CN109063571A - Artificial bee colony BP neural network signal recognition method based on WAVELET PACKET DECOMPOSITION - Google Patents
Artificial bee colony BP neural network signal recognition method based on WAVELET PACKET DECOMPOSITION Download PDFInfo
- Publication number
- CN109063571A CN109063571A CN201810727380.7A CN201810727380A CN109063571A CN 109063571 A CN109063571 A CN 109063571A CN 201810727380 A CN201810727380 A CN 201810727380A CN 109063571 A CN109063571 A CN 109063571A
- Authority
- CN
- China
- Prior art keywords
- solution
- signal
- neural network
- wavelet packet
- bee colony
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/086—Learning methods using evolutionary algorithms, e.g. genetic algorithms or genetic programming
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- Biophysics (AREA)
- Physiology (AREA)
- Biomedical Technology (AREA)
- Signal Processing (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Complex Calculations (AREA)
Abstract
The invention proposes a kind of artificial bee colony BP neural network signal recognition method based on WAVELET PACKET DECOMPOSITION, it is characterized in that, the following steps are included: carrying out WAVELET PACKET DECOMPOSITION to original signal, obtain the information on original signal different frequency range, focus on the energy of original signal in a few coefficients in transform domain, signal energy on different scale is solved to come, and is arranged according to scale sequence, feature vector is formed;Simple BP neural network classifier is optimized using artificial bee colony algorithm, is classified using the BP neural network classifier after optimization to feature vector, improves the classification performance of algorithm, and the validity of verification algorithm.The present invention calculates simply, in the signal identification under low signal-to-noise ratio, has higher Classification and Identification rate.Even if 94% or more still can be reached to the discrimination of signal under the signal-to-noise ratio of 5dB.
Description
Technical field
The present invention relates to a kind of radar signal feature methods.
Background technique
Blipology is all widely used in civil and military field, is that a kind of typical pattern-recognition is asked
Topic.In Modern Communication System, communication environment is complicated and changeable, and with the development that the communication technology is at full speed, the Modulation Types of signal become
More complicated multiplicity can not effectively carry out signal so that simple signal recognition method is difficult to meet actual demand
Identification.Therefore, how to improve the accuracy of signal identification in electromagnetic environment complicated and changeable is Modern Communication System research
It is crucial.
Currently, the major way of signal identification is characterized extraction and classifier design, this to a certain extent, realize compared with
Signal identification under low signal-to-noise ratio also provides good theoretical foundation for the identification in modern communication technology.But with communication
The increasingly complexity of environment, more and more difficult compared with identification signal accurate in low signal-to-noise ratio environment.In existing technical method, adopt
With traditional signal characteristic parameter and classifier design method, calculating is relatively easy, easy to accomplish.But low signal-to-noise ratio environment
In, it is difficult to signal characteristic parameter is accurately extracted, is suitable for noise and compares signal identification under epipodium border.This simple classification
Though device design can be realized the classification to signal, often accuracy is not high enough, and the classification accuracy rate meeting under low signal-to-noise ratio
It has a greatly reduced quality.As can be seen that related scholar increasingly payes attention to signal identification from the document that domestic and foreign scholars in recent years deliver
Theory, various modern signal processing technologies are all opened including wavelet theory, fractal theory, artificial neural network etc.
Begin or has been applied in the research to the theory.
Summary of the invention
The technical problem to be solved by the present invention is more effectively carrying out feature extraction to signal under the conditions of compared with low signal-to-noise ratio
With classification.
In order to solve the above-mentioned technical problem, the technical solution of the present invention is to provide a kind of based on the artificial of WAVELET PACKET DECOMPOSITION
Bee colony BP neural network signal recognition method, which comprises the following steps:
Step 1 carries out WAVELET PACKET DECOMPOSITION to original signal, obtains the information on original signal different frequency range, makes original letter
Number energy focused in a few coefficients in transform domain, the signal energy on different scale is solved to come, and according to scale
Sequence arranges, and forms feature vector;
Step 2 optimizes simple BP neural network classifier using artificial bee colony algorithm, uses the BP mind after optimization
Classify through network classifier to feature vector, improves the classification performance of algorithm, and the validity of verification algorithm.
Preferably, the step 1 the following steps are included:
Step 101 carries out time domain pretreatment to original signal;
Step 102 carries out N layers of WAVELET PACKET DECOMPOSITION to by pretreated signal, by WAVELET PACKET DECOMPOSITION to the low of signal
Frequency part carries out continuous decomposition, obtains the wavelet packet coefficient of slight part;
Step 103 carries out energy value calculating to node each on scale N;
The energy value calculated is saved and is exported by step 104, constitutes described eigenvector.
Preferably, in the step 2, using artificial bee colony algorithm to simple BP neural network classifier optimize including
Following steps:
One step 201, creation BP neural network.
Step 202, the parameter for initializing artificial bee colony algorithm, the size N including bee colonyc, gathering honey bee quantity Ne, follow
The quantity N of beeo, solution number Ns, limiting value limit, maximum cycle MCN and D tie up initial solution Xi(i=1 ..., Ns),
Wherein, D ties up initial solution Xi(i=1 ..., Ns) represent the connection weight and threshold values of created BP neural network in step 201;
Step 203 calculates each D dimension initial solution XiFitness value f (Xi):
In formula, MSEiIndicate i-th of D dimension initial solution XiBP network mean square error;
Step 204, gathering honey bee are with current memory solution for according to progress new explanation VijSearch:
Vij=Xij+ rand (- 1,1) (Xij-Xkj)
In formula, XijIndicating old solution, i indicates i-th of old solution, and j and k are random number, and j ∈ { 1,2 ..., D }, k ∈ 1,
2 ..., NsAnd k ≠ i, gathering honey bee is using greedy back-and-forth method, if the fitness value of new explanation writes down the old of update greater than old solution
Solution, on the contrary add 1 in the update frequency of failure of old solution;
Step 205, the probable value P for calculating each solutioni:
It follows bee using these probable values as foundation, the search of new explanation is carried out from the neighborhood of existing solution;
If step 206, more new explanation XiThe frequency of failure be more than preset limiting value, just illustrate that this solution can not continue quilt
It optimizes, it is given up, replaced with the new explanation that following formula generates:
Xi=Xmin+ rand (0,1) (Xmax-Xmin)
In formula, XminIndicate the minimum value of solution, XmaxIndicate the maximum value of solution;
If step 207, the number of iterations have been more than maximum cycle MCN, training terminates, conversely, return step 204;
Step 208, the weight and threshold values that obtained optimal solution is transformed into BP network test nerve net with data simulation
Network.
Preferably, in the step 202, the size N of bee colonyc, gathering honey bee quantity Ne, follow the quantity N of beeo, solution
Number NsMeet following relationship:
Nc=2Ns=Ne+No, Ne=No。
Preferably, in the step 202, the dimension D of each D dimension initial solution meets following equations:
D=Ninput×Nhidden+Nhidden+Nhidden×Noutput+Noutput, in formula, Ninput、Nhidden、NoutputIt is respectively
Input layer, hidden layer, output layer neuron number.
The present invention proposes artificial bee colony algorithm, to traditional BP neural network classification on the basis of traditional BP neural network
Device optimizes, and more accurately classifies to realize to signal, and the algorithm calculates simply, easy to accomplish.
Artificial bee colony algorithm of the present invention is the optimization algorithm based on bee colony intelligent behavior, it is mainly used for
Continuous optimization problems are solved, are evolved relative to genetic algorithm (GA), particle group optimizing (PSO), differential evolution (DE) and population
Algorithm (PS-EA) etc. has preferably optimization performance.Therefore, the neural network classifier phase optimized using artificial bee colony algorithm
For general neural network classifier, preferably signal can be analyzed, realized under low signal-to-noise ratio environment to letter
Number accurately identify.
