CN113556194A - Wireless signal region strength detection method based on deep learning - Google Patents
Wireless signal region strength detection method based on deep learning Download PDFInfo
- Publication number
- CN113556194A CN113556194A CN202110821215.XA CN202110821215A CN113556194A CN 113556194 A CN113556194 A CN 113556194A CN 202110821215 A CN202110821215 A CN 202110821215A CN 113556194 A CN113556194 A CN 113556194A
- Authority
- CN
- China
- Prior art keywords
- signal
- region
- deep learning
- dimensional
- threshold
- 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
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B17/00—Monitoring; Testing
- H04B17/30—Monitoring; Testing of propagation channels
- H04B17/309—Measuring or estimating channel quality parameters
- H04B17/318—Received signal strength
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computing Systems (AREA)
- General Physics & Mathematics (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Quality & Reliability (AREA)
- Electromagnetism (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a wireless signal area strength detection method based on deep learning, which comprises the following steps: obtaining a two-dimensional signal spectrogram through short-time Fourier transform; cutting, filling and deforming to obtain a language graph sample; making a data set label; training a customized data set by using a deep learning one-dimensional compression model to obtain a model weight file; pre-treating; predicting the preprocessed picture to obtain the output of the strength probability of the regional signal; and obtaining the position information of the signal and the image quality prediction value of the signal to obtain an optimal area. The invention can process signal detection under complex conditions, and has strong robustness; strong and weak probability output is carried out on each region of the signal, detection of the signal is realized, pertinence is strong, and false detection rate is low; not only the position information of the signal but also the quality evaluation of each region of the signal can be obtained.
Description
Technical Field
The invention belongs to the technical field of wireless signal processing, and particularly relates to a wireless signal region strength detection method based on deep learning.
Background
Radio communication is widely used in many fields such as outing, weather, post and telecommunications, military, and transportation to transmit information such as images, data, languages, and characters. Due to factors such as path attenuation, atmospheric noise, time delay, ionospheric fading, multipath effect and the like, short-wave signals are distorted and weakened in the propagation process, and the effect of radio communication is influenced. Meanwhile, in a radio communication scenario, a person is often required to observe a received map through experience to identify a signal and mark a position.
At present, the deep learning technology develops well in the aspect of image recognition, and the deep learning technology and the atlas recognition are combined, so that the workload of the participators is greatly reduced. Although the general deep learning detection algorithm has a certain effect on the identification of signals, the algorithm is not completely suitable for the detection of radio signals. The characteristics of a radio signal and a common object are greatly different, the radio signal may be a whole formed by several regions, the difference of texture characteristics of each region is large, even the same signal is subjected to different external interferences at different moments, so that the textures of the signal are different, and some weaker signals may only have incomplete characteristics. Because the marking mode of the sample is the frame selection marking in the general deep learning detection algorithm of the radio signal, the marking mode greatly influences the detection effect.
When the general deep learning target detection method is used for detecting the signal language graph, the background noise is inevitably framed in the frame selection mark due to the conditions of discontinuity, defect, low signal to noise ratio and the like of an actual signal, and the method can cause high false detection rate, so that the general deep learning mark and detection method are not suitable for detecting the signal.
Disclosure of Invention
In order to solve the above problems, an object of the present invention is to provide a method for detecting a wireless signal region strength based on deep learning, including:
the method comprises the following steps: carrying out short-time Fourier transform on the signal data to obtain a two-dimensional signal spectrogram;
step two: cutting, filling and deforming the two-dimensional signal spectrogram to obtain a spectrogram sample;
step three: averagely dividing each sample into N areas, grading the signal areas according to the strength of the signal areas, and marking each area of a training sample according to the grade to manufacture a data set label;
step four: training a customized data set by using a deep learning one-dimensional compression model, and stopping training after a loss value is converged to obtain a model weight file;
step five: preprocessing a prediction signal through the first step and the second step;
step six: predicting the preprocessed pictures by using a deep learning one-dimensional compression network and a trained model weight file, wherein each picture is predicted to obtain a group of corresponding region signal strength probability outputs with the length of M;
step seven: mapping the region signal strength probability to an original prediction picture, setting a signal threshold, setting an optimal region threshold as the difference between the maximum probability value and the signal threshold, considering the region signal as a signal when the maximum probability value is greater than the signal threshold, considering the region signal as background noise when the maximum probability value is less than the signal threshold, and integrating the region probabilities to obtain the position information of the signal; obtaining the image quality prediction value of each section of signal by averaging the prediction values of each section of signal area; and taking the longest part of the signal which is larger than the optimal region threshold value as the optimal region of the signal.
The invention has the beneficial effects that:
1. in actual real environment, weak signals and incomplete signals account for a great proportion, the traditional signal detection method is generally only suitable for the situation that the signals are good, the radio signal area strong and weak detection method based on deep learning is used, the signals can be detected under the situation that the signal-to-noise ratio is high, and the detection method can also have good effect when the signals are weak and incomplete, can process signal detection under complex conditions, and is strong in robustness;
2. the method for detecting the strength of the radio signal region based on deep learning is used, a language map of the radio signal region is compressed into a one-dimensional characteristic vector according to the characteristics of the signal, and strength probability output is carried out on each region of the signal, so that the detection of the signal is realized, the pertinence is strong, and the false detection rate is low;
3. the method for detecting the radio signal area based on deep learning uses a special data set, the mark information in the data set contains information of signal quality, after the method is used, not only the position information of the signal can be obtained, but also the quality evaluation of each area of the signal can be obtained, the probability output is carried out on the quality of the signal in each area, and the signal returns to the area with the optimal quality, and particularly, the method has good effect under the condition that the signal is weak or incomplete.
Drawings
FIG. 1 is a flow chart of the present invention;
FIG. 2 is a schematic flow chart of a detection method for a deep learning-based one-dimensional compression network and a weight file;
FIG. 3 is a schematic view of a customized data set production flow;
FIG. 4 is a schematic diagram of a signal prediction process;
FIG. 5 is a table of one-dimensional compression network structure of deep learning images;
Detailed Description
The invention will be further described with reference to the accompanying drawings in which:
as shown in fig. 1, the present invention is a method for detecting the intensity of a wireless signal region based on deep learning, which comprises the following steps:
the method comprises the following steps: carrying out short-time Fourier transform on the signal data to obtain a two-dimensional signal spectrogram;
step two: cutting, filling and deforming the two-dimensional signal spectrogram to obtain a spectrogram sample;
step three: averagely dividing each sample into N areas, grading the signal areas according to the strength of the signal areas, and marking each area of a training sample according to the grade to manufacture a data set label;
step four: training a customized data set by using a deep learning one-dimensional compression model, and stopping training after a loss value is converged to obtain a model weight file;
step five: preprocessing a prediction signal through the first step and the second step;
step six: predicting the preprocessed pictures by using a deep learning one-dimensional compression network and a trained model weight file, wherein each picture is predicted to obtain a group of corresponding region signal strength probability outputs with the length of M;
step seven: mapping the region signal strength probability to an original prediction picture, setting a signal threshold, setting an optimal region threshold as the difference between the maximum probability value and the signal threshold, considering the region signal as a signal when the maximum probability value is greater than the signal threshold, considering the region signal as background noise when the maximum probability value is less than the signal threshold, and integrating the region probabilities to obtain the position information of the signal; obtaining the image quality prediction value of each section of signal by averaging the prediction values of each section of signal area; and taking the longest part of the signal which is larger than the optimal region threshold value as the optimal region of the signal.
Further, each morphogram sample has a size of 256 × 512 × 1, a height of 256, and a width of 512.
