CN114358096A - Deep learning Morse code identification method and device based on step-by-step threshold judgment - Google Patents
Deep learning Morse code identification method and device based on step-by-step threshold judgment Download PDFInfo
- Publication number
- CN114358096A CN114358096A CN202210274909.0A CN202210274909A CN114358096A CN 114358096 A CN114358096 A CN 114358096A CN 202210274909 A CN202210274909 A CN 202210274909A CN 114358096 A CN114358096 A CN 114358096A
- Authority
- CN
- China
- Prior art keywords
- audio signals
- morse code
- signal
- class
- neural network
- 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
- 238000000034 method Methods 0.000 title claims abstract description 57
- 238000013135 deep learning Methods 0.000 title claims abstract description 28
- 230000005236 sound signal Effects 0.000 claims abstract description 194
- 238000013528 artificial neural network Methods 0.000 claims abstract description 54
- 238000012545 processing Methods 0.000 claims abstract description 35
- 238000007781 pre-processing Methods 0.000 claims abstract description 7
- 230000000306 recurrent effect Effects 0.000 claims abstract description 5
- 230000006870 function Effects 0.000 claims description 18
- 230000000750 progressive effect Effects 0.000 claims description 16
- 238000001914 filtration Methods 0.000 claims description 14
- 230000015654 memory Effects 0.000 claims description 12
- 239000013598 vector Substances 0.000 claims description 10
- 238000013507 mapping Methods 0.000 claims description 8
- 238000005070 sampling Methods 0.000 claims description 8
- 238000000926 separation method Methods 0.000 claims description 8
- 238000004590 computer program Methods 0.000 claims description 6
- 238000012549 training Methods 0.000 claims description 6
- 230000009467 reduction Effects 0.000 claims description 4
- 108091026890 Coding region Proteins 0.000 claims description 3
- 238000010586 diagram Methods 0.000 description 9
- 238000004891 communication Methods 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 230000007787 long-term memory Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/02—Preprocessing
- G06F2218/04—Denoising
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- 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
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L15/00—Apparatus or local circuits for transmitting or receiving dot-and-dash codes, e.g. Morse code
-
- 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)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Software Systems (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Signal Processing (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computer Networks & Wireless Communication (AREA)
- Compression, Expansion, Code Conversion, And Decoders (AREA)
Abstract
The invention discloses a Morse code deep learning identification method based on step-by-step threshold judgment, which comprises the steps of obtaining Morse code audio signals, preprocessing the Morse code audio signals, and obtaining preprocessed audio signals; judging the preprocessed audio signals, and classifying the preprocessed audio signals into first-class audio signals and second-class audio signals according to a judgment result; adding a recursion gate for analysis and judgment on the basis of a recurrent neural network, and constructing the neural network for generating a Morse data tag; according to the input condition of the neural network, coding the first type of audio signals and the second type of audio signals to obtain coded data; inputting the data after the coding processing into a neural network for processing, and outputting to obtain a Morse data tag; carrying out inverse processing on the Morse code data label to identify the Morse code; the identification method of the invention ensures the identification efficiency and accuracy of Morse code.
Description
Technical Field
The invention belongs to the technical field of communication signal processing, and particularly relates to a method and a device for recognizing deep learning Morse codes based on step-by-step threshold judgment.
Background
Morse code is a signal code that is on and off at intervals, and expresses different English letters, numbers and punctuation marks through different arrangement sequences. It was invented in 1837 as an early form of digital communication, also known under the name of its inventor as morse code or morse code. Unlike modern digital communications, morse code uses only binary codes of zero and one two states, whose codes include five: short-duration point signals "-, long signals" -, which represent pauses between points and dashes, moderate pauses between each word, and long pauses between sentences; for the variable speed Morse code recognition, the reference interval needs to be continuously updated and adjusted according to the received Morse code data, so that the point code, the stroke code, the word interval and the word interval in a section of Morse signal can be judged, and the corresponding characters can be obtained through inquiry and recognition.
Based on this, there is no method for efficiently and accurately identifying Morse codes based on deep learning techniques.
Disclosure of Invention
Aiming at the defects of the prior art, the technical problems to be solved by the invention are as follows: how to provide a Morse code deep learning identification method and device based on step-by-step threshold judgment to efficiently and accurately identify Morse codes.
