason, features cannot be quantitatively grasped.
The present invention has been made to solve the above problems, and has as its object to quantitatively grasp changes in various frequency features, with higher precision, which are contained in a time-series signal and obtained by performing a wavelet transform of the time-series signal.
[Means of Solution to the Problem]
The present invention has been made to solve such problems. According to the first method of the present inv
Kazato Hirohiko
Noma Takashi
Sasaoka Hideki
Blakely Sokoloff Taylor & Zafma
Nguyen John
Yamatake Corporation
Young Brian
LandOfFree
Data conversion method, data converter, and program storage... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Data conversion method, data converter, and program storage..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Data conversion method, data converter, and program storage... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2547535