Furthermore, the input of the deep learning one-dimensional compression model is 256 × 512 × 1, the output is 1 × 32 × 1, the network firstly uses 64 convolution kernels to compress the picture into a one-dimensional feature vector through convolution to obtain a 1 × 510 × 64 feature matrix, then performs pooling to obtain a 1 × 255 × 64 feature matrix, then performs high-dimensional feature extraction on the feature matrix, obtains a 1 × 32 × 512 feature matrix after three times of convolution pooling, and finally performs flattening and fully connects to a 512-dimensional feature vector, and then fully connects to a 32-dimensional feature vector.
Specifically, a method for detecting the strength of a wireless signal area based on deep learning comprises the following specific steps:
step one, carrying out short-time Fourier transform on signal data to obtain a two-dimensional signal spectrogram;
step two, cutting, filling and deforming the two-dimensional signal pictograph to obtain a pictograph sample, wherein the size of each sample is 256 × 512 × 1, the height of each sample is 256, and the width of each sample is 512;
averagely dividing each sample into N areas, grading the signal areas according to the strength of the signal areas, and marking each area of the training samples according to the grades to prepare a data set label;
step four, training a customized data set by using a deep learning one-dimensional compression model, stopping training after a loss value is converged, and obtaining a model weight file;
step five, preprocessing the prediction signal through the step one and the step two;
predicting the preprocessed pictures by using a deep learning one-dimensional compression network and a trained model weight file, wherein each picture can obtain a group of corresponding region signal strength probability outputs with the length of 32 through prediction;
step seven, mapping the region signal strength probability to the original prediction picture, setting a signal threshold value as a, setting an optimal region threshold value as the difference between the maximum probability value and the a, considering the region signal as a signal when the signal threshold value is larger than the signal threshold value, and considering the region signal as background noise when the signal threshold value is smaller than the signal threshold value, and integrating the region probabilities to obtain the position information of the signal; obtaining the image quality prediction value of each section of signal by averaging the prediction values of each section of signal area; and taking the longest part of the region with the signal, which is larger than the optimal threshold value, as the optimal region of the signal.
In this embodiment, as shown in fig. 1, which is a schematic view of an overall flow of the detection method of the present invention, the detection method needs to be based on a deep learning one-dimensional compression network and a weight file, the weight file is obtained by inputting a customized data set into the deep learning one-dimensional compression network for training, and a flow of manufacturing the customized data set is shown in fig. 2; performing short-time Fourier transform on the training signal to obtain a two-dimensional signal morphogram, performing shearing filling to obtain a training sample, and performing region labeling to obtain a data set, as shown in FIG. 3; then, the predicted signal can be used to detect the signal position and obtain the signal quality information through the network model and the weight file, and the detection process is shown in fig. 4.
In this embodiment, the present invention designs a deep learning-based one-dimensional compression network, and the network structure is shown in fig. 5. The input of the network is 256 × 512 × 1, the output is 1 × 32 × 1, the network firstly uses 64 convolution kernels to compress the picture into a one-dimensional feature vector through convolution to obtain a 1 × 510 × 64 feature matrix, then performs pooling to obtain a 1 × 255 × 64 feature matrix, then performs high-dimensional feature extraction on the feature matrix, obtains a 1 × 32 × 512 feature matrix through three times of convolution pooling, and finally performs flattening and fully connects to a 512-dimensional feature vector and then fully connects to a 32-dimensional feature vector.
The invention has the beneficial effects that:
1. in actual real environment, weak signals and incomplete signals account for a great proportion, the traditional signal detection method is generally only suitable for the situation that the signals are good, the radio signal area strong and weak detection method based on deep learning is used, the signals can be detected under the situation that the signal-to-noise ratio is high, and the detection method can also have good effect when the signals are weak and incomplete, can process signal detection under complex conditions, and is strong in robustness;
2. the method for detecting the strength of the radio signal region based on deep learning is used, a language map of the radio signal region is compressed into a one-dimensional characteristic vector according to the characteristics of the signal, and strength probability output is carried out on each region of the signal, so that the detection of the signal is realized, the pertinence is strong, and the false detection rate is low;
3. the method for detecting the radio signal area based on deep learning uses a special data set, the mark information in the data set contains information of signal quality, after the method is used, not only the position information of the signal can be obtained, but also the quality evaluation of each area of the signal can be obtained, the probability output is carried out on the quality of the signal in each area, and the signal returns to the area with the optimal quality, and particularly, the method has good effect under the condition that the signal is weak or incomplete.
The technical solution of the present invention is not limited to the limitations of the above specific embodiments, and all technical modifications made according to the technical solution of the present invention fall within the protection scope of the present invention.
Claims (3)
1. A wireless signal region strength detection method based on deep learning is characterized by comprising the following steps:
the method comprises the following steps: carrying out short-time Fourier transform on the signal data to obtain a two-dimensional signal spectrogram;
step two: cutting, filling and deforming the two-dimensional signal spectrogram to obtain a spectrogram sample;
step three: averagely dividing each sample into N areas, grading the signal areas according to the strength of the signal areas, and marking each area of a training sample according to the grade to manufacture a data set label;
step four: training a customized data set by using a deep learning one-dimensional compression model, and stopping training after a loss value is converged to obtain a model weight file;
step five: preprocessing a prediction signal through the first step and the second step;
step six: predicting the preprocessed pictures by using a deep learning one-dimensional compression network and a trained model weight file, wherein each picture is predicted to obtain a group of corresponding region signal strength probability outputs with the length of M;
step seven: mapping the region signal strength probability to an original prediction picture, setting a signal threshold, setting an optimal region threshold as the difference between the maximum probability value and the signal threshold, considering the region signal as a signal when the maximum probability value is greater than the signal threshold, and considering the region signal as background noise when the maximum probability value is less than the signal threshold, and integrating the region probabilities to obtain the position information of the region signal; obtaining the image quality prediction value of each section of signal by averaging the prediction values of each section of signal area; and taking the longest part of the signal which is larger than the optimal region threshold value as the optimal region of the signal.
2. The method according to claim 1, wherein the size of each morphogram sample is 256 × 512 × 1, the height is 256, and the width is 512.
3. The method for detecting the intensity of the wireless signal region based on the deep learning according to claim 1, wherein the input of the deep learning one-dimensional compression model is 256 × 512 × 1, the output is 1 × 32 × 1, the network first uses 64 convolution kernels to perform convolution compression on the picture into the one-dimensional feature vector to obtain 1 × 510 × 64 feature matrix, performs pooling to obtain 1 × 255 × 64 feature matrix, performs high-dimensional feature extraction on the feature matrix, performs three-dimensional convolution pooling to obtain 1 × 32 × 512 feature matrix, and finally performs flattening and full connection to 512-dimensional feature vectors and full connection to 32-dimensional feature vectors.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110821215.XA CN113556194B (en) | 2021-07-20 | 2021-07-20 | Wireless signal region strength detection method based on deep learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110821215.XA CN113556194B (en) | 2021-07-20 | 2021-07-20 | Wireless signal region strength detection method based on deep learning |
Publications (2)
Publication Number | Publication Date |
---|---|
CN113556194A true CN113556194A (en) | 2021-10-26 |
CN113556194B CN113556194B (en) | 2022-11-29 |
Family
ID=78103604
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110821215.XA Active CN113556194B (en) | 2021-07-20 | 2021-07-20 | Wireless signal region strength detection method based on deep learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113556194B (en) |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20110286555A1 (en) * | 2010-05-24 | 2011-11-24 | Postech Academy-Industry Foundation | Method and apparatus for detecting presence of signal in wireless communication system based on cr technology |
CN109886307A (en) * | 2019-01-24 | 2019-06-14 | 西安交通大学 | A kind of image detecting method and system based on convolutional neural networks |
CN109886060A (en) * | 2019-02-21 | 2019-06-14 | 苏州天准科技股份有限公司 | A kind of bar code area positioning method based on deep learning |
CN112740269A (en) * | 2020-05-13 | 2021-04-30 | 华为技术有限公司 | Target detection method and device |
CN112866156A (en) * | 2021-01-15 | 2021-05-28 | 浙江工业大学 | Radio signal clustering method and system based on deep learning |
CN112949658A (en) * | 2021-03-10 | 2021-06-11 | 清华大学 | Deep learning method with stable performance |
-
2021
- 2021-07-20 CN CN202110821215.XA patent/CN113556194B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20110286555A1 (en) * | 2010-05-24 | 2011-11-24 | Postech Academy-Industry Foundation | Method and apparatus for detecting presence of signal in wireless communication system based on cr technology |
CN109886307A (en) * | 2019-01-24 | 2019-06-14 | 西安交通大学 | A kind of image detecting method and system based on convolutional neural networks |
CN109886060A (en) * | 2019-02-21 | 2019-06-14 | 苏州天准科技股份有限公司 | A kind of bar code area positioning method based on deep learning |
CN112740269A (en) * | 2020-05-13 | 2021-04-30 | 华为技术有限公司 | Target detection method and device |
CN112866156A (en) * | 2021-01-15 | 2021-05-28 | 浙江工业大学 | Radio signal clustering method and system based on deep learning |
CN112949658A (en) * | 2021-03-10 | 2021-06-11 | 清华大学 | Deep learning method with stable performance |
Also Published As
Publication number | Publication date |
---|---|
CN113556194B (en) | 2022-11-29 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109165623B (en) | Rice disease spot detection method and system based on deep learning | |
CN113780296A (en) | Remote sensing image semantic segmentation method and system based on multi-scale information fusion | |
CN114973002A (en) | Improved YOLOv 5-based ear detection method | |
CN107527023A (en) | Classification of Polarimetric SAR Image method based on super-pixel and topic model | |
CN113469278B (en) | Strong weather target identification method based on deep convolutional neural network | |
CN112699717A (en) | SAR image generation method and generation device based on GAN network | |
CN102855485B (en) | The automatic testing method of one grow wheat heading | |
CN108171119B (en) | SAR image change detection method based on residual error network | |
CN114119642A (en) | Method, device, equipment and storage medium for extracting water body in flood disaster | |
CN112950780A (en) | Intelligent network map generation method and system based on remote sensing image | |
CN114612769A (en) | Integrated sensing infrared imaging ship detection method integrated with local structure information | |
CN103593853A (en) | Remote-sensing image multi-scale object-oriented classification method based on joint sparsity representation | |
CN113256624A (en) | Continuous casting round billet defect detection method and device, electronic equipment and readable storage medium | |
CN110516552B (en) | Multi-polarization radar image classification method and system based on time sequence curve | |
CN111339927B (en) | Intelligent recognition system for working state of personnel in electric power business hall | |
CN111291818B (en) | Non-uniform class sample equalization method for cloud mask | |
CN115302963A (en) | Bar code printing control method, system and medium based on machine vision | |
CN111707998A (en) | Sea surface floating small target detection method based on connected region characteristics | |
CN115035381A (en) | Lightweight target detection network of SN-YOLOv5 and crop picking detection method | |
CN113556194B (en) | Wireless signal region strength detection method based on deep learning | |
CN115546658B (en) | Night cloud detection method combining quality improvement and CNN improvement of data set | |
CN113256563A (en) | Method and system for detecting surface defects of fine product tank based on space attention mechanism | |
CN115830302B (en) | Multi-scale feature extraction fusion power distribution network equipment positioning identification method | |
CN117351371A (en) | Remote sensing image target detection method based on deep learning | |
CN113947723B (en) | High-resolution remote sensing scene target detection method based on size balance FCOS |
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 |