In order to solve the technical problems, the invention adopts the following technical scheme: a deep learning Morse code identification method based on step-by-step threshold judgment comprises the following steps:
acquiring Morse code audio signals, and preprocessing the Morse code audio signals to obtain preprocessed audio signals;
judging the preprocessed audio signals, and classifying the preprocessed audio signals into first-class audio signals and second-class audio signals according to a judgment result;
adding a recursion gate for analysis and judgment according to the recurrent neural network, and constructing a neural network for generating a Morse data tag;
according to the input condition of the neural network, coding the first type of audio signals and the second type of audio signals to obtain coded data;
inputting the data after the coding processing into a neural network for processing, and outputting to obtain a Morse data tag;
and (5) carrying out inverse processing on the Moss code data label to identify the Moss code.
In some embodiments, in the above method for recognizing a morse code based on deep learning by stage threshold decision, acquiring a morse code audio signal, and preprocessing the morse code audio signal to obtain a preprocessed audio signal includes:
carrying out noise reduction processing on the Morse code audio signal to obtain a de-noised signal;
fourier transform is carried out on the Morse code audio signal to obtain a preliminary filtering result;
and further processing the preliminary filtering result through a Butterworth filter and a Kalman filter to obtain a processed audio signal.
In some embodiments, in the above method for recognizing a deep learning morse code based on step-by-step threshold decision, the step of determining a preprocessed audio signal, classifying the preprocessed audio signal into a first type audio signal and a second type audio signal according to a determination result, and executing the step of determining includes:
setting a signal judgment threshold according to the signal characteristics within the preset time;
judging the processed audio signal according to the signal judgment threshold value;
if the processed audio signal is larger than the signal judgment threshold, judging the audio signal to be classified;
and if the processed audio signal is less than or equal to the signal judgment threshold value, judging that no signal exists.
In some embodiments, in the above method for recognizing a deep learning morse code based on step-by-step threshold decision, the step of determining a preprocessed audio signal, classifying the preprocessed audio signal into a first type audio signal and a second type audio signal according to a determination result, and executing classification on the audio signal to be classified includes:
determining the duration relation of the first type of audio signal and the second type of audio signal according to an international standard;
setting a signal duration threshold and a judgment threshold point number according to the signal judgment threshold;
reading a preset number of first-class audio signals in the audio signals to be classified;
setting a sampling time interval, and acquiring a duration equation of a first type of audio signal when the number of the first type of audio signal in the sampling time interval and in a preset number range is larger than the number of a judgment threshold;
acquiring a mean equation and a variance equation of a preset number of first-class audio signals according to the signal duration threshold;
determining a first type of Gaussian distribution according to a duration equation, a mean equation and a variance equation;
determining a second class of Gaussian distribution according to the duration relation of the first class of audio signals and the second class of audio signals;
obtaining a first class probability density function and a second class probability density function according to the first class Gaussian distribution and the second class Gaussian distribution;
obtaining a classification threshold value according to the first class probability density function and the second class probability density function;
dividing the audio signals with the duration time larger than a classification threshold value in the audio signals to be classified into first-class audio signals;
and dividing the audio signals with the duration less than or equal to the classification threshold value in the audio signals to be classified into second-class audio signals.
In some embodiments, in the above method for deep learning morse code recognition based on step-by-step threshold decision, before performing encoding processing on a first type of audio signal and a second type of audio signal according to an input condition of a neural network to obtain encoded data, the method includes:
and performing character separation on the preprocessed audio signals according to the first class of audio signals and the second class of audio signals.
In some embodiments, in the method for deep learning morse code identification based on step-by-step threshold decision, the encoding processing is performed on the first type of audio signal and the second type of audio signal according to an input condition of a neural network, so as to obtain encoded data, and the method includes:
and setting the coding formats of the first type of audio signals and the second type of audio signals according to the influence of the length of the input sequence on the decoding accuracy of the neural network, and performing coding processing to obtain a coding sequence and a class label.
In some embodiments, in the above method for deep learning Morse code identification based on step-by-step threshold decision, before encoding a first type of audio signal and a second type of audio signal according to an input condition of a neural network, the method includes:
and mapping the class labels into One-hot vectors by adopting One-hot coding.
In some embodiments, in the above method for deep learning Morse code identification based on step-by-step threshold decision, the data after encoding processing is input into a neural network for processing, and a Morse code data tag is obtained by outputting, including:
and inputting the One-hot vector and the class label as training data into a neural network for operation, and outputting a Morse code data label.
In some embodiments, in the above method for recognizing a morse code based on deep learning by using a step-by-step threshold decision, the inverse processing of the morse code data label to recognize the morse code includes:
and (5) carrying out inverse mapping on the Moss code data label by adopting One-hot coding, and identifying the Moss code.
A computer apparatus, the computer apparatus comprising: a processor for implementing the steps of the above method when executing the computer program stored in the memory.
