Interference source identification method and system based on vehicle-mounted mobile monitoring data analysis

Abstract

The application provides a method and system for identifying interference sources based on vehicle-mounted mobile monitoring data analysis, and relates to the technical field of data processing.In the application, first, first monitoring time domain data corresponding to first monitoring frequency domain data and second monitoring time domain data corresponding to second monitoring frequency domain data are determined; then, based on the coded semantic information in the first monitoring frequency domain data, the coded semantic information in the first monitoring time domain data is subjected to semantic constraint to form first monitoring constraint features; further, based on the coded semantic information in the second monitoring frequency domain data, the coded semantic information in the second monitoring time domain data is subjected to semantic constraint to form second monitoring constraint features; finally, the fusion features of the first monitoring constraint features and the second monitoring constraint features are subjected to semantic decoding to form target interference source identification results.Based on the above, the problem of relatively low reliability of interference source identification in the prior art can be improved.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and more specifically, to a method and system for identifying interference sources based on vehicle-mounted mobile monitoring data analysis. Background Technology

[0002] With the rapid development of wireless communication technology in modern society, radio interference has gradually become a significant challenge affecting communication quality, navigation and positioning, and security monitoring. In complex urban environments, traditional radio interference identification technologies have many limitations, and conventional methods often fail to meet the demands for efficient and accurate identification. Identifying targets, interference, or illegal radio signal sources is a core task in the field of radio monitoring. However, current technologies primarily rely on comparing and matching the salient features of the main signal with those of the interference signal sample to determine whether it is an interference signal. This results in relatively low reliability in identification. Summary of the Invention

[0003] In view of this, the purpose of this application is to provide a method and system for interference source identification based on vehicle-mounted mobile monitoring data analysis, so as to improve the problem of relatively low reliability of interference source identification in the prior art.

[0004] To achieve the above objectives, this application adopts the following technical solution:

[0005] An interference source identification method based on vehicle-mounted mobile monitoring data analysis includes:

[0006] First monitoring time domain data and second monitoring time domain data are extracted from the vehicle mobile monitoring data obtained from the vehicle monitoring equipment. The first monitoring time domain data and the second monitoring time domain data are used to reflect the time domain information of the target signal at two different locations.

[0007] Determine the first monitoring frequency domain data corresponding to the first monitoring time domain data and the second monitoring frequency domain data corresponding to the second monitoring time domain data, respectively;

[0008] Based on the encoded semantic information in the first monitoring frequency domain data, semantic constraints are applied to the encoded semantic information in the first monitoring time domain data to form a first monitoring constraint feature.

[0009] Based on the encoded semantic information in the second monitoring frequency domain data, semantic constraints are applied to the encoded semantic information in the second monitoring time domain data to form a second monitoring constraint feature.

[0010] The fused features of the first monitoring constraint feature and the second monitoring constraint feature are semantically decoded to form a target interference source identification result, wherein the target interference source identification result is used to reflect whether the target signal belongs to an interference source.

[0011] In a preferred embodiment of this application, in the above-described interference source identification method based on vehicle-mounted mobile monitoring data analysis, the step of semantically constraining the coded semantic information in the first monitoring time domain data based on the coded semantic information in the first monitoring frequency domain data to form a first monitoring constraint feature includes:

[0012] Convolutional mining is performed on the first monitoring frequency domain data to form the first monitoring frequency domain features, and convolutional mining is performed on the first monitoring time domain data to form the first monitoring time domain features;

[0013] The first monitoring frequency domain feature is enhanced to form a first monitoring enhancement feature;

[0014] Based on the first monitoring enhancement feature, semantic constraints are applied to the first monitoring time-domain feature to form the first monitoring constraint feature.

[0015] In a preferred embodiment of this application, in the above-described interference source identification method based on vehicle-mounted mobile monitoring data analysis, the step of enhancing the first monitoring frequency domain features to form a first enhanced monitoring feature includes:

[0016] For each frequency point in the first monitoring amplitude spectrum included in the first monitoring frequency domain data, it is determined whether the amplitude of that frequency point jumps, and the data characterizing whether the amplitude of each frequency point jumps is subjected to convolution mining to form amplitude jump features.

[0017] For each frequency point in the first monitoring phase spectrum included in the first monitoring frequency domain data, it is determined whether the phase of that frequency point has jumped, and the data characterizing whether the phase of each frequency point has jumped is subjected to convolution mining to form phase jump features.

[0018] Based on the amplitude jump feature and the phase jump feature respectively, the first monitoring amplitude feature and the first monitoring phase feature included in the first monitoring frequency domain feature are focused and mined to form the first amplitude focus feature and the first phase focus feature;

[0019] Based on the first amplitude focusing feature and the first phase focusing feature, the first monitoring enhancement feature is determined.

[0020] In a preferred embodiment of this application, in the above-described interference source identification method based on vehicle-mounted mobile monitoring data analysis, the step of semantically constraining the first monitoring time-domain features based on the first monitoring enhancement features to form first monitoring constraint features includes:

[0021] The first monitoring enhancement feature is subjected to feature space transformation to form a first monitoring transformation feature, and the first monitoring transformation feature is subjected to self-attention processing to form a monitoring attention feature;

[0022] The monitoring attention features are nonlinearly activated to form monitoring activation features;

[0023] The monitoring activation feature and the first monitoring time domain feature are multiplied together to form the first monitoring constraint feature.

[0024] In a preferred embodiment of this application, in the above-described interference source identification method based on vehicle-mounted mobile monitoring data analysis, the step of semantically constraining the coded semantic information in the second monitoring time domain data based on the coded semantic information in the second monitoring frequency domain data to form a second monitoring constraint feature includes:

[0025] Convolution mining is performed on the second monitoring frequency domain data to form the second monitoring frequency domain features, and convolution mining is performed on the second monitoring time domain data to form the second monitoring time domain features;

[0026] The second monitoring frequency domain feature is enhanced to form a second monitoring enhancement feature;

[0027] Based on the second monitoring enhancement feature, semantic constraints are applied to the second monitoring time-domain feature to form the second monitoring constraint feature.

[0028] In a preferred embodiment of this application, in the above-described interference source identification method based on vehicle-mounted mobile monitoring data analysis, the step of semantically decoding the fused features of the first monitoring constraint feature and the second monitoring constraint feature to form the target interference source identification result includes:

[0029] The first monitoring constraint feature and the second monitoring constraint feature are cross-fused to form a cross-fused monitoring feature.

[0030] During the semantic forward fusion process, based on the first monitoring constraint feature, the monitoring cross-fusion feature is subjected to deep fusion processing to form a monitoring deep fusion feature;

[0031] During the semantic backward fusion process, based on the second monitoring constraint features, the monitoring deep fusion features are subjected to deep fusion processing to form monitoring target fusion features;

[0032] The fused features of the monitored target are semantically decoded to form the target interference source identification result.

[0033] In a preferred embodiment of this application, in the above-mentioned interference source identification method based on vehicle-mounted mobile monitoring data analysis, the step of performing deep fusion processing on the monitoring cross-fusion features based on the first monitoring constraint features to form monitoring deep fusion features during the semantic forward fusion process includes:

[0034] In the first stage, the first monitoring constraint feature is subjected to self-attention processing to form the first attention feature of the first stage, and the first attention feature of the first stage is mapped to form the first gating adjustment feature of the first stage, and the monitoring cross-fusion feature is adjusted based on the first gating adjustment feature of the first stage to form the first monitoring adjustment feature of the first stage.

[0035] In the second stage, the first attention feature and the first monitoring and regulation feature of the first stage are downsampled to form the first attention sampling feature and the first monitoring sampling feature of the second stage. Based on the first attention sampling feature of the second stage, a mapping is performed to form the first gating regulation feature of the second stage. Based on the first gating regulation feature of the second stage, the first monitoring sampling feature of the second stage is regulated to form the first monitoring regulation feature of the second stage.

[0036] In each subsequent stage, the first attention sampling feature and the first monitoring regulation feature of the previous stage are downsampled to form the first attention sampling feature and the first monitoring sampling feature of the current stage. Based on the first attention sampling feature of the current stage, a mapping is performed to form the first gating regulation feature of the current stage. Based on the first gating regulation feature of the current stage, the first monitoring sampling feature of the current stage is regulated to form the first monitoring regulation feature of the current stage.

[0037] The first monitoring and regulation feature of the last stage is used as the monitoring deep fusion feature.

[0038] In a preferred embodiment of this application, in the above-mentioned interference source identification method based on vehicle-mounted mobile monitoring data analysis, the step of performing deep fusion processing on the monitoring deep fusion features based on the second monitoring constraint features to form monitoring target fusion features during the semantic backward fusion process includes:

[0039] In the first stage, the compressed features of the second monitoring constraint features are subjected to self-attention processing to form the second attention features of the first stage; and the second attention features of the first stage are mapped to form the second gating adjustment features of the first stage; and the monitoring deep fusion features are adjusted based on the second gating adjustment features of the first stage to form the second monitoring adjustment features of the first stage.

[0040] In the second stage, the second attention feature and the second monitoring and regulation feature of the first stage are upsampled to form the second attention sampling feature and the second monitoring sampling feature of the second stage. Based on the second attention sampling feature of the second stage, a mapping is performed to form the second gating regulation feature of the second stage. Based on the second gating regulation feature of the second stage, the second monitoring sampling feature of the second stage is regulated to form the second monitoring regulation feature of the second stage.

[0041] In each subsequent stage, the second attention sampling feature and the second monitoring regulation feature in the previous stage are upsampled to form the second attention sampling feature and the second monitoring sampling feature of the current stage, and the second attention sampling feature of the current stage is mapped to form the second gating regulation feature of the current stage, and the second monitoring sampling feature of the current stage is regulated based on the second gating regulation feature of the current stage to form the second monitoring regulation feature of the current stage.

[0042] The second monitoring and regulation feature of the last stage is used as the monitoring target fusion feature.

[0043] In a preferred embodiment of this application, in the above-described interference source identification method based on vehicle-mounted mobile monitoring data analysis, the step of cross-fusion processing the first monitoring constraint feature and the second monitoring constraint feature to form a cross-fusion monitoring feature includes:

[0044] Based on the first monitoring constraint feature, the second monitoring constraint feature is subjected to cross-attention processing to form a first cross-fusion feature, and based on the attention parameters formed during the cross-attention processing, the first weight parameter of the first cross-fusion feature is determined.

[0045] Based on the second monitoring constraint feature, the first monitoring constraint feature is subjected to cross-attention processing to form a second cross-fusion feature, and based on the attention parameters formed during the cross-attention processing, the second weight parameter of the second cross-fusion feature is determined.

[0046] Based on the first weight parameter and the second weight parameter, the first cross-fusion feature and the second cross-fusion feature are weighted and averaged to form the monitoring cross-fusion feature.

[0047] Based on the above, this application also provides an interference source identification system based on vehicle-mounted mobile monitoring data analysis, comprising:

[0048] Memory, used to store computer programs;

[0049] The processor connected to the memory is used to execute the computer program stored in the memory to implement the above-described method for identifying interference sources based on vehicle-mounted mobile monitoring data analysis.

[0050] The interference source identification method and system based on vehicle-mounted mobile monitoring data analysis provided in this application firstly extracts first monitoring time-domain data and second monitoring time-domain data from the vehicle-mounted mobile monitoring data acquired by the vehicle-mounted monitoring device; secondly, it determines the first monitoring frequency-domain data corresponding to the first monitoring time-domain data and the second monitoring frequency-domain data corresponding to the second monitoring time-domain data; then, based on the encoded semantic information in the first monitoring frequency-domain data, it performs semantic constraints on the encoded semantic information in the first monitoring time-domain data to form a first monitoring constraint feature; further, based on the encoded semantic information in the second monitoring frequency-domain data, it performs semantic constraints on the encoded semantic information in the second monitoring time-domain data to form a second monitoring constraint feature; finally, it performs semantic decoding on the fusion feature of the first monitoring constraint feature and the second monitoring constraint feature to form a target interference source identification result. Based on the above, on the one hand, due to the different external influences at different locations, the accuracy of the collected data is relatively low. Therefore, vehicle-mounted monitoring equipment can acquire time-domain information from two different locations, and the data from the two locations can complement each other, reducing errors caused by incomplete or inaccurate data from a single location. For example, a signal at one location may be distorted due to environmental factors (such as obstruction), but by fusing data from another location, the robustness of signal identification can be improved. Furthermore, data from different locations may reflect different interference source characteristics or signal propagation characteristics. By fusing data from these two locations, spatial changes in the signal can be better captured, thereby improving the accuracy of identification. On the other hand, since the time-domain encoded semantic information is also based on the frequency-domain encoded semantic information, the semantic representation accuracy of the formed monitoring constraint features can be higher, making full and effective use of the potential semantic information in the acquired data. Therefore, the reliability of interference source identification can be effectively improved, and reliable target interference source identification results can be obtained, thereby addressing the problem of relatively low reliability in interference source identification in existing technologies. Attached Figure Description

[0051] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings.

[0052] Figure 1 This is a structural block diagram of an interference source identification system based on vehicle-mounted mobile monitoring data analysis, provided in an embodiment of this application.

[0053] Figure 2 This is a flowchart illustrating the interference source identification method based on vehicle-mounted mobile monitoring data analysis provided in an embodiment of this application.

[0054] Figure 3 A schematic diagram illustrating feature enhancements provided for embodiments of this application.

[0055] Figure 4 A schematic diagram illustrating semantic