CN106503649A - A kind of short-wave signal detection recognition method based on computer vision - Google Patents
A kind of short-wave signal detection recognition method based on computer vision Download PDFInfo
- Publication number
- CN106503649A CN106503649A CN201610917977.9A CN201610917977A CN106503649A CN 106503649 A CN106503649 A CN 106503649A CN 201610917977 A CN201610917977 A CN 201610917977A CN 106503649 A CN106503649 A CN 106503649A
- Authority
- CN
- China
- Prior art keywords
- short
- frequency
- wave signal
- time
- signal
- 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.)
- Granted
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
- G06F2218/08—Feature extraction
- G06F2218/10—Feature extraction by analysing the shape of a waveform, e.g. extracting parameters relating to peaks
-
- 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
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Signal Processing (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Image Analysis (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
The invention discloses a kind of short-wave signal detection recognition method based on computer vision, according to the time-frequency matrix character and the grader building process of machine learning of short-wave signal, a kind of short-wave signal detection and method of discrimination based on computer vision of invention, compared to traditional signal detection discriminant approach based on manual type and other existing signal detection method of discrimination, this method can achieve Aulomatizeted Detect and identification of the computer to short-wave signal, while more other signal detection method of discrimination, noise circumstance residing for signal is required lower, it is applied to the noise as little as channel circumstance of 0dB.By having carried out a large amount of real data tests under the conditions of different channels, the detection of short-wave signal differentiates accuracy more than 90%.
Description
Technical field
The present invention relates to technical field of computer vision and machine learning field, more particularly, to one kind to being based on computer
The short-wave signal feature that vision is extracted via SVM (Support Vector Machine) classification and then realizes that short-wave signal is positioned
Detection method with common type identification.
Background technology
Short-wave signal is one of most commonly seen signal set in radio communication.According to Consultative Committee on International Radio (CCIR)
(CCIR) division, shortwave are defined as wavelength in 100m~10m, and frequency is the electromagnetic wave of 3MHz~30MHz, is carried out using shortwave
Radio communication become short wave communication, also known as high frequency (HF) communicate.In being usually used, in order to make full use of shortwave closely
The advantage of communication, the actually used frequency range of short wave communication are 1.5MHz~30MHz.
Technology according to the present invention is the detection identification of the short-wave signal based on computer vision.Wherein, computer vision
Introducing particularly important.
Computer vision refers to and replace human eye to be identified target with computer, track, measure, so as to realize similar to
One computer technology of human visual function.Which mainly contains the content of two aspects of image procossing and pattern recognition:Figure
As processing procedure is the position fixing process of target, main realization carries out target location analysis to the visual information for being input into computer, often
Method includes image denoising, dimensionality reduction etc.;Mode identification procedure refers to the various forms of (numerical value to characterizing things or phenomenon
, word and logical relation) information processed and analyzed, things or phenomenon being described, be recognized, is classified reconciliation
The process that releases, is the important component part of information science and artificial intelligence.Common mode identification procedure is based on machine learning
Grader realize.Which mainly includes data training set structure, the feature extraction of training data, the features training of grader
Etc. step, eventually through testing classification device for known and unknown data classifier performance, the performance of classification of assessment device is good and bad.
At present, short-wave signal detection is also main based on manually scouting mode, i.e., by business personnel manually to short-wave reception
The signal that machine is received is intercepted.When a certain suspicious or neutral signal is listened to, in addition it is also necessary to by instrument to signal wave
Shape carries out visual analyzing, and receives control accordingly according to acquired signal parameter development and keep equipment.The process is to business personnel
Short-wave signal intercept that skill requirement is very high, artificial training cost is very big.
The present invention is directed to the cost of labor height of traditional artificial short-wave signal detection method, takes high deficiency, proposes
A kind of short-wave signal detection recognition method based on computer vision, by computer to common class signal in short-wave signal
Carry out positioning and the analysis of automatization, it is achieved that the autonomous type identification of the computer of short-wave signal, effectively increase signal detection
Ageing.