The invention has the following beneficial effects:
the technical scheme provided by the invention adopts a step-by-step threshold judgment method for Morse code audio signals, and comprises the steps of judging whether Morse code signals exist in preprocessed audio signals through a first-stage threshold, reducing noise interference, ensuring the accuracy of identification, and further determining a first-class audio signal and a second-class audio signal through a second-stage threshold judgment; constructing a third-level threshold judgment, and performing character separation on the preprocessed audio signals according to the first-class audio signals and the second-class audio signals; and finally, constructing a neural network for generating Morse code data labels, and identifying Morse codes according to the Morse code data labels, so that the Morse code identification, calculation and decoding efficiency and accuracy are ensured.
Drawings
FIG. 1 is a general flow diagram of the present invention.
FIG. 2 is a basic structure diagram of RNN.
Fig. 3 is a sub-flowchart of step S200.
Fig. 4 is a signal waveform diagram of an initially acquired morse code.
Fig. 5 is a signal waveform diagram of morse code after preliminary filtering.
Fig. 6 is a signal waveform diagram of a final filtered morse code.
Fig. 7 is a flowchart of the determination step in step S300.
Fig. 8 is a schematic diagram of setting a signal determination threshold.
Fig. 9 is a flowchart of the classification step in step S300.
Fig. 10 is a schematic diagram of the duration of acquiring a long signal and a point signal.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings.
In the specific implementation: a deep learning morse code recognition method based on progressive threshold decision, as shown in fig. 1, includes the following steps:
s100: acquiring Morse code audio signals, and preprocessing the Morse code audio signals to obtain preprocessed audio signals;
in order to reduce noise interference as much as possible and improve the identification accuracy, the signal needs to be preprocessed before passing through the neural network, so as to obtain a preprocessed audio signal after noise reduction.
S200: judging the preprocessed audio signals, and classifying the preprocessed audio signals into first-class audio signals and second-class audio signals according to a judgment result;
in order to identify the signal, the obtained preprocessed audio signal should be judged and identified first, because the noise aliasing in the signal cannot be completely removed, one of the meanings of judging and identifying is to distinguish the signal from the noise in terms of amplitude, then the audio signal judged to have the signal is selected as the signal to be classified for subsequent classification of the audio signal, because the morse code generally comprises two types of audio signals, the first type of audio signal is a short point signal, the second type of audio signal is a long signal which is kept for a certain time, and according to the international standard, the time length of one long signal can be determined to be equal to the time length of three point signals without intervals, so the two types of audio signals can be distinguished by calculating the statistical characteristic of the actual audio signal duration and setting a classification threshold.
S300: adding a recursion gate for analysis and judgment according to the recurrent neural network, and constructing a neural network for generating a Morse data tag;
recurrent Neural Networks (RNNs) are one type of Neural network. It can transmit the state circularly in the network itself, so as to receive more extensive time series structure input, such as non-segmented handwriting recognition, speech recognition, etc., and the basic structure of RNN is shown in FIG. 2.
A recursion gate called as a forgetting gate is added on the basis of the RNN for judging whether information is useful or not, and the problems that the weight exponential explosion or gradient disappears and long-time dependence is difficult to capture in recursion existing in the RNN can be solved. Through the gating state, the internal transmission state can be controlled, information needing long-term memory is reserved, unimportant information is forgotten, and therefore the long-term dependence relationship between input values is established.
S400: according to the input condition of the neural network, coding the first type of audio signals and the second type of audio signals to obtain coded data;
in order to adapt to the network structure of the neural network, input and output are structured, and point signals and long signals are encoded to obtain encoded data including an encoding sequence for input to the neural network in consideration of the influence of the length of an input sequence on the accuracy of neural Gaolu decoding.
S500: inputting the data after the coding processing into a neural network for processing, and outputting to obtain a Morse data tag;
s600: and (5) carrying out inverse processing on the Moss code data label to identify the Moss code.
And carrying out inverse mapping on the Moss code data label to realize the identification of the Moss code.
In some embodiments, in the above method for deep learning morse code identification based on progressive threshold decision, as shown in fig. 3, the step S200 further includes:
s201: carrying out noise reduction processing on the Morse code audio signal to obtain a de-noised signal;
s202: fourier transform is carried out on the Morse code audio signal to obtain a preliminary filtering result;
in order to reduce noise interference as much as possible and improve the identification accuracy, the audio signal needs to be filtered before passing through the neural network;
fig. 4 is a waveform diagram of the acquired morse code signal, which is characterized in that firstly, fourier transform is performed on a morse code audio signal, then frequency point information with the largest complex variance is selected to be reserved, and the rest information is discarded to realize preliminary filtering, and fig. 5 is a waveform diagram of the signal after the preliminary filtering.
S203: and further processing the preliminary filtering result through a Butterworth filter and a Kalman filter to obtain a processed audio signal.
