CN117725521A - Majority vote based classification method with adaptive voting time window length - Google Patents
Majority vote based classification method with adaptive voting time window length Download PDFInfo
- Publication number
- CN117725521A CN117725521A CN202311750179.8A CN202311750179A CN117725521A CN 117725521 A CN117725521 A CN 117725521A CN 202311750179 A CN202311750179 A CN 202311750179A CN 117725521 A CN117725521 A CN 117725521A
- Authority
- CN
- China
- Prior art keywords
- time window
- voting
- classification
- length
- switching
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 19
- 230000003044 adaptive effect Effects 0.000 title claims description 6
- 238000001514 detection method Methods 0.000 claims abstract description 28
- 230000035945 sensitivity Effects 0.000 description 5
- 238000010586 diagram Methods 0.000 description 2
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides a classification method based on majority vote with self-adaptive voting time window length, comprising the following steps: initializing the voting time window length and switching detection time window length; classification: the classifier outputs a classification result every other time unit and increases the length of the voting time window; classifying based on majority voting in the current voting time window, and outputting classification results; judging whether a switching detection time window exists in the voting time window, if so, judging whether category switching occurs in the switching detection time window, if so, resetting the length of the voting time window to an initial value, and returning to the classification step, if not, directly returning to the classification step; the invention realizes the self-adaptive adjustment of the voting time window length, reduces the system delay caused by majority voting in the class switching process, improves the accuracy of the majority voting method, and is suitable for signal classification tasks with time dependence of voice signals, physiological signals and the like.
Description
Technical Field
The invention relates to a signal classification technology, in particular to a classification technology based on majority voting.
Background
In some signal classification tasks with time dependency, such as some applications of speech recognition classification, gesture classification, etc., the single classification result output by the classifier is often lower in accuracy, and the main reason is that the single classification result is from isolated analysis of a small segment of signal, and the time dependency of the signal is not fully utilized, so that the confidence is also lower.
Majority voting is a typical means of handling data stream classification. For the classification task with time dependency, as the classification proceeds with time, a voting time window can be defined for the classification result of each time unit output by the classifier, and the majority vote is performed in the voting time window, and then the result pointed by the majority vote is used as the final classification result. The majority voting can remarkably improve the confidence coefficient of the classification result, improves the classification accuracy, and has wide application in the post-classification processing algorithm.
For classification processing of signals with time dependency relationship such as voice signals and biological signals, a voting time window defined by a current majority voting method is always of a fixed window length, and in general, the longer the window length of the voting time window is, the higher the accuracy of classification results is, and the higher the classification confidence is. For example, the gesture motion of the human body is classified and identified by collecting the electromyographic signals, and the electromyographic signals of each time unit have one motion classification, in most cases, the human body cannot switch the motion at high frequency, and the human body can keep one gesture motion for a period of time, but a plurality of time units in the period possibly have some other motion classifications due to interference. When a human body acts in a continuous gesture, when the voting time window is longer, the error caused by interference can be easily eliminated under the condition that the classification result of most time units is correct. However, if the voting time window is fixed for a long time, when the gesture motion is switched, the motion classification corresponding to the new motion may have a delay in switching the final result due to the previous vote of the time unit of the most old motion. Extending the voting time window blindly can lead to a decrease in system sensitivity during class switching. Therefore, the majority voting method with a fixed voting time window length is difficult to realize high confidence and high sensitivity at the same time.
Disclosure of Invention
The invention aims to solve the technical problem of providing a classification method based on majority vote, which is used for considering confidence and sensitivity by flexibly adjusting the voting time window length.
The invention adopts the technical proposal that the classifying method based on majority vote with self-adaptive voting time window length comprises the following steps:
initializing: initializing the voting time window length and switching detection time window length;
classification: the classifier outputs a classification result every other time unit and increases the length of the voting time window by one time unit; classifying based on majority voting in the current voting time window, and outputting classification results;
and a switching detection time window detection step: judging whether a switching detection time window exists in the voting time window, if so, entering a switching detection step, if not, returning to a classification step; the condition that the switching detection time window exists in the voting time window is that the current voting time window is longer than the switching detection time window, and the switching detection time window is determined by retracting one switching detection time window from the current time unit and ending the switching detection time window;
and a switching detection step: judging whether category switching occurs in the switching detection time window, if so, resetting the voting time window length to an initial value, and returning to the classification step; if not, directly returning to the classification step.
