CN108871334A - A kind of RFID indoor orientation method and system based on machine learning - Google Patents

A kind of RFID indoor orientation method and system based on machine learning Download PDF

Info

Publication number
CN108871334A
CN108871334A CN201810595369.XA CN201810595369A CN108871334A CN 108871334 A CN108871334 A CN 108871334A CN 201810595369 A CN201810595369 A CN 201810595369A CN 108871334 A CN108871334 A CN 108871334A
Authority
CN
China
Prior art keywords
machine learning
model
module
phase data
prediction model
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
Application number
CN201810595369.XA
Other languages
Chinese (zh)
Inventor
谭洪舟
张�浩
曾衍瀚
廖裕兴
陈曦恒
王嘉奇
方巍
陈翔
张鑫
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Sun Yat Sen University
SYSU CMU Shunde International Joint Research Institute
Research Institute of Zhongshan University Shunde District Foshan
National Sun Yat Sen University
Original Assignee
SYSU CMU Shunde International Joint Research Institute
Research Institute of Zhongshan University Shunde District Foshan
National Sun Yat Sen University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by SYSU CMU Shunde International Joint Research Institute, Research Institute of Zhongshan University Shunde District Foshan, National Sun Yat Sen University filed Critical SYSU CMU Shunde International Joint Research Institute
Priority to CN201810595369.XA priority Critical patent/CN108871334A/en
Publication of CN108871334A publication Critical patent/CN108871334A/en
Pending legal-status Critical Current

Links

Classifications

    • GPHYSICS
    • G01MEASURING; TESTING
    • G01CMEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
    • G01C21/00Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00
    • G01C21/20Instruments for performing navigational calculations
    • G01C21/206Instruments for performing navigational calculations specially adapted for indoor navigation

Abstract

The invention discloses a kind of RFID indoor orientation method and system based on machine learning, wherein method include:Multifrequency carrier wave is sent to target, obtains phase data;Pre-process phase data;Prediction model is established using classification regression algorithm;The coordinate of target is obtained according to prediction model.Compared to traditional technology, positioning accuracy of the present invention is higher, precision ranging, data processing and modeling analysis can be carried out to target, to obtain accurate, stable positioning result.

Description

A kind of RFID indoor orientation method and system based on machine learning
Technical field
The present invention relates to fixation and recognition field, especially a kind of RFID indoor orientation method based on machine learning and it is System.
Background technique
RFID is the abbreviation of Radio Frequency Identification, i.e. Radio Frequency Identification Technology.RFID radio frequency is known It is not a kind of contactless automatic identification technology, it passes through radiofrequency signal automatic identification target object and obtains related data, Identify that work without manual intervention, is operable with various adverse circumstances;RFID technique can recognize high-speed moving object and can be simultaneously Identify multiple labels, it is swift and convenient to operate.RFID indoor positioning technologies for simplicity, are generally adopted by unrelated with ranging at present Location algorithm, since it does not carry out accurate ranging and data processing, positioning accuracy is lower, obtained positioning result The probability of inaccuracy greatly increases.
Summary of the invention
To solve the above-mentioned problems, the object of the present invention is to provide a kind of RFID indoor orientation method based on machine learning And system, precision ranging and data processing can be carried out to target, therefore positioning accuracy is higher.
In order to make up for the deficiencies of the prior art, the technical solution adopted by the present invention is that:
A kind of RFID indoor orientation method based on machine learning, includes the following steps:
Multifrequency carrier wave is sent to target, obtains phase data;
Pre-process phase data;
Prediction model is established using classification regression algorithm;
The coordinate of target is obtained according to prediction model.
Further, phase data is pre-processed, is included the following steps:
Normal distribution model RSSI is established, by i-th of phase difference piDensity function be expressed as
Wherein, μ and σ is respectively the mean value and standard deviation of RSSI, m For phase data total amount;
The range of f (x) and corresponding x are set, screening obtains that f (x) and corresponding x is made to meet the phase data of area requirement.
Further, the range of the f (x) and corresponding x are 0.6≤f (x)≤1, the 0.15 σ+μ≤σ of x≤3.09+μ.
Further, the classification regression algorithm includes in random forest, multiple linear regression model, model tree and random tree It is one or more.
Further, using classification regression algorithm establish prediction model with according to prediction model obtain target coordinate it Between, it further include step:It is tested assessment using ten times of cross-validation methods to all prediction models, selects wherein that position error is most It is small as final prediction model.
A kind of RFID indoor locating system based on machine learning, including:
Acquisition module obtains phase data for sending multifrequency carrier wave to target;
Preprocessing module, for pre-processing phase data;
Modeling module, for establishing prediction model using classification regression algorithm;
Module is obtained, for obtaining the coordinate of target according to prediction model.
Further, the preprocessing module includes:
Condition module, for establishing normal distribution model RSSI, by i-th of phase difference piDensity function be expressed as
Wherein, μ and σ is respectively the mean value and standard deviation of RSSI, M is phase data total amount;
Screening module, for setting the range of f (x) and corresponding x, screening obtains that f (x) and corresponding x is made to meet area requirement Phase data.
Further, the range of f (x) and corresponding x are 0.6≤f (x)≤1, the 0.15 σ+μ≤σ of x≤3.09+μ.
Further, the classification regression algorithm includes in random forest, multiple linear regression model, model tree and random tree It is one or more.
Further, inspection module is additionally provided between modeling module and acquisition module;The inspection module, for using Ten times of cross-validation methods test assessment to all prediction models, and it is the smallest as final prediction to select wherein position error Model.
The beneficial effects of the invention are as follows:By sending different frequency carrier wave to target and calculating the phase difference of return, according to Common knowledge although it is understood that target range;And it is handled for the data measured, to improve the validity and standard of data True property;Then prediction model is established using classification regression algorithm, and the data measured is substituted into the model, can accurately understood To coordinates of targets.Therefore, positioning accuracy of the present invention is higher, can carry out precision ranging, data processing and modeling analysis to target, from And obtain accurate, stable positioning result.
Detailed description of the invention
Present pre-ferred embodiments are provided, with reference to the accompanying drawing with the embodiment that the present invention will be described in detail.
Fig. 1 is step flow diagram of the invention;
Fig. 2 is the schematic diagram that the present invention sends multifrequency carrier wave to target;
Fig. 3 is the algorithm principle figure of random forest of the present invention.
Specific embodiment
Referring to Fig.1, a kind of RFID indoor orientation method based on machine learning of the invention, includes the following steps:
Multifrequency carrier wave is sent to target, obtains phase data;
Pre-process phase data;
Prediction model is established using classification regression algorithm;
The coordinate of target is obtained according to prediction model.
Further, pre-processing used by phase data is Gauss filter method, to filter out abnormal phase data, Satisfactory phase data is obtained, following steps are specifically included:
Normal distribution model RSSI is established, by i-th of phase difference piDensity function be expressed as
Wherein, μ and σ is respectively the mean value and standard deviation of RSSI, M is phase data total amount;
The range of f (x) and corresponding x are set, screening obtains that f (x) and corresponding x is made to meet the phase data of area requirement.
Further, the range of the f (x) and corresponding x are 0.6≤f (x)≤1, the 0.15 σ+μ≤σ of x≤3.09+μ.
Further, the classification regression algorithm includes in random forest, multiple linear regression model, model tree and random tree It is one or more.
Further, using classification regression algorithm establish prediction model with according to prediction model obtain target coordinate it Between, it further include step:It is tested assessment using ten times of cross-validation methods to all prediction models, selects wherein that position error is most It is small as final prediction model.
A kind of RFID indoor locating system based on machine learning, including:
Acquisition module obtains phase data for sending multifrequency carrier wave to target;
Preprocessing module, for pre-processing phase data;
Modeling module, for establishing prediction model using classification regression algorithm;
Module is obtained, for obtaining the coordinate of target according to prediction model.
Further, the preprocessing module includes:
Condition module, for establishing normal distribution model RSSI, by i-th of phase difference piDensity function be expressed as
Wherein, μ and σ is respectively the mean value and standard deviation of RSSI, M is phase data total amount;
Screening module, for setting the range of f (x) and corresponding x, screening obtains that f (x) and corresponding x is made to meet area requirement Phase data.
Further, the range of f (x) and corresponding x are 0.6≤f (x)≤1, the 0.15 σ+μ≤σ of x≤3.09+μ.
Further, the classification regression algorithm includes random forest (i.e. RF), multiple linear regression model (i.e. MLR), mould One of type tree (i.e. M5P) and random tree (i.e. RT) are a variety of.
Further, inspection module is additionally provided between modeling module and acquisition module;The inspection module, for using Ten times of cross-validation methods test assessment to all prediction models, and it is the smallest as final prediction to select wherein position error Model.
Specifically, referring to Fig. 2, the positioning method based on FD-PDOA sends different frequencies to same target by reader The carrier wave of rate, by subtracting each other the phase information of the carrier signal of these different frequencies with obtain phase difference and calculate away from From:
Assuming that transmitting two carrier signals, frequency is respectively f1And f2, since the light velocity is c, according to phase difference and distance Between relationship, then have
Two formulas above simultaneous, can obtain
Wherein Δ θ=θ12, Δ f=f1-f2, θ1And θ2It indicates in carrier frequency to be respectively f1And f2When phase change Amount.
For 865MHz-956MHz, the range of corresponding effective distance exists the operating frequency range for the reader that the present invention uses Between 15.7cm-17.3cm, therefore the method for using single frequency carrier can not preferably meet the requirement of indoor positioning, then this hair It is bright that phase difference is obtained using FD-PDOA method, i.e., phase difference is obtained at different frequencies, is then subtracted each other two-by-two again;Preferably, it selects Four readers taken can operating frequency be 920.625MHz, 921.875MHz, 923.125MHz and 924.375MHz respectively.
Wherein, random forest (RandomForest, RF) is a kind of assembled classifier algorithm, and Meta algorithm is by more decisions Tree composition, can solve classification or regression problem.Each tree generates respective classification results, for classification problem, final result It will be determined by ballot method with the most sorting item of number of votes obtained, for regression problem, according to the method for average with all subclassification results Average value as final result.
Decision tree in random forest is mainly to be realized using recursive mode:First by root node, be divided into a left side Then right two subtrees continue to generate corresponding left and right subtree from subtree, so go down, until every stalk tree recursive generation is new Subtree simultaneously reaches leaf node.Optimal fork attribute is selected, training sample is subjected to optimal dividing, is pass when constructing decision Key, fork attribute carry out branch using gini index.
Assuming that pi(=1,2 ..., | y |) indicate proportion of the i-th class sample in current sample I, the base of sample set I Buddhist nun's value is represented by
In conjunction with above formula, the gini index of attribute x is represented by
Geordie value illustrates two different classes of probability of extraction in sample set, and Geordie value is smaller, i.e., gini index is got over It is small, illustrate that the purity in sample set is higher.
Random forests algorithm thought:It referring to Fig. 3, is sampled first using bootstrap, from original training set, extracts k Sub- training set;Secondly, establishing k decision tree (i.e. CART algorithm) respectively to k sub- training sets;Finally, for test sample collection Each of sample to be sorted k kind classification results will be generated according to k decision-tree model, by being carried out to these classification results It votes or average value selects, determine its final classification.
Multiple linear regression model (multivariable linear regression, MLR) is a kind of by multiple categories The linear combination of property is come the algorithm predicted.It is assumed that variable ytWith k-1 explanatory variable xtj, exist between j=1,2 ... k-1 Linear relationship, then multiple linear regression model is expressed as:
yt01xt12xt2+…+βk-1xtk-1t
Wherein ytIt is explained variable (i.e. dependent variable), xtj, j=1,2 ... k-1 are explanatory variable (i.e. independent variable), μt It is stochastic error, βi, i=0,1,2 ... k-1 are regression parameters.
As a given sample (yt, xt1, xt2,…xtk-1), when t=1,2 ... T, above-mentioned model is represented by,
Equation group is indicated with matrix form, it is as follows.
Enable Y=(y1 y2 … yT)′(T×1),
β=(β1 β2 … βk-1)′(k×1), μ=(μ1 μ2 … μT)′(T×1)
Then above formula can be write as:Y=X β+μ
Phase data is obtained in the actual environment and establishes prediction model, after prediction model is established, is needed to prediction Model is verified, to select the lesser prediction model of error.Ten times of cross-validation methods are a kind of models more better than residual error method The principle of appraisal procedure, this method is as follows:Data set is divided into 10 subsets, every time using 9 subsets as training set, is left one This method is repeated 10 times by the data of a subset as test set, finally calculates all 10 mean errors.Ten times of intersections are tested Card result is as shown in Table 1, and x, y are respectively the horizontal axis of reference axis, the longitudinal axis.
The position error of 10 times of cross validations of table
MLR RF M5P RT
x/m 0.460 0.347 0.414 0.492
y/m 0.477 0.361 0.478 0.506
It can be seen that, the position error by the RF prediction model established is minimum, only 0.35m or so, secondly from table It is RT, M5P and MLR, therefore RF is determined as final prediction model based on the prediction model that random forest is established, is based on this Final prediction model obtains coordinates of targets.
Since the coordinate for obtaining target according to final prediction model is any technique commonly known, phase is almost related only to The iteration or substitution of potential difference data or range data, and therefore not to repeat here.
Presently preferred embodiments of the present invention and basic principle is discussed in detail in the above content, but the invention is not limited to Above embodiment, those skilled in the art should be recognized that also have on the premise of without prejudice to spirit of the invention it is various Equivalent variations and replacement, these equivalent variations and replacement all fall within the protetion scope of the claimed invention.

Claims (10)

1. a kind of RFID indoor orientation method based on machine learning, which is characterized in that include the following steps:
Multifrequency carrier wave is sent to target, obtains phase data;
Pre-process phase data;
Prediction model is established using classification regression algorithm;
The coordinate of target is obtained according to prediction model.
2. a kind of RFID indoor orientation method based on machine learning according to claim 1, which is characterized in that pretreatment Phase data includes the following steps:
Normal distribution model RSSI is established, by i-th of phase difference piDensity function be expressed as
Wherein, μ and σ is respectively the mean value and standard deviation of RSSI, M is phase data total amount;
The range of f (x) and corresponding x are set, screening obtains that f (x) and corresponding x is made to meet the phase data of area requirement.
3. a kind of RFID indoor orientation method based on machine learning according to claim 2, which is characterized in that the f (x) and the range of corresponding x is the σ of 0.6≤f (x)≤1,0.15+μ≤σ of x≤3.09+μ.
4. a kind of RFID indoor orientation method based on machine learning according to claim 1, which is characterized in that described point Class regression algorithm includes one of random forest, multiple linear regression model, model tree and random tree or a variety of.
5. a kind of RFID indoor orientation method based on machine learning according to claim 1 or 4, which is characterized in that Prediction model is established using classification regression algorithm and is obtained between the coordinate of target according to prediction model, further includes step:Using Ten times of cross-validation methods test assessment to all prediction models, and it is the smallest as final prediction to select wherein position error Model.
6. a kind of RFID indoor locating system based on machine learning, which is characterized in that including:
Acquisition module obtains phase data for sending multifrequency carrier wave to target;
Preprocessing module, for pre-processing phase data;
Modeling module, for establishing prediction model using classification regression algorithm;
Module is obtained, for obtaining the coordinate of target according to prediction model.
7. a kind of RFID indoor locating system based on machine learning according to claim 6, which is characterized in that described pre- Processing module includes:
Condition module, for establishing normal distribution model RSSI, by i-th of phase difference piDensity function be expressed as
Wherein, μ and σ is respectively the mean value and standard deviation of RSSI, M is phase data total amount;
Screening module, for setting the range of f (x) and corresponding x, screening obtains that f (x) and corresponding x is made to meet the phase of area requirement Potential difference data.
8. a kind of RFID indoor locating system based on machine learning according to claim 7, which is characterized in that f (x) and The range of corresponding x is the σ of 0.6≤f (x)≤1,0.15+μ≤σ of x≤3.09+μ.
9. a kind of RFID indoor locating system based on machine learning according to claim 6, which is characterized in that described point Class regression algorithm includes one of random forest, multiple linear regression model, model tree and random tree or a variety of.
10. a kind of RFID indoor locating system based on machine learning according to claim 6 or 9, which is characterized in that Inspection module is additionally provided between modeling module and acquisition module;The inspection module, for using ten times of cross-validation methods pair All prediction models are tested assessment, and it is the smallest as final prediction model to select wherein position error.
CN201810595369.XA 2018-06-11 2018-06-11 A kind of RFID indoor orientation method and system based on machine learning Pending CN108871334A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810595369.XA CN108871334A (en) 2018-06-11 2018-06-11 A kind of RFID indoor orientation method and system based on machine learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810595369.XA CN108871334A (en) 2018-06-11 2018-06-11 A kind of RFID indoor orientation method and system based on machine learning

Publications (1)

Publication Number Publication Date
CN108871334A true CN108871334A (en) 2018-11-23

Family

ID=64338594

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810595369.XA Pending CN108871334A (en) 2018-06-11 2018-06-11 A kind of RFID indoor orientation method and system based on machine learning

Country Status (1)

Country Link
CN (1) CN108871334A (en)

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102998654A (en) * 2012-11-23 2013-03-27 无锡儒安科技有限公司 Method and device of label phase information and based on passive wireless radio frequency recognition
CN107360548A (en) * 2017-08-21 2017-11-17 山东师范大学 Indoor article localization method and system based on RFID
CN107490783A (en) * 2016-06-10 2017-12-19 天津力芯伟业科技有限公司 A kind of RFID localization methods based on ELM
WO2017222510A1 (en) * 2016-06-22 2017-12-28 Intel Corporation Communication device and a method for localization

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102998654A (en) * 2012-11-23 2013-03-27 无锡儒安科技有限公司 Method and device of label phase information and based on passive wireless radio frequency recognition
CN107490783A (en) * 2016-06-10 2017-12-19 天津力芯伟业科技有限公司 A kind of RFID localization methods based on ELM
WO2017222510A1 (en) * 2016-06-22 2017-12-28 Intel Corporation Communication device and a method for localization
CN107360548A (en) * 2017-08-21 2017-11-17 山东师范大学 Indoor article localization method and system based on RFID

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
刘克强等: "基于随机森林算法的多维情境特征活动识别", 《测绘通报》 *
胡宏泽: "基于ZigBee网络的移动节点RFID定位技术的研究与实现", 《中国优秀硕士学位论文全文数据库 信息科技辑 》 *
胡宏泽等: "一种基于频域的相位差无源标签测距方法", 《数据采集与处理》 *

Similar Documents

Publication Publication Date Title
CN111191732B (en) Target detection method based on full-automatic learning
CN106373118B (en) The complex curved surface parts point cloud compressing method of border and local feature can be effectively retained
CN104931960B (en) Trend message and radar target situation information full flight path segment data correlating method
CN108846259A (en) A kind of gene sorting method and system based on cluster and random forests algorithm
CN102324038B (en) Plant species identification method based on digital image
CN108537790B (en) Different-source image change detection method based on coupling translation network
CN113344019A (en) K-means algorithm for improving decision value selection initial clustering center
CN106023159B (en) Facilities vegetable leaf portion scab image partition method and system
CN106599915A (en) Vehicle-mounted laser point cloud classification method
CN109165672A (en) A kind of Ensemble classifier method based on incremental learning
CN109003266A (en) A method of based on fuzzy clustering statistical picture quality subjective evaluation result
CN112986925A (en) Radar pulse sequence sorting method based on image characteristics
CN103268465B (en) The method for quickly identifying of label classification in a kind of radio-frequency recognition system
CN104809471A (en) Hyperspectral image residual error fusion classification method based on space spectrum information
CN104680185A (en) Hyperspectral image classification method based on boundary point reclassification
CN107403205B (en) RFID warehouse goods package plane positioning method based on random forest
CN113452802A (en) Equipment model identification method, device and system
CN105654042B (en) The proving temperature character identifying method of glass-stem thermometer
CN113128584B (en) Mode-level unsupervised sorting method of multifunctional radar pulse sequence
CN110458020A (en) A kind of clothes fashion search method based on Shape context
CN109977131A (en) A kind of house type matching system
CN111105041B (en) Machine learning method and device for intelligent data collision
CN103076595B (en) Abnormal type identifying method for multivariate discrete radar emitter signal
CN108871334A (en) A kind of RFID indoor orientation method and system based on machine learning
CN109919626B (en) High-risk bank card identification method and device

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
RJ01 Rejection of invention patent application after publication

Application publication date: 20181123

RJ01 Rejection of invention patent application after publication