After the signal is subjected to Fourier denoising, the characteristics of the signal are still not intuitive, the signal needs to be subjected to a Butterworth filter and a Kalman filter, so that the envelope of the signal is extracted, the final filtering is realized, and the finally obtained Morse code signal oscillogram is shown in FIG. 6, so that the filtering effect is obvious.
In practical situations, the characteristic quantities such as the amplitude, the duration and the like of the signal are not constant, and the Morse code identification in the scheme is to judge the signal according to the statistical characteristics of the signal within a period of time and input the signal into the constructed neural network for identification, so that whether the audio signal exists needs to be judged before, and if the audio signal exists, the audio signal is listed as the audio signal to be classified for classification.
In some embodiments, in the above method for recognizing deep learning morse codes based on progressive threshold decision, the step of determining in step S300 is executed as shown in fig. 7, and includes:
s310: setting a signal judgment threshold according to the signal characteristics within the preset time;
s311: judging the processed audio signal according to the signal judgment threshold value;
s312: if the processed audio signal is larger than the signal judgment threshold, judging the audio signal to be classified;
s313: and if the processed audio signal is less than or equal to the signal judgment threshold value, judging that no signal exists.
Defining the processed audio signal as
In order to recognize a signal, a signal judgment threshold should be set first
When a certain time goes on
The data of (a) are:
if it is determined that there is a signal, it is used as the signal to be classified, otherwise, when there is a signal at a certain time
The data of (a) are:
and judging that no signal exists.
As shown in fig. 8, wherein the signal judgment threshold value
The size of the signal is set according to the signal characteristics in a period of time, and the setting process is as follows: the window is long
The amplitude of all sampling points in the window is averaged, and the average value is defined as a threshold value for judging whether a signal in the current window exists or not;
it should be noted that the window length should be set according to actual conditions, and if the window length is set too small, an extreme condition that no signal or all signals are present in the window is easily determined, so that the threshold is too large or too small; if the window length is set to be too large and even exceeds the length of the signal, too many blank signal-free areas can be collected, so that the judgment is wrong;
since noise aliased in the signal cannot be completely removed, setting the signal determination threshold has a benefit in that the signal can be distinguished from the noise in amplitude.
In some embodiments, in the above method for recognizing deep learning morse codes based on progressive threshold decision, as shown in fig. 9, the step S300 of classifying the audio signal to be classified includes:
s320: determining the duration relation of the first type of audio signal and the second type of audio signal according to an international standard;
s321: setting a signal duration threshold and a judgment threshold point number according to the signal judgment threshold;
s322: reading a preset number of first-class audio signals in the audio signals to be classified;
s323: setting a sampling time interval, and acquiring a duration equation of a first type of audio signal when the number of the first type of audio signal in the sampling time interval and in a preset number range is larger than the number of a judgment threshold;
s324: acquiring a mean equation and a variance equation of a preset number of first-class audio signals according to the signal duration threshold;
s325: determining a first type of Gaussian distribution according to a duration equation, a mean equation and a variance equation;
s326: determining a second class of Gaussian distribution according to the duration relation of the first class of audio signals and the second class of audio signals;
s327: obtaining a first class probability density function and a second class probability density function according to the first class Gaussian distribution and the second class Gaussian distribution;
s328: obtaining a classification threshold value according to the first class probability density function and the second class probability density function;
as shown in fig. 10, the threshold value is judged when a signal is given
On the basis of (2) defining the duration of the signal
The duration of the long signal is the time length of the signal exceeding the threshold
；
In order to reduce the decision error, the present embodiment analyzes and processes a plurality of point signals and a long signal;
firstly, for a certain section of the audio signal which is processed by filtering, reading the audio signal
A dot signal;
set the sampling time interval to
If it is at first
A (a)
) The number of points when the point signal is greater than the signal judgment threshold is
Then, consider the duration formula of the signal at this point:
；
to obtain the above
Mean value of the duration of the point signal
Sum variance
Comprises the following steps:
in the audio signal, the duration of each point signal is substantially similar, so
Can be regarded as obeying the mean value of
And variance of
Of gaussian distribution random variables, i.e.
Similarly, the duration of the long signal
Also satisfies a Gaussian distribution of random variables, i.e.
；
Obtaining a probability density function of the point signal based on the above-mentioned Gaussian distribution of the point signal
Obtaining a probability density function of the long signal according to the Gaussian distribution of the long signal
。
S329: dividing the audio signals with the duration time larger than a classification threshold value in the audio signals to be classified into first-class audio signals;
s330: and dividing the audio signals with the duration less than or equal to the classification threshold value in the audio signals to be classified into second-class audio signals.
When in use
Then, an intersection point appears in the point signal probability density function and the long signal probability density function, the intersection point value is a classification threshold, and the unit of the classification threshold is time;
thus, the decision of the signal category is converted into a binary classification problem,
wherein,
in the form of a signal class,shortthe representative point signal is a signal representative of the point,longrepresenting long signals, i.e.
In case the duration is less than the classification threshold, it is considered as a point signal;
in the case where the duration is greater than the classification threshold, a long signal is considered.
In some embodiments, in the above deep learning morse code identification method based on progressive threshold decision, before performing step S400, the following steps are performed:
and performing character separation on the preprocessed audio signals according to the first class of audio signals and the second class of audio signals.
In order for Morse code to be accurately recognized, it is necessary that characters be separated from each other; according to the international standard, the durations of the intervals between the dot signal and the long signal, between the characters and between the words are different from each other, as shown in table 1:
TABLE 1 duration of blank intervals
Wherein,
is a basic unit of time. The dot length sequence can be broken by the difference of interval duration, and each character is extracted. As shown in FIG. 8, the duration of the interval is defined as the time period during which the signal amplitude is less than the signal determination threshold, e.g., as in FIG. 8
And
respectively representing the duration between dot length signals and between characters;
classifying the different time intervals: first read
Dot length intervals, the duration of each interval being calculated
， 
And then calculate this
Mean and variance of individual data. Setting two threshold values based on the obtained probability density function
And
(ii) a The former is used to distinguish between dot-long signal intervals and other intervals, and the latter is used to distinguish between character intervals and word intervals. According to the judgment, the dot length sequence can be divided into a plurality of characters, each character can be divided into a plurality of characters, and the character separation is completed through the steps.
In some embodiments, in the above method for deep learning morse code identification based on progressive threshold decision, step S400 includes:
and setting the coding formats of the first type of audio signals and the second type of audio signals according to the influence of the length of the input sequence on the decoding accuracy of the neural network, and performing coding processing to obtain a coding sequence and a class label.
After the above processing is completed, the point signal and the long signal are encoded in consideration of the influence of the length of the input sequence on the decoding accuracy of the neural network, wherein the encoding format is specified as shown in table 2:
TABLE 2 coding formats for Point signals and Long signals
In the international convention, the Morse code corresponding to the numbers 0-9 has a division of long and short codes. For numbers 1-9, the long code format is used; for the number 0, a short code format is used. Since the dot signal and the long signal in table 2 have different coding lengths and the number of the digital morse code dot signal and the number of the long signal are different, the sequence generated by the above coding operation is a variable length sequence having a length of 7 bits (number 0) as a minimum and 31 bits (numbers 1 and 9) as a maximum. Therefore, all the sequences are converted into fixed-length sequences with the length of 32 bits through 0 complementing operation, and finally the fixed-length sequences which can be input into the neural network are obtained, and the coding sequences input into the neural network for identifying Morse codes are shown in the following table 3.
TABLE 3 comparison of proto-Morse code-coding sequences
In some embodiments, in the above deep learning morse code identification method based on progressive threshold decision, before performing step S500, the following steps are performed:
and mapping the class labels into One-hot vectors by adopting One-hot coding.
One-hot encoding is an encoding for converting class variables into data types convenient for computers to perform classification tasks, wherein Morse code recognition by using a neural network can be regarded as a classification task, and the classification task is a typical machine learning task. And the classifier judges the category of the newly observed sample based on the training data of the marked category. The feature vectors input into the classifier should be numbers and normalized to some similar range;
therefore, the feature vector is processed by adopting One-hot coding, and the feature vector can be coded by adopting One-hot coding
Conversion of one possible value into
A binary feature whose value set is
In addition, the One-hot coding maps values of discrete features to Euclidean space, and a certain value of the discrete features corresponds to a certain point in the Euclidean space, so that feature distance and cosine similarity in a classification task can be conveniently calculated.
Therefore, taking the identification number as an example, the data has 10 kinds of class labels, which correspond to the numbers 0-9, as shown in table 4, the class labels 0-9 are mapped to One-hot vectors, and then can be input into the neural network as the labels of the training data for Morse code identification.
TABLE 4 Category tag and One-hot encoding
In some embodiments, in the above method for deep learning morse code identification based on progressive threshold decision, step S500 includes:
and inputting the One-hot vector and the class label as training data into a neural network for operation, and outputting a Morse code data label.
After the class labels and the One-hot codes are input into the neural network to carry out Morse code recognition, the neural network can distinguish the class labels and output the class labels corresponding to the numbers.
In some embodiments, in the above method for deep learning morse code identification based on progressive threshold decision, step S600 includes:
and (5) carrying out inverse mapping on the Moss code data label by adopting One-hot coding, and identifying the Moss code.
And finally, carrying out One-to-One corresponding inverse mapping on the class label of the output number according to the class label of the One-hot code to obtain original code data and realize the identification of Morse code.
Further, the identification method is experimentally verified, the adopted experimental data are 7 segments of Morse code audio signals with different signal-to-noise ratios, and after the Morse code audio signals are preprocessed, the operations such as judgment, classification, character separation and the like are carried out, so that an input sequence matched with the constructed neural network input format is obtained. The neural network was constructed based on a Tensorflow framework with 64 layers and outputs classified into 10 classes (corresponding to numbers 0-9). The results of the tests on Morse code decoding are shown in Table 5:
TABLE 5 neural network test results identifying Morse code
The results show that after preprocessing the Morse code filtering signal, the Morse code is identified by utilizing the neural network seen by the dog, the obtained decoding result has high accuracy, and the good decoding performance is realized for the audio signal with low signal-to-noise ratio.
In summary, the present invention first obtains the morse code audio signal to be identified and constructs the neural network for identifying the morse code audio signal, and then the audio signal is input to the neural network for operation after the filtering operation, the signal judgment, the signal classification, the character separation and the encoding process, and the data of the audio signal for training the neural network corresponds to the tags one by one. For the audio frequency used for testing, the neural network can distinguish the audio frequency only by inputting the filtered signal, outputs the Morse code data label of the corresponding number, and finally inversely maps the Morse code data label to obtain the original code data and realize the identification of the Morse code.
The technical scheme provided by the invention adopts a step-by-step threshold judgment method for Morse code audio signals, and comprises the steps of judging whether Morse code signals exist in preprocessed audio signals through a first-stage threshold, reducing noise interference and ensuring the accuracy of identification; determining a first class of audio signals and a second class of audio signals through second-level threshold judgment; constructing a third-level threshold judgment, and performing character separation on the preprocessed audio signals according to the first-class audio signals and the second-class audio signals; and finally, constructing a neural network for generating Morse code data labels, and identifying Morse codes according to the Morse code data labels, so that the Morse code identification, calculation and decoding efficiency and accuracy are ensured.
A computer apparatus, the computer apparatus comprising: a processor for implementing the steps of the above method when executing the computer program stored in the memory.
In a second aspect, an embodiment of the present invention provides a computer apparatus, including: a processor for implementing the steps of the method of constructing a knowledge-graph as described above when executing a computer program stored in the memory. The processor may be a Central Processing Unit (CPU) or other form of processing unit having data processing capabilities and/or instruction execution capabilities, and may control other components in the computer to perform desired functions. The memory may include one or more computer program products that may include various forms of computer-readable storage media, such as volatile memory and/or non-volatile memory. Volatile memory can include, for example, Random Access Memory (RAM), cache memory (or the like). The non-volatile memory may include, for example, Read Only Memory (ROM), a hard disk, flash memory, and the like. One or more computer program instructions may be stored on a computer-readable storage medium and executed by a processor to implement the above method steps of the various embodiments of the present application and/or other desired functions.
The above is only a preferred embodiment of the present invention, and it should be noted that several modifications and improvements made by those skilled in the art without departing from the technical solution should also be considered as falling within the scope of the claims.
Claims (10)
1. A deep learning Morse code identification method based on step-by-step threshold judgment is characterized in that: the method comprises the following steps:
acquiring Morse code audio signals, and preprocessing the Morse code audio signals to obtain preprocessed audio signals;
judging the preprocessed audio signals, and classifying the preprocessed audio signals into first-class audio signals and second-class audio signals according to a judgment result;
adding a recursion gate for analysis and judgment on the basis of a recurrent neural network, and constructing the neural network for generating a Morse data tag;
according to the input condition of the neural network, coding the first type of audio signals and the second type of audio signals to obtain coded data;
inputting the data after the coding processing into the neural network for processing, and outputting to obtain a Morse code data tag;
and carrying out inverse processing on the Morse code data label to identify the Morse code.
2. The method for recognizing the deep learning Morse code based on the progressive threshold decision as claimed in claim 1, wherein: the acquiring Morse code audio signals, preprocessing the Morse code audio signals to obtain preprocessed audio signals, includes:
carrying out noise reduction processing on the Morse code audio signal to obtain a de-noised signal;
fourier transform is carried out on the Morse code audio signal to obtain a preliminary filtering result;
and further processing the preliminary filtering result through a Butterworth filter and a Kalman filter to obtain a processed audio signal.
3. The method for recognizing the deep learning Morse code based on the progressive threshold decision as claimed in claim 2, wherein: the step of judging the preprocessed audio signals, classifying the preprocessed audio signals into a first type of audio signals and a second type of audio signals according to a judgment result, and executing the judgment comprises the following steps:
setting a signal judgment threshold according to the signal characteristics within the preset time;
judging the processed audio signal according to the signal judgment threshold value;
if the processed audio signal is larger than the signal judgment threshold, judging the processed audio signal as an audio signal to be classified;
and if the processed audio signal is less than or equal to the signal judgment threshold value, judging that no signal exists.
4. The method for deep learning Morse code identification based on progressive threshold decision as claimed in claim 3, wherein: the step of judging the preprocessed audio signals, classifying the preprocessed audio signals into a first type of audio signals and a second type of audio signals according to a judgment result, and executing classification on the audio signals to be classified comprises the following steps:
determining a duration relationship of the first type of audio signal and the second type of audio signal according to an international standard;
setting a signal duration threshold and a judgment threshold point number according to the signal judgment threshold;
reading a preset number of first-class audio signals in the audio signals to be classified;
setting a sampling time interval, and acquiring a duration equation of a first type of audio signal when the number of the first type of audio signal in the sampling time interval and in a preset number range is larger than the number of a judgment threshold;
acquiring a mean equation and a variance equation of a preset number of first-class audio signals according to the signal duration threshold;
determining a first type of Gaussian distribution according to the duration equation, the mean equation and the variance equation;
determining a second class of Gaussian distribution according to the duration relation between the first class of audio signals and the second class of audio signals;
respectively obtaining a first class probability density function and a second class probability density function according to the first class Gaussian distribution and the second class Gaussian distribution;
obtaining a classification threshold value according to the first class probability density function and the second class probability density function;
dividing the audio signals with the duration time larger than the classification threshold value in the audio signals to be classified into first-class audio signals;
and dividing the audio signals with the duration less than or equal to the classification threshold value in the audio signals to be classified into second-class audio signals.
5. The method of claim 4, wherein the method for recognizing Morse code based on progressive threshold decision comprises: before encoding the first-class audio signal and the second-class audio signal according to the input condition of the neural network to obtain encoded data, the method comprises the following steps:
and performing character separation on the preprocessed audio signals according to the first class of audio signals and the second class of audio signals.
6. The method of claim 5, wherein the method comprises the steps of: the encoding the first-class audio signal and the second-class audio signal according to the input condition of the neural network to obtain encoded data includes:
and setting the coding formats of the first type of audio signals and the second type of audio signals according to the influence of the length of the input sequence on the decoding accuracy of the neural network, and performing coding processing to obtain a coding sequence and a class label.
7. The method of claim 6, wherein the method for recognizing Morse code based on progressive threshold decision comprises: before the encoding processing is performed on the first type audio signal and the second type audio signal according to the input condition of the neural network, the method comprises the following steps:
and mapping the class label into a One-hot vector by adopting One-hot coding.
8. The method of claim 7, wherein the method for recognizing Morse code based on progressive threshold decision comprises: inputting the data after the coding processing into the neural network for processing, and outputting to obtain a Morse code data tag, wherein the Morse code data tag comprises:
and inputting the One-hot vector and the class label as training data into the neural network for operation, and outputting a Morse code data label.
9. The method of claim 8, wherein the method for recognizing Morse code based on progressive threshold decision comprises: the step of performing inverse processing on the Morse code data label to identify the Morse code comprises the following steps:
and carrying out inverse mapping on the Morse code data label by adopting the One-hot code, and identifying the Morse code.
