Intelligent suppression management system and method for radio frequency signal interference based on image recognition

Abstract

The invention discloses a radio frequency signal interference intelligent suppression management system and method based on image recognition. The system comprises a multi-source sensing module, a fusion analysis module, a dynamic suppression module and an execution feedback module. The multi-source sensing module forms a radio frequency original signal set containing frequency spectrum features and time domain waveforms; the fusion analysis module receives the radio frequency original signal set and extracts an interference source feature vector in the radio frequency original signal set through a pre-trained convolutional neural network to generate a fusion decision signal; the dynamic suppression module receives the fusion decision signal and activates a corresponding algorithm in a preset suppression strategy library according to the interference type identifier; and the execution feedback module receives the real-time control signal and drives the tunable filter, the phased-array antenna and the power amplifier to execute interference suppression operation. According to the intelligent suppression management system and method for the radio-frequency signal interference based on image recognition, the problems of fuzzy positioning of the radio-frequency interference source, single suppression strategy and lack of environmental perception capability can be solved.

Description

Technical Field

[0001] The present invention relates to the technical field of intersection of wireless communications and computer vision, and in particular to an intelligent radio frequency signal interference suppression management system and method based on image recognition. Background Art

[0002] With the widespread adoption of 5G communications, the Internet of Things, and drone technology, RF interference incidents in complex electromagnetic environments are increasing exponentially. Traditional interference suppression systems rely on pure RF signal processing, using spectrum analyzers to detect abnormal frequencies and employing fixed thresholds to trigger frequency-domain filtering or power suppression. These methods suffer from three fundamental flaws: First, RF signatures alone cannot distinguish between natural and malicious interference, resulting in a false alarm rate exceeding 30% for transient interference such as drone remote control signals and spark discharges from industrial equipment. Second, positioning technology based on direction of arrival estimation is affected by multipath effects, resulting in positioning errors often exceeding 15 degrees in urban environments, causing the beamforming null to deviate from the actual interference source. Third, static rule bases cannot adapt to dynamic interference scenarios. When the interference source moves or the modulation method suddenly changes, the preset suppression strategy becomes ineffective.

[0003] While recent research has attempted to incorporate artificial intelligence, most solutions have only optimized RF signal classification models and have failed to overcome the bottleneck of physical environment perception. For example, systems that use convolutional neural networks to identify modulation types, while increasing interference classification accuracy to 85%, still lack spatial localization capabilities and are unable to guide precise suppression. Other dynamic spectrum access solutions based on reinforcement learning, while capable of frequency avoidance, are unable to combat blocking broadband interference.

[0004] At its core, the existing technology is limited by its disconnection between electromagnetic and physical space, making it impossible to establish closed-loop control where "what you see is what you suppress." This invention, through the deep integration of computer vision and RF processing, achieves for the first time the machine vision recognition of the physical entity of the interference source and the real-time inversion of its spatial coordinates, providing a new path to overcome this technical bottleneck. Summary of the Invention

[0005] In view of the shortcomings of the above-mentioned prior art, the purpose of the present invention is to provide an intelligent radio frequency signal interference suppression management system and method based on image recognition, which is used to solve the problems of fuzzy positioning of radio frequency interference sources, single suppression strategy, and lack of environmental perception ability. The present invention uses a multi-source perception module to synchronously collect radio frequency signals and spatial video streams, uses a fusion analysis module to extract visual target features and radio frequency spectrum features and perform spatiotemporal alignment to generate an interference decision signal with a three-dimensional position tag; the dynamic suppression module calls the space-frequency joint algorithm to generate beamforming and filtering parameters according to the interference type, the execution feedback module evaluates the suppression effect in real time and optimizes the decision model in a closed loop, and the instruction storage module coordinates the loading of system-wide parameters to achieve accurate identification and adaptive suppression of interference sources.

[0006] The present invention provides an intelligent radio frequency signal interference suppression and management system based on image recognition, comprising: Multi-source perception module, which synchronously collects RF frequency domain signals and video stream data of the corresponding physical space to form a set of RF original signals containing spectral characteristics and time domain waveforms; The fusion analysis module receives the original RF signal set and extracts the interference source feature vectors through a pre-trained convolutional neural network. It also extracts the spectral feature vectors of the RF signal through a fast Fourier transform. The two types of feature vectors are aligned in time and space and then input into the multimodal decision model to generate a fusion decision signal. Dynamic suppression module: The dynamic suppression module receives the fusion decision signal and activates the corresponding algorithm in the preset suppression strategy library according to the interference type identification. When the confidence rating exceeds the set threshold, it generates a real-time control signal containing frequency avoidance parameters, beamforming weight matrix and power adjustment instructions; The execution feedback module receives real-time control signals and drives the tunable filter, phased array antenna and power amplifier to perform interference suppression operations. At the same time, it collects the suppressed ambient RF signals to generate feedback monitoring signals and transmits them back to the fusion analysis module.

[0007] In one embodiment of the present invention, the multi-source perception module includes a broadband RF receiving unit and a multi-spectral imaging unit. The broadband RF receiving unit captures the time domain waveform containing interference characteristics in the target frequency band through a tunable local oscillator circuit and generates a set of original RF signals. The multi-spectral imaging unit uses a mechanical gimbal equipped with a high-definition camera to actively track and scan the direction of the source of the RF signal, and generates a set of original visual signals containing infrared radiation characteristics through an image sensor. The two types of signal sets are time stamp aligned in a hardware synchronization manner and then transmitted to the fusion analysis module.

[0008] In one embodiment of the present invention, when the fusion analysis module performs multimodal feature fusion, it first extracts the dynamic target contour features of continuous frame images in the visual original signal set through a three-dimensional convolutional neural network, and at the same time uses wavelet packet transform to decompose the transient pulse components in the radio frequency original signal set. The extracted visual motion trajectory features and radio frequency pulse envelope features are input into the spatiotemporal registration model, and the two types of features are unified into the same geographic coordinate system through a coordinate mapping matrix to generate a fusion decision signal with a three-dimensional position label.

[0009] In one embodiment of the present invention, the dynamic suppression module includes a strategy selection engine and a parameter optimizer. The strategy selection engine calls the space-frequency joint suppression algorithm in the suppression strategy library according to the interference type identifier in the fusion decision signal. The parameter optimizer dynamically adjusts the beamforming constraints and frequency domain filtering order in the algorithm based on the confidence rating to generate a real-time control signal including the phased array antenna weighting coefficient and the center frequency parameter of the band-stop filter.

[0010] In one embodiment of the present invention, the execution feedback module is provided with an interference suppression effect evaluation unit, which constructs a feedback monitoring signal including spectrum purity index and communication quality improvement by real-time acquisition of the target signal-to-noise ratio change at the phased array antenna receiving end and the residual interference power spectrum density at the output end of the tunable filter. The signal is transmitted back to the fusion analysis module in a closed-loop form for correcting the weight parameters of the multimodal decision model.

[0011] In one embodiment of the present invention, the instruction storage module adopts a distributed architecture to store the neural network model parameters, including the multi-layer convolution kernel tensor of the visual feature extraction network, the time-frequency analysis operator of the radio frequency signal processing network, and the decision tree rule set of the multimodal decision model. The central instruction distribution signal loads the parameter subsets required by different functional modules on demand through a high-speed serial bus.

[0012] In one embodiment of the present invention, the spatiotemporal registration model uses a combined algorithm of affine transformation and perspective transformation to convert the image pixel coordinates in the original visual signal set into longitude and latitude coordinates in the geodetic coordinate system. At the same time, the angle of arrival information contained in the original RF signal set is converted into spatial vector coordinates through the RF arrival direction estimation algorithm. Finally, the error compensation alignment of the two types of coordinate systems is achieved through least squares fitting.

[0013] In one embodiment of the present invention, the joint space-frequency suppression algorithm includes a beam nulling forming unit and a dynamic spectrum sensing unit. The beam nulling forming unit calculates the directivity nulling pointing angle of the phased array antenna based on the three-dimensional position of the interference source. The dynamic spectrum sensing unit detects the bandwidth change of the interference signal in real time and generates adaptive notch filter parameters. The two work together to generate composite control instructions for suppressing spatial interference and frequency domain interference.

[0014] In one embodiment of the present invention, the interference suppression effect evaluation unit constructs a composite evaluation index including the time domain correlation coefficient and the frequency domain mutual information entropy, and quantifies the spectrum purity scale value in the feedback monitoring signal by comparing the decrease rate of the time domain mutual correlation coefficient and the change in the frequency domain power spectrum mutual information entropy between the target signal and the interference signal before and after suppression.

[0015] The present invention also provides a method for intelligently suppressing and managing radio frequency signal interference based on image recognition, comprising: S1: Synchronously collect RF frequency domain signals and video stream data of the corresponding physical space to form a set of RF original signals containing spectrum characteristics and time domain waveforms; S2: Receives the original RF signal set and extracts the interference source feature vectors using a pre-trained convolutional neural network. It also extracts the RF signal's spectrum feature vectors using a fast Fourier transform. These two feature vectors are then spatially and temporally aligned and fed into a multimodal decision model to generate a fused decision signal. S3: Receives the fusion decision signal and activates the corresponding algorithm in the preset suppression strategy library based on the interference type identifier. When the confidence rating exceeds the set threshold, it generates a real-time control signal containing frequency avoidance parameters, beamforming weight matrix, and power adjustment instructions. S4: Receive the real-time control signal and drive the tunable filter, phased array antenna and power amplifier to perform interference suppression operations, while collecting the suppressed ambient RF signal to generate a feedback monitoring signal and transmit it back to step S2.

[0016] The image recognition-based RF signal interference intelligent suppression management system and method provided by the present invention synchronously collects RF signals and spatial video streams through a multi-source perception module, uses a fusion analysis module to extract visual target features and RF spectrum features and perform spatiotemporal alignment to generate an interference decision signal with a three-dimensional position tag; the dynamic suppression module calls the space-frequency joint algorithm to generate beamforming and filtering parameters according to the interference type, the execution feedback module evaluates the suppression effect in real time and optimizes the decision model in a closed loop, and the instruction storage module coordinates the loading of system-wide parameters to achieve accurate identification and adaptive suppression of interference sources. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0018] Figure 1 This is the system architecture diagram of the RF signal interference intelligent suppression management system based on image recognition; Figure 2 A schematic diagram showing the workflow of an intelligent radio frequency signal interference suppression management system based on image recognition; Figure 3 The present invention is a flow chart of a method for intelligent suppression management of radio frequency signal interference based on image recognition. DETAILED DESCRIPTION

[0019] The following describes the embodiments of the present invention through specific examples. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments. The details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the following embodiments and features in the embodiments can be combined with each other unless they conflict.

[0020] It should be noted that the illustrations provided in the following embodiments are merely schematic illustrations of the basic concept of the present invention. Therefore, the illustrations only show components related to the present invention and are not drawn according to the number, shape, and size of components in actual implementation. In actual implementation, the type, quantity, and proportion of each component may be changed arbitrarily, and the component layout may also be more complex.

[0021] In the following description, numerous details are discussed to provide a more thorough explanation of the embodiments of the present invention. However, it will be apparent to those skilled in the art that the embodiments of the present invention may be practiced without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring the embodiments of the present invention.

[0022] See Figure 1-3 , which shows the image recognition-based intelligent radio frequency signal interference suppression management system and method of the present invention. The image recognition-based intelligent radio frequency signal interference suppression management system of the present invention includes a multi-source perception module, a fusion analysis module, a dynamic suppression module, and an execution feedback module. The multi-source perception module synchronously collects radio frequency frequency domain signals and video stream data of the corresponding physical space to form a set of original radio frequency signals containing spectral features and time domain waveforms. The fusion analysis module receives the original radio frequency signal set and extracts the interference source feature vectors therein through a pre-trained convolutional neural network. It also extracts the spectral feature vectors of the radio frequency signal through a fast Fourier transform. The two types of feature vectors are aligned in time and space and then input into a multimodal decision model to generate a fusion decision signal. The dynamic suppression module receives the fusion decision signal and activates the corresponding algorithm in the preset suppression strategy library based on the interference type identifier. When the confidence rating exceeds the set threshold, it generates a real-time control signal containing frequency avoidance parameters, beamforming weight matrix, and power adjustment instructions. The execution feedback module receives the real-time control signal and drives the tunable filter, phased array antenna, and power amplifier to perform interference suppression operations. At the same time, it collects the suppressed ambient radio frequency signal to generate a feedback monitoring signal and transmits it back to the fusion analysis module.

[0023] like Figure 1As shown in the figure, the multi-source perception module consists of a hardware collaborative acquisition system consisting of a broadband RF receiving unit and a multispectral imaging unit. The broadband RF receiving unit is implemented using a software-defined radio architecture. Its core is a tunable local oscillator circuit and a high-speed analog-to-digital converter chain. It captures the instantaneous time-domain waveform containing interference characteristics within the target frequency band through dynamic reconstruction of the local oscillator frequency. Simultaneously, it uses digital down-conversion technology to generate baseband orthogonal components, ultimately outputting a set of raw RF signals containing a complex time-domain sampling sequence and a spectrum amplitude spectrum. The multispectral imaging unit consists of a mechanical gimbal, a visible-light and infrared dual-channel camera, and an image coprocessor. The mechanical gimbal drives the camera toward the signal source area based on the estimated RF direction of arrival. The visible light channel captures a high-resolution RGB image sequence, and the infrared channel simultaneously captures thermal radiation distribution maps. The image coprocessor performs non-uniformity correction and dynamic range compression to generate a set of raw visual signals that fuses visible light texture and infrared radiation characteristics. The two signal sets are spatially and temporally aligned through a hardware synchronization mechanism. Specifically, a high-precision clock source trained by the Global Positioning System (GPS) generates a unified timestamp. Both the RF sampling pulse and the camera exposure signal are triggered by this clock source, ensuring strict time synchronization between the RF signal and the visual image at any given moment. The timestamp is then embedded in the signal packet header and transmitted to the fusion analysis module. The wideband RF receiving unit also features an anti-saturation protection circuit. When a strong interference signal is detected, a digitally controlled attenuator chain is automatically activated to prevent receiver blockage. The attenuation value is dynamically adjusted based on the interference power and written as metadata to the original RF signal set.

[0024] Furthermore, the fusion analysis module uses a cascaded processing flow to perform multimodal feature fusion. The visual processing chain first inputs the original visual signal set into a three-dimensional convolutional neural network, which contains a spatiotemporal separation convolution layer and a long-short-term memory gating mechanism. The network extracts the motion trajectory characteristics and shape contour feature vectors of the dynamic target through a sequence of ten consecutive image frames. The motion trajectory characteristics include the displacement vector and acceleration estimate of the target in the image plane, while the shape contour features include the target edge gradient histogram and regional invariant moment descriptor. The RF processing chain also launches the wavelet packet transform engine in parallel, performing a four-layer decomposition of the time domain waveform of the original RF signal set, extracting the envelope morphology characteristics and sub-band energy distribution characteristics of the transient pulse component. The envelope morphology characteristics include the pulse rise time, fall time, and quantized overshoot value, while the sub-band energy distribution characteristics record the normalized power ratio of the wavelet coefficients in each frequency band. After receiving the above-mentioned feature vectors, the spatiotemporal registration model starts the coordinate mapping process: in the visual coordinate system conversion stage, the Zhang Zhengyou calibration method is used to calculate the camera intrinsic parameter matrix. Combined with the gimbal pitch angle and azimuth angle data, the image pixel coordinates are converted into three-dimensional coordinates in a local rectangular coordinate system with the device as the origin through perspective projection transformation. In the radio frequency coordinate system conversion stage, a multiple signal classification algorithm is used to calculate the interference signal arrival angle, and the spatial vector direction is inverted based on the geometric structure of the phased array antenna array. Finally, the spatial mapping relationship between the two types of coordinate systems is established through least squares fitting. The Lie group SE(3) transformation matrix is used to unify the radio frequency space vector and the visual three-dimensional coordinates into the geodetic coordinate system, and a fusion decision signal with latitude, longitude, elevation labels and interference type confidence is output. The confidence calculation uses a fuzzy logic system to comprehensively evaluate the feature matching degree and coordinate residuals.

[0025] Specifically, the dynamic suppression module includes an adaptive control loop consisting of a strategy selection engine and a parameter optimizer. After receiving the fusion decision signal, the strategy selection engine retrieves a space-frequency joint suppression algorithm template from the suppression strategy library based on the interference type identifier. If narrowband communication interference is identified, the frequency-domain suppression template is invoked, activating an adaptive notch filter based on the minimum mean square error criterion. If broadband blocking interference is identified, the spatial-domain suppression template is invoked, activating a beamformer based on the linearly constrained minimum variance criterion. If frequency-hopping interference is identified, the space-frequency joint template is invoked, initiating coordinated control of beam nulling and dynamic spectrum sensing. The parameter optimizer dynamically adjusts algorithm parameters based on the confidence rating. When the confidence rating is above the first threshold, a strict constraint mode is used, with the beamforming constraint set to a gain of no more than -30 decibels in the direction of the interference source and an eighth-order elliptic function filter. When the confidence rating is below the first threshold but above the second threshold, a loose constraint mode is activated, with the beamforming constraint relaxed to -20 decibels and the filter order reduced to a fourth-order Chebyshev filter. When the confidence rating falls below the second threshold, the system switches to monitoring mode, recording interference signatures without performing active mitigation. The real-time control signal output by the parameter optimizer includes the phased array antenna weighting matrix and a set of bandstop filter parameters. The weighting matrix consists of complex values, the real part of which controls the antenna element excitation amplitude and the imaginary part controls the phase delay, creating a deep null in the direction of the interference source by changing the array pattern. The bandstop filter parameter set includes three core parameters: center frequency, stopband width, and attenuation depth. The center frequency is dynamically set based on the centroid of the interference signal spectrum, the stopband width is configured as three times the interference signal bandwidth, and the attenuation depth is proportional to the interference power. The module also has a strategy effectiveness prediction unit, which uses Monte Carlo simulation to evaluate the degree of improvement in the system bit error rate under different parameter combinations, and gives priority to parameter solutions that reduce the bit error rate by more than 50%.

[0026] In one embodiment of the present invention, the execution feedback module constructs a closed-loop verification mechanism through an interference suppression effect evaluation unit, which includes a target signal analysis channel and a residual interference monitoring channel. The target signal analysis channel is connected to the baseband processor at the receiving end of the phased array antenna, demodulating the original symbol stream of the communication link in real time. By calculating the mean Euclidean distance of the decision error vector within 100 consecutive symbol periods, it derives the actual change in the target signal-to-noise ratio. The residual interference monitoring channel is connected to the power detection circuit at the output of the tunable filter, using a sliding window fast Fourier transform to generate an instantaneous power spectrum density diagram, extracting the ratio of the power integral value within the interference mainlobe frequency band to the background noise power in the sidelobe area. The above-mentioned dual-channel data input effect fusion processor constructs a feedback monitoring signal including a spectrum purity index and a communication quality improvement degree. The spectrum purity index is defined as the normalized difference between the residual interference power and the target signal power. Its calculation requires compensation for the gain fluctuation introduced by the change in the antenna pattern. The communication quality improvement degree is obtained by comparing the area under the bit error rate curve before and after suppression. The feedback monitoring signal is transmitted back to the feature matching engine of the fusion analysis module through the Gigabit Ethernet interface. The engine performs a correlation analysis on the actual suppression effect and the expected value. If the deviation of the spectrum purity index exceeds the set tolerance range, the weight parameter correction process of the multimodal decision model is triggered. Specifically, the stochastic gradient descent method is used to adjust the kernel weight coefficient of the spatiotemporal separation convolution layer in the three-dimensional convolutional neural network, so that the subsequent fusion decision signal is more in line with the actual interference scenario.

[0027] like Figure 2As shown, the instruction storage module utilizes non-volatile memory and dynamic random access memory (DRAM) to construct a two-level distributed storage architecture. The non-volatile memory utilizes three-dimensional stacked flash memory chips to store the basic parameter library, including the convolution kernel tensor set for the visual feature extraction network, the wavelet packet decomposition tree structure for the RF signal processing network, and the decision tree rule set for the multimodal decision model. The convolution kernel tensor set is stored in blocks by network level, with error checking code appended to each tensor block to ensure data integrity. The dynamic random access memory (DRAM) serves as a high-speed cache, storing dynamic parameters generated during system operation, including the real-time updated beamforming weight matrix, adaptive filter coefficient sequence, and the current interference source trajectory prediction state vector. Centralized command distribution signals schedule data flow via the PCIe high-speed serial bus. When the fusion analysis module initiates feature extraction, the bus controller loads the convolution kernel tensor sub-set for the corresponding network layer from the non-volatile memory into the designated address space of the dynamic random access memory on demand. When the dynamic suppression module generates a real-time control signal, the bus prioritizes transmitting the latest beamforming weight matrix from the dynamic random access memory to the phased array antenna controller. The data scheduling process uses a memory mapping mechanism to translate physical addresses into virtual addresses. A direct memory access controller manages the data transfer path, ensuring that parameter loading delays do not exceed one-fifth of the system sampling period. The non-volatile memory features a wear-leveling algorithm that dynamically adjusts the write frequency of flash memory blocks to extend the life of the storage media. Its erase count status is embedded as a health indicator in the metadata field of the central command distribution signal. The error compensation alignment of the spatiotemporal registration model uses a sliding window mechanism to handle dynamic targets. Its core is a trajectory prediction-correction system based on a Kalman filter. The prediction phase uses the target displacement vector extracted from the raw visual signal set to establish a state equation. The system state vector is defined as the target's position coordinates, velocity components, and acceleration components in the geodetic coordinate system. A time-recursive equation is used to estimate the trajectory coordinate sequence for the next three frames. The correction phase integrates RF time difference of arrival measurements: three non-coplanar antenna subarrays are selected from the phased array antenna array. The precise time difference between the arrival of the same interference signal at each subarray is measured. The three-dimensional spatial distance difference of the target relative to the antenna array is inverted using the electromagnetic wave propagation velocity. This generates an overdetermined set of equations to solve the observed value of the target's current position. The predicted coordinates and observed values are input into the Kalman gain calculation unit, which dynamically adjusts the weight coefficients based on the prediction covariance matrix and the observed noise variance, ultimately outputting the optimal estimated coordinates after error compensation. A sliding window mechanism maintains a fixed-length historical coordinate sequence, extracting trajectory curvature features through polynomial fitting. When the prediction residuals exceed a threshold for five consecutive times, the process noise covariance is automatically increased to enhance system sensitivity. The compensated coordinate sequence is appended with a timestamp index and written into the fusion decision signal. Its spatial position accuracy is determined by the joint Cramer-Rao lower bound of the visual positioning error and the radio frequency ranging error, achieving centimeter-level positioning accuracy in typical application scenarios.

[0028] Furthermore, the joint space-frequency suppression algorithm is implemented through the hardware collaboration of the beam nulling unit and the dynamic spectrum sensing unit. The beam nulling unit receives the three-dimensional coordinates of the interference source in the fusion decision signal, calculates the array manifold vector based on the geometric configuration of the phased array antenna array, and uses the linear constrained minimum variance algorithm to solve the optimal weighting coefficient. The specific process is: construct an equality constraint matrix with zero gain in the direction of the interference source, while keeping the gain in the direction of the target signal at unity value. The Lagrange multiplier method is used to solve the set of complex weight coefficients that minimize the array output power. Its mathematical essence is the generalized inverse operation of the covariance matrix. The dynamic spectrum sensing unit monitors the intermediate frequency output signal of the RF receiver in parallel, divides the target frequency band into 32 subchannels through a complex modulated filter group, detects the mutation points of the power spectrum of each subchannel in real time, and dynamically generates the band-stop filter parameters according to the spectrum diffusion range when the interference signal bandwidth changes. The two units exchange information through a collaborative controller: the beam nulling unit transmits the nulling angle range to the dynamic spectrum sensing unit, which adjusts the subchannel monitoring density accordingly, increasing the subchannel division density fourfold within the frequency band corresponding to the nulling direction. The interference bandwidth information identified by the dynamic spectrum sensing unit is fed back to the beam nulling unit, triggering frequency-dependent compensation of the array weighting coefficients. The resulting composite control command consists of a spatial suppression component, which is reflected in the real amplitude distribution and imaginary phase gradient of the antenna weighting coefficient matrix, and a frequency suppression component, which is reflected in the list of stopband center frequencies and the set of roll-off coefficients of the tunable filter.

[0029] like Figure 3 As shown, the present invention is an intelligent suppression management method for radio frequency signal interference based on image recognition. S1: Synchronously collect radio frequency frequency domain signals and video stream data of corresponding physical spaces to form a radio frequency original signal set containing spectrum features and time domain waveforms; S2: Receive the radio frequency original signal set and extract the interference source feature vector therein through a pre-trained convolutional neural network, and at the same time extract the spectrum feature vector of the radio frequency signal through fast Fourier transform, align the two types of feature vectors in time and space, and input them into the multimodal decision model to generate a fusion decision signal; S3: Receive the fusion decision signal, activate the corresponding algorithm in the preset suppression strategy library according to the interference type identifier, and generate a real-time control signal containing frequency avoidance parameters, beamforming weight matrix and power adjustment instructions when the confidence rating exceeds the set threshold; S4: Receive the real-time control signal and drive the tunable filter, phased array antenna and power amplifier to perform interference suppression operations, and at the same time collect the suppressed environmental radio frequency signal to generate a feedback monitoring signal back to step S2.

[0030] The image-recognition-based RF signal interference intelligent suppression management system and method of the present invention synchronously collects RF signals and spatial video streams through a multi-source perception module, utilizes a fusion analysis module to extract visual target features and RF spectrum features and perform spatiotemporal alignment to generate an interference decision signal with a three-dimensional position tag; the dynamic suppression module calls the space-frequency joint algorithm to generate beamforming and filtering parameters according to the interference type; the execution feedback module evaluates the suppression effect in real time and optimizes the decision model in a closed loop; the instruction storage module coordinates the loading of system-wide parameters to achieve accurate identification and adaptive suppression of interference sources.

[0031] Therefore, the image recognition-based RF signal interference intelligent suppression management system and method of the present invention solves the problems of fuzzy RF interference source positioning, single suppression strategy, and lack of environmental perception ability. The above embodiments are only illustrative of the principles and effects of the present invention, and are not used to limit the present invention. Anyone familiar with this technology can modify or change the above embodiments without violating the spirit and scope of the present invention. Therefore, all equivalent modifications or changes made by those with ordinary knowledge in the technical field without departing from the spirit and technical ideas disclosed by the present invention should still be covered by the claims of the present invention.

Claims

1. The radio frequency signal interference intelligent suppression and management system based on image recognition is characterized by: include: A multi-source sensing module, which synchronously collects radio frequency frequency domain signals and video stream data of corresponding physical spaces to form a set of radio frequency original signals including spectral characteristics and time domain waveforms; A fusion analysis module receives the original RF signal set and extracts interference source feature vectors therefrom through a pre-trained convolutional neural network. It also extracts spectrum feature vectors of the RF signal through a fast Fourier transform (FFT). The two feature vectors are spatially and temporally aligned and then input into a multimodal decision model to generate a fusion decision signal. a dynamic suppression module, which receives the fusion decision signal, activates a corresponding algorithm in a preset suppression strategy library according to the interference type identifier, and generates a real-time control signal including frequency avoidance parameters, a beamforming weight matrix, and a power adjustment instruction when the confidence rating exceeds a set threshold; An execution feedback module receives the real-time control signal and drives the tunable filter, phased array antenna and power amplifier to perform interference suppression operations, while collecting the suppressed ambient RF signal to generate a feedback monitoring signal and transmit it back to the fusion analysis module.

2. The radio frequency signal interference intelligent suppression and management system based on image recognition according to claim 1 is characterized in that: The multi-source perception module includes a broadband RF receiving unit and a multispectral imaging unit. The broadband RF receiving unit captures the time domain waveform containing interference characteristics in the target frequency band through a tunable local oscillator circuit and generates a set of original RF signals. The multispectral imaging unit uses a mechanical pan-tilt platform equipped with a high-definition camera to actively track and scan the direction of the RF signal source, and generates a set of visual original signals containing infrared radiation characteristics through an image sensor. The two types of signal sets are time-stamp aligned in a hardware synchronization manner and then transmitted to the fusion analysis module.

3. The radio frequency signal interference intelligent suppression and management system based on image recognition according to claim 1 is characterized in that: When the fusion analysis module performs multimodal feature fusion, it first extracts the dynamic target contour features of continuous frame images in the visual original signal set through a three-dimensional convolutional neural network, and at the same time uses wavelet packet transform to decompose the transient pulse components in the radio frequency original signal set. The extracted visual motion trajectory features and radio frequency pulse envelope features are input into the spatiotemporal registration model, and the two types of features are unified into the same geographic coordinate system through a coordinate mapping matrix to generate a fusion decision signal with a three-dimensional position label.

4. The radio frequency signal interference intelligent suppression and management system based on image recognition according to claim 1, characterized in that: The dynamic suppression module includes a strategy selection engine and a parameter optimizer. The strategy selection engine calls the space-frequency joint suppression algorithm in the suppression strategy library according to the interference type identifier in the fusion decision signal. The parameter optimizer dynamically adjusts the beamforming constraints and frequency domain filtering order in the algorithm based on the confidence rating to generate a real-time control signal including the phased array antenna weighting coefficient and the center frequency parameter of the band-stop filter.

5. The radio frequency signal interference intelligent suppression and management system based on image recognition according to claim 1 is characterized in that: The execution feedback module is provided with an interference suppression effect evaluation unit, which constructs a feedback monitoring signal including spectrum purity index and communication quality improvement by real-time acquisition of the target signal-to-noise ratio change at the phased array antenna receiving end and the residual interference power spectrum density at the output end of the tunable filter. The signal is transmitted back to the fusion analysis module in a closed-loop form for correcting the weight parameters of the multimodal decision model.

6. The radio frequency signal interference intelligent suppression and management system based on image recognition according to claim 1, characterized in that: The instruction storage module uses a distributed architecture to store neural network model parameters, including the multi-layer convolution kernel tensor of the visual feature extraction network, the time-frequency analysis operator of the radio frequency signal processing network, and the decision tree rule set of the multimodal decision model. The central instruction distribution signal loads the parameter subsets required by different functional modules on demand through a high-speed serial bus.

7. The radio frequency signal interference intelligent suppression and management system based on image recognition according to claim 3 is characterized in that: The spatiotemporal registration model adopts a combined algorithm of affine transformation and perspective transformation to convert the image pixel coordinates in the original visual signal set into longitude and latitude coordinates in the geodetic coordinate system. At the same time, the angle of arrival information contained in the original radio frequency signal set is converted into spatial vector coordinates through the radio frequency arrival direction estimation algorithm. Finally, the error compensation alignment of the two types of coordinate systems is achieved through least squares fitting.

8. The radio frequency signal interference intelligent suppression and management system based on image recognition according to claim 4, characterized in that: The joint space-frequency suppression algorithm includes a beam nulling unit and a dynamic spectrum sensing unit. The beam nulling unit calculates the nulling pointing angle of the phased array antenna according to the three-dimensional position of the interference source. The dynamic spectrum sensing unit detects the bandwidth changes of the interference signal in real time and generates adaptive notch filter parameters. The two work together to generate composite control instructions for suppressing spatial interference and frequency domain interference.

9. The radio frequency signal interference intelligent suppression and management system based on image recognition according to claim 5, characterized in that: The interference suppression effect evaluation unit constructs a composite evaluation index including the time domain correlation coefficient and the frequency domain mutual information entropy, and quantifies the spectrum purity scale value in the feedback monitoring signal by comparing the decrease rate of the time domain mutual correlation coefficient of the target signal and the interference signal before and after suppression and the change in the frequency domain power spectrum mutual information entropy.

10. The method for intelligently suppressing and managing radio frequency signal interference based on image recognition according to the radio frequency signal interference intelligent suppression and management system based on image recognition according to claims 1 to 9, comprising: S1: Synchronously collect RF frequency domain signals and video stream data of the corresponding physical space to form a set of RF original signals containing spectrum characteristics and time domain waveforms; S2: Receive the original RF signal set and extract the interference source feature vector therein through a pre-trained convolutional neural network. Simultaneously, extract the spectrum feature vector of the RF signal through a fast Fourier transform. The two types of feature vectors are spatially and temporally aligned and then input into a multimodal decision model to generate a fusion decision signal. S3: receiving the fusion decision signal, activating a corresponding algorithm in a preset suppression strategy library according to the interference type identifier, and generating a real-time control signal including frequency avoidance parameters, beamforming weight matrix, and power adjustment instructions when the confidence rating exceeds a set threshold; S4: Receive the real-time control signal and drive the tunable filter, phased array antenna and power amplifier to perform interference suppression operations, and at the same time collect the suppressed ambient RF signal to generate a feedback monitoring signal and transmit it back to step S2.

Images