CN108446726B - Vehicle cab recognition classification method based on information gain rate Yu fisher linear discriminant - Google Patents
Vehicle cab recognition classification method based on information gain rate Yu fisher linear discriminant Download PDFInfo
- Publication number
- CN108446726B CN108446726B CN201810205247.5A CN201810205247A CN108446726B CN 108446726 B CN108446726 B CN 108446726B CN 201810205247 A CN201810205247 A CN 201810205247A CN 108446726 B CN108446726 B CN 108446726B
- Authority
- CN
- China
- Prior art keywords
- vehicle
- class
- classification
- lane
- scatter matrix
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/285—Selection of pattern recognition techniques, e.g. of classifiers in a multi-classifier system
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/08—Detecting or categorising vehicles
Abstract
The vehicle cab recognition classification method based on information gain rate Yu fisher linear discriminant that the invention discloses a kind of, this method are suitable for classifying to microwave remote sensor acquisition model data.It comprises the steps of: and finds out the corresponding lane information entropy production rate of each vehicle;Using lane information entropy production rate as Scatter Matrix weight in class, the vehicle classification device of new fisher linear discriminant is obtained;Classification results are obtained according to classifier;Information gain rate and fisher linear discriminant are applied to microwave data vehicle classification by the present invention, vehicle classification precision is improved with this, in addition, will be added as features in classification at a distance from microwave remote sensor and tested vehicle herein, vehicle classification precision is further increased.
Description
Technical field
The present invention relates to a kind of methods of vehicle cab recognition classification, in particular to linear based on information gain rate and fisher
The vehicle cab recognition classification method of differentiation.
Background technique
The classification standard of Chinese vehicle cab recognition is according to Department of Transportation in " highway communication condition survey motor vehicle model point
According to division as foundation in class ".Wherein automobile is divided into compact car, in-between car, large car and super-huge vehicle by first-level class
Four classes.Vehicle cab recognition is in transport investigation, and roading, the field of traffic such as road early warning are widely used, thus become
The hot spot of various countries' research.Traditional vehicle cab recognition data source mainly passes through ground induction coil, laser, and the methods of image obtains, but
The disadvantages of it is at high cost to be that these methods have, and maintenance is complicated, big by external environmental interference.Using microwave acquisition model data have at
This is low, and easy to maintain, be affected by the external environment small advantage.
Microwave remote sensor Doppler effect obtains vehicle commander and speed.Doppler effect refer to object radiation wavelength because
The relative motion of wave source and observer and generate variation, before the wave source of movement, wave is compressed, and wavelength becomes shorter, frequency
Become a kind of higher phenomenon.Microwave remote sensor ceaselessly emits microwave, can reflect or scatter immediately after microwave encounters vehicle
Wave, since there are Doppler effect, reflection or scattered waves will generate Doppler frequency shift, using generate frequency displacement wave and this vibration wave into
Row mixing handles the movement velocity that vehicle can be obtained, the time and vehicle that microwave passes through vehicle using electronic circuit appropriate
Speed, which is multiplied, obtains the length of vehicle, meanwhile, microwave can also obtain lane information, the range information of microwave remote sensor and vehicle.I
By microwave obtain data be referred to as model data.The vehicle commander that conventional method obtains microwave remote sensor, speed, lane information
It as the feature of classifier, is added in classifier using Fisher linear discriminant method and is classified to vehicle, but classified
Accuracy rate is undesirable.
Summary of the invention
The present invention is of the existing technology in order to solve the problems, such as, proposes one kind based on information gain rate and fisher line
Property differentiate vehicle cab recognition classification method.Information gain rate is introduced into fisher linear discriminant method, new point is formed
Class method improves the vehicle classification precision of microwave data to overcome defect of the existing technology.
The present invention is the vehicle cab recognition based on information gain rate Yu fisher linear discriminant realizing goal of the invention and proposing
Classification method comprising the steps of:
Step 1: acquiring model data using microwave, the model data includes vehicle commander, speed, lane data, sensor
At a distance from vehicle, the model data then acquired according to microwave finds out the corresponding lane information entropy production rate of each vehicle: first
The corresponding lane information entropy of each vehicle first is found out with information entropy theory, lane information is calculated to vehicle by information gain standard
The information gain of type data sample;In order to avoid there is preference in the information gain criterion lane more to desirable vehicle, in each vehicle
The corresponding lane information entropy production rate of each vehicle is further found out on the basis of the corresponding lane information entropy production of type.
1. finding out the corresponding lane information entropy of each vehicle with information entropy theory, lane information entropy can be indicated are as follows:
H indicates the comentropy of specified vehicle in formula, and U specifies the sample set of model data, and n is specific lane, and range is logical
It is often number of track-lines, it is the corresponding set of data samples of n that x, which specifies vehicle and lane, and P (x) indicates certain lane vehicle in model data sample
The probability that this concentration specifies vehicle to occur.
2. lane information is calculated to the information gain of model data sample by information gain standard, model data sample
Information gain can indicate are as follows:
Gain in formula (S is a) specified vehicle lane to the information gain of model data sample, | S | for specified model data
Sample set matrix, a refer to lane, | Sn| it is specified vehicle in specific lane set of data samples matrix, H (S) indicates specified vehicle
The comentropy of type, H (Sn) it is comentropy of the specified vehicle in specific lane data.
3. in order to avoid there is preference in the information gain criterion lane more to desirable vehicle, in the corresponding lane of each vehicle
Information gain rate is further found out on the basis of information gain, is indicated are as follows:
Gain_ratio (S, a)=Gain (S, a)/IV (a)
IV (a) is the eigenvalue of vehicle, and the value number of vehicle is more, then IV (a) is bigger, is thus avoided that information increases
The beneficial criterion IV (a) more to desirable number has preference.
Step 2: entropy information ratio of profit increase being combined with Scatter Matrix in class, new vehicle classification device is obtained: finding out vehicle
Scatter Matrix and class scatter matrix in the class of sample set;The corresponding lane information entropy production of each vehicle that step 1 is obtained
Rate, which is added in the class of corresponding class as weight, in Scatter Matrix, forms Scatter Matrix in new class;Using will in new class dissipate
It spends matrix to substitute into Fisher linear classifier, forms new vehicle classification device.
1. seeking Scatter Matrix in the data sample class of specified vehicle classification i, Scatter Matrix S in classiIt indicates are as follows:
M in formulaiFor vehicle classification i (its value range is vehicle classification number) sample mean vector.XiFor the vehicle of classification i
Set of data samples, (x-mi) T be (x-mi) transposed matrix.
Ask specified model data class scatter matrix, class scatter matrix SbIt indicates are as follows:
Sb=(mi-mj)(mi-mj)T
I in formula, j are vehicle classification, mi, mjRespectively correspond to the sample mean vector of class.
2. the corresponding lane information entropy production rate of each vehicle that step 1 is acquired is dissipated as in the class of model data sample
The weight for spending matrix, finds out Scatter Matrix in total class.Scatter Matrix S in total classwIt indicates are as follows:
Sw=Gain_ratio (Ui,a)*Si+Gain_ratio(Uj,a)*Sj
Ui, UjFor the corresponding set of data samples of vehicle classification i, j, Si, SjFor divergence square in the corresponding class of vehicle classification i, j
Battle array.
3. it is theoretical according to Fisher linear classification, the two class model datas classified will be needed to project to a plane, projected
Afterwards, as intensive as possible inside Different categories of samples, Different categories of samples is as far as possible far away.After this means that projection, class scatter matrix
It is the bigger the better, class scatter matrix is the bigger the better.That is class scatter matrix and the ratio of class scatter matrix is the bigger the better.
Stb=wTSbw
Stw=wTSww
S in formulatb, StwScatter Matrix in class scatter matrix and total class after respectively projecting, w are that Fisher is linear
Classification projection vector, wTFor the transposition of w.The ratio of class scatter matrix and class scatter matrix seeks JF(w) maximum value.So that
JF(w) maximized w.JF(w) maximized w can be indicated are as follows:
For the inverse matrix of Scatter Matrix in the sample class before projection, mi, mjRespectively vehicle classification i, j (its value model
Enclose for vehicle classification number) sample mean vector.
W is updated in Fisher linear classifier, as new vehicle classification device.With traditional Fisher linear classification
Device is different, and new classifier takes Scatter Matrix calculation method in the completely new total class of one kind.
Present invention is specifically directed to the methods that microwave data vehicle classification proposes.It has the feature that microwave remote sensor 1)
It is added as features in classification at a distance from tested vehicle, fully considers that different automobile types to the factor of lane preference, improve
Vehicle classification precision.
2) using lane information entropy production rate as Scatter Matrix weight in class, the vehicle of new fisher linear discriminant is obtained
Classifier improves vehicle classification precision.The present invention is used directly for the vehicle classification of microwave data.
The present invention combines information gain rate with fisher linear discriminant, is more there is the classifier of discriminating power,
So as to improve the nicety of grading of classifier.In addition, microwave remote sensor is added as features at a distance from tested vehicle herein
In classification, vehicle classification precision is further increased.
Specific embodiment
Below with reference to embodiment, invention is further described in detail.
Embodiment:
1, on a highway by microwave remote sensor installation, obtain the vehicle commander of vehicle, speed, lane data, sensor with
The information such as the distance of vehicle.The vehicle of these vehicles is obtained by the method for manual identified, to guarantee the accuracy of vehicle, and will
These data are together with vehicle, training sample set of a part as vehicle cab recognition classification.
2, the vehicle cab recognition classification method based on information gain rate and fisher linear discriminant can only solve two class of vehicle
Classification problem needs vehicle classification to be compact car, in-between car, four class of large car and super-huge vehicle in the actual implementation process.
It is one-to-one using multiple groups on the basis of based on the vehicle cab recognition classification method of information gain rate and fisher linear discriminant
Method, solution type identify that four class classification problems, specific implementation step are as follows:
1. by vehicle cab recognition classification based training sample set compact car and in-between car data classification take out, managed with comentropy
By compact car and the corresponding lane information entropy of in-between car is found out respectively, lane information is calculated to vehicle number by information gain standard
According to the compact car and in-between car information gain of sample;In order to avoid the information gain criterion lane more to desirable vehicle has partially
It is good, information gain rate is further found out on the basis of entropy production.Scatter Matrix in the trolley class of vehicle sample set is found out respectively
With Scatter Matrix in the class of in-between car and their class scatter matrix;By compact car and the corresponding lane information entropy of in-between car
Ratio of profit increase is added in corresponding class in Scatter Matrix as weight, forms Scatter Matrix in new total class;Using will be new total
Scatter Matrix substitutes into Fisher linear classifier in class, forms new compact car and in-between car classifier.
2. with the step in 1., obtain respectively compact car and large car classifier, compact car and super-huge vehicle classifier, in
Type vehicle and large car classifier, in-between car and super-huge vehicle classifier, large car and super-huge vehicle classifier.
3. model data to be put into 6 classifiers 1. and 2. generated, 6 classification results are obtained, take ballot method, are chosen
The most vehicle of classification results number, the final vehicle as the model data.
3, the test sample collection that vehicle cab recognition is classified, makes according to traditional Fisher linear classifier and step 2 respectively
Classified with the vehicle cab recognition classification method based on information gain rate and fisher linear discriminant, and is concentrated just with test sample
True vehicle comparison, obtains the classification accuracy of two kinds of classifiers.By experiment, traditional Fisher linear classifier classification is accurate
Rate is 81.23%, and it is accurate to be classified using the vehicle cab recognition classification method based on information gain rate and fisher linear discriminant
Rate is increased to 92.58%.Experiment shows compared to traditional fisher linear discriminant method, based on information gain rate with
The vehicle cab recognition classification method of fisher linear discriminant is significantly improved to classification accuracy.
Claims (3)
1. a kind of vehicle cab recognition classification method based on information gain rate Yu fisher linear discriminant acquires vehicle using microwave
Data, the model data include the range information of vehicle commander, speed, lane, vehicle and sensor;It is characterized by: according to micro-
The model data of wave acquisition, establishes vehicle classification device according to the following steps:
Step 1. finds out the corresponding lane information entropy production rate of each vehicle
The corresponding lane information entropy of each vehicle is found out with information entropy theory first, further finds out the corresponding vehicle of each vehicle
Road information gain rate;
Step 2. combines information gain rate with Scatter Matrix in class, establishes new vehicle classification device
Find out Scatter Matrix and class scatter matrix in the class of vehicle sample set;The corresponding vehicle of each vehicle that step 1 is obtained
Road information gain rate is added in the class of corresponding class as weight in Scatter Matrix, forms Scatter Matrix in new class;It will be new
Class in Scatter Matrix substitute into Fisher linear classifier in, form new vehicle classification device;
Step 3. sorts out vehicle according to new vehicle classification device.
2. the vehicle cab recognition classification method according to claim 1 based on information gain rate Yu fisher linear discriminant,
It is characterized by: the corresponding lane information entropy production rate of each vehicle is found out described in step 1, including
1) the corresponding lane information entropy H of each vehicle is found out:
H indicates the comentropy of specified vehicle in formula, and U specifies the sample set of model data, and n is specific lane, in the range of lane
Number, it is the corresponding set of data samples of n that x, which specifies vehicle and lane, and P (x) indicates that certain lane vehicle refers in model data sample set
Determine the probability of vehicle appearance;
2) information gain of the measuring and calculating lane information to model data sample:
Gain in formula (S is a) specified vehicle lane to the information gain of model data sample, | S | for specified model data sample
Collecting matrix, a refers to lane, | Sn| it is specified vehicle in specific lane set of data samples matrix, H (S) indicates specified vehicle
Comentropy, H (Sn) it is comentropy of the specified vehicle in specific lane data;
3) the corresponding lane information entropy production rate of each vehicle is found out:
Gain_ratio (S, a)=Gain (S, a)/IV (a)
IV (a) is the eigenvalue of vehicle.
3. the vehicle cab recognition classification method according to claim 1 based on information gain rate Yu fisher linear discriminant,
It is characterized by: the step 2 combines information gain rate with Scatter Matrix in class, new vehicle classification device is established, is had
Body the following steps are included:
1) Scatter Matrix in the class of specified model data sample is sought:
M in formulaiFor the sample mean vector for specifying class;M in formulaiFor vehicle classification sample mean vector, its value range of i is vehicle
Type classification number;XiFor the model data sample set of classification i, (x-mi)TFor (x-mi) transposed matrix;
Seek specified model data class scatter matrix:
Sb=(mi-mj)(mi-mj)T;
J is vehicle classification, m in formulajFor the sample mean vector of corresponding class;
2) the corresponding lane information entropy production rate of each vehicle acquired step 1 is as divergence square in the class of model data sample
The weight of battle array, finds out Scatter Matrix in total class:
Sw=Gain_ratio (Ui,a)*Si+Gain_ratio(Uj,a)*Sj
Ui, UjFor the corresponding set of data samples of vehicle classification i, j, Si, SjFor Scatter Matrix in the corresponding class of vehicle classification i, j;
3) theoretical according to Fisher linear classification, it would be desirable to which that two class model datas of classification project to a plane
Stb=wTSbw
Stw=wTSww
S in formulatb, StwScatter Matrix in class scatter matrix and total class after respectively projecting, w are the throwing of Fisher linear classification
Shadow vector;
Seek JF(w) maximum value, so that JF(w) maximized w are as follows:
For the inverse matrix of Scatter Matrix in the sample class before projection, mi, mjRespectively vehicle classification i, j sample mean vector;
W is updated in Fisher linear classifier, as new vehicle classification device.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810205247.5A CN108446726B (en) | 2018-03-13 | 2018-03-13 | Vehicle cab recognition classification method based on information gain rate Yu fisher linear discriminant |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810205247.5A CN108446726B (en) | 2018-03-13 | 2018-03-13 | Vehicle cab recognition classification method based on information gain rate Yu fisher linear discriminant |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108446726A CN108446726A (en) | 2018-08-24 |
CN108446726B true CN108446726B (en) | 2019-07-19 |
Family
ID=63195031
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810205247.5A Active CN108446726B (en) | 2018-03-13 | 2018-03-13 | Vehicle cab recognition classification method based on information gain rate Yu fisher linear discriminant |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108446726B (en) |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110580286A (en) * | 2019-08-09 | 2019-12-17 | 中山大学 | Text feature selection method based on inter-class information entropy |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103063202A (en) * | 2012-12-30 | 2013-04-24 | 同济大学 | Cyanobacteria biomass spatial-temporal change monitoring and visualization method based on remote sensing image |
CN104573714A (en) * | 2014-12-31 | 2015-04-29 | 南京理工大学 | Self-adaptation parameter-free feature extraction method |
CN104809233A (en) * | 2015-05-12 | 2015-07-29 | 中国地质大学(武汉) | Attribute weighting method based on information gain ratios and text classification methods |
CN106778473A (en) * | 2016-11-20 | 2017-05-31 | 南宁市浩发科技有限公司 | A kind of model recognizing method |
Family Cites Families (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP1661022A1 (en) * | 2003-09-05 | 2006-05-31 | Agency for Science, Technology and Research | Methods of processing biological data |
CN100481083C (en) * | 2005-07-26 | 2009-04-22 | 索尼株式会社 | Information processor, and feature extraction method |
KR100825756B1 (en) * | 2006-12-05 | 2008-04-29 | 한국전자통신연구원 | Method for feature extraction and its apparatus |
CN101441716A (en) * | 2008-11-27 | 2009-05-27 | 上海交通大学 | Integral and local characteristic fuse recognition system facing to identification |
CN106951914B (en) * | 2017-02-22 | 2020-06-09 | 江苏大学 | Method for identifying vinegar variety by electronic nose for optimizing fuzzy identification vector extraction |
-
2018
- 2018-03-13 CN CN201810205247.5A patent/CN108446726B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103063202A (en) * | 2012-12-30 | 2013-04-24 | 同济大学 | Cyanobacteria biomass spatial-temporal change monitoring and visualization method based on remote sensing image |
CN104573714A (en) * | 2014-12-31 | 2015-04-29 | 南京理工大学 | Self-adaptation parameter-free feature extraction method |
CN104809233A (en) * | 2015-05-12 | 2015-07-29 | 中国地质大学(武汉) | Attribute weighting method based on information gain ratios and text classification methods |
CN106778473A (en) * | 2016-11-20 | 2017-05-31 | 南宁市浩发科技有限公司 | A kind of model recognizing method |
Non-Patent Citations (4)
Title |
---|
A hybrid data mining model of feature selection algorithms and ensemble learning classifiers for credit scoring;Fatemeh Nemati Koutanaei;《Retailing and Consumer Services》;20150716;11-23 |
Object-based image analysis of optical and radar variables for wetland evaluation;Laura Dingle Robertson 等;《Remote Sensing》;20151123;5811–5841 |
一种新的保类内核Fisher判别法及说话人辨别应用;郑建炜,王万良;《计算机科学》;20100731;第37卷(第7期);243-247 |
基于PCA-LDA-SVM的多普勒雷达车型识别算法;方菲菲,余稳;《数据采集与处理》;20120131;第27卷(第1期);111-116 |
Also Published As
Publication number | Publication date |
---|---|
CN108446726A (en) | 2018-08-24 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110208793B (en) | Auxiliary driving system, method, terminal and medium based on millimeter wave radar | |
CN110427986B (en) | Target classification method of kernel support vector machine based on millimeter wave radar point cloud characteristics | |
Wu et al. | Automatic vehicle classification using roadside LiDAR data | |
Pavlic et al. | Classification of images in fog and fog-free scenes for use in vehicles | |
CN105574552A (en) | Vehicle ranging and collision early warning method based on monocular vision | |
DE102014223363A1 (en) | Method and device for localizing a motor vehicle in a fixed reference card | |
CN105242271A (en) | Object classification for vehicle radar systems | |
Prophet et al. | Semantic segmentation on automotive radar maps | |
CN113412505A (en) | System and method for ordered representation and feature extraction of point clouds obtained by detection and ranging sensors | |
CN101136141A (en) | Vehicle type classification method based on single frequency continuous-wave radar | |
CN102463990A (en) | System and method for tracking objects | |
CN111175718B (en) | Automatic target recognition method and system for ground radar combining time-frequency domains | |
CN113516052B (en) | Imaging millimeter wave radar point cloud target classification method based on machine learning | |
Chaabani et al. | Estimating meteorological visibility range under foggy weather conditions: A deep learning approach | |
CN105913083A (en) | Dense SAR-SIFT and sparse coding-based SAR classification method | |
Akita et al. | Object tracking and classification using millimeter-wave radar based on LSTM | |
CN113378751A (en) | Traffic target identification method based on DBSCAN algorithm | |
CN108446726B (en) | Vehicle cab recognition classification method based on information gain rate Yu fisher linear discriminant | |
Kim et al. | Target classification using combined YOLO-SVM in high-resolution automotive FMCW radar | |
CN116205024A (en) | Self-adaptive automatic driving dynamic scene general generation method for high-low dimension evaluation scene | |
DE102019130096A1 (en) | Method for determining radar information, simulation method, classification method, computer program and calculation system | |
CN114120150A (en) | Road target detection method based on unmanned aerial vehicle imaging technology | |
DE102018222672A1 (en) | Determination of the orientation of objects with radar and other electromagnetic interrogation radiation | |
Matsuda et al. | A Method for Detecting Street Parking Using Dashboard Camera Videos. | |
Ngo | A methodology for validation of a radar simulation for virtual testing of autonomous driving |
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 | ||
TR01 | Transfer of patent right | ||
TR01 | Transfer of patent right |
Effective date of registration: 20191105 Address after: 210000, No. 6, Changjiang back street, Xuanwu District, Jiangsu, Nanjing Patentee after: Nanjing Osai Environmental Protection Technology Co., Ltd. Address before: 212031 298 South Xu Road, run Zhou District, Zhenjiang, Jiangsu Patentee before: Zhenjiang Yun Chen Information Technology Co., Ltd. |