CN103606007B - Target identification method based on Internet of Things and device - Google Patents
Target identification method based on Internet of Things and device Download PDFInfo
- Publication number
- CN103606007B CN103606007B CN201310590706.3A CN201310590706A CN103606007B CN 103606007 B CN103606007 B CN 103606007B CN 201310590706 A CN201310590706 A CN 201310590706A CN 103606007 B CN103606007 B CN 103606007B
- Authority
- CN
- China
- Prior art keywords
- internet
- things
- neutral net
- target
- measured value
- 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
Abstract
The invention discloses a kind of target identification method based on Internet of Things, including: obtain the measured value that in Internet of things system, predeterminated target is gathered by each sensor tip;Using the described measured value of acquisition as the input quantity of BP neutral net, described BP neutral net is trained;Wherein, the tanh transfer function of described BP neutral net uses multinomial to approach;According to the described BP neutral net after training, described predeterminated target is carried out pattern recognition.Approximation by polynomi-als is incorporated in Internet of Things field by the present invention, the approximation by polynomi-als of transfer function has been drawn in conjunction with Gaussian beam, ensure computational accuracy and reduce computation complexity, neural network model is able under the resource constrained environment of sensor quickly realize, thus substantially increase real-time, and then realize target recognition based on Internet of Things well.
Description
Technical field
The present invention relates to technology of Internet of things field, particularly relate to a kind of target identification method based on Internet of Things and a kind of base
Target Identification Unit in Internet of Things.
Background technology
In actual Internet of things system, various features amount is monitored, and these information are merged, obtain target
Concordance identification, all these depend on sensing system.The learning capacity intrinsic because of neutral net and the ability of adaptation, BP god
Be widely used in Practical Project through network, such as weather prognosis, image procossing, automatically control, Combinatorial Optimization, video divide
The field such as cut.But BP neutral net requires that hardware has stronger computing capability, and this causes the sensing that some computing capabilitys are weak
Device hardware cannot use.Therefore, when input feature vector amount dimension is bigger, neutral net not only structure is complicated, and the training time is significantly
Extend, and real-time is the most bad.So, in the neutral net of Multi-sensor Fusion experiment platform for IOT, relate to transfer
The calculating of function, computation complexity is high, thus significantly limit Multi-sensor Fusion Internet of Things answering in terms of target recognition
Use prospect.
Field of target recognition in Internet of Things, no matter traditional classic BP neutral net or the relevant BP improved are neural
Network, being required to hardware has stronger computing capability.Sensor in Internet of Things, belongs to a kind of and calculates resource-constrained ring
Border.There is no CPU on sensor, therefore cannot call the hyperbolic tangent function inside programming language storehouse, therefore hardware computing capability
Weak, it is impossible to undertake related operation, seriously constrain the target recognition ability of Internet of Things.
Summary of the invention
Based on this, the invention provides a kind of target identification method based on Internet of Things and a kind of target based on Internet of Things
Identify device.
A kind of target identification method based on Internet of Things, comprises the following steps:
Obtain the measured value that in Internet of things system, predeterminated target is gathered by each sensor tip;
Using the described measured value of acquisition as the input quantity of BP neutral net, described BP neutral net is trained;Its
In, the tanh transfer function of described BP neutral net uses multinomial to approach;
According to the described BP neutral net after training, described predeterminated target is carried out pattern recognition.
Compared with general technology, approximation by polynomi-als is incorporated into Internet of Things by present invention target identification method based on Internet of Things
In field, draw the approximation by polynomi-als of transfer function in conjunction with Gaussian beam, it is ensured that computational accuracy also reduces computation complexity so that
Neural network model is able to quickly realize under the resource constrained environment of sensor, thus substantially increases real-time, and then very
Target recognition is realized well based on Internet of Things.
A kind of Target Identification Unit based on Internet of Things, including acquisition module, training module and identification module;
Described acquisition module, for obtaining the measurement that in Internet of things system, predeterminated target is gathered by each sensor tip
Value;
Described training module, for the described measured value that will obtain as the input quantity of BP neutral net, to described BP god
It is trained through network;Wherein, the tanh transfer function of described BP neutral net uses multinomial to approach;
Described identification module, for according to the described BP neutral net after training, carrying out pattern knowledge to described predeterminated target
Not.
Compared with general technology, approximation by polynomi-als is incorporated into Internet of Things by present invention Target Identification Unit based on Internet of Things
In field, draw the approximation by polynomi-als of transfer function in conjunction with Gaussian beam, it is ensured that computational accuracy also reduces computation complexity so that
Neural network model is able to quickly realize under the resource constrained environment of sensor, thus substantially increases real-time, and then very
Target recognition is realized well based on Internet of Things.
Accompanying drawing explanation
Fig. 1 is the schematic flow sheet of present invention target identification method based on Internet of Things;
Fig. 2 is the schematic flow sheet of the preferred embodiment realizing present invention target identification method based on Internet of Things;
Fig. 3 is the structural representation of present invention Target Identification Unit based on Internet of Things.
Detailed description of the invention
By further illustrating the technological means and the effect of acquirement that the present invention taked, below in conjunction with the accompanying drawings and the most real
Execute example, to technical scheme, carry out clear and complete description.
Refer to Fig. 1, for the schematic flow sheet of present invention target identification method based on Internet of Things.
Present invention target identification method based on Internet of Things, comprises the following steps:
S101 obtains the measured value that in Internet of things system, predeterminated target is gathered by each sensor tip;
Described BP neutral net as the input quantity of BP neutral net, is instructed by S102 by the described measured value of acquisition
Practice;Wherein, the tanh transfer function of described BP neutral net uses multinomial to approach;
S103, according to the described BP neutral net after training, carries out pattern recognition to described predeterminated target.
In step S101, obtain the measured value that in Internet of things system, predeterminated target is gathered by each sensor tip, make
For one of them embodiment, described measured value includes temperature, humidity, pressure and flow.
Measured value can include the various measured values that in Internet of things system, predeterminated target is gathered by sensor tip, measured value
Type the most extensive, the later stage is the highest to the accuracy of target recognition.
In actual Internet of things system, various features amount (such as temperature, humidity, pressure, flow etc.) is monitored, and right
These information merge, and obtain compatibility of goals and explain and describe, all these depend on sensing system.
In step s 102, in the neutral net of Multi-sensor Fusion experiment platform for IOT, the former transfer used
Function is as follows:
As x=9,1-tanh (9)=3.05 × 10-8.And use float(floating number) precision time, computer institute can table
The minimum interval shown is 1.1921 × 10-7.This can key in eps(' single ' in matlab) confirm.Again because of tanh
X () is odd function, i.e. tanh (x)=-tanh (-x),Have only to approach ex, x ∈ [0,9].According to exponential function
exCharacteristic, it is known that:
Wherein,It is Gaussian beam, represents the maximum integer being not more than x,Then represent the smallest positive integral not less than x.
If e1,e2,…,e9Calculate in advance, and leave in a depositor array as constant, be designated as a={a1,…,a9}.Root
According to (1), to ex, the calculating of x ∈ [0,9], it is only necessary to calculate ex,x∈[-0.5,0.5].Use Taylor polynomial to launch approximation to ask
Solve.When x from Taylor expansion point more away from, truncated error is the biggest.Above-mentioned skill is limited to | x |≤0.5 range of variables, significantly helps
In reducing error.Along with exponent number increases, upper error reduces rapidly, and actual error declines faster.Such as n=7, rnThe upper bound
Have been reduced to 1.1 × 10-7, less than the float minimum interval of computer.
Similar analysis understand, when independent variable in [-0.5,0] time, i.e. x ∈ [0,0.5].e-xTaylor polynomial expansion can be carried out
As follows:
Thus understand truncated error and be
For succinctly, it is defined as follows symbol:
Thus can obtain:
For the most fixing x, have:
According to unlimited progression,Additionally, { umMonotonicity is obvious, so { rnIt is alternately
Function series sum, its general term dullness goes to zero, namely alternating harmonic series.Character according to alternating harmonic series:
Visible exIndependent variable have lower upper error when [-0.5,0] than in [0,0.5].More than according to, work as n=7
Time, rnThe upper bound has descended to 0.97 × 10-7.Therefore, in the present invention, it is only necessary to take n=7 and i.e. can reach required precision.
Now it is defined as follows function
WhereinWithCan read from the depositor array the most stored.By the definition with superior function, have:
Maximum error occurs at x=8.5, and is compared as follows with exact value:
e8.5=4914.768840.
Visible, even if at maximum error, also there are seven significance bits, this is employing floating number precision time institute energy in computer
The most significance bits represented.So, it is appreciated that transfer function tanh (x) of BP neutral net is used
Maximum error when approaching occurs at x=0.5, and no more than 1.2 × 10-7。
As one of them embodiment, described multinomial is as follows:
Wherein,
Wherein,For Gaussian beam, represent the maximum integer being not more than x,Represent the smallest positive integral not less than x.WithCan read from the depositor array stored.
Use the tanh transfer function of approximation by polynomi-als neutral net, not only ensure that approximation accuracy, the most greatly
The computation complexity reduced, its calculating process relates to computing and the most i.e. only relates to polynomial addition and multiplication, makes Internet of Things put down
Platform has more real-time.
In step s 103, as one of them embodiment, described predeterminated target includes image and video.
By Internet of things system, can be to various predeterminated target implementation pattern identifications.Artificial neural network is by substantial amounts of base
This neuron is connected with each other and forms, it is possible to carry out distributed variable-frequencypump and non-linear conversion, has powerful study and summary
The function concluded, has an ability of robustness, parallel processing, it is ensured that given precision.Should in BP neutral net and platform
The relation in, have the advantage that 1, inputting and exporting can keep Nonlinear Monotone relation;2, this function unusual light, symbol
Close relevant gradient to solve;3, to neutral net zmodem.Therefore, present invention target identification method based on Internet of Things is special
It is applicable to the pattern recognition to image and video, it is easy to promote.
As a example by BP Algorithm, the method for the application present invention is approached.Thus can be based on multinomial
Approach the Multi-sensor Fusion Internet of Things experimental procedure of BP neutral net:
Using the various features amounts such as multisensor test vibration, temperature, humidity, pressure, flow, amount measurement obtained is made
For input quantity;
Use approximation by polynomi-als BP neutral net to be trained, learn, conclude, obtain the parameter being correlated with;
Utilize the parameter obtained, target is carried out pattern recognition or classification.
Refer to Fig. 2, for realizing the flow process of a preferred embodiment of present invention target identification method based on Internet of Things
Schematic diagram.
Compared with general technology, approximation by polynomi-als is incorporated into Internet of Things by present invention target identification method based on Internet of Things
In field, draw the approximation by polynomi-als of transfer function in conjunction with Gaussian beam, it is ensured that computational accuracy also reduces computation complexity so that
Neural network model is able to quickly realize under the resource constrained environment of sensor, thus substantially increases real-time, and then very
Target recognition is realized well based on Internet of Things.
Refer to Fig. 3, for the structural representation of present invention Target Identification Unit based on Internet of Things.
Present invention Target Identification Unit based on Internet of Things, including acquisition module 301, training module 302 and identification module
303;
Described acquisition module 301, for obtaining the survey that in Internet of things system, predeterminated target is gathered by each sensor tip
Value;
Described training module 302, for the described measured value that will obtain as the input quantity of BP neutral net, to described BP
Neutral net is trained;Wherein, the tanh transfer function of described BP neutral net uses multinomial to approach;
Described identification module 303, for according to the described BP neutral net after training, carrying out pattern to described predeterminated target
Identify.
As one of them embodiment, described measured value includes temperature, humidity, pressure and flow.
Measured value can include the various measured values that in Internet of things system, predeterminated target is gathered by sensor tip, measured value
Type the most extensive, the later stage is the highest to the accuracy of target recognition.
As one of them embodiment, described multinomial is as follows:
Wherein,
Wherein,For Gaussian beam, represent the maximum integer being not more than x,Represent the smallest positive integral not less than x.
Use the tanh transfer function of approximation by polynomi-als neutral net, not only ensure that approximation accuracy, the most greatly
The computation complexity reduced, its calculating process relates to computing and the most i.e. only relates to polynomial addition and multiplication, makes Internet of Things put down
Platform has more real-time.
As one of them embodiment, described predeterminated target includes image and video.
By Internet of things system, can be to various predeterminated target implementation pattern identifications.Artificial neural network is by substantial amounts of base
This neuron is connected with each other and forms, it is possible to carry out distributed variable-frequencypump and non-linear conversion, has powerful study and summary
The function concluded, has an ability of robustness, parallel processing, it is ensured that given precision.Should in BP neutral net and platform
The relation in, have the advantage that 1, inputting and exporting can keep Nonlinear Monotone relation;2, this function unusual light, symbol
Close relevant gradient to solve;3, to neutral net zmodem.Therefore, present invention target identification method based on Internet of Things is special
It is applicable to the pattern recognition to image and video, it is easy to promote.
Compared with general technology, approximation by polynomi-als is incorporated into Internet of Things by present invention Target Identification Unit based on Internet of Things
In field, draw the approximation by polynomi-als of transfer function in conjunction with Gaussian beam, it is ensured that computational accuracy also reduces computation complexity so that
Neural network model is able to quickly realize under the resource constrained environment of sensor, thus substantially increases real-time, and then very
Target recognition is realized well based on Internet of Things.
Embodiment described above only have expressed the several embodiments of the present invention, and it describes more concrete and detailed, but also
Therefore the restriction to the scope of the claims of the present invention can not be interpreted as.It should be pointed out that, for those of ordinary skill in the art
For, without departing from the inventive concept of the premise, it is also possible to make some deformation and improvement, these broadly fall into the guarantor of the present invention
Protect scope.Therefore, the protection domain of patent of the present invention should be as the criterion with claims.
Claims (6)
1. a target identification method based on Internet of Things, it is characterised in that comprise the following steps:
Obtain the measured value that in Internet of things system, predeterminated target is gathered by each sensor tip;
Using the described measured value of acquisition as the input quantity of BP neutral net, described BP neutral net is trained;Wherein, institute
The tanh transfer function stating BP neutral net uses the multinomial containing Gaussian beam to approach;Described multinomial is as follows:
Wherein,
Wherein,For Gaussian beam, represent the maximum integer being not more than x,Represent the smallest positive integral not less than x;
According to the described BP neutral net after training, described predeterminated target is carried out pattern recognition.
Target identification method based on Internet of Things the most according to claim 1, it is characterised in that at described acquisition Internet of Things
In system in the step of the measured value that predeterminated target is gathered by each sensor tip, described measured value includes temperature, humidity, pressure
Power and flow.
Target identification method based on Internet of Things the most according to claim 1, it is characterised in that described to described predetermined
Target carries out in the step of pattern recognition, and described predeterminated target includes image and video.
4. a Target Identification Unit based on Internet of Things, it is characterised in that include acquisition module, training module and identification mould
Block;
Described acquisition module, for obtaining the measured value that in Internet of things system, predeterminated target is gathered by each sensor tip;
Described training module, for the described measured value that will obtain as the input quantity of BP neutral net, to described BP nerve net
Network is trained;Wherein, the tanh transfer function of described BP neutral net uses the multinomial containing Gaussian beam to force
Closely;Described multinomial is as follows:
Wherein,
Wherein,For Gaussian beam, represent the maximum integer being not more than x,Represent the smallest positive integral not less than x;
Described identification module, for according to the described BP neutral net after training, carrying out pattern recognition to described predeterminated target.
Target Identification Unit based on Internet of Things the most according to claim 4, it is characterised in that described measured value includes temperature
Degree, humidity, pressure and flow.
Target Identification Unit based on Internet of Things the most according to claim 4, it is characterised in that described predeterminated target includes
Image and video.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201310590706.3A CN103606007B (en) | 2013-11-20 | 2013-11-20 | Target identification method based on Internet of Things and device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201310590706.3A CN103606007B (en) | 2013-11-20 | 2013-11-20 | Target identification method based on Internet of Things and device |
Publications (2)
Publication Number | Publication Date |
---|---|
CN103606007A CN103606007A (en) | 2014-02-26 |
CN103606007B true CN103606007B (en) | 2016-11-16 |
Family
ID=50124227
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201310590706.3A Active CN103606007B (en) | 2013-11-20 | 2013-11-20 | Target identification method based on Internet of Things and device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN103606007B (en) |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1945602A (en) * | 2006-07-07 | 2007-04-11 | 华中科技大学 | Characteristic selecting method based on artificial nerve network |
CN101329169A (en) * | 2008-07-28 | 2008-12-24 | 中国航空工业第一集团公司北京航空制造工程研究所 | Neural network modeling approach of electron-beam welding consolidation zone shape factor |
CN101561427A (en) * | 2009-05-15 | 2009-10-21 | 江苏大学 | Pig house environment harmful gas multi-point measurement system based on CAN field bus |
CN102749471A (en) * | 2012-07-13 | 2012-10-24 | 兰州交通大学 | Short-term wind speed and wind power prediction method |
CN102759430A (en) * | 2012-06-28 | 2012-10-31 | 北京自动化控制设备研究所 | BP (Back Propagation) neural network based high-precision correction and test method for resonance cylinder pressure sensor |
-
2013
- 2013-11-20 CN CN201310590706.3A patent/CN103606007B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1945602A (en) * | 2006-07-07 | 2007-04-11 | 华中科技大学 | Characteristic selecting method based on artificial nerve network |
CN101329169A (en) * | 2008-07-28 | 2008-12-24 | 中国航空工业第一集团公司北京航空制造工程研究所 | Neural network modeling approach of electron-beam welding consolidation zone shape factor |
CN101561427A (en) * | 2009-05-15 | 2009-10-21 | 江苏大学 | Pig house environment harmful gas multi-point measurement system based on CAN field bus |
CN102759430A (en) * | 2012-06-28 | 2012-10-31 | 北京自动化控制设备研究所 | BP (Back Propagation) neural network based high-precision correction and test method for resonance cylinder pressure sensor |
CN102749471A (en) * | 2012-07-13 | 2012-10-24 | 兰州交通大学 | Short-term wind speed and wind power prediction method |
Also Published As
Publication number | Publication date |
---|---|
CN103606007A (en) | 2014-02-26 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107273936B (en) | GAN image processing method and system | |
CN104599292B (en) | A kind of anti-noise moving object detection algorithm decomposed based on low-rank matrix | |
CN107368845A (en) | A kind of Faster R CNN object detection methods based on optimization candidate region | |
CN108763683B (en) | New WENO format construction method under trigonometric function framework | |
RU2012102394A (en) | METHOD FOR CALCULATING PHYSICAL VALUES, METHOD FOR NUMERICAL ANALYSIS, PROGRAM FOR CALCULATING PHYSICAL VALUES, PROGRAM FOR NUMERICAL ANALYSIS, DEVICE FOR CALCULATING PHYSICAL VALUES AND DEVICES FOR NUMERICAL ANALYSIS | |
CN108959794A (en) | A kind of structural frequency response modification methodology of dynamics model based on deep learning | |
CN104573621A (en) | Dynamic gesture learning and identifying method based on Chebyshev neural network | |
CN108596274A (en) | Image classification method based on convolutional neural networks | |
CN106127125A (en) | Distributed DTW human body behavior intension recognizing method based on human body behavior characteristics | |
CN113259331A (en) | Unknown abnormal flow online detection method and system based on incremental learning | |
CN104616324A (en) | Target tracking method based on adaptive appearance model and point-set distance metric learning | |
CN108628164A (en) | A kind of semi-supervised flexible measurement method of industrial process based on Recognition with Recurrent Neural Network model | |
CN105095857A (en) | Face data enhancement method based on key point disturbance technology | |
Kear et al. | Computational aerodynamic optimisation of vertical axis wind turbine blades | |
CN104182771A (en) | Time series data graphics analysis method based on automatic coding technology with packet loss | |
CN105138983A (en) | Pedestrian detection method based on weighted part model and selective search segmentation | |
CN104794455A (en) | Dongba hieroglyphic recognizing method | |
Lin et al. | Optimal CNN-based semantic segmentation model of cutting slope images | |
CN110288026A (en) | A kind of image partition method and device practised based on metric relation graphics | |
CN111159956B (en) | Feature-based flow field discontinuity capturing method | |
CN103606007B (en) | Target identification method based on Internet of Things and device | |
CN108898527B (en) | Traffic data filling method of generative model based on destructive measurement | |
Meddage et al. | Predicting transient wind loads on tall buildings in three-dimensional spatial coordinates using machine learning | |
Guo et al. | Design of a prediction system based on the dynamical feed-forward neural network | |
Cui et al. | Applications of Phase Field Methods in Modeling Fatigue Fracture and Performance Improvement Strategies: A Review |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C14 | Grant of patent or utility model | ||
GR01 | Patent grant |