Based on the traditional majority voting method for the length of the fixed voting time window, the invention carries out class switching detection by adding the switching detection time window, thereby realizing the self-adaptive adjustment of the length of the voting time window. When the method does not detect the class switching, the classification result with high accuracy and high confidence is obtained by prolonging the voting time window length. And when the class switching is detected, the voting time window is reset, so that the system is ensured to have high response sensitivity.
The invention has the advantages that the length of the voting time window is flexibly adjusted according to the actual classification task, the system delay caused by majority voting in the class switching process is reduced, the accuracy of the majority voting method is improved, and the invention is suitable for the signal classification task with time dependency relationship such as voice signals, physiological signals and the like.
Drawings
FIG. 1 is a schematic diagram of adaptive adjustment of the voting time window length;
FIG. 2 is a flow chart of a classification result majority voting method workflow with adaptive voting time window length.
Detailed Description
The invention adds a switching detection time window in the traditional voting time window, and detects class switching through the switching detection time window, thereby adaptively adjusting the voting time window length. A schematic diagram of adaptive adjustment of the voting time window length is shown in FIG. 1, and an overall method workflow is shown in FIG. 2.
S0, initializing voting time window length t 1 Switching detection time window length t 2 Switching threshold Th, taking the interval time of the classifier outputting the primary classification result as a time unit tau; t is t 1 ＝0，t 2 And the set value of Th is determined according to the requirement of the system sensitivity;
s1, outputting a classification result by the classifier every other time unit tau, and updating the length t of a voting time window 1 ＝t 1 +τ, using the classification result of the current time unit as one of the voting items in the voting time window;
s2, according to the voting time windowMajority voting is carried out on voting items in a voting time window, namely each classification result is one voting item, wherein n voting items correspond to the classification result c k Are all identical and sort results c in a voting time window k The votes of (a) are the most numerous than the other classification results, i.e. classification result c k The most votes are obtained, and the classification result c with the most votes is obtained k Outputting the final classification result as the current time unit;
s3, judging the length t of the current voting time window 1 Whether or not it is greater than the switch detection time window length t 2 If yes, set up [ t ] in voting time window 1 -t 2 ,t 1 ]The segment window is a switching detection time window, and then step S4 is carried out; if not, returning to the step S1;
s4, counting the classification result c with the most ticket obtained in the current switching detection time window k Number of corresponding voting itemsJudging->If the current category is larger than the switching threshold Th, if so, the current category is considered to be not switched, and the step S1 is returned; if not, the current category is considered to have been switched, and the voting time window length t is restarted 1 And (5) setting zero, and returning to the step S1.
Claims (3)
1. A method of classifying a majority vote with an adaptive voting time window length, comprising the steps of:
initializing: initializing the voting time window length and switching detection time window length;
classification: the classifier outputs a classification result every other time unit and increases the length of the voting time window by one time unit; classifying based on majority voting in the current voting time window, and outputting classification results; judging whether class switching occurs in the switching detection time window, if so, resetting the voting time window length to an initial value, and classifying based on majority voting; if not, classification based directly on majority vote.
2. The method of claim 1, wherein the specific method for determining that the class switch occurs within the switch detection time window is: the initialization step also sets a switching threshold;
in the step of switching detection, judging whether the number of the current classification results in the switching detection time window is larger than a switching threshold value, if so, not switching the category, and if not, switching the category.
3. The method of claim 1, wherein the initial value of the voting time window length is 0.
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311250854 | 2023-09-26 | ||
| CN2023112508540 | 2023-09-26 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117725521A true CN117725521A (en) | 2024-03-19 |
Family
ID=90208504
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311750179.8A Pending CN117725521A (en) | 2023-09-26 | 2023-12-19 | Majority vote based classification method with adaptive voting time window length |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117725521A (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP3492945A1 (en) * | 2017-12-01 | 2019-06-05 | Origin Wireless, Inc. | Method, apparatus, and system for periodic motion detection and monitoring |
| US20190303713A1 (en) * | 2018-03-30 | 2019-10-03 | Regents Of The University Of Minnesota | Discovery of shifting patterns in sequence classification |
| CN110363157A (en) * | 2019-07-17 | 2019-10-22 | 杭州电子科技大学 | Ectoskeleton mixing brain-computer interface control method based on time encoding |
| CN112099619A (en) * | 2020-08-11 | 2020-12-18 | 东南大学 | Time window length self-adaptive selection method of mixed sight brain-computer interface |
| CN113314209A (en) * | 2021-06-11 | 2021-08-27 | 吉林大学 | Human body intention identification method based on weighted KNN |
-
2023
- 2023-12-19 CN CN202311750179.8A patent/CN117725521A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP3492945A1 (en) * | 2017-12-01 | 2019-06-05 | Origin Wireless, Inc. | Method, apparatus, and system for periodic motion detection and monitoring |
| US20190303713A1 (en) * | 2018-03-30 | 2019-10-03 | Regents Of The University Of Minnesota | Discovery of shifting patterns in sequence classification |
| CN110363157A (en) * | 2019-07-17 | 2019-10-22 | 杭州电子科技大学 | Ectoskeleton mixing brain-computer interface control method based on time encoding |
| CN112099619A (en) * | 2020-08-11 | 2020-12-18 | 东南大学 | Time window length self-adaptive selection method of mixed sight brain-computer interface |
| CN113314209A (en) * | 2021-06-11 | 2021-08-27 | 吉林大学 | Human body intention identification method based on weighted KNN |
Non-Patent Citations (2)
| Title |
|---|
| LEE T: ""Adaptive Window Method Based on FBCCA for Optimal SSVEP Recognition"", 《IEEE TRANS NEURAL SYST REHABIL ENG》, 28 October 2022 (2022-10-28), pages 1 - 15 * |
| 张英杰: ""基于半定长滑动窗口数据的供水管网漏损检测"", 《湖南大学学报(自然科学版)》, 31 October 2022 (2022-10-31), pages 1 - 8 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US20190244062A1 (en) | Gesture recognition method, gesture recognition system, and performing device therefore | |
| JP2019133639A5 (en) | ||
| WO2019200782A1 (en) | Sample data classification method, model training method, electronic device and storage medium | |
| Yeck et al. | Leveraging deep learning in global 24/7 real‐time earthquake monitoring at the National Earthquake Information Center | |
| US20080126556A1 (en) | System and method for classifying data streams using high-order models | |
| CN101814147A (en) | Method for realizing classification of scene images | |
| Berchtold et al. | An extensible modular recognition concept that makes activity recognition practical | |
| Rasolzadeh et al. | Response binning: Improved weak classifiers for boosting | |
| EP3624113A1 (en) | Apparatus for processing a signal | |
| CN116821809A (en) | Vital sign data acquisition system based on artificial intelligence | |
| CN112652320A (en) | Sound source positioning method and device, computer readable storage medium and electronic equipment | |
| US8326457B2 (en) | Apparatus for detecting user and method for detecting user by the same | |
| CN117725521A (en) | Majority vote based classification method with adaptive voting time window length | |
| CN116895286B (en) | Printer fault monitoring method and related device | |
| EP3847646A1 (en) | An audio processing apparatus and method for audio scene classification | |
| US8560469B2 (en) | Method for a pattern discovery and recognition | |
| CN115083439A (en) | Vehicle whistling sound identification method, system, terminal and storage medium | |
| US6115702A (en) | Automatic determination of report granularity | |
| CN112308153A (en) | Smoke and fire detection method and device | |
| JPH0843520A (en) | Pulse-signal sorting apparatus | |
| CN113033615B (en) | Radar signal target real-time association method based on online micro-cluster clustering | |
| Tumer et al. | A framework for estimating performance improvements in hybrid pattern classifiers | |
| CN111340174A (en) | Adjusting method, device, equipment and medium of cost sensitive neural network | |
| JP7335379B1 (en) | LEARNING APPARATUS, LEARNING METHOD, AND PROGRAM | |
| JP7335378B1 (en) | Message classifier, message classifier method, and program |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |