Microwave photon radar multi-band signal fusion method based on machine learning

Abstract

The invention provides a microwave photon radar multi-band signal fusion method based on machine learning. The method comprises the specific steps of multi-band signal generation and acquisition, signal preprocessing and tagging, deep learning model construction and training, and fusion signal imaging processing. And a self-attention mechanism and a deep learning model of a U-Net architecture are adopted, so that the precision and calculation efficiency of signal fusion are effectively improved. According to simulation and experiment verification, the microwave photon radar multi-band signal fusion method based on machine learning has remarkable advantages in the aspects of target detection and high-resolution imaging, the dependence of the radar on continuous wide spectrum resources can be reduced, and an efficient and reliable solution is provided for radar application in a complex environment.

Description

Technical Field

[0001] The present invention belongs to the technical field of radar signal processing, and in particular relates to a microwave photon radar multi-band signal fusion method based on machine learning. Background Art

[0002] The widespread application of radar technology in modern military, aerospace, and civilian fields has placed higher demands on its detection accuracy, resolution, and environmental adaptability. Particularly in complex environments, the resolution and anti-interference capabilities of traditional radar systems often fall short of these requirements. In recent years, microwave photon radar technology has garnered widespread attention due to its wideband signal support, high-resolution imaging, and anti-interference capabilities, becoming a key area of ​​radar technology development.

[0003] Microwave photon imaging radars based on stepped-frequency signals can significantly improve the range resolution of radar systems by continuously scanning across multiple frequency bands and employing synthetic broadband methods. For example, in microwave photon radars operating in the Ka band and below, synthetic broadband technology enables sub-centimeter range resolution. This technological breakthrough is crucial for high-precision target detection and imaging, particularly in identifying small targets and their boundary features. However, when using stepped-frequency signals with no gaps in frequency, the resulting large synthesized spectrum is highly susceptible to external interference. For example, strong interference in complex electromagnetic environments can cause signal distortion or frequency band loss, thus compromising imaging quality. Traditional broadband signal processing methods struggle to maintain stability in interference environments. Machine learning techniques, particularly deep learning-based feature extraction and fusion models, offer inherent advantages in processing complex signals and nonlinear features. By introducing deep learning methods, efficient feature extraction and fusion of multi-band radar signals can be achieved, effectively addressing the shortcomings of traditional methods. Based on the combination of stepped frequency signals and spectrum recognition technology, this paper makes full use of machine learning methods and proposes a multi-band signal fusion method for microwave photon radar based on machine learning. The purpose is to address the limitations of traditional methods in adaptability, signal fusion accuracy and computational efficiency in interference environments, and realize high-resolution, interference-resistant radar imaging and target recognition. Summary of the Invention

[0004] The purpose of the present invention is to provide a multi-band signal fusion method for microwave photon radar based on machine learning, aiming to address the shortcomings of existing microwave photon radar technology in high-resolution detection and adaptation to complex electromagnetic environments. By utilizing machine learning technology, adaptive fusion of multiple discontinuous frequency band signals is achieved, thereby improving the resolution and anti-interference capability of the radar system and reducing the dependence on broadband continuous spectrum.

[0005] To achieve the above objectives, the present invention provides a method for multi-band signal fusion of microwave photon radar based on machine learning, comprising the following steps:

[0006] Step 1: Multi-band signal generation and acquisition: Use stepped frequency signal and sparse frequency band signal technology to generate multi-band signals, eliminate interfered frequency bands through spectrum sensing, and use opportunistic frequency bands for radar signal detection and data acquisition;

[0007] Step 2: Signal preprocessing and labeling: denoising, time synchronization, and frequency band normalization are performed on the collected multi-band signals, and key features of the signals are labeled;

[0008] Step 3: Deep learning model construction and training, based on the U-Net encoder-decoder architecture and combined with a deep learning model with a self-attention mechanism, is used to extract multi-band signal features and perform spectrum completion;

[0009] Step 4: Signal fusion and imaging processing: Use a deep learning model to complete the missing spectrum, perform short-time Fourier transform and pulse compression on the completed wide-spectrum signal to achieve high-resolution imaging of the target;

[0010] Step 5: Experimental verification and optimization: compare the spectral error between the completed signal and the real signal, and optimize the model parameters to improve signal fusion accuracy and processing efficiency.

[0011] Preferably, the multi-band signal generation includes:

[0012] Determine the target detection scenario and select the appropriate signal type; generate simulation signals based on the carrier frequency range, bandwidth, and pulse repetition frequency; introduce interference models into the simulation signals to simulate complex electromagnetic environments.

[0013] Preferably, the signal preprocessing includes:

[0014] Perform time alignment and denoising on the collected multi-band signals; generate multi-dimensional labels based on target features and signal characteristics; and perform discrete encoding processing on continuous numerical labels.

[0015] Preferably, the self-attention mechanism of the deep learning model generates multi-head attention features by calculating the relevance of each position to other positions in the input sequence to enhance feature extraction capabilities.

[0016] Preferably, the signal fusion step comprises:

[0017] The missing frequency bands are completed through a deep learning model; the main frequency components are extracted by short-time Fourier transform of the completed broadband signal; and the pulse compression method is used to extract the distance information of the target and generate imaging data.

[0018] Preferably, the experimental verification and optimization steps use indicators such as spectrum reconstruction error and target resolution to evaluate model performance, and use data enhancement technology to expand the training data set to improve the adaptability of the model.

[0019] Therefore, the present invention employs the aforementioned machine learning-based microwave photon radar multi-band signal fusion method to address the shortcomings of existing microwave photon radar technology in high-resolution detection and adaptability to complex electromagnetic environments. By leveraging machine learning, it achieves adaptive fusion of signals from multiple discontinuous frequency bands, improving the radar system's resolution and anti-interference capabilities while reducing its reliance on wideband continuous spectrum. This method offers significant advantages in target detection and high-resolution imaging, while also reducing the radar's reliance on continuous wideband spectrum resources, providing an efficient and reliable solution for radar applications in complex environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 This is a flow chart of a multi-band signal fusion method for microwave photon radar based on machine learning in the present invention;

[0021] Figure 2 For deep learning models. DETAILED DESCRIPTION

[0022] The technical solution of the present invention is further described below with reference to the accompanying drawings and embodiments.

[0023] Unless otherwise defined, technical or scientific terms used in the present invention shall have the same meaning as commonly understood by one of ordinary skill in the art to which the present invention belongs.

[0024] Example 1

[0025] like Figure 1 and Figure 2 As shown in FIG, a multi-band signal fusion method for microwave photon radar based on machine learning specifically includes the following steps:

[0026] (1) Multi-band signal generation and acquisition. Use stepped frequency signal and sparse frequency band signal technology to generate multi-band signals. Combined with spectrum recognition technology to detect the electromagnetic environment, eliminate the interfered frequency bands, and only use the opportunistic frequency bands for radar signal detection and data acquisition.

[0027] Determine the target detection scenario and select the appropriate radar signal type (such as stepped frequency signal, linear frequency modulation signal, or sparse frequency band signal); set signal parameters according to actual application requirements, including carrier frequency range, stepped frequency, bandwidth, pulse repetition frequency (PRF), and signal duration.

[0028] Use MATLAB to generate simulation signals with different frequency distributions, set interference models, and simulate interference signals in complex electromagnetic environments, including in-band interference and broadband interference.

[0029] Add signal characteristic parameters (such as target distance, speed, radar cross-section, etc.) to generate echo signals, set the signal-to-noise ratio (SNR) range, add Gaussian white noise or a noise model commonly used in actual observations, and obtain a simulated signal.

[0030] (2) Signal preprocessing and labeling. The collected multi-band signals are preprocessed, including signal denoising, time synchronization, and frequency band normalization. The key features of the signals (such as carrier frequency, bandwidth, and signal-to-noise ratio) are labeled for subsequent machine learning model training.

[0031] Signal preprocessing: Check the collected signal data, remove strongly interfered or incomplete signal segments, time-align multi-band signals, use synchronization flags or reference signals in the signal for timing correction, and randomly crop, interpolate, or add noise to the signal to expand data diversity and enhance model robustness.

[0032] Signal tags are designed based on the signal's physical characteristics and the actual detection requirements. Common tags include target characteristics: distance, speed, and direction; signal characteristics: carrier frequency, bandwidth, pulse width, and signal-to-noise ratio (SNR); and environmental information: interference intensity, background noise level, and target motion status.

[0033] Label encoding: Quantize target feature labels, discretizing continuous values ​​(such as target distance and speed) into discrete values ​​at set intervals. Discrete features (such as target type and interference category) are directly encoded as integers or one-hot encoding.

[0034] (3) Deep learning model construction and training, in which the network structure is constructed using the U-Net encoder-decoder architecture. Combined with the attention mechanism, it can further improve the model's feature focusing ability and information processing efficiency:

[0035] A deep learning model based on the U-Net encoder-decoder architecture is constructed. The encoder extracts multi-scale features of the input signal through multi-layer convolution operations. Each layer consists of convolution, batch normalization, and the ReLU activation function.

[0036] The decoder upsamples the multi-scale features extracted by the encoder layer by layer, while combining the encoder's skip connection (SkipConnection).

[0037] The self-attention mechanism introduces the ability to focus on specific regions or features, improving the network's efficiency in extracting key features. Combined with U-Net, an attention module can be inserted between the encoder and decoder.

[0038] The self-attention mechanism calculates the relevance of each position in the input sequence to other positions:

[0039] Self-attention formula query (Query), key (Key), value (Value) generation, multi-head self-attention enhances the model representation capability.

[0040] The calculation loss function includes signal reconstruction loss (mean square error), adversarial loss, attention regularization loss, and total loss function:

[0041] (4) Signal fusion and imaging processing.

[0042] Spectrum sensing performs spectrum analysis on collected multi-band signals to identify frequency missing areas.

[0043] The missing area is marked by setting an energy threshold T and marking the frequency segment below the threshold in the spectrum as the missing area.

[0044] Mark missing frequency regions and complete the missing frequency bands through deep learning models

[0045] Use the deep learning model in step 3 to complete the missing frequency bands.

[0046] The model is trained using complete data with true spectra as labels, and the loss function is the spectrum reconstruction error.

[0047] To enhance the model's ability to recover spectral details, sparse regularization is added.

[0048] Imaging processing: Short-time Fourier transform (STFT) analysis is performed on the completed broadband signal to extract the main frequency components and target time domain characteristics for imaging initialization.

[0049] Pulse compression: Calculate the reference signal s ref (t) and the complement signal s fused (t), extract the cross-correlation peak and obtain the distance information of the target.

[0050] Image generation: Synthetic aperture processing uses the range-Doppler method to generate a two-dimensional image. In the range dimension, a fast Fourier transform is performed on each time slice. In the Doppler dimension, a fast Fourier transform is performed on multiple time slice signals to convert the target signal into a two-dimensional plane image.

[0051] The range-Doppler method is used to generate a two-dimensional image. Fast Fourier transform (FFT) is performed on each time slice to calculate the range dimension signal. The formula is:

[0052] Then, the multi-time slice signal is subjected to FFT to obtain Doppler dimension information and synthesize a two-dimensional image.

[0053] (5) Experimental Verification and Performance Optimization: Evaluate the spectral error between the completed signal and the true signal, the minimum resolvable distance of point targets, and evaluate performance under different interference environments. Adjust the model architecture to improve completion accuracy and use data augmentation techniques to expand the diversity of the training dataset.

[0054] The spectral error between the completed signal and the real signal is evaluated through simulation experiments.

[0055] Therefore, the present invention adopts the above-mentioned microwave photon radar multi-band signal fusion method based on machine learning to realize the adaptive fusion of multiple discontinuous frequency band signals, improve the resolution and anti-interference capability of the radar system, and reduce the dependence on broadband continuous spectrum.

[0056] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit the same. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solutions of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A microwave photon radar multi-band signal fusion method based on machine learning, characterized in that: The following steps are involved: Step 1: Multi-band signal generation and acquisition: Use stepped frequency signal and sparse frequency band signal technology to generate multi-band signals, eliminate interfered frequency bands through spectrum sensing, and use opportunistic frequency bands for radar signal detection and data acquisition; Step 2: Signal preprocessing and labeling: denoising, time synchronization, and frequency band normalization are performed on the collected multi-band signals, and key features of the signals are labeled; Step 3: Deep learning model construction and training, based on the U-Net encoder-decoder architecture and combined with a deep learning model with a self-attention mechanism, is used to extract multi-band signal features and perform spectrum completion; Step 4: Signal fusion and imaging processing: Use a deep learning model to complete the missing spectrum, perform short-time Fourier transform and pulse compression on the completed wide-spectrum signal to achieve high-resolution imaging of the target; Step 5: Experimental verification and optimization: compare the spectral error between the completed signal and the real signal, and optimize the model parameters to improve signal fusion accuracy and processing efficiency.

2. The method for multi-band signal fusion of microwave photon radar based on machine learning according to claim 1, characterized in that: Multi-band signal generation, including: Determine the target detection scenario and select the appropriate signal type; generate simulation signals based on the carrier frequency range, bandwidth, and pulse repetition frequency; introduce interference models into the simulation signals to simulate complex electromagnetic environments.

3. The method for multi-band signal fusion of microwave photon radar based on machine learning according to claim 1, characterized in that: Signal preprocessing, including: Perform time alignment and denoising on the collected multi-band signals; generate multi-dimensional labels based on target features and signal characteristics; and perform discrete encoding processing on continuous numerical labels.

4. The method for multi-band signal fusion of microwave photon radar based on machine learning according to claim 1, characterized in that: The self-attention mechanism of the deep learning model generates multi-head attention features to enhance feature extraction capabilities by calculating the relevance of each position in the input sequence to other positions.

5. The method for multi-band signal fusion of microwave photon radar based on machine learning according to claim 1, characterized in that: The signal fusion steps include: The missing frequency bands are completed through a deep learning model; the main frequency components are extracted by short-time Fourier transform of the completed broadband signal; and the pulse compression method is used to extract the distance information of the target and generate imaging data.

6. The method for multi-band signal fusion of microwave photon radar based on machine learning according to claim 1, characterized in that: The experimental verification and optimization steps use indicators such as spectrum reconstruction error and target resolution to evaluate model performance, and use data augmentation technology to expand the training data set to improve the adaptability of the model.

Images