Intelligent identification method and system for ultra-short wave signal with low signal-to-noise ratio

Through high power spectrum method and deep learning technology, combined with Mobile-Net neural network, the problem of low recognition accuracy of ultra-short wave signal under low signal-to-noise ratio is solved, and more efficient identification and simplified identification process is achieved, reducing hardware requirements and maintenance costs.

CN119004163APending Publication Date: 2024-11-22ZHEJIANG YANGMING TECHNOLOGY DEVELOPMENT CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410947817.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-16
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

The existing methods are difficult to effectively identify ultrashort wave signals in low signal-to-noise environments, resulting in low recognition accuracy, mainly because the feature extraction of ultrashort wave signals under low signal-to-noise ratio is not obvious.

Method used

High power spectrum method is used to extract the features of ultrashort wave signals, combined with deep learning technology, and the extracted features are identified and classified using the Mobile-Net neural network model. Through data preprocessing, the training set, test set and verification set are divided, and the training set is trained. and verify neural network models to improve recognition rates.

Benefits of technology

By simplifying the preprocessing steps, reducing the fluctuation of threshold values, reducing maintenance costs, and improving hardware compatibility, the recognition accuracy and recognition efficiency of low signal-to-noise ratio ultra-short wave signals are significantly improved.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119004163A_ABST
    Figure CN119004163A_ABST
Patent Text Reader

Abstract

The invention relates to an intelligent identification method and system for an ultra-short wave signal with a low signal-to-noise ratio, and the method comprises the steps: carrying out the data preprocessing of a signal sample in a data set, converting the ultra-short wave signal into a corresponding high-power spectrogram through frequency spectrum transformation, extracting the features in the high-power spectrogram through a neural network, and carrying out the extraction of the features in the high-power spectrogram; performing ultra-short wave signal identification and classification on the features extracted by the neural network; dividing the data set into a training set, a test set and a verification set, and dividing the preprocessed ultrashort wave data; training and testing the training data of the training set and the test data of the test set by adopting a Mobile-Net neural network model; and verifying the recognition accuracy of the ultimately trained network model to the ultrashort wave signal. According to the method, the features of the ultra-short wave signals are extracted through the high-power spectrum method, then the features of the ultra-short wave signals are extracted again through deep learning, and the recognition rate of the ultra-short wave signals is effectively improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the technical field of signal processing, and in particular relates to an intelligent recognition method and system for ultrashort wave signals with low signal-to-noise ratio. Background Art

[0002] Ultra-short wave signal recognition plays a very important role in electronic countermeasures, software radio, etc., and has been a research hotspot in the field of signal processing for many years. Ultra-short wave communication mainly uses radio waves in the 30MHz to 300MHz band to transmit information.

[0003] Ultrashort waves are widely used in civil and military fields. The main civil applications include navigation, television, FM broadcasting, radar, ionospheric scattering communication, fixed and mobile communication services. Ultrashort wave communication is widely used in the above services due to its advantages of flexibility, simplicity, reliability, strong anti-destruction, and low development and maintenance costs. In the military field, ultrashort wave radio stations can transmit signals in complex electromagnetic environments with their strong penetration and anti-interference capabilities, achieve long-distance communication, and can also be used in military reconnaissance and intelligence collection. Since ultrashort wave communication is not restricted by geography and has a large coverage range, it is often used for maritime communication, providing ships with a good means of communication and ensuring the safety of maritime navigation. Ultrashort wave communication is also often used in emergency communications. In natural disasters, major accidents or other emergency situations, ultrashort wave radio stations can achieve ultra-long-distance communication through unique transmission methods and frequency bands, providing important communication means for rescue work.

[0004] With the development of deep learning, a large number of scholars have begun to introduce deep learning into the modulation recognition of communication signals, which has led to the rapid development of the modulation recognition technology of communication signals and achieved good results. Inspired by this, some scholars have begun to introduce this idea into the recognition of ultrashort wave signals and have achieved certain results. However, from the current research results, the existing methods are not suitable for the recognition of ultrashort wave signals in low signal-to-noise ratio environments. The main reason is that the features of ultrashort wave signals extracted by these methods are not obvious under low signal-to-noise ratio, resulting in low accuracy of ultrashort wave signal recognition under low signal-to-noise ratio. Therefore, in a low signal-to-noise ratio environment, the recognition of ultrashort wave signals becomes a challenging task. Therefore, studying an effective intelligent recognition method for ultrashort wave signals with low signal-to-noise ratio has important practical significance and application value. Summary of the invention

[0005] The purpose of the present invention is to provide an intelligent recognition method and system for ultrashort wave signals with low signal-to-noise ratio. The characteristics of ultrashort wave signals are first extracted by a high-order spectrum method, and then the characteristics of ultrashort wave signals are extracted again by deep learning, so as to effectively improve the recognition rate of ultrashort wave signals.

[0006] The present invention provides an intelligent recognition method for ultrashort wave signals with low signal-to-noise ratio, comprising the following steps:

[0007] Step 1: perform data preprocessing on the signal samples in the data set, convert the ultrashort wave signal into the corresponding high-order spectrum through spectrum transformation, use the neural network to extract the features in the high-order spectrum, and identify and classify the ultrashort wave signal based on the features extracted by the neural network;

[0008] Step 2: Divide the data set into a training set, a test set, and a validation set. The pre-processed ultrashort wave data is divided into a training set accounting for 70%, a test set accounting for 20%, and a validation set accounting for 10%.

[0009] Step 3: Use the Mobile-Net neural network model to train and test the training data of the training set and the test data of the test set;

[0010] Step 4: Verify the accuracy of the finally trained network model in recognizing ultra-short wave signals.

[0011] Furthermore, the step 1 specifically includes:

[0012] The ultrashort wave signal received by the receiver is divided into L segments, and the length of each segment is the same, which is N. The calculation formula of the high-order spectrum of the i-th segment ultrashort wave signal is as follows:

[0013]

[0014] Among them, U represents the normalization factor, d(t) represents the windowing function, the windowing function selects the Hanning window, and M represents the power of the high-order spectrum, which represents the second power, the fourth power, and the eighth power;

[0015] The high-order spectrum estimates obtained in each segment are averaged to obtain the overall high-order spectrum estimate of the ultrashort wave signal. The specific formula is as follows:

[0016]

[0017] The generated high-order spectrum of the ultrashort wave signal is sent as training data to the neural network training to extract features for recognition and classification.

[0018] Furthermore, the total amount of the data set in step 2 is 20,000 pictures, each picture is 224*224 in size, and the signal-to-noise ratio of the ultrashort wave signal data set ranges from -5dB to 5dB.

[0019] Furthermore, the training steps of the Mobile-Net neural network model in step 3 include:

[0020] The pre-processed training data and test data are sent to the input layer in the form of images. The features in the high-order spectrum are extracted through deep separable convolution. The ultrashort wave signal is classified according to the extracted features, and finally the classification result is output through the Softmax layer.

[0021] Furthermore, the step 4 includes:

[0022] Feed the validation data into the trained network model;

[0023] Load the trained network model;

[0024] Output the results and accuracy of the validation data.

[0025] The present invention also provides an intelligent recognition system for ultrashort wave signals with low signal-to-noise ratio, comprising:

[0026] The data preprocessing module is used to perform data preprocessing on the signal samples in the data set, convert the ultrashort wave signal into the corresponding high-order spectrum through spectrum transformation, use the neural network to extract the features in the high-order spectrum, and identify and classify the ultrashort wave signal based on the features extracted by the neural network;

[0027] The data set division module is used to divide the data set into a training set, a test set and a validation set, and divide the pre-processed ultrashort wave data into a training set accounting for 70%, a test set accounting for 20%, and a validation set accounting for 10%;

[0028] The deep convolutional neural network training module is used to train and test the training data of the training set and the test data of the test set using the Mobile-Net neural network model;

[0029] The verification module of the deep convolutional neural network is used to verify the accuracy of the finally trained network model in recognizing ultra-short wave signals.

[0030] Furthermore, the data preprocessing module divides the ultrashort wave signal received by the receiver into L segments, each segment has the same cutting length, which is N. The calculation formula of the high-order spectrum of the i-th segment ultrashort wave signal is as follows:

[0031]

[0032] Among them, U represents the normalization factor, d(t) represents the windowing function, the windowing function selects the Hanning window, and M represents the power of the high-order spectrum, which represents the second power, the fourth power, and the eighth power;

[0033] The high-order spectrum estimates obtained in each segment are averaged to obtain the overall high-order spectrum estimate of the ultrashort wave signal. The specific formula is as follows:

[0034]

[0035] The generated high-order spectrum of the ultrashort wave signal is sent as training data to the neural network training to extract features for recognition and classification.

[0036] Furthermore, the total amount of the data set is 20,000 pictures, each picture is 224*224 in size, wherein the signal-to-noise ratio of the ultrashort wave signal data set ranges from -5dB to 5dB.

[0037] Furthermore, the training module of the deep convolutional neural network sends the pre-processed training data and test data to the input layer in the form of images, extracts features in the high-order spectrum through deep separable convolution, classifies the ultrashort wave signal according to the extracted features, and finally outputs the classification results through the Softmax layer.

[0038] Furthermore, the verification module of the deep convolutional neural network sends the verification data to the trained network model; loads the trained network model; and outputs the result and accuracy of the verification data.

[0039] By means of the above scheme, the intelligent recognition method and system for ultrashort wave signals with low signal-to-noise ratio have the following technical effects:

[0040] 1) The present invention performs simple high-order spectrum processing on ultrashort wave signals, simplifies the complexity of preprocessing, and solves the problem of high computational complexity of traditional algorithms for identifying ultrashort wave signals with low signal-to-noise ratio.

[0041] 2) The present invention has intelligent recognition, and autonomously learns the characteristics of ultrashort wave signals through a network model, and distinguishes ultrashort wave signals according to comprehensive characteristics, which can greatly reduce the fluctuation of threshold values, and solves the problem that the traditional algorithm sets a threshold value to distinguish ultrashort wave signals. Since the threshold value is easily affected by the signal-to-noise ratio, the threshold value fluctuates greatly and errors are prone to occur.

[0042] 3) The present invention only needs to retrain the new data and then replace the trained network model, which solves the problem that the traditional algorithm has high maintenance cost and once the environment changes, the threshold value is easily affected, requiring personnel to improve the corresponding algorithm in time.

[0043] 4) The present invention adopts the Mobile-Net lightweight neural network, which not only has a short training time but also has low hardware requirements. The network model is easy to deploy in the project, which solves the problem that the training of traditional neural networks takes a lot of time and has high hardware requirements, and is difficult to be integrated into the project for use.

[0044] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention and implement it according to the contents of the specification, the following is a detailed description of the preferred embodiments of the present invention in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 This is a flow chart of intelligent recognition of ultrashort wave signals with low signal-to-noise ratio according to the present invention;

[0046] Figure 2 This is a diagram of depth-separable convolution calculation results in one embodiment of the present invention;

[0047] Figure 3 This is a diagram of standard convolution calculation results in one embodiment of the present invention;

[0048] Figure 4 Schematic diagram of the structure of the Mobile-Net neural network in one embodiment of the present invention. DETAILED DESCRIPTION

[0049] The specific implementation of the present invention is further described in detail below in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

[0050] Ginseng Figure 1 As shown, this embodiment uses a high-power spectrum and intelligent recognition method to improve the problem of low recognition rate of ultrashort wave signals under low signal-to-noise ratio. The specific steps are as follows:

[0051] Step 1: Through the data preprocessing module, the signal samples in the data set need to be processed by high-order spectrum first. The high-order spectrum algorithm mainly converts the ultrashort wave signal into the corresponding high-order spectrum through spectrum transformation. From the high-order spectrum, it can be intuitively seen that different signals have different high-order spectra. However, it is quite difficult to construct the relevant characteristic parameters in the high-order spectrum through formulas. Therefore, in order to reduce the amount of calculation for constructing characteristic parameters, this time we choose to convert the ultrashort wave signal into a high-order spectrum, and then directly send the high-order spectrum generated by the ultrashort wave signal to the neural network, use the neural network to extract the features in the high-order spectrum, and identify and classify the ultrashort wave signal according to the features extracted by the neural network.

[0052] The ultrashort wave signal received by the receiver is processed by high-order spectrum. The main method used is the Welch high-order spectrum estimation algorithm, which divides the ultrashort wave signal received by the receiver into L segments, and the length of each segment is the same, which is N. The calculation formula of the high-order spectrum of the i-th segment ultrashort wave signal is as follows:

[0053]

[0054] In the above expression, U represents the normalization factor, d(t) represents the windowing function, the windowing function is the Hanning window, and M represents the power of the high-order spectrum, which can represent the second power, the fourth power, and the eighth power. Then, the high-order spectrum estimation obtained by segmentation is averaged to obtain the overall high-order spectrum estimation of the ultrashort wave signal. The specific formula is as follows:

[0055]

[0056] After the ultrashort wave signal is converted into the corresponding high-order spectrum, the high-order spectrum of different types of ultrashort wave signals will show different characteristics in peak size, number of single spectral lines, and shape of the overall spectrum. Therefore, the generated high-order spectrum of the ultrashort wave signal can be sent as training data to the neural network training to extract features, thereby achieving the purpose of recognition and classification.

[0057] Step 2: Use the data set partitioning module to divide the data set into training set, test set and validation set. The pre-processed ultrashort wave data is divided into 70% training set, 20% test set and 10% validation set. The total data set is 20,000 images, each of which is 224*224 in size. The signal-to-noise ratio of the ultrashort wave signal data set ranges from -5dB to 5dB.

[0058] Step 3: Through the deep convolutional neural network training module, the Mobile-Net neural network model is selected. This network model is a lightweight convolutional neural network. As we all know, the training time of convolutional neural networks is long, the amount of calculation is large, and the hardware requirements are high. Considering the convenience of engineering implementation and hardware cost, the Mobile-Net neural network model can solve the above problems well. Compared with the traditional convolutional neural network, the convolution of this network model uses depth-separable convolution, which simplifies the amount of calculation in the convolution process.

[0059] Step 4: Use the verification module of the deep convolutional neural network to verify the accuracy of the final trained network model in identifying ultrashort wave signals. The specific steps can be divided into the following steps: the first step is to send the verification data to the trained network model; the second step is to load the trained network model into the verification module; the third step is to output the results and accuracy of the verification data.

[0060] like Figure 2 As shown, the detailed calculation process of the depth-separable convolution is as follows: Its convolution kernel can be expressed as D k ×D k ×M, and the convolution kernel of point-by-point convolution can be expressed as 1×1×M. There are N feature maps in total, so the parameter amount of the convolution can be expressed as

[0061] D k ×D k×M+M×D

[0062] The amount of calculation can be expressed as:

[0063] D k ×D k ×M×D W ×D H +M×N×D W ×D H

[0064] like Figure 3 As shown, the number of parameters of the standard convolution can be expressed as:

[0065] D k ×D k ×M×N

[0066] The amount of calculation can be expressed as:

[0067] D k ×D k ×M×N×D W ×D H

[0068] pass Figure 2 and Figure 3 By comparison, it can be seen that the computational complexity of depthwise separable convolution is less than that of standard convolution. The specific reduction can be expressed as

[0069]

[0070] like Figure 4 As shown in the figure, the training steps of the Mobile-Net neural network model are as follows: first, the pre-processed training data and test data are sent to the input layer in the form of images, then the features in the high-order spectrum are extracted through deep separable convolution, and then the ultrashort wave signal is classified according to the extracted features, and finally the classification result is output through the Softmax layer.

[0071] The Mobile-Net network model is actually a streamlined structure, which only uses depthwise separable convolution instead of standard convolution. Depthwise separable convolution is decomposed into two parts, namely pointwise convolution and depthwise convolution. In the deep convolutional network, each input path uses a single filter, which is then folded point by point with the result of the depthwise convolution. This can achieve the function of filtering and can also be used for the final merging.

[0072] The design of the Mobile-Net network structure mainly involves the design of the input layer, the number of convolution kernels, the size of the convolution kernels, the design of the convolution step, the design of the pooling layer, the design of the fully connected layer and the output layer. The network model has a total of 28 levels of network structure.

[0073] The intelligent recognition method for low signal-to-noise ratio ultrashort wave signals has the following technical effects:

[0074] 1) The present invention performs simple high-order spectrum processing on ultrashort wave signals, simplifies the complexity of preprocessing, and solves the problem of high computational complexity of traditional algorithms for identifying ultrashort wave signals with low signal-to-noise ratio.

[0075] 2) The present invention has intelligent recognition, and autonomously learns the characteristics of ultrashort wave signals through a network model, and distinguishes ultrashort wave signals according to comprehensive characteristics, which can greatly reduce the fluctuation of threshold values, and solves the problem that the traditional algorithm sets a threshold value to distinguish ultrashort wave signals. Since the threshold value is easily affected by the signal-to-noise ratio, the threshold value fluctuates greatly and errors are prone to occur.

[0076] 3) The present invention only needs to retrain the new data and then replace the trained network model, which solves the problem that the traditional algorithm has high maintenance cost and once the environment changes, the threshold value is easily affected, requiring personnel to improve the corresponding algorithm in time.

[0077] 4) The present invention adopts the Mobile-Net lightweight neural network, which not only has a short training time but also has low hardware requirements. The network model is easy to deploy in the project, which solves the problem that the training of traditional neural networks takes a lot of time and has high hardware requirements, and is difficult to be integrated into the project for use.

[0078] The above description is only a preferred embodiment of the present invention and does not limit the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. An intelligent recognition method for ultrashort wave signals with low signal-to-noise ratio, characterized in that: The following steps are involved: Step 1: perform data preprocessing on the signal samples in the data set, convert the ultrashort wave signal into the corresponding high-order spectrum through spectrum transformation, use the neural network to extract the features in the high-order spectrum, and identify and classify the ultrashort wave signal based on the features extracted by the neural network; Step 2: Divide the data set into a training set, a test set, and a validation set. The pre-processed ultrashort wave data is divided into a training set accounting for 70%, a test set accounting for 20%, and a validation set accounting for 10%. Step 3: Use the Mobile-Net neural network model to train and test the training data of the training set and the test data of the test set; Step 4: Verify the accuracy of the finally trained network model in recognizing ultra-short wave signals.

2. The intelligent recognition method for ultrashort wave signals with low signal-to-noise ratio according to claim 1, characterized in that: The step 1 specifically includes: The ultrashort wave signal received by the receiver is divided into L segments, and the length of each segment is the same, which is N. The calculation formula of the high-order spectrum of the i-th segment ultrashort wave signal is as follows: Among them, U represents the normalization factor, d(t) represents the windowing function, the windowing function selects the Hanning window, and M represents the power of the high-order spectrum, which represents the second power, the fourth power, and the eighth power; The high-order spectrum estimates obtained in each segment are averaged to obtain the overall high-order spectrum estimate of the ultrashort wave signal. The specific formula is as follows: The generated high-order spectrum of the ultrashort wave signal is sent as training data to the neural network training to extract features for recognition and classification.

3. The intelligent identification method for ultrashort wave signals with low signal-to-noise ratio according to claim 2, characterized in that: The total amount of the data set in step 2 is 20,000 images, each image is 224*224 in size, and the signal-to-noise ratio of the ultrashort wave signal data set ranges from -5dB to 5dB.

4. The intelligent identification method for ultrashort wave signals with low signal-to-noise ratio according to claim 3 is characterized in that: The training steps of the Mobile-Net neural network model in step 3 include: The pre-processed training data and test data are sent to the input layer in the form of images. The features in the high-order spectrum are extracted through deep separable convolution. The ultrashort wave signal is classified according to the extracted features, and finally the classification result is output through the Softmax layer.

5. The intelligent identification method for ultrashort wave signals with low signal-to-noise ratio according to claim 4, characterized in that: The fourth step comprises: Feed the validation data into the trained network model; Load the trained network model; Output the results and accuracy of the validation data.

6. An intelligent recognition system for ultrashort wave signals with low signal-to-noise ratio, characterized in that: include: The data preprocessing module is used to perform data preprocessing on the signal samples in the data set, convert the ultrashort wave signal into the corresponding high-order spectrum through spectrum transformation, use the neural network to extract the features in the high-order spectrum, and identify and classify the ultrashort wave signal based on the features extracted by the neural network; The data set division module is used to divide the data set into a training set, a test set and a validation set, and divide the pre-processed ultrashort wave data into a training set accounting for 70%, a test set accounting for 20%, and a validation set accounting for 10%; The deep convolutional neural network training module is used to train and test the training data of the training set and the test data of the test set using the Mobile-Net neural network model; The verification module of the deep convolutional neural network is used to verify the accuracy of the finally trained network model in recognizing ultra-short wave signals.

7. According to the intelligent recognition system for ultrashort wave signals with low signal-to-noise ratio according to claim 6, the data preprocessing module divides the ultrashort wave signal received by the receiver into L segments, and the cutting length of each segment is consistent, which is N. The calculation formula of the high-order spectrum of the i-th segment ultrashort wave signal is as follows: in, U represents the normalization factor, d(t) represents the windowing function, the windowing function selects the Hanning window, and M represents the power of the high-order spectrum, which represents the second power, the fourth power, and the eighth power; The high-order spectrum estimates obtained in each segment are averaged to obtain the overall high-order spectrum estimate of the ultrashort wave signal. The specific formula is as follows: The generated high-order spectrum of the ultrashort wave signal is sent as training data to the neural network training to extract features for recognition and classification.

8. The intelligent recognition system for ultrashort wave signals with low signal-to-noise ratio according to claim 7, characterized in that: The total amount of the data set is 20,000 pictures, each picture is 224*224 in size, and the signal-to-noise ratio of the ultrashort wave signal data set ranges from -5dB to 5dB.

9. The intelligent recognition system for ultrashort wave signals with low signal-to-noise ratio according to claim 8, characterized in that: The training module of the deep convolutional neural network sends the pre-processed training data and test data to the input layer in the form of images, extracts features in the high-order spectrum through deep separable convolution, classifies the ultrashort wave signal according to the extracted features, and finally outputs the classification results through the Softmax layer.

10. The intelligent recognition system for ultrashort wave signals with low signal-to-noise ratio according to claim 9, characterized in that: The verification module of the deep convolutional neural network sends the verification data to the trained network model; loads the trained network model; and outputs the result and accuracy of the verification data.