10. A computer device, characterized by: the computer device includes: a processor for implementing the steps of the method according to any one of claims 1 to 9 when executing a computer program stored in a memory.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210274909.0A CN114358096B (en) | 2022-03-21 | 2022-03-21 | Deep learning Morse code identification method and device based on step-by-step threshold judgment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210274909.0A CN114358096B (en) | 2022-03-21 | 2022-03-21 | Deep learning Morse code identification method and device based on step-by-step threshold judgment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114358096A true CN114358096A (en) | 2022-04-15 |
| CN114358096B CN114358096B (en) | 2022-06-07 |
Family
ID=81094348
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210274909.0A Active CN114358096B (en) | 2022-03-21 | 2022-03-21 | Deep learning Morse code identification method and device based on step-by-step threshold judgment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114358096B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117421540A (en) * | 2023-12-19 | 2024-01-19 | 北京中电慧声科技有限公司 | Automatic Morse code identification method and device |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106650605A (en) * | 2016-10-20 | 2017-05-10 | 北京工业大学 | Morse signal automatic detection decoding method based on machine learning |
| WO2018199997A1 (en) * | 2017-04-28 | 2018-11-01 | Hewlett-Packard Development Company, L.P. | Audio classifcation with machine learning model using audio duration |
| CN109861932A (en) * | 2019-02-15 | 2019-06-07 | 中国人民解放军战略支援部队信息工程大学 | A kind of shortwave Morse's message automatic identifying method based on intelligent image analysis |
| CN110011942A (en) * | 2019-02-15 | 2019-07-12 | 中国人民解放军战略支援部队信息工程大学 | Morse's message intelligent measurement recognition methods based on deep learning |
| CN111079665A (en) * | 2019-12-20 | 2020-04-28 | 长沙深之瞳信息科技有限公司 | Morse code automatic identification method based on Bi-LSTM neural network |
| CN111092832A (en) * | 2019-12-20 | 2020-05-01 | 长沙深之瞳信息科技有限公司 | Method for making Morse code deep learning training data |
-
2022
- 2022-03-21 CN CN202210274909.0A patent/CN114358096B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106650605A (en) * | 2016-10-20 | 2017-05-10 | 北京工业大学 | Morse signal automatic detection decoding method based on machine learning |
| WO2018199997A1 (en) * | 2017-04-28 | 2018-11-01 | Hewlett-Packard Development Company, L.P. | Audio classifcation with machine learning model using audio duration |
| CN109861932A (en) * | 2019-02-15 | 2019-06-07 | 中国人民解放军战略支援部队信息工程大学 | A kind of shortwave Morse's message automatic identifying method based on intelligent image analysis |
| CN110011942A (en) * | 2019-02-15 | 2019-07-12 | 中国人民解放军战略支援部队信息工程大学 | Morse's message intelligent measurement recognition methods based on deep learning |
| CN111079665A (en) * | 2019-12-20 | 2020-04-28 | 长沙深之瞳信息科技有限公司 | Morse code automatic identification method based on Bi-LSTM neural network |
| CN111092832A (en) * | 2019-12-20 | 2020-05-01 | 长沙深之瞳信息科技有限公司 | Method for making Morse code deep learning training data |
Non-Patent Citations (2)
| Title |
|---|
| 贾克斌等: "基于宽带时频图的多路摩尔斯信号自动检测", 《北京工业大学学报》 * |
| 陈莹莹等: "莫尔斯报务信号的检测和识别", 《计算机应用与软件》 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117421540A (en) * | 2023-12-19 | 2024-01-19 | 北京中电慧声科技有限公司 | Automatic Morse code identification method and device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114358096B (en) | 2022-06-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110209823B (en) | Multi-label text classification method and system | |
| CN111126386B (en) | Sequence domain adaptation method based on countermeasure learning in scene text recognition | |
| CN109189767B (en) | Data processing method and device, electronic equipment and storage medium | |
| CA2969593C (en) | Method for text recognition and computer program product | |
| CN106980856B (en) | Formula identification method and system and symbolic reasoning calculation method and system | |
| KR102104605B1 (en) | Method and apparatus of automatic recognition of symbols in piping and instrument diagrams using ai technology | |
| CN108363701B (en) | Named entity identification method and system | |
| CN112669851A (en) | Voice recognition method and device, electronic equipment and readable storage medium | |
| CN116127953B (en) | Chinese spelling error correction method, device and medium based on contrast learning | |
| CN108710907B (en) | Handwritten data classification method, model training method, device, equipment and medium | |
| CN116991875B (en) | SQL sentence generation and alias mapping method and device based on big model | |
| CN114358096B (en) | Deep learning Morse code identification method and device based on step-by-step threshold judgment | |
| CN111079665A (en) | Morse code automatic identification method based on Bi-LSTM neural network | |
| CN110866499A (en) | Handwritten text recognition method, system, device and medium | |
| CN111191033B (en) | Open set classification method based on classification utility | |
| CN113821637A (en) | Long text classification method and device, computer equipment and readable storage medium | |
| CN105335689A (en) | Character recognition method and apparatus | |
| CN115953123A (en) | Method, device and equipment for generating robot automation flow and storage medium | |
| CN116627821A (en) | Source code defect detection method and related device | |
| CN113688953B (en) | Industrial control signal classification method, device and medium based on multilayer GAN network | |
| CN116798044A (en) | Text recognition method and device and electronic equipment | |
| CN114863574A (en) | Handwritten signature recognition method, device, equipment, medium and program product | |
| CN113934833A (en) | Training data acquisition method, device and system and storage medium | |
| CN118312890B (en) | Method for training keyword recognition model, method and device for recognizing keywords | |
| CN117195895B (en) | Relation extraction method and system based on semantic enhancement mechanism |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |