CN116910469A - Lightning signal processing method based on three-channel ResNet - Google Patents
Lightning signal processing method based on three-channel ResNet Download PDFInfo
- Publication number
- CN116910469A CN116910469A CN202310775223.4A CN202310775223A CN116910469A CN 116910469 A CN116910469 A CN 116910469A CN 202310775223 A CN202310775223 A CN 202310775223A CN 116910469 A CN116910469 A CN 116910469A
- Authority
- CN
- China
- Prior art keywords
- channel
- lightning
- time
- layer
- frequency
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000003672 processing method Methods 0.000 title claims abstract description 22
- 238000010586 diagram Methods 0.000 claims abstract description 43
- 238000012545 processing Methods 0.000 claims abstract description 30
- 238000000034 method Methods 0.000 claims abstract description 28
- 238000004458 analytical method Methods 0.000 claims abstract description 15
- 230000008569 process Effects 0.000 claims abstract description 8
- 238000009432 framing Methods 0.000 claims abstract description 4
- 238000005070 sampling Methods 0.000 claims description 14
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 claims description 9
- 238000001914 filtration Methods 0.000 claims description 9
- 238000006243 chemical reaction Methods 0.000 claims description 6
- 210000002569 neuron Anatomy 0.000 claims description 6
- 230000001052 transient effect Effects 0.000 claims description 5
- 230000005684 electric field Effects 0.000 claims description 4
- 238000005314 correlation function Methods 0.000 claims description 3
- 230000004927 fusion Effects 0.000 claims description 3
- 230000017105 transposition Effects 0.000 claims description 3
- 238000000605 extraction Methods 0.000 abstract description 4
- 230000009286 beneficial effect Effects 0.000 abstract description 3
- 230000006870 function Effects 0.000 description 13
- 238000000354 decomposition reaction Methods 0.000 description 5
- 238000001514 detection method Methods 0.000 description 5
- 238000011160 research Methods 0.000 description 5
- 230000008901 benefit Effects 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 238000004891 communication Methods 0.000 description 2
- 238000010276 construction Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000001228 spectrum Methods 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 230000007123 defense Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000009977 dual effect Effects 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012549 training Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- 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
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Measuring Frequencies, Analyzing Spectra (AREA)
Abstract
The invention relates to the technical field of lightning signal processing, in particular to a lightning signal processing method based on three channels ResNet. The method solves the problems of single lightning information extraction dimension and single processing method. The technical proposal is as follows: the method comprises the following steps: step one, framing processing of lightning signals; step two, constructing a first channel time-frequency diagram; step three, constructing a second channel time-frequency diagram; step four, constructing a third channel time-frequency diagram; fifthly, three-channel network characteristic processing; step six, the operation process of the ResNet lightning signal processing method based on three-channel time-frequency analysis. The beneficial effects of the invention are as follows: the invention can improve the accuracy and the robustness of lightning signal processing.
Description
Technical Field
The invention relates to the technical field of lightning signal processing, in particular to a lightning signal processing method based on three channels ResNet.
Background
Lightning is one of the serious main natural disasters, can cause forest and oil depot fires, cause power supply and communication information system faults or damages, and has great threat to aerospace, mines, some important and sensitive high-technology equipment and the like. After eighties, the damage caused by lightning is significantly increased, and particularly in the fields with close relation with high and new technologies, such as aerospace, national defense, communication, electric power, computer, electronic industry and the like, the probability of being struck by lightning is greatly increased due to wide application of large-scale and ultra-large-scale integrated circuits extremely sensitive to lightning electromagnetic interference.
Along with the development of detection and acquisition equipment and a positioning algorithm in recent years, the lightning positioning precision is continuously improved, which is beneficial to the development of thunderstorm prediction, tracking and lightning protection. The paper "a business lightning potential forecasting scheme" (meteorological science "2008) researches a lightning short-term forecasting scheme in Guangzhou area by adopting lightning positioning data, and has higher reference value for lightning potential forecasting.
However, the lightning positioning calculation depends on the time precision of lightning pulse, which has high requirements on the quality of the lightning original signal collected by the measuring station, and once noise pulse is mixed in the lightning signal, the positioning precision is reduced. The unknown environmental noise spectrum is wide, and the lightning signal spectrum is rich, so that the noise signal is difficult to filter by adopting a conventional frequency domain filter. The scientific research team usually adopts wavelet (dual-density dual-tree wavelet transformed lightning signal denoising research (laser and infrared (2013)) and empirical mode decomposition (Empirical Mode Decomposition, EMD) (ANew Method ofThree-Dimensional Location for Low-Frequency Electric Field DetectionArray) (Journal ofGeophysical Research (2019)) to denoise lightning signals, but the algorithms are more difficult to set parameters, such as wavelet for different types of signals, proper wavelet basis needs to be selected for transformation, and larger errors can be generated if improper wavelet basis is selected; the EMD algorithm is an iterative calculation method, so that the algorithm is very sensitive to the selection of an initial decomposition process and the setting of initial parameters, and some random noise or accidental events can have a great effect on the result. If an unsuitable initial decomposition parameter is selected, unstable processing results may be caused, and modal decomposition results may be distorted.
If a statistical learning method is adopted, the lightning signal can be directly extracted according to the lightning characteristics, but the statistical information extracted by the method is also easily affected by noise, so that the performance of the model is reduced, and the quality of the lightning signal is reduced. The time-frequency analysis method can separate most of noise in the frequency domain and visualize waveform information in the time-frequency domain, so that the method is gradually applied to lightning original signal processing. However, the current research is limited to the application of the single-dimensional single-channel technology in lightning location (journal of electric wave science (2019)), and the single-dimensional time-frequency processing usually loses part of information to affect the processing effect.
In summary, the prior art has some drawbacks, including single dimension of information extraction, single processing method, and simple fitting function.
Disclosure of Invention
The invention aims to provide a lightning signal processing method based on three-channel ResNet.
In order to achieve the aim of the invention, the invention adopts the technical scheme that:
a lightning signal processing method based on three channels ResNet includes the following steps:
step one, framing processing of lightning signals; the signals are segmented, so that the increase of time cost caused by long signals is avoided, and the detection efficiency and the detection precision are improved.
Step two, constructing a first channel time-frequency diagram; the channel captures the local time-frequency characteristic of the lightning signal.
Step three, constructing a second channel time-frequency diagram; the channel captures transient frequency information of the low frequency portion of the lightning signal.
Step four, constructing a third channel time-frequency diagram; the channel provides accurate and rapid lightning signal frequency domain variation trend under the condition of good SNR.
Fifthly, three-channel network characteristic processing; the advantages of various time-frequency methods are comprehensively considered, the high-dimensional data formed by the various time-frequency methods is reduced to a low-dimensional space, the redundancy and the computational complexity of the data are reduced, and meanwhile, important characteristic information is reserved.
Step six, the operation process of the ResNet lightning signal processing method based on three-channel time-frequency analysis. And a more complex and accurate identification model is provided for lightning signals in complex weather environments. The model can learn more complex and abstract characteristic representation, so that the lightning identification performance is improved.
The first step is as follows:
and acquiring lightning original signals of a plurality of measuring stations through a fast and slow antenna or a magnetic antenna, wherein the signals are lightning electric field signals or magnetic field signals. N sampling points are selected as the standard length of time-frequency processing, and a data segment formed by the N sampling points is called a frame; to avoid that a lightning signal is present between two frames, which affects the analysis and feature extraction of the signal, the latter frame contains M samples of the previous frame.
The second step is specifically as follows:
let the lightning signal of the first channel be x, the S-transformation result of the first channel is:
wherein f is frequency; η is the time variable of the signal x (η); t is the time of S (t, f) after S conversion of the signal x; w (η -t, f) is a gaussian window function expressed mathematically as:
when the lightning signal x is in discrete form, it is denoted x [ kT ], where k=0, 1, …, N-1; t is the sampling interval; the discrete representation of the S-transform is:
where n= fNT, m is a discrete index in the time domain, and j is an imaginary unit.
The third step is as follows:
let X (f) be the fourier transform of the second channel flash signal X, WVD be defined as the fourier transform of the signal center covariance function, then in the time domain signal, the WVD transform result of the second channel is:
wherein τ is an integral variable; * Represents the complex conjugate number;is a transient correlation function of the lightning signal;
in the frequency domain signal, WVD is defined as:
where θ is the integral variable in the frequency domain.
The fourth step is specifically as follows:
multiplying the signal of the third channel with a time-limited window function h (t), representing the non-stationary course of the lightning signal as a superposition of a plurality of short-time stationary signals; the STFT of a signal can be expressed as:
STFT(t,f)=∫x(τ)h(τ-t)e -j2πfτ dτ
where h (τ -t) is the analysis window function, and STFT is the conventional Fourier transform when h (t) is constant equal to 1.
The fifth step is specifically as follows:
let the output of the three channels be x m Wherein m=1, 2,3 corresponds to the first channel, the second channel, and the third channel, respectively. The first level of convolution operation is described as:
wherein ,inputting a jth time-frequency diagram of a mth channel first layer; />Outputting a jth time-frequency diagram of the mth channel first-1 layer; />A convolution kernel between the ith and the jth time-frequency diagrams of the mth channel ith layer; />Bias for the jth time-frequency diagram of the mth channel first layer; i epsilon M j The time-frequency diagram is a time-frequency diagram of the connection between the previous layer of the mth channel and the jth time-frequency diagram of the current layer;
the process of one downsampling is expressed as:
wherein ,multiplicative bias for the jth time-frequency diagram of the mth channel jth layer; down is the sample point addition
The mathematical description of the residual block is expressed as:
wherein ,output for the residual block of the mth channel layer 1+1,>input to the mth channel, layer I, W l m Weight of the mth channel first layer, < ->
The sixth step is specifically as follows:
respectively calculating time-frequency information of three channels for lightning signals through the second step, the third step and the fourth step, and constructing a three-channel time-frequency diagram;
randomly initializing network weights and biases;
step five, calculating the convolution layers and residual blocks of all channels layer by layer, and adding to obtain the characteristic output value;
and calculating a sampling layer of each channel, and fusing at a full-connection layer. Each neuron in the full-connection layer is connected with all neurons of the previous layer, and the input characteristics are subjected to linear combination and nonlinear conversion by using weights, so that characteristic fusion is performed;
obtaining a filtering result of the lightning signal through softmax;
if there is a sample class c and there is a sample u, the probability that softmax belongs to class c is:
wherein ,wc Is a class c weight vector; c is the total number of samples; (. Cndot. T Is a transposition operation;
the decision function of softmax is expressed as:
wherein ,the method comprises the following steps: when the function reaches the maximum, c takes on the value.
Compared with the prior art, the invention has the beneficial effects that:
1. the time-frequency analysis method has the advantage of reducing the influence of noise space on model construction in subsequent processing by effectively separating noises of lightning signals on different frequencies. By independently processing different frequency components, the quality and accuracy of the signals can be improved, the interference of noise on the signals can be reduced, and more reliable input can be provided for subsequent processing. The invention can improve the signal-to-noise ratio gain by at least 5 dB.
2. The invention adopts a lightning signal multichannel processing method to describe the lightning signal from different angles. Compared with the traditional single method, the multi-channel processing method can more comprehensively capture various characteristics of the lightning original signal. By integrating the information of different channels, more comprehensive and accurate signal description can be obtained, and the problem of incomplete signal description in the traditional method is avoided, so that the analysis and understanding capability of lightning signals is improved.
3. The invention adopts a ResNet network structure, introduces a jump connection mode, effectively relieves the gradient vanishing problem in the traditional deep network, and improves the accuracy of lightning signal processing. The jump connection allows the information to jump over some layers directly in the network, preserving more detail and feature information, facilitating efficient propagation of gradients and training of models. In this way, the ResNet network of the invention can better maintain gradient flow when processing lightning signals, and the convergence speed of the model and the accuracy of the result are improved. Compared with the traditional method, the lightning signal identification method can achieve the lightning signal identification accuracy of about 98 percent.
In general, compared with the traditional method, the method has new improvements and innovations in the aspects of denoising, feature map construction, lightning signal automatic processing and the like, and can improve the accuracy and the robustness of the lightning signal processing. By comprehensively utilizing the technologies of time-frequency analysis, multichannel processing and jump connection, the invention brings new breakthrough to the field of lightning signal processing and provides an effective solution for realizing more accurate and reliable processing results.
Drawings
The accompanying drawings are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate the invention and together with the embodiments of the invention, serve to explain the invention.
Fig. 1 is a schematic diagram of a multichannel res net network architecture according to the present invention.
FIG. 2 is a diagram showing the filtering effect of the present invention; (a) A time-frequency diagram of the signal before filtering and (b) a time-frequency diagram of the signal after filtering.
Detailed Description
The present invention will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present invention more apparent. Of course, the specific embodiments described herein are for purposes of illustration only and are not intended to limit the invention.
Example 1
Fig. 1 shows a three-channel flash signal processing flow. Firstly, lightning signals are processed through S transformation, WVD and STFT respectively, and the processed signals are mapped with identity through multilayer convolution; projecting the output result to the sampling layer by an addition operation (addition sign in the figure is the addition operation); then, taking the sampling layer results of the three channels as the input of a full-connection layer; finally, a lightning signal filtering model is modeled.
Fig. 2 shows a graph of a time-frequency analysis of lightning signals before and after processing. By this filtering method, clutter of the signal is effectively filtered out. The filter improves the quality and reliability of the signal by selectively attenuating or eliminating clutter components in the signal. The three-channel design enables the model to have higher clutter rejection capability, so that the filtered signals are cleaner and clearer, and unnecessary interference and noise are reduced. The filtering result can improve the readability, the understandability and the usability of the signal, and is helpful for further lightning signal analysis work.
A lightning signal processing method based on three channels ResNet includes the following steps:
step one, framing processing of lightning signals; the signals are segmented, so that the increase of time cost caused by long signals is avoided, and the detection efficiency and the detection precision are improved.
Step two, constructing a first channel time-frequency diagram; the channel captures the local time-frequency characteristic of the lightning signal.
Step three, constructing a second channel time-frequency diagram; the channel captures transient frequency information of the low frequency portion of the lightning signal.
Step four, constructing a third channel time-frequency diagram; the channel provides accurate and rapid lightning signal frequency domain variation trend under the condition of good SNR.
Fifthly, three-channel network characteristic processing; the advantages of various time-frequency methods are comprehensively considered, the high-dimensional data formed by the various time-frequency methods is reduced to a low-dimensional space, the redundancy and the computational complexity of the data are reduced, and meanwhile, important characteristic information is reserved.
Step six, the operation process of the ResNet lightning signal processing method based on three-channel time-frequency analysis. And a more complex and accurate identification model is provided for lightning signals in complex weather environments. The model can learn more complex and abstract characteristic representation, so that the lightning identification performance is improved.
The first step is as follows:
and acquiring lightning original signals of a plurality of measuring stations through a fast and slow antenna or a magnetic antenna, wherein the signals are lightning electric field signals or magnetic field signals. N sampling points are selected as the standard length of time-frequency processing, and a data segment formed by the N sampling points is called a frame; to avoid that a lightning signal is present between two frames, which affects the analysis and feature extraction of the signal, the latter frame contains M samples of the previous frame.
The second step is specifically as follows:
let the first flash channel signal be x, then the S-transform result of the first channel is:
wherein f is frequency; η is the time variable of the signal x (η); t is the time of S (t, f) after S conversion of the signal x; w (η -t, f) is a gaussian window function expressed mathematically as:
when the lightning signal x is in discrete form, it is denoted x [ kT ], where k=0, 1, …, N-1; t is the sampling interval; the discrete representation of the S-transform is:
where n= fNT, m is a discrete index in the time domain, and j is an imaginary unit.
The third step is as follows:
let X (f) be the fourier transform of the second lightning channel signal X, WVD be defined as the fourier transform of the signal center covariance function, then in the time domain signal the WVD transform result of the second channel is:
wherein τ is an integral variable; * Represents the complex conjugate number;is a transient correlation function of the lightning signal;
in the frequency domain signal, WVD is defined as:
where θ is the integral variable in the frequency domain.
The fourth step is specifically as follows:
multiplying the signal of the third channel with a time-limited window function h (t), representing the non-stationary course of the lightning signal as a superposition of a plurality of short-time stationary signals; the STFT of a signal can be expressed as:
STFT(t,f)=∫x(τ)h(τ-t)e -j2πfτ dτ
where h (τ -t) is the analysis window function, and STFT is the conventional Fourier transform when h (t) is constant equal to 1.
Let the output of the three channels be x m Wherein m=1, 2,3 corresponds to the first channel, the second channel, and the third channel, respectively. The first level of convolution operation is described as:
wherein ,inputting a jth time-frequency diagram of a mth channel first layer; />Outputting a jth time-frequency diagram of the mth channel first-1 layer; />A convolution kernel between the ith and the jth time-frequency diagrams of the mth channel ith layer; />Bias for the jth time-frequency diagram of the mth channel first layer; i epsilon M j The time-frequency diagram is a time-frequency diagram of the connection between the previous layer of the mth channel and the jth time-frequency diagram of the current layer;
the process of one downsampling is expressed as:
wherein ,multiplicative bias for the jth time-frequency diagram of the mth channel jth layer; down is the sample point addition
The mathematical description of the residual block is expressed as:
wherein ,output for the residual block of the mth channel layer 1+1,>input to the mth channel, layer I, W l m Weight of the mth channel first layer, < ->
The sixth step is specifically as follows:
respectively calculating time-frequency information of three channels for lightning signals through the second step, the third step and the fourth step, and constructing a three-channel time-frequency diagram;
randomly initializing network weights and biases;
step five, calculating the convolution layers and residual blocks of all channels layer by layer, and adding to obtain the characteristic output value;
and calculating a sampling layer of each channel, and fusing at a full-connection layer. Each neuron in the full-connection layer is connected with all neurons of the previous layer, and the input characteristics are subjected to linear combination and nonlinear conversion by using weights, so that characteristic fusion is performed;
obtaining a filtering result of the lightning signal through softmax;
if there is a sample class c and there is a sample u, the probability that softmax belongs to class c is:
wherein ,wc Is a class c weight vector; c is the total number of samples; (. Cndot. T Is a transposition operation;
the decision function of softmax is expressed as:
wherein ,the method comprises the following steps: when the function reaches the maximum, c takes on the value.
The foregoing description of the preferred embodiments of the invention is not intended to limit the invention to the precise form disclosed, and any such modifications, equivalents, and alternatives falling within the spirit and scope of the invention are intended to be included within the scope of the invention.
Claims (7)
1. A lightning signal processing method based on three-channel ResNet is characterized by comprising the following steps:
step one, framing processing of lightning signals;
step two, constructing a first channel time-frequency diagram;
step three, constructing a second channel time-frequency diagram;
step four, constructing a third channel time-frequency diagram;
fifthly, three-channel network characteristic processing;
step six, the operation process of the ResNet lightning signal processing method based on three-channel time-frequency analysis.
2. The lightning signal processing method based on three-channel ResNet according to claim 1, wherein the first step is specifically:
acquiring lightning original signals of a plurality of measuring stations through a fast and slow antenna or a magnetic antenna, wherein the signals are lightning electric field signals or magnetic field signals; n sampling points are selected as the standard length of time-frequency processing, and a data segment formed by the N sampling points is called a frame; the latter frame contains M samples of the previous frame.
3. The lightning signal processing method based on three-channel ResNet according to claim 2, wherein the second step is specifically:
let the lightning signal of the first channel be x, the S-transformation result of the first channel is:
wherein f is frequency; η is the time variable of the signal x (η); t is the time of S (t, f) after S conversion of the signal x; w (η -t, f) is a gaussian window function expressed mathematically as:
when the lightning signal x is in discrete form, it is denoted x [ kT ], where k=0, 1, …, N-1; t is the sampling interval; the discrete representation of the S-transform is:
where n= fNT, m is a discrete index in the time domain, and j is an imaginary unit.
4. A lightning signal processing method based on three-channel res net according to claim 3, wherein the third step is specifically:
let X (f) be the fourier transform of the second channel flash signal X, WVD be defined as the fourier transform of the signal center covariance function, then in the time domain signal, the WVD transform result of the second channel is:
wherein τ is an integral variable; * Represents the complex conjugate number;is a transient correlation function of the lightning signal;
in the frequency domain signal, WVD is defined as:
where θ is the integral variable in the frequency domain.
5. The lightning signal processing method based on three-channel ResNet as claimed in claim 4, wherein said step four is:
multiplying the signal of the third channel with a time-limited window function h (t), representing the non-stationary course of the lightning signal as a superposition of a plurality of short-time stationary signals; the STFT of a signal can be expressed as:
where h (τ -t) is the analysis window function, and STFT is the conventional Fourier transform when h (t) is constant equal to 1.
6. The lightning signal processing method based on three-channel ResNet according to claim 5, wherein the fifth step is specifically:
let the output of the three channels be x m Where m=1, 2, and 3 correspond to the first channel, the second channel, and the third channel, respectively, then the convolution operation of the first layer is described as:
wherein ,inputting a jth time-frequency diagram of a mth channel first layer; />Outputting a jth time-frequency diagram of the mth channel first-1 layer; />A convolution kernel between the ith and the jth time-frequency diagrams of the mth channel ith layer; />Bias for the jth time-frequency diagram of the mth channel first layer; i epsilon M j The time-frequency diagram is a time-frequency diagram of the connection between the previous layer of the mth channel and the jth time-frequency diagram of the current layer;
the process of one downsampling is expressed as:
wherein ,multiplicative bias for the jth time-frequency diagram of the mth channel jth layer; down is the sample point addition;
the mathematical description of the residual block is expressed as:
wherein ,output for the residual block of the mth channel layer 1+1,>input to the mth channel, layer I, W l m Weight of the mth channel first layer, < ->
7. The lightning signal processing method based on three-channel ResNet as claimed in claim 6, wherein step six is specifically:
respectively calculating time-frequency information of three channels for lightning signals through the second step, the third step and the fourth step, and constructing a three-channel time-frequency diagram;
randomly initializing network weights and biases;
step five, calculating the convolution layers and residual blocks of all channels layer by layer, and adding to obtain the characteristic output value;
calculating sampling layers of all channels, and fusing at a full-connection layer; each neuron in the full-connection layer is connected with all neurons of the previous layer, and the input characteristics are subjected to linear combination and nonlinear conversion by using weights, so that characteristic fusion is performed;
obtaining a filtering result of the lightning signal through softmax;
if there is a sample class c and there is a sample u, the probability that softmax belongs to class c is:
wherein ,wc Is a class c weight vector; c is the total number of samples; (. Cndot. T Is a transposition operation;
the decision function of softmax is expressed as:
wherein ,the method comprises the following steps: when the function reaches the maximum, c takes on the value.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310775223.4A CN116910469A (en) | 2023-06-28 | 2023-06-28 | Lightning signal processing method based on three-channel ResNet |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310775223.4A CN116910469A (en) | 2023-06-28 | 2023-06-28 | Lightning signal processing method based on three-channel ResNet |
Publications (1)
Publication Number | Publication Date |
---|---|
CN116910469A true CN116910469A (en) | 2023-10-20 |
Family
ID=88363945
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202310775223.4A Pending CN116910469A (en) | 2023-06-28 | 2023-06-28 | Lightning signal processing method based on three-channel ResNet |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN116910469A (en) |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2018036280A1 (en) * | 2016-08-26 | 2018-03-01 | 深圳大学 | Method of using mobile terminal to detect and cooperatively position lightning |
CN112529045A (en) * | 2020-11-20 | 2021-03-19 | 济南信通达电气科技有限公司 | Weather image identification method, equipment and medium related to power system |
CN113655295A (en) * | 2021-10-21 | 2021-11-16 | 南京信息工程大学 | Lightning intensity identification method based on radar detection data |
CN114755745A (en) * | 2022-05-13 | 2022-07-15 | 河海大学 | Hail weather identification and classification method based on multi-channel depth residual shrinkage network |
CN114866172A (en) * | 2022-07-05 | 2022-08-05 | 中国人民解放军国防科技大学 | Interference identification method and device based on inverse residual deep neural network |
-
2023
- 2023-06-28 CN CN202310775223.4A patent/CN116910469A/en active Pending
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2018036280A1 (en) * | 2016-08-26 | 2018-03-01 | 深圳大学 | Method of using mobile terminal to detect and cooperatively position lightning |
CN112529045A (en) * | 2020-11-20 | 2021-03-19 | 济南信通达电气科技有限公司 | Weather image identification method, equipment and medium related to power system |
CN113655295A (en) * | 2021-10-21 | 2021-11-16 | 南京信息工程大学 | Lightning intensity identification method based on radar detection data |
CN114755745A (en) * | 2022-05-13 | 2022-07-15 | 河海大学 | Hail weather identification and classification method based on multi-channel depth residual shrinkage network |
CN114866172A (en) * | 2022-07-05 | 2022-08-05 | 中国人民解放军国防科技大学 | Interference identification method and device based on inverse residual deep neural network |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111783558A (en) | Satellite navigation interference signal type intelligent identification method and system | |
CN109890043B (en) | Wireless signal noise reduction method based on generative countermeasure network | |
CN102799892B (en) | Mel frequency cepstrum coefficient (MFCC) underwater target feature extraction and recognition method | |
CN102664010B (en) | Robust speaker distinguishing method based on multifactor frequency displacement invariant feature | |
CN104655423A (en) | Rolling bearing fault diagnosis method based on time-frequency domain multidimensional vibration feature fusion | |
CN112735460B (en) | Beam forming method and system based on time-frequency masking value estimation | |
CN109543643A (en) | Carrier signal detection method based on one-dimensional full convolutional neural networks | |
Ma et al. | Single‐channel blind source separation for vibration signals based on TVF‐EMD and improved SCA | |
CN106451498A (en) | Low frequency oscillation modal identification method based on improved generalized morphological filtering | |
CN109658944B (en) | Helicopter acoustic signal enhancement method and device | |
CN107329115B (en) | LFM modulated parameter estimating method based on GCRBF network | |
Mao et al. | Electric network frequency based audio forensics using convolutional neural networks | |
CN116576956A (en) | Multisource vibration signal separation method based on distributed optical fiber acoustic wave sensing | |
CN113642417B (en) | Denoising method for partial discharge signal of insulated overhead conductor based on improved wavelet algorithm | |
Liu et al. | Maximum correntropy criterion-based blind deconvolution and its application for bearing fault detection | |
CN113238193A (en) | Multi-component combined reconstruction SAR echo broadband interference suppression method | |
Ma et al. | DBN based automatic modulation recognition for ultra-low SNR RFID signals | |
CN113052099A (en) | SSVEP classification method based on convolutional neural network | |
Limin et al. | Low probability of intercept radar signal recognition based on the improved AlexNet model | |
Zhu et al. | Aiming to Complex Power Quality Disturbances: A Novel Decomposition and Detection Framework | |
CN116910469A (en) | Lightning signal processing method based on three-channel ResNet | |
CN114598577B (en) | Multi-band signal fusion filtering method for 5G communication system | |
CN107341519B (en) | Support vector machine identification optimization method based on multi-resolution analysis | |
Mou et al. | Motion classification for radar moving target via STFT and convolution neural network | |
DelSole | Low-frequency variations of surface temperature in observations and simulations |
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 |