CN109740635B - Classification recognition feature mapping method based on two classifiers - Google Patents
Classification recognition feature mapping method based on two classifiers Download PDFInfo
- Publication number
- CN109740635B CN109740635B CN201811512493.1A CN201811512493A CN109740635B CN 109740635 B CN109740635 B CN 109740635B CN 201811512493 A CN201811512493 A CN 201811512493A CN 109740635 B CN109740635 B CN 109740635B
- Authority
- CN
- China
- Prior art keywords
- msf
- characteristic
- dimension
- feature
- classification
- 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.)
- Active
Links
- 238000013507 mapping Methods 0.000 title claims abstract description 30
- 238000000034 method Methods 0.000 title claims abstract description 25
- 238000012549 training Methods 0.000 claims abstract description 13
- 238000005315 distribution function Methods 0.000 claims description 8
- 102100035373 Cyclin-D-binding Myb-like transcription factor 1 Human genes 0.000 claims description 5
- 101000804518 Homo sapiens Cyclin-D-binding Myb-like transcription factor 1 Proteins 0.000 claims description 5
- 238000000605 extraction Methods 0.000 claims description 3
- 238000006243 chemical reaction Methods 0.000 claims description 2
- 238000012545 processing Methods 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 3
- 230000007613 environmental effect Effects 0.000 description 3
- 230000008859 change Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000004088 simulation Methods 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000008092 positive effect Effects 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000005236 sound signal Effects 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
Landscapes
- Geophysics And Detection Of Objects (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention relates to a classification and identification feature mapping method based on two classifiers, which comprises the following steps: respectively counting Gaussian distribution conditions of each dimension characteristic of various target characteristics for training classifier coefficients; respectively counting Gaussian distribution conditions of each dimension characteristic of various target characteristics under the condition that the characteristic deviation occurs relative to various target characteristics of training classifier coefficients; and respectively constructing a distribution mapping relation for each dimension characteristic in each classifier under the two conditions of characteristic deviation according to the Gaussian distribution condition obtained by statistics. The invention can avoid the situation that the classification recognition accuracy is greatly reduced due to the characteristic deviation.
Description
Technical Field
The invention relates to the technical field of digital signal processing, in particular to a classification recognition feature mapping method based on two classifiers.
Background
The classification recognition technology is an important digital signal processing technology and is widely applied to the technical field of digital signal processing. The feature extraction and optimization are the first step of the classification recognition technology and are the most important step, and the suitability of feature selection directly influences the design complexity of the whole recognition system and determines the recognition accuracy of the system. Common features mainly comprise time domain features, frequency domain features, parameterized model features, high-order spectrum features, time-frequency domain features, nonlinear features and the like, and particularly the time-frequency domain features play an increasingly important role in non-stationary signal processing, and a plurality of achievements are achieved. However, since the nonstationary signals in the actual signal processing process are affected by many factors, the signal characteristics of the signals of the same target type under different environmental conditions are also very different, for example, the characteristics of the earthquake signal can change along with the change of geological conditions, and the characteristics of the sound signal are greatly affected by wind. Therefore, the classification recognition system is very good in classification recognition effect under certain environmental conditions, and the performance is drastically reduced when the classification recognition system is changed to another environmental condition.
Disclosure of Invention
The invention aims to solve the technical problem of providing a classification and identification feature mapping method based on two classifiers, which avoids the situation that the classification and identification accuracy is greatly reduced due to feature deviation.
The technical scheme adopted for solving the technical problems is as follows: the classification recognition feature mapping method based on the two classifiers comprises the following steps:
(1) Respectively counting Gaussian distribution conditions of each dimension characteristic of various target characteristics for training classifier coefficients;
(2) Respectively counting Gaussian distribution conditions of each dimension characteristic of various target characteristics under the condition that the characteristic deviation occurs relative to various target characteristics of training classifier coefficients;
(3) And respectively constructing a distribution mapping relation for each dimension characteristic in each classifier under the two conditions of characteristic deviation according to the Gaussian distribution condition obtained by statistics.
And (3) adopting the Gaussian distribution situation counted by the following method in the step (1) and the step (2): defining the number of target features of one class as N, feature dimensionM, characteristic sequence F 1 ,...,F N (F i ＝[F i1 ,...,F iM ]) Then the kth dimension feature sequence is F 1k ,...,F Nk Calculating the mean value of the k-th dimension characteristic sequenceCalculating standard deviation of the kth dimension feature sequence
The step (3) specifically comprises the following substeps:
(31) Defining the mean value of the kth dimension characteristic sequences of two types of targets used for training classifier coefficients in one classifier as MTF k1 And MTF k2 Standard deviation is STDTF k1 And STDTF k2 The average value of the kth dimension characteristic sequences of the two types of targets under the condition of characteristic offset is MSF k1 And MSF k2 Standard deviation is STDSF k1 And STDSF k2 ；
(32) Define the x-axis flip parameter as dp= (MTF) k1 -MTF k2 )×(MSF k1 -MSF k2 ) If the x-axis flipping parameter DP < 0, then the x-axis is flipped, i.e. x= -x, MSF k1 ＝-MSF k1 ，MSF k2 ＝-MSF k2 Otherwise, directly entering the step (33);
(33) Constructing Gaussian distribution functions of kth dimension feature sequences of two types of targets according to the calculated mean and standard deviation
(34) Calculating a Gaussian distribution function DF 1 (x) With DF 2 (x) Is the intersection point x of (2) C ；
(35) Defining the distribution map region determination parameter as dmtf=sign (MTF) k1 -MTF k2 ) For example, distribution map region determination parameter DMTF X (x-x C ) The distribution mapping relation of the kth dimension characteristic is more than 0Otherwise, the k-th dimension characteristic distribution mapping relation is +.>
If the number of the calculated intersection points in the step (34) is greater than 1, selecting the intersection point to be in the MSF k1 、MSF k2 The x value between them serves as the demarcation point for the x axis.
Advantageous effects
Due to the adoption of the technical scheme, compared with the prior art, the invention has the following advantages and positive effects: according to the invention, the Gaussian distribution conditions of each dimension feature under the conditions of before and after feature deviation are respectively counted, then a distribution mapping relation is respectively constructed for each dimension feature in each classifier according to the distribution conditions obtained by counting, and the feature with the feature deviation is corrected through the constructed distribution mapping relation. The method can effectively correct the classification recognition features under the condition of feature deviation, thereby avoiding the occurrence of the condition that the classification recognition accuracy is greatly reduced due to the feature deviation.
Drawings
Fig. 1 is a flow chart of the present invention.
Detailed Description
The invention will be further illustrated with reference to specific examples. It is to be understood that these examples are illustrative of the present invention and are not intended to limit the scope of the present invention. Further, it is understood that various changes and modifications may be made by those skilled in the art after reading the teachings of the present invention, and such equivalents are intended to fall within the scope of the claims appended hereto.
The embodiment of the invention relates to a classification recognition feature mapping method based on two classifiers, which comprises the following steps: respectively counting Gaussian distribution conditions of each dimension characteristic of various target characteristics for training classifier coefficients; respectively counting Gaussian distribution conditions of each dimension characteristic of each type of target characteristic under the condition that characteristic deviation (non-time variation) occurs to each type of target characteristic relative to the training classifier coefficient; and respectively constructing a distribution mapping relation for each dimension characteristic in each classifier under the two conditions of before and after characteristic deviation according to the distribution condition obtained by statistics.