Existing technical method be usually signal is directly carried out feature extraction and use common BP neural network algorithm into
Row classification, often complexity is higher for the good method of noiseproof feature, for calculating relatively simple algorithm, under low signal-to-noise ratio often
It is difficult to reach satisfied recognition effect.The technical solution that the invention is proposed mentions signal characteristic using WAVELET PACKET DECOMPOSITION
It takes, energy value calculating is carried out to each node after decomposition, passes through the BP neural network classifier optimized based on artificial bee colony algorithm
Classify, 2ASK, 2FSK, 2DPSK signal plus different distributions noise is identified, the validity of verification algorithm, with
Realize the purpose for improving the accuracy of signal identification under more low signal-to-noise ratio.
The present invention is asked for increasingly complicated communication electromagnetism environment, not high this of normal signal recognizer recognition correct rate
Topic proposes a kind of new signal Recognition Algorithm based on artificial bee colony algorithm and WAVELET PACKET DECOMPOSITION, calculates simply, in low noise
In signal identification than under, there is higher Classification and Identification rate.Even if under the signal-to-noise ratio of 5dB, still to the discrimination of signal
94% or more can be reached.
Detailed description of the invention
Fig. 1 is that BP algorithm emulates classification results figure;
Fig. 2 is that BP algorithm emulates classification results figure under 15dB;
Fig. 3 is that 5dB believes that lower BP algorithm emulates classification results;
Fig. 4 is that Optimal BP Algorithm emulates classification results;
Fig. 5 is that Optimal BP Algorithm emulates classification results under 15dB;
Fig. 6 is that Optimal BP Algorithm emulates classification results under 5dB;
Fig. 7 is the comparison of worst accuracy;
Fig. 8 is the comparison of optimal accuracy;
Fig. 9 is optimal accuracy comparison under 5dB;
Figure 10 is worst accuracy comparison under 5dB.
Specific embodiment
Present invention will be further explained below with reference to specific examples.It should be understood that these embodiments are merely to illustrate the present invention
Rather than it limits the scope of the invention.In addition, it should also be understood that, after reading the content taught by the present invention, those skilled in the art
Member can make various changes or modifications the present invention, and such equivalent forms equally fall within the application the appended claims and limited
Range.
The invention firstly uses feature extraction is carried out to signal using WAVELET PACKET DECOMPOSITION, recycle artificial bee colony algorithm to letter
Single BP neural network classifier optimizes, and improves the simple BP neural network classifier of tradition to the accuracy of Modulation recognition.
WAVELET PACKET DECOMPOSITION is that wavelet packet is indicated with parsing tree, i.e., analyzes input signal using the wavelet conversion repeatedly iterated
Detail section, it is a kind of method of finer signal analysis, improves the time resolution of signal.
Wavelet packet is defined as follows:
Given scaling function φ (t) and wavelet function ψ (t), scaling relation are as follows:
In formula (1), (2), h0kAnd h1kIt is the filter coefficient in multiresolution analysis respectively, parameter k is positioning index,
Parameter t is time parameter.
Define following recurrence relation:
Parameter n is the number of oscillation, wn(2t-k) indicates the wavelet packet functions set in (2t-k) scale.As n=0, w0
(t)=φ0(t), w1(t)=ψ (t).Function set { w defined aboven(t)}n∈ZFor w0(t)=φ0(t) small echo determined by
It wraps, as a result, wavelet packet { wn(t)}n∈zBeing includes scaling function w0(t) and mother wavelet function w1(t) one including has centainly
The set of the function of connection.
Artificial bee colony algorithm is that Dervis Karaboga in 2005 obtains inspiration from the intelligent behavior of honeybee and puts forward
's.Honeybee is that nectar source information transmitting is carried out by way of jive.When jive, signified direction is represented
Direction of the nectar source with respect to the sun, the frequency representation that is waved by honeybee of distance in nectar source, the length of swing duration refer to
The number of nectar source honey amount shown.In artificial bee colony algorithm, the position in nectar source is exactly a possible solution in optimization problem, flower
Honey number represent the quality or grade of fit of solution corresponding to it.One artificial bee colony is made of 3 parts: gathering honey bee follows
Bee, investigation bee.The quantity N of gathering honey beeeWith the quantity N for following bee0It is identical, and is equal to the number N of solutions.Each solution
It is a D dimensional vector, wherein D represents the number for needing to optimize parameter.In the algorithm, there are 3 parameters as control: solution
Number Ns, limiting value limit, maximum cycle MCN.
Firstly, initial population, i.e. N is randomly generated in algorithmsA solution X1..., XNs, setting limit value limit and largest loop
Number MCN.By initialization procedure, honeybee starts to carry out cyclic search: each gathering honey bee is using greedy back-and-forth method, from the neighbour of solution
The new solution of domain search, if the grade of fit of new explanation is greater than the grade of fit of old solution, gathering honey bee forgets old solution, remembers new explanation.Calculate this
The probable value P solved a biti, follow probable value P of the bee again with these solutionsiThe solution remembered by foundation and gathering honey bee is new in neighborhood search
Solution, searching method still uses greedy back-and-forth method.If some newest solution cannot be updated again, (number for updating failure is greater than
Limiting value limit), then this solution can be abandoned by investigation bee, be replaced with new solution.Reciprocation cycle, until maximum cycle.
Only one investigation bee is set in each circulation.
Based on above-mentioned basis, a kind of artificial bee colony BP neural network signal knowledge based on WAVELET PACKET DECOMPOSITION provided by the invention
Other method, comprising the following steps:
The first step, the signal characteristic extracting methods based on WAVELET PACKET DECOMPOSITION
Information after carrying out WAVELET PACKET DECOMPOSITION to signal, on available original signal different frequency range.Become using wavelet packet
It changes, focuses on the energy of original signal in a few coefficients in transform domain, the signal energy on different scale is solved to come,
And arranged according to scale sequence, forming feature vector is that subsequent signal identification is provided fundamental basis, and specific step is as follows for algorithm:
Step 101 carries out time domain pretreatment to original signal;
Step 102 carries out 4 layers of WAVELET PACKET DECOMPOSITION to by pretreated signal, by WAVELET PACKET DECOMPOSITION to the low of signal
Frequency part carries out continuous decomposition, obtains the wavelet packet coefficient of slight part;
Step 103 carries out energy value calculating to node each on scale 4;
The energy value calculated is saved and is exported by step 104, constitutes described eigenvector so that classifier is classified
Identification.
Second step, the BP neural network classifier based on artificial bee colony algorithm optimization
The present invention merges artificial bee colony algorithm with BP neural network classifier, is found most using artificial bee colony algorithm
Excellent network weight and threshold values, has the extensive mapping ability of neural network and the global iterative of artificial bee colony algorithm concurrently and part is searched
The characteristics of rope.Specific step is as follows for Optimizing BP Network classifier design based on artificial bee colony algorithm:
One step 201, creation BP neural network.
Step 202, the parameter for initializing artificial bee colony algorithm, the size N including bee colonyc, gathering honey bee quantity Ne, follow
The quantity N of beeo, solution number Ns, limiting value limit, maximum cycle MCN and D tie up initial solution Xi(i=1 ..., Ns),
Wherein:
The size N of bee colonyc, gathering honey bee quantity Ne, follow the quantity N of beeo, solution number NsMeet following relationship:
Nc=2Ns=Ne+No, Ne=No;
D ties up initial solution Xi(i=1 ..., Ns) represent the connection weight and valve of created BP neural network in step 201
Value.The dimension D of each D dimension initial solution meets following equations:
D=Ninput×Nhidden+Nhidden+Nhiaden×Noutput+Noutput, in formula, Ninput、Nhidden、NoutputIt is respectively
Input layer, hidden layer, output layer neuron number;The value of initial solution is the number between (- 1,1) being randomly generated;
Step 203 calculates each D dimension initial solution XiFitness value f (Xi):
In formula, MSEiIndicate i-th of D dimension initial solution XiBP network mean square error;Obviously, as grade of fit MSEiReach 1
When be optimal state.
Step 204, gathering honey bee are with current memory solution for according to progress new explanation VijSearch:
Vij=Xij+ rand (- 1,1) (Xij-Xkj)
In formula, XijIndicating old solution, i indicates i-th of old solution, and j and k are random number, and j ∈ { 1,2 ..., D }, k ∈ 1,
2 ..., NsAnd k ≠ i, gathering honey bee is using greedy back-and-forth method, if the fitness value of new explanation writes down the old of update greater than old solution
Solution, on the contrary add 1 in the update frequency of failure of old solution;
Step 205, the probable value P for calculating each solutioni:
It follows bee using these probable values as foundation, the search of new explanation is carried out from the neighborhood of existing solution;
If step 206, more new explanation XiThe frequency of failure be more than preset limiting value, just illustrate that this solution can not continue quilt
It optimizes, it is given up, replaced with the new explanation that following formula generates:
Xi=Xmin+ rand (0,1) (Xmax-Xmin)
In formula, XminIndicate the minimum value of solution, XmaxIndicate the maximum value of solution;
If step 207, the number of iterations have been more than maximum cycle MCN, training terminates, conversely, return step 204;
Step 208, the weight and threshold values that obtained optimal solution is transformed into BP network test nerve net with data simulation
Network.
After carrying out WAVELET PACKET DECOMPOSITION to signal and solve energy value, energy feature data are as shown in table 1:
1 2ASK, 2FSK, 2DPSK signal extraction characteristic of table
Recycle traditional BP neural network classifier and the improved BP neural network classifier based on artificial bee colony algorithm
Classify respectively to above data, classification results are as shown in table 2:
The different signal-to-noise ratio classification accuracy rate comparisons of table 2
Simulation result shows there is preferable classification effect using the BP neural network classifier of artificial bee colony algorithm optimization
Fruit.BP neural network classifier is as shown in Figure 1 to Figure 3 to the accuracy of Modulation recognition, the BP optimized using artificial bee colony algorithm
Neural network classifier is as shown in Figures 4 to 6 to the accuracy of Modulation recognition.From the comparison of table 2 as can be seen that for difference
2ASK, 2FSK, 2DPSK signal that the noise sequence of different distributions is added under signal-to-noise ratio utilize the BP nerve of artificial bee colony optimization
The accuracy of network algorithm classification is better than the accuracy of simple BP neural network algorithm classification, even if under very low signal-to-noise ratio,
Discrimination still with higher.Due to the difference of each additional noise, lead to the knowledge under the noise of different signal-to-noise ratio to signal
There are certain fluctuations for other result, calculate recognition result and optimal identification result worst in identification process under different signal-to-noise ratio,
And comparison diagram is done, as shown in Figure 7 to 10.Which show the BP neural network classifiers pair under different signal-to-noise ratio, optimizing front and back
Modulation recognition is optimal and the comparison of worst accuracy.
Claims (5)
1. a kind of artificial bee colony BP neural network signal recognition method based on WAVELET PACKET DECOMPOSITION, which is characterized in that including following
Step:
Step 1 carries out WAVELET PACKET DECOMPOSITION to original signal, obtains the information on original signal different frequency range, makes original signal
Energy focuses in a few coefficients in transform domain, the signal energy on different scale is solved to come, and according to scale sequence
Arrangement forms feature vector;
Step 2 optimizes simple BP neural network classifier using artificial bee colony algorithm, uses the BP nerve net after optimization
Network classifier classifies to feature vector, improves the classification performance of algorithm, and the validity of verification algorithm.
2. a kind of artificial bee colony BP neural network signal recognition method based on WAVELET PACKET DECOMPOSITION as described in claim 1,
Be characterized in that, the step 1 the following steps are included:
Step 101 carries out time domain pretreatment to original signal;
Step 102 carries out N layers of WAVELET PACKET DECOMPOSITION to by pretreated signal, by WAVELET PACKET DECOMPOSITION to the low frequency portion of signal
Divide and carry out continuous decomposition, obtains the wavelet packet coefficient of slight part;
Step 103 carries out energy value calculating to node each on scale N;
The energy value calculated is saved and is exported by step 104, constitutes described eigenvector.
3. a kind of artificial bee colony BP neural network signal recognition method based on WAVELET PACKET DECOMPOSITION as described in claim 1,
It is characterized in that, in the step 2, simple BP neural network classifier is optimized including following step using artificial bee colony algorithm
It is rapid:
One step 201, creation BP neural network.
Step 202, the parameter for initializing artificial bee colony algorithm, the size N including bee colonyc, gathering honey bee quantity Ne, follow bee
Quantity No, solution number Ns, limiting value Iimit, maximum cycle MCN and D tie up initial solution Xi(i=1 ..., Ns), wherein
D ties up initial solution Xi(i=1 ..., Ns) represent the connection weight and threshold values of created BP neural network in step 201;
Step 203 calculates each D dimension initial solution XiFitness value f (Xi):
In formula, MSEiIndicate i-th of D dimension initial solution XiBP network mean square error;
Step 204, gathering honey bee are with current memory solution for according to progress new explanation VijSearch:
Vij=Xij+ rand (- 1,1) (Xij-Xkj)
In formula, XijIndicating old solution, i indicates i-th of old solution, and j and k are random number, and j ∈ { 1,2 ..., D }, k ∈ 1,2 ...,
NsAnd k ≠ i, gathering honey bee writes down the old solution of update if the fitness value of new explanation is greater than old solution using greedy back-and-forth method, on the contrary
Add 1 in the update frequency of failure of old solution;
Step 205, the probable value P for calculating each solutioni:
It follows bee using these probable values as foundation, the search of new explanation is carried out from the neighborhood of existing solution;
If step 206, more new explanation XiThe frequency of failure be more than preset limiting value, it is optimised just to illustrate that this solution can not continue
, it is given up, is replaced with the new explanation that following formula generates:
Xi=Xmin+ rand (0,1) (Xmax-Xmin)
In formula, XminIndicate the minimum value of solution, XmaxIndicate the minimum value of solution;
If step 207, the number of iterations have been more than maximum cycle MCN, training terminates, conversely, return step 204;
Step 208, the weight and threshold values that obtained optimal solution is transformed into BP network test neural network with data simulation.
4. a kind of artificial bee colony BP neural network signal recognition method based on WAVELET PACKET DECOMPOSITION as claimed in claim 3,
It is characterized in that, in the step 202, the size N of bee colonyc, gathering honey bee quantity Ne, follow the quantity N of beeo, solution number NsIt is full
It is enough lower relationship:
Nc=2Ns=Ne+No, Ne=No。
5. a kind of artificial bee colony BP neural network signal recognition method based on WAVELET PACKET DECOMPOSITION as claimed in claim 3,
It is characterized in that, in the step 202, the dimension D of each D dimension initial solution meets following equations:
D=Ninput×Nhidden+Nhidden+Nhidden×Noutput+Noutput, in formula, Ninput、Nhidden、NoutputIt is input respectively
Layer, hidden layer, output layer neuron number.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810727380.7A CN109063571A (en) | 2018-07-04 | 2018-07-04 | Artificial bee colony BP neural network signal recognition method based on WAVELET PACKET DECOMPOSITION |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810727380.7A CN109063571A (en) | 2018-07-04 | 2018-07-04 | Artificial bee colony BP neural network signal recognition method based on WAVELET PACKET DECOMPOSITION |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109063571A true CN109063571A (en) | 2018-12-21 |
Family
ID=64819180
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810727380.7A Pending CN109063571A (en) | 2018-07-04 | 2018-07-04 | Artificial bee colony BP neural network signal recognition method based on WAVELET PACKET DECOMPOSITION |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109063571A (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109617888A (en) * | 2018-12-24 | 2019-04-12 | 湖北大学 | A kind of anomalous traffic detection method neural network based and system |
CN109977972A (en) * | 2019-03-29 | 2019-07-05 | 东北大学 | A kind of intelligent characteristic recognition methods based on STEP |
CN112183462A (en) * | 2020-10-23 | 2021-01-05 | 东北大学 | Method for controlling intelligent mouse device by means of surface myoelectricity of stump |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105225223A (en) * | 2015-08-27 | 2016-01-06 | 南京市计量监督检测院 | Based on the damage Detection of Smart Composite Structure method of wavelet analysis and BP neural network |
-
2018
- 2018-07-04 CN CN201810727380.7A patent/CN109063571A/en active Pending
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105225223A (en) * | 2015-08-27 | 2016-01-06 | 南京市计量监督检测院 | Based on the damage Detection of Smart Composite Structure method of wavelet analysis and BP neural network |
Non-Patent Citations (5)
Title |
---|
XIAOLEI ZHU等: "Automatic Recognition of Communication Signal Modulation based on Neural Network", 《2016 IEEE INTERNATIONAL CONFERENCE ON ELECTRONIC INFORMATION AND COMMUNICATION TECHNOLOGY (ICEICT)》 * |
史先铭等: "改进ABC-BP模型在数字调制识别中的应用", 《江南大学学报(自然科学版)》 * |
徐俊艳等: "基于小波和神经网络的雷达目标识别方法研究", 《山东农业大学学报(自然科学版)》 * |
李坤等: "基于小波变换的电力线通信信号识别研究", 《电子技术应用》 * |
雷恒恒等: "基于小波包变换的雷达辐射源信号识别", 《电子信息对抗技术》 * |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109617888A (en) * | 2018-12-24 | 2019-04-12 | 湖北大学 | A kind of anomalous traffic detection method neural network based and system |
CN109617888B (en) * | 2018-12-24 | 2021-05-07 | 湖北大学 | Abnormal flow detection method and system based on neural network |
CN109977972A (en) * | 2019-03-29 | 2019-07-05 | 东北大学 | A kind of intelligent characteristic recognition methods based on STEP |
CN109977972B (en) * | 2019-03-29 | 2023-02-07 | 东北大学 | Intelligent feature identification method based on STEP |
CN112183462A (en) * | 2020-10-23 | 2021-01-05 | 东北大学 | Method for controlling intelligent mouse device by means of surface myoelectricity of stump |
CN112183462B (en) * | 2020-10-23 | 2023-07-25 | 东北大学 | Method for electrically controlling intelligent mouse device by means of surface myoelectricity of residual limb |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107622104B (en) | Character image identification and marking method and system | |
Peng et al. | Reverse graph learning for graph neural network | |
WO2020108474A1 (en) | Picture classification method, classification identification model generation method and apparatus, device, and medium | |
CN109063571A (en) | Artificial bee colony BP neural network signal recognition method based on WAVELET PACKET DECOMPOSITION | |
Bai et al. | Unsupervised adversarial instance-level image retrieval | |
CN102779510A (en) | Speech emotion recognition method based on feature space self-adaptive projection | |
CN111414964A (en) | Image security identification method based on defense sample | |
CN111047054A (en) | Two-stage countermeasure knowledge migration-based countermeasure sample defense method | |
Wei et al. | Automatic modulation recognition using neural architecture search | |
Zhang et al. | Targeted attack of deep hashing via prototype-supervised adversarial networks | |
CN115048983A (en) | Counterforce sample defense method of artificial intelligence system based on data manifold topology perception | |
Li et al. | Optimizing convolutional neural network performance by mitigating underfitting and overfitting | |
Jiang et al. | Classification for incomplete data using classifier ensembles | |
Wang et al. | Out-of-distributed semantic pruning for robust semi-supervised learning | |
CN117593398A (en) | Remote sensing image generation method based on diffusion model | |
Zeng et al. | Term-based pooling in convolutional neural networks for text classification | |
CN113434668B (en) | Deep learning text classification method and system based on model fusion | |
CN110704575B (en) | Dynamic self-adaptive binary hierarchical vocabulary tree image retrieval method | |
Ma | The Research of Stock Predictive Model based on the Combination of CART and DBSCAN | |
CN114120049A (en) | Long tail distribution visual identification method based on prototype classifier learning | |
Chen et al. | MagicGAN: multiagent attacks generate interferential category via GAN | |
Cui et al. | Weighted particle swarm clustering algorithm for self-organizing maps | |
CN114595336A (en) | Multi-relation semantic solution model based on Gaussian mixture model | |
Zheng et al. | Unlabeled text classification optimization algorithm based on active self-paced learning | |
Tang et al. | Efficient image classification using sparse coding and random forest |
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: 20181221 |
|
RJ01 | Rejection of invention patent application after publication |