Content of the invention
The technical problem to be solved is for the short-wave signal manual detection identification in noisy communication channel
The high problem of time-consuming high, complexity, there is provided a kind of computer vision feature modeling by signal is simultaneously judged by SVM classifier
Short-wave signal Aulomatizeted Detect and knowledge method for distinguishing, effectively improve in the case where short-wave signal detection identification accuracy is ensured
The process ageing.
The present invention solve the technical scheme that adopted of above-mentioned technical problem for:A kind of short-wave signal based on computer vision
Detection and localization is extracted with time-frequency characteristics, and the SVM classifier based on machine learning, the time-frequency characteristics of signal is mated, most
Aulomatizeted Detect and the identification of short-wave signal are realized eventually.Technical scheme flow process is as shown in Figure 1.
The method of the present invention is comprised the steps of:
1) to signal time-frequency data matrix to be detected, according to frequency direction per 300kHz piecemeals, frequency during time-frequency is obtained
According to submatrix.
2) to each time-frequency data submatrix, following signal framing detection is carried out, concrete operations are as follows:
2.1) time-frequency data submatrix is projected on the frequency axis and obtains one-dimensional row and vector V, V are represented
Number energy cumulative, as shown in Fig. 2 wherein Fig. 2 (a) is time-frequency data submatrix figure, Fig. 2 (b) is based on one-dimensional row and vector V
The drop shadow curve of drafting.
2.2) distribution histogram of each element in V is analyzed.Locating threshold T=(V_med-V_min)+V_med, its
Middle V_med represents that the mode of V element, V_min represent the minima of element in V.
2.3) the locating threshold T in previous step is combined as thresholding, and the 2.1) drop shadow curve in step, bent with reference to projection
It is higher than the frequency separation of thresholding in line, corresponding frequency separation in time-frequency matrix is intercepted, currently processed time-frequency is obtained
The set of the time frequency signal submatrix of short-wave signal present in data submatrix.
3) the corresponding time frequency signal submatrix set of each time-frequency data submatrix for obtaining first two steps merges, and is made
Time frequency signal set of matrices for the whole short-wave signals for including in the signal time-frequency data matrix to be detected of input.
4) set constituted by the time frequency signal matrix of each short-wave signal, carries out signal characteristic abstraction one by one.
4.1) kurtosis feature
The time frequency signal matrix of short-wave signal is projected on the frequency axis and obtains one-dimension array Y.Y is carried out at windowed traversal
Reason, as shown in Figure 3.Calculate the difference of the corresponding numerical value of point of proximity respectively, and substituted in equation below.
The value of each point in Y is replaced with corresponding Du, kurtosis curve Y ' is finally given.The value of each point in Y is replaced
For corresponding Du, kurtosis curve Y ' is finally given.
Packet sequencing is subsequently carried out:Using the null value in kurtosis curve as partitioning standards, kurtosis curve is divided into multiple
Spike group, the statistics corresponding kurtosis curve regions maximum of each group obtain the row that spike group maximum value sequence carries out from big to small
Sequence.
Subsequently the spike group maximum value sequence for sorting from big to small obtained in packet sequencing is divided into two classes by given threshold,
Threshold value for the corresponding kurtosis curve of maximum sharpness group maximum divided by constant C (empirical value C=10^3), take group sequence smaller portions
As local spike group maximum value sequence, and to the part spike group maximum value sequence.
Ratio asked successively to two adjacent values of spike group maximum value sequence, statistics obtain maximum ratio corresponding two most
Spike group maximum value sequence, using two maximums as division limits, is divided into two parts, takes sequence order less by big value position
Part as effective spike group maximum value sequence, count the maximum number that the partial sequence includes, as effective spike
Count, and the result as kurtosis feature is exported;
4.2) spatial distribution characteristic
The time frequency signal matrix of short-wave signal is projected on the frequency axis and obtains one-dimension array Y.Evaluation domain interval | V (i) |
∈ [a, b], if interval size is k (k=b-a), divides to k and (by studying the relation of k value and resolution, obtains corresponding relation
Array), the scatterplot quantity in each interval is counted, the ratio that the scatterplot number included in each interval accounts for total scatterplot number is calculated.
Each interval scatterplot proportionality percentage array is exported as spatial distribution characteristic.
4.3) symmetric characteristics
The time frequency signal matrix of short-wave signal is projected on the frequency axis and obtains one-dimension array Y.Evaluation domain interval [a, b],
Thresholding C is set, the part in curve higher than C is effective coverage.Wherein, C=0.2* (b-a)+a.With curve peak as starting
Point, position when looking for curve to decline to for the first time below C to both sides seek starting point with both sides end point respectively as end point
Apart from d_right and d_left.The ratio of d_right and d_left is sought, when symmetrical, the result is close to 1.
4.4) Variance feature
The time frequency signal matrix of short-wave signal is projected on the frequency axis and obtains one-dimension array Y, extract maximum in array Y
Position is corresponding frequency row Y (N), calculates the variance of Y (N) rows, exports as Variance feature.
4.5) rectangle morphological characteristic
For each short-wave signal time-frequency data matrix, even rectangular degree Pr is calculated.Computational methods are Pr=As/Ar (As
It is the area of connected domain S in image, Ar is the minimum enclosed rectangle area for surrounding the connected domain).
Corresponding to single short-wave signal time-frequency data matrix 4.1) integrate to 4.5) feature, obtain the short-wave signal
The eigenmatrix of time-frequency data matrix.
5) each short-wave signal Input matrix for 4) obtaining is obtained grader in the good SVM classifier of training in advance
Output, according to grader output result, is marked to short-wave signal type.
Described 5) in, used the SVM classifier that has trained, the kurtosis feature extracted based on machine vision, space point
Cloth feature, symmetric characteristics, Variance feature, rectangle morphological characteristic, are identified to short-wave signal type
Compared with prior art, it is an advantage of the current invention that:Traditional short-wave signal detection recognition method relies primarily on handss
Work mode detects that cost of labor is high, and uses duration.Meanwhile, the more already present short-wave signal detection for being related to computer technology
Recognition methodss application scenarios are more harsh, and for the sensitivity of noise is very high, and identification types are more single.Fully profit of the invention
With the information of short-wave signal time-frequency matrix, in conjunction with computer vision technique, short-wave signal time-frequency matrix is carried out manifold
Extract, and select efficient machine learning SVM classifier to be trained and recognize.Whole process is compared to traditional short-wave signal
Detection method, while accuracy rate is ensured, effectively reduces cost of labor and the used time of detection, improves short-wave signal detection
Recognized is ageing.
Test result indicate that, under noise level lower limit is for the noisy channel of 0dB, the average detected of short-wave signal and identification are just
Really rate is up to more than 90%.
Description of the drawings
Fig. 1 is the flow chart of the inventive method;
Fig. 2 is time-frequency data submatrix perspective view on the frequency axis, and wherein Fig. 2 (a) is time-frequency data submatrix figure,
Fig. 2 (b) is the drop shadow curve that is drawn based on one-dimensional row and vector V.
Fig. 3 is the windowing process schematic diagram of kurtosis feature.
Specific embodiment
The present invention is further elaborated below in conjunction with accompanying drawing.
The present invention devises a kind of shortwave based on machine vision for the short-wave signal detection in noisy channel with identification
Automatic signal detection recognition methodss.In actual use, computer by by this method come the noisy short-wave signal to being input into
Carry out detection identification.Method of the present invention step is as follows:
The first step:To signal time-frequency data matrix to be detected, according to frequency direction per 300kHz piecemeals, when obtaining time-frequency
Frequency is according to submatrix.
Second step:To each time-frequency data submatrix, following signal framing detection is carried out, concrete operations are as follows:
1) time-frequency data submatrix is projected on the frequency axis and obtains one-dimensional row and vector V, V represent signal in frequency direction
Energy cumulative, as shown in Fig. 2 wherein Fig. 2 (a) is time-frequency data submatrix figure, Fig. 2 (b) is to be painted based on one-dimensional row and vector V
The drop shadow curve of system.
2) distribution histogram of each element in V is analyzed.Locating threshold T=(V_med-V_min)+V_med, wherein
V_med represents that the mode of V element, V_min represent the minima of element in V.
3) the locating threshold T in previous step is combined as thresholding, and the 2.1) drop shadow curve in step, with reference to drop shadow curve
In higher than thresholding frequency separation, corresponding frequency separation in time-frequency matrix is intercepted, currently processed when frequency is obtained
Set according to the time frequency signal submatrix of short-wave signal present in submatrix.
3rd step:The corresponding time frequency signal submatrix set of each time-frequency data submatrix that first two steps are obtained merges,
Obtain the time frequency signal set of matrices of the whole short-wave signals for including in the signal time-frequency data matrix to be detected as input.
4th step:The set constituted by the time frequency signal matrix of each short-wave signal, carries out signal characteristic abstraction one by one.
1) kurtosis feature
The time frequency signal matrix of short-wave signal is projected on the frequency axis and obtains one-dimension array Y.Y is carried out at windowed traversal
Reason, as shown in Figure 3.Calculate the difference of the corresponding numerical value of point of proximity respectively, and substituted in equation below.
The value of each point in Y is replaced with corresponding Du, kurtosis curve Y ' is finally given.The value of each point in Y is replaced
For corresponding Du, kurtosis curve Y ' is finally given.
Packet sequencing is subsequently carried out:Using the null value in kurtosis curve as partitioning standards, kurtosis curve is divided into multiple
Spike group, the statistics corresponding kurtosis curve regions maximum of each group obtain the row that spike group maximum value sequence carries out from big to small
Sequence.
Subsequently the spike group maximum value sequence for sorting from big to small obtained in packet sequencing is divided into two classes by given threshold,
Threshold value for the corresponding kurtosis curve of maximum sharpness group maximum divided by constant C (empirical value C=10^3), take group sequence smaller portions
As local spike group maximum value sequence, and to the part spike group maximum value sequence.
Ratio asked successively to two adjacent values of spike group maximum value sequence, statistics obtain maximum ratio corresponding two most
Spike group maximum value sequence, using two maximums as division limits, is divided into two parts, takes sequence order less by big value position
Part as effective spike group maximum value sequence, count the maximum number that the partial sequence includes, as effective spike
Count, and the result as kurtosis feature is exported;
2) spatial distribution characteristic
The time frequency signal matrix of short-wave signal is projected on the frequency axis and obtains one-dimension array Y.Evaluation domain interval | V (i) |
∈ [a, b], if interval size is k (k=b-a), divides to k and (by studying the relation of k value and resolution, obtains corresponding relation
Array), the scatterplot quantity in each interval is counted, the ratio that the scatterplot number included in each interval accounts for total scatterplot number is calculated.
Each interval scatterplot proportionality percentage array is exported as spatial distribution characteristic.
3) symmetric characteristics
The time frequency signal matrix of short-wave signal is projected on the frequency axis and obtains one-dimension array Y.Evaluation domain interval [a, b],
Thresholding C is set, the part in curve higher than C is effective coverage.Wherein, C=0.2* (b-a)+a.With curve peak as starting
Point, position when looking for curve to decline to for the first time below C to both sides seek starting point with both sides end point respectively as end point
Apart from d_right and d_left.The ratio of d_right and d_left is sought, when symmetrical, the result is close to 1.
4) Variance feature
The time frequency signal matrix of short-wave signal is projected on the frequency axis and obtains one-dimension array Y, extract maximum in array Y
Position is corresponding frequency row Y (N), calculates the variance of Y (N) rows, exports as Variance feature.
5) rectangle morphological characteristic
For each short-wave signal time-frequency data matrix, even rectangular degree Pr is calculated.Computational methods are Pr=As/Ar (As
It is the area of connected domain S in image, Ar is the minimum enclosed rectangle area for surrounding the connected domain).
Corresponding to single short-wave signal time-frequency data matrix 4.1) integrate to 4.5) feature, obtain the short-wave signal
The eigenmatrix of time-frequency data matrix.
5th step:Each short-wave signal Input matrix that 4th step is obtained is obtained in the good SVM classifier of training in advance
Export to grader, according to grader output result, short-wave signal type is marked.
In order to check the performance of method proposed by the invention, the method for the present invention is entered to the time-frequency data for gathering in a large number
Test is gone.Test platform is visual studio 2010, and system configuration is windows 10 (16G internal memories, CPU are i5).
Test process is:The short-wave signal data .dat file that actual samples are obtained is input to the C language journey based on the inventive method
In sequence, according to the short-wave signal type of detection identification output, the short-wave signal collection of the short-wave signal type of the input data is built
Close, by manual review, count the accuracy of short-wave signal classification.Survey through carrying out substantial amounts of time-frequency data to the inventive method
Examination, verifies the actual performance of inventive algorithm.Test result shows, signal detection anti-noise of the inventive algorithm in noisy channel
Acoustic energy power is strong, can carry out effective detection and localization to short-wave signal under noisy channel of the noise level lower limit for 0dB.With
When, the actual implementation process of the inventive method carried out automatically based on computer, it is to avoid because testing staff artificial unstable because
Detection recognition accuracy fluctuation caused by plain, the average detected recognition correct rate of the short-wave signal of this method is up to more than 90%.
There is stable detection discrimination, ageing good.
Table 1 carries out test once based on actual channel based on the inventive method, and obtained divides according to short-wave signal type
The short-wave signal set of class.Wherein, the signal number that the corresponding numeral of signal number is obtained for this method classification, is artificial in bracket
The correct signal number for meeting the short-wave signal type determined after check.By counting, collect described in table 1 and amount to comprising signal
1106, wherein correct category signal 1031, accuracy is 93.21%.
Short-wave signal set of the table 1 based on short-wave signal type
Table 2 compared for the inventive method and be detected with Traditional Man shortwave detection recognition method, existing time domain short-wave signal
Recognition methodss Average Accuracy, accuracy rate stability, process time difference.As the inventive method is based on computer vision
Short-wave signal time-frequency characteristics have been carried out with effective extraction, the characteristic information of redundancy has effectively been eliminated, therefore in terms of accuracy rate compared with
Traditional Man detection method is had a distinct increment with existing tim e- domain detection recognition methodss.Automatically process as a result of computer,
Therefore accuracy rate stability is not affected with personnel's fatigue by the time, and more traditional manual detection recognition methodss are more stable.?
In terms of process time, due to being to carry out detection identification to the feature of the short-wave signal for having refined, the data volume for therefore processing is compared with people
Work detection method is less with Time Domain Analysis, and the SVM classifier for being simultaneously introduced machine learning carries out grader identification, processes
Time is shorter.
The contrast of table 2 the inventive method and other shortwave detection recognition methods
Method type | Traditional artificial shortwave detection identification | The detection identification of time domain shortwave | Shortwave detection identification based on computer vision |
Average Accuracy | 60% | 70% | 93.21% |
Accuracy rate stability | ± 20% | ± 5% | ± 5% |
Process time | > 1s | < 500ms | < 200ms |
Claims (3)
1. a kind of short-wave signal detection recognition method based on computer vision, it is characterised in that:The method is for a kind of based on meter
The short-wave signal detection and localization of calculation machine vision and time-frequency characteristics are extracted, and the SVM classifier based on machine learning, to signal when
Frequency feature is mated, and finally realizes Aulomatizeted Detect and the identification of short-wave signal;
This method is comprised the steps of:
1) to signal time-frequency data matrix to be detected, according to frequency direction per 300kHz piecemeals, time-frequency time-frequency data is obtained
Matrix;
2) to each time-frequency data submatrix, following signal framing detection is carried out, concrete operations are as follows:
2.1) time-frequency data submatrix is projected on the frequency axis and obtains one-dimensional row and vector V, V represent signal energy in frequency direction
That measured is cumulative;
2.2) distribution histogram of each element in V is analyzed;
2.3) the locating threshold T in previous step is combined as thresholding, and the 2.1) drop shadow curve in step, with reference in drop shadow curve
Higher than the frequency separation of thresholding, corresponding frequency separation in time-frequency matrix is intercepted, currently processed time-frequency data are obtained
The set of the time frequency signal submatrix of short-wave signal present in submatrix;
3) the corresponding time frequency signal submatrix set of each time-frequency data submatrix for obtaining first two steps merges, and obtains as defeated
The time frequency signal set of matrices of the whole short-wave signals for including in the signal time-frequency data matrix to be detected for entering;
4) set constituted by the time frequency signal matrix of each short-wave signal, carries out signal characteristic abstraction one by one;
4.1) kurtosis feature
The time frequency signal matrix of short-wave signal is projected on the frequency axis and obtains one-dimension array Y;Windowed traversal process is carried out to Y;
Calculate the difference of the corresponding numerical value of point of proximity respectively, and substituted in equation below;
The value of each point in Y is replaced with corresponding Du, kurtosis curve Y ' is finally given;It is right that the value of each point in Y is replaced with
The Du for answering, finally gives kurtosis curve Y ';
Packet sequencing is subsequently carried out:Using the null value in kurtosis curve as partitioning standards, kurtosis curve is divided into multiple spikes
Group, the statistics corresponding kurtosis curve regions maximum of each group obtain the sequence that spike group maximum value sequence carries out from big to small;
Subsequently the spike group maximum value sequence for sorting from big to small obtained in packet sequencing is divided into two classes, threshold value by given threshold
For the corresponding kurtosis curve of maximum sharpness group maximum divided by constant C (empirical value C=10^3), take a group sequence smaller portions conduct
Local spike group maximum value sequence, and to the part spike group maximum value sequence;
Ratio, statistics is asked to obtain corresponding two maximums of maximum ratio successively two adjacent values of spike group maximum value sequence
Spike group maximum value sequence, using two maximums as division limits, is divided into two parts, takes the less portion of sequence order by position
Be allocated as effective spike group maximum value sequence, count the maximum number that the partial sequence includes, as effective spike number, and
Result as kurtosis feature is exported;
4.2) spatial distribution characteristic
The time frequency signal matrix of short-wave signal is projected on the frequency axis and obtains one-dimension array Y;Evaluation domain interval | V (i) | ∈ [a,
B], if interval size is k (k=b-a), (by studying the relation of k value and resolution, obtaining corresponding relation array) is divided to k,
The scatterplot quantity in each interval is counted, the ratio that the scatterplot number included in each interval accounts for total scatterplot number is calculated;By each
Interval scatterplot proportionality percentage array is exported as spatial distribution characteristic;
4.3) symmetric characteristics
The time frequency signal matrix of short-wave signal is projected on the frequency axis and obtains one-dimension array Y;Evaluation domain interval [a, b], sets
Thresholding C, the part in curve higher than C are effective coverage;Wherein, C=0.2* (b-a)+a;With curve peak as starting point, to
The distance of starting point and both sides end point is sought respectively as end point in position when both sides look for curve to decline to for the first time below C
D_right and d_left;The ratio of d_right and d_left is sought, when symmetrical, the result is close to 1;
4.4) Variance feature
The time frequency signal matrix of short-wave signal is projected on the frequency axis and obtains one-dimension array Y, extract maximum value position in array Y
Corresponding frequency row Y (N), calculates the variance of Y (N) rows, exports as Variance feature;
4.5) rectangle morphological characteristic
For each short-wave signal time-frequency data matrix, even rectangular degree Pr is calculated;Computational methods are that (As is figure to Pr=As/Ar
The area of connected domain S as in, Ar are the minimum enclosed rectangle areas for surrounding the connected domain);
Corresponding to single short-wave signal time-frequency data matrix 4.1) integrate to 4.5) feature, obtain the short-wave signal time-frequency
The eigenmatrix of data matrix;
5) by step 4) each short-wave signal Input matrix for obtaining, in the good SVM classifier of training in advance, obtains grader
Output, according to grader output result, is marked to short-wave signal type.
2. a kind of short-wave signal detection recognition method based on computer vision according to claim 1, it is characterised in that
Described 2.2) in employ based on histogrammic threshold value thresholding computational methods, computing formula is T=(V_med-V_min)+V_
Med, wherein V_med represent that the mode of V element, V_min represent the minima of element in V.
3. a kind of short-wave signal detection recognition method based on computer vision according to claim 1, it is characterised in that
Described 5) in, used the SVM classifier that has trained, the kurtosis feature extracted based on machine vision, spatial distribution characteristic, right
Title property feature, Variance feature, rectangle morphological characteristic, are identified to short-wave signal type.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610917977.9A CN106503649B (en) | 2016-10-20 | 2016-10-20 | A kind of short-wave signal detection recognition method based on computer vision |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610917977.9A CN106503649B (en) | 2016-10-20 | 2016-10-20 | A kind of short-wave signal detection recognition method based on computer vision |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106503649A true CN106503649A (en) | 2017-03-15 |
CN106503649B CN106503649B (en) | 2019-06-07 |
Family
ID=58319308
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610917977.9A Active CN106503649B (en) | 2016-10-20 | 2016-10-20 | A kind of short-wave signal detection recognition method based on computer vision |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106503649B (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110631541A (en) * | 2019-10-31 | 2019-12-31 | 湖北民族大学 | Steel product shape measurement linearity detection and correction method |
CN112004468A (en) * | 2018-02-23 | 2020-11-27 | 波士顿科学国际有限公司 | Method for evaluating vessels using continuous physiological measurements |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20050232512A1 (en) * | 2004-04-20 | 2005-10-20 | Max-Viz, Inc. | Neural net based processor for synthetic vision fusion |
CN101398846A (en) * | 2008-10-23 | 2009-04-01 | 上海交通大学 | Image, semantic and concept detection method based on partial color space characteristic |
CN101631102A (en) * | 2009-04-10 | 2010-01-20 | 北京理工大学 | Interference pattern recognition technology of frequency hopping system |
US8369567B1 (en) * | 2010-05-11 | 2013-02-05 | The United States Of America As Represented By The Secretary Of The Navy | Method for detecting and mapping fires using features extracted from overhead imagery |
CN103968933A (en) * | 2014-04-09 | 2014-08-06 | 西安电子科技大学 | Fuzzy domain characteristics based optical fiber vibration signal identifying method |
CN105354592A (en) * | 2015-10-22 | 2016-02-24 | 上海无线电设备研究所 | Classification based optimal time-frequency distribution design and target identification method |
CN105574495A (en) * | 2015-12-14 | 2016-05-11 | 中国电子科技集团公司第二十八研究所 | Interval-type radar radiation source signal feature type recognition capability analyzing method |
-
2016
- 2016-10-20 CN CN201610917977.9A patent/CN106503649B/en active Active
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20050232512A1 (en) * | 2004-04-20 | 2005-10-20 | Max-Viz, Inc. | Neural net based processor for synthetic vision fusion |
CN101398846A (en) * | 2008-10-23 | 2009-04-01 | 上海交通大学 | Image, semantic and concept detection method based on partial color space characteristic |
CN101631102A (en) * | 2009-04-10 | 2010-01-20 | 北京理工大学 | Interference pattern recognition technology of frequency hopping system |
US8369567B1 (en) * | 2010-05-11 | 2013-02-05 | The United States Of America As Represented By The Secretary Of The Navy | Method for detecting and mapping fires using features extracted from overhead imagery |
CN103968933A (en) * | 2014-04-09 | 2014-08-06 | 西安电子科技大学 | Fuzzy domain characteristics based optical fiber vibration signal identifying method |
CN105354592A (en) * | 2015-10-22 | 2016-02-24 | 上海无线电设备研究所 | Classification based optimal time-frequency distribution design and target identification method |
CN105574495A (en) * | 2015-12-14 | 2016-05-11 | 中国电子科技集团公司第二十八研究所 | Interval-type radar radiation source signal feature type recognition capability analyzing method |
Non-Patent Citations (3)
Title |
---|
XU M等: ""A New Time-Frequency Spectrogram Analysis of FH Signals by Image Enhancement and Mathematical Morphology"", 《INTERNATIONAL CONFERENCE ON IMAGE AND GRAPHICS》 * |
张汝波等: ""强噪声背景下莫尔斯信号的自动检测与识别"", 《哈尔滨工程大学学报》 * |
董延坤等: ""一种短波宽带综合检测分类算法"", 《信息工程大学学报》 * |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112004468A (en) * | 2018-02-23 | 2020-11-27 | 波士顿科学国际有限公司 | Method for evaluating vessels using continuous physiological measurements |
CN112004468B (en) * | 2018-02-23 | 2023-11-14 | 波士顿科学国际有限公司 | Method for evaluating vessels using continuous physiological measurements |
CN110631541A (en) * | 2019-10-31 | 2019-12-31 | 湖北民族大学 | Steel product shape measurement linearity detection and correction method |
CN110631541B (en) * | 2019-10-31 | 2021-07-23 | 湖北民族大学 | Steel product shape measurement linearity detection and correction method |
Also Published As
Publication number | Publication date |
---|---|
CN106503649B (en) | 2019-06-07 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109782274B (en) | Water damage identification method based on time-frequency statistical characteristics of ground penetrating radar signals | |
CN108186011A (en) | Atrial fibrillation detection method, device and readable storage medium storing program for executing | |
CN110148162A (en) | A kind of heterologous image matching method based on composition operators | |
CN104715261A (en) | FMRI dynamic brain function sub-network construction and parallel connection SVM weighted recognition method | |
CN108629380A (en) | A kind of across scene wireless signal cognitive method based on transfer learning | |
CN104794729B (en) | SAR image change detection based on conspicuousness guiding | |
CN108469560A (en) | A kind of objective complexity evaluation methodology of electromagnetic interference based on quick S-transformation time frequency space model | |
CN106597365B (en) | A kind of sophisticated electronic signal time difference positioning method based on time domain cluster | |
CN111428561B (en) | Broadband electromagnetic interference source identification method and system, storage medium and equipment | |
CN103455820A (en) | Method and system for detecting and tracking vehicle based on machine vision technology | |
CN109410184A (en) | Live streaming pornographic image detection method based on dense confrontation network semi-supervised learning | |
CN104680185B (en) | Hyperspectral image classification method based on boundary point reclassification | |
CN106611160A (en) | CNN (Convolutional Neural Network) based image hair identification method and device | |
CN110991397B (en) | Travel direction determining method and related equipment | |
CN103971106A (en) | Multi-view human facial image gender identification method and device | |
CN109886242A (en) | A kind of method and system that pedestrian identifies again | |
CN109684986B (en) | Vehicle analysis method and system based on vehicle detection and tracking | |
CN104794440A (en) | False fingerprint detection method based on multi-block multi-scale LBP | |
CN108537787A (en) | A kind of quality judging method of facial image | |
CN111934796B (en) | LTE uplink interference automatic labeling method and system | |
CN112668612A (en) | Partial discharge signal clustering analysis method based on grids | |
CN104680545A (en) | Method for detecting existence of salient objects in optical images | |
CN106503649A (en) | A kind of short-wave signal detection recognition method based on computer vision | |
CN111046025B (en) | Unmanned aerial vehicle signal detection method and device | |
CN115343685A (en) | Multi-dimensional ground penetrating radar detection method, device and equipment applied to disease identification |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
TR01 | Transfer of patent right | ||
TR01 | Transfer of patent right |
Effective date of registration: 20221114 Address after: 100012 817, Floor 8, No. 101, Floor 3 to 8, Building 17, Rongchuang Road, Chaoyang District, Beijing Patentee after: Beijing Ge Lei Information Technology Co.,Ltd. Address before: 100124 No. 100 Chaoyang District Ping Tian Park, Beijing Patentee before: Beijing University of Technology |