The method for counting Gaussian distribution conditions comprises the following steps: defining the number of target features as N, the feature dimension as M and the feature sequence as F 1 ,...,F N (F i ＝[F i1 ,...,F iM ]) Then the kth dimension feature sequence is F 1k ,...,F Nk Calculating the mean value of the k-th dimension characteristic sequenceCalculating standard deviation of the kth dimension feature sequence
When a distribution mapping relation is constructed, firstly defining the mean value of the kth dimension characteristic sequences of two types of targets used for training classifier coefficients in a classifier as MTF k1 And MTF k2 Standard deviation is STDTF k1 And STDTF k2 The average value of the kth dimension characteristic sequences of the two types of targets under the condition of characteristic offset is MSF k1 And MSF k2 Standard deviation is STDSF k1 And STDSF k2 And calculated according to the method described above. Next, the x-axis flip parameter is defined as dp= (MTF k1 -MTF k2 )×(MSF k1 -MSF k2 ) If the x-axis flipping parameter DP < 0, then the x-axis is flipped, i.e. x= -x, MSF k1 ＝-MSF k1 ，MSF k2 ＝-MSF k2 The method comprises the steps of carrying out a first treatment on the surface of the Then, constructing Gaussian distribution functions of the kth dimension feature sequences of the two types of targets according to the calculated mean and standard deviationAnd calculate the Gaussian distribution function DF 1 (x) With DF 2 (x) Is the intersection point x of (2) C If the calculated intersection point number is greater than 1, selecting the intersection point to be in MSF k1 、MSF k2 The x value between them serves as the demarcation point for the x axis. Finally, a distribution map region determination parameter is defined as dmtf=sign (MTF) k1 -MTF k2 ) Determination of parameters DMTF×, e.g. distribution map region(x-x C ) The distribution mapping relation of the kth dimension characteristic is more than 0Otherwise, the k-th dimension characteristic distribution mapping relation is
The invention is further illustrated by a specific example.
In this embodiment, the earthquake motion signal is used as a research object, and since the earthquake motion signal is greatly affected by the geological conditions, the characteristics of the earthquake motion signals generated under different geological conditions are greatly different, that is, the characteristic deviation phenomenon is generated, and the geological conditions are not changed with time. In a simulation experiment, a method of carrying out FFT conversion on signals and then solving a cepstrum is adopted to carry out feature extraction, a feature dimension M=511, a fuzzy classifier is adopted to carry out classification identification, a membership function of the fuzzy classification is a Gaussian function, classification identification targets are wheel vehicles and tracked vehicles, the signal features for training classification identification coefficients are assumed to come from a geological condition 0, and k=1, the number of the steps is equal to 511.
As the membership function of the fuzzy classification is a Gaussian function, the mean MTF of each dimension characteristic of two kinds of target characteristics for training the classifier coefficient can be obtained from the trained classification recognition coefficient k1 、MTF k2 And standard deviation STDTF k1 、STDTF k2 。
Calculating the mean value MSF of each dimension characteristic sequence of two types of targets of geological conditions with characteristic shift k1 、MSF k2 And standard deviation STDSF k1 、STDSF k2 。
Calculating parameters DP for determining whether each dimension x-axis is flipped k ＝(MTF k1 -MTF k2 )×(MSF k1 -MSF k2 ) According to DP k Whether or not less than 0 determines whether or not each dimension x-axis is flipped, i.e., x k ＝-x k ，MSF k1 ＝-MSF k1 ，MSF k2 ＝-MSF k2 。
Construction of Gaussian distribution function of each dimension characteristic sequence of two kinds of targetsAnd->
Calculating the distribution function DF of two kinds of targets in each dimension characteristic k1 (x k ) With DF k2 (x k ) Is the intersection point x of (2) Ck The calculated number of intersections may be greater than 1, and the intersections are selected to be in MSF k1 、MSF k2 X between k Value as x k The demarcation point of the shaft.
Calculating parameter DMTF of decision distribution mapping area in each dimension characteristic k ＝sign(MTF k1 -MTF k2 ) Each dimension feature builds a distribution mapping relation: such as DMTF k ×(x k -x Ck )＞0，Otherwise
And carrying out feature mapping on the classification recognition features with the feature offset according to the constructed distribution mapping relation, and then sending the mapped features into a classifier for recognition and classification.
The vibration signals of the wheeled vehicle and the tracked vehicle, which are collected under 3 geological conditions with characteristic deviation, are simulated by adopting a classification recognition method without characteristic mapping and a classification recognition method with characteristic mapping respectively, and the correct recognition rate of each type of targets under the 3 geological conditions of the two methods is counted, wherein the correct recognition rate is shown in a table 1.
Table 1 statistical table of simulation results
It is easy to find that the invention respectively calculates the Gaussian distribution conditions of each dimension feature under the conditions of before and after feature deviation, then respectively constructs a distribution mapping relation for each dimension feature in each classifier according to the distribution conditions obtained by calculation, and corrects the feature with the feature deviation through the constructed distribution mapping relation. The method can effectively correct the classification recognition features under the condition of feature deviation, thereby avoiding the occurrence of the condition that the classification recognition accuracy is greatly reduced due to the feature deviation.
Claims (3)
1. A classification and identification method for earthquake signals is characterized by comprising the following steps:
performing characteristic extraction by adopting a method of performing FFT conversion on signals and then solving a cepstrum;
inputting the extracted characteristics into a classifier to realize classification and identification of the earthquake motion signals;
the classifier corrects the offset characteristics through a classification and identification characteristic mapping method during classification and identification; the classification recognition feature mapping method comprises the following steps:
(1) Respectively counting Gaussian distribution conditions of each dimension characteristic of various target characteristics for training classifier coefficients;
(2) Respectively counting Gaussian distribution conditions of each dimension characteristic of various target characteristics under the condition that the characteristic deviation occurs relative to various target characteristics of training classifier coefficients;
(3) Respectively constructing a distribution mapping relation for each dimension feature in each classifier under the two conditions of before and after feature deviation according to the Gaussian distribution condition obtained by statistics, and specifically comprising the following steps:
(31) Defining the mean value of the kth dimension characteristic sequences of two types of targets used for training classifier coefficients in one classifier as MTF k1 And MTF k2 Standard deviation is STDTF k1 And STDTF k2 The average value of the kth dimension characteristic sequences of the two types of targets under the condition of characteristic offset is MSF k1 And MSF k2 Standard deviation is STDSF k1 And STDSF k2 ；
(32) Define the x-axis flip parameter as dp= (MTF) k1 -MTF k2 )×(MSF k1 -MSF k2 ) If the x-axis flipping parameter DP < 0, then the x-axis is flipped, i.e. x= -x, MSF k1 ＝-MSF k1 ，MSF k2 ＝-MSF k2 Otherwise, directly entering the step (33);
(33) Constructing Gaussian distribution functions of kth dimension feature sequences of two types of targets according to the calculated mean and standard deviation
(34) Calculating a Gaussian distribution function DF 1 (x) With DF 2 (x) Is the intersection point x of (2) C ；
(35) Defining the distribution map region determination parameter as dmtf=sign (MTF) k1 -MTF k2 ) For example, distribution map region determination parameter DMTF X (x-x C ) The distribution mapping relation of the kth dimension characteristic is more than 0Otherwise, the k-th dimension characteristic distribution mapping relation is
2. The method of claim 1, wherein the step (1) and the step (2) are both performed by using gaussian distribution statistics by the following method: defining the number of target features as N, the feature dimension as M and the feature sequence as F 1 ,...,F N (F i ＝[F i1 ,...,F iM ]) Then the kth dimension feature sequence is F 1k ,...,F Nk Calculating the mean value of the k-th dimension characteristic sequenceCalculating standard deviation of the kth dimension feature sequence
3. The method of classification and identification of seismic signals according to claim 1, wherein if the number of calculated intersections in step (34) is greater than 1, selecting the intersection to be in the MSF k1 、MSF k2 The x value between them serves as the demarcation point for the x axis.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811512493.1A CN109740635B (en) | 2018-12-11 | 2018-12-11 | Classification recognition feature mapping method based on two classifiers |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811512493.1A CN109740635B (en) | 2018-12-11 | 2018-12-11 | Classification recognition feature mapping method based on two classifiers |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109740635A CN109740635A (en) | 2019-05-10 |
| CN109740635B true CN109740635B (en) | 2023-12-26 |
Family
ID=66358694
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811512493.1A Active CN109740635B (en) | 2018-12-11 | 2018-12-11 | Classification recognition feature mapping method based on two classifiers |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109740635B (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103544963A (en) * | 2013-11-07 | 2014-01-29 | 东南大学 | Voice emotion recognition method based on core semi-supervised discrimination and analysis |
| CN108647602A (en) * | 2018-04-28 | 2018-10-12 | 北京航空航天大学 | A kind of aerial remote sensing images scene classification method based on image complexity judgement |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8380647B2 (en) * | 2009-08-14 | 2013-02-19 | Xerox Corporation | Training a classifier by dimension-wise embedding of training data |
| US8429102B2 (en) * | 2011-03-31 | 2013-04-23 | Mitsubishi Electric Research Laboratories, Inc. | Data driven frequency mapping for kernels used in support vector machines |
-
2018
- 2018-12-11 CN CN201811512493.1A patent/CN109740635B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103544963A (en) * | 2013-11-07 | 2014-01-29 | 东南大学 | Voice emotion recognition method based on core semi-supervised discrimination and analysis |
| CN108647602A (en) * | 2018-04-28 | 2018-10-12 | 北京航空航天大学 | A kind of aerial remote sensing images scene classification method based on image complexity judgement |
Non-Patent Citations (2)
| Title |
|---|
| 水声目标识别中的K-D树KNN-SVM分类器研究;黄杰;朱广平;;海洋技术学报(01);全文 * |
| 用11/2维谱结合小波包能量提取地震动信号特征;张亚东;华春蓉;董大伟;闫兵;鲁志文;谢逍;;噪声与振动控制(01);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109740635A (en) | 2019-05-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111783558A (en) | Satellite navigation interference signal type intelligent identification method and system | |
| CN112287784B (en) | Radar signal classification method based on deep convolutional neural network and feature fusion | |
| CN110417694A (en) | A kind of modulation mode of communication signal recognition methods | |
| CN107123432A (en) | A kind of Self Matching Top N audio events recognize channel self-adapted method | |
| CN106501800A (en) | Based on tracking before the MIMO radar target detection of cost reference particle filter | |
| CN103994820B (en) | A kind of moving target recognition methods based on micropore diameter microphone array | |
| CN105913090A (en) | SAR image object classification method based on SDAE-SVM | |
| CN105678047A (en) | Wind field characterization method with empirical mode decomposition noise reduction and complex network analysis combined | |
| CN115270872A (en) | Radar radiation source individual small sample learning and identifying method, system, device and medium | |
| CN106940688B (en) | A kind of mathematical model modeling method of EMD chaff component identification | |
| CN115743101A (en) | Vehicle track prediction method, and track prediction model training method and device | |
| CN105654518A (en) | Trace template self-adaption method based on variance estimation | |
| CN108596398A (en) | Time Series Forecasting Methods and device based on condition random field Yu Stacking algorithms | |
| CN102521402B (en) | Text filtering system and method | |
| CN111783336A (en) | Uncertain structure frequency response dynamic model correction method based on deep learning theory | |
| CN105929380A (en) | Full-waveform laser radar data denoising method for satellite laser altimeter | |
| CN109740635B (en) | Classification recognition feature mapping method based on two classifiers | |
| CN103207997B (en) | Kernel density estimation-based license plate character segmentation method | |
| CN103440226A (en) | EMD (Empirical Mode Decomposition) endpoint effect suppression method based on HMM (Hidden Markov Model) correction and neural network extension | |
| CN109995690B (en) | Neural network self-optimization method for MFSK digital signal subclass modulation recognition | |
| CN105117537A (en) | Weight comparison based resampling method for particle filter system | |
| CN111181574A (en) | End point detection method, device and equipment based on multi-layer feature fusion | |
| CN114626412A (en) | Multi-class target identification method and system for unattended sensor system | |
| CN103903267B (en) | Image partition method based on average template and student's t mixed models | |
| CN103049887B (en) | Curvelet transformation domain image denoising method based on fuzzy support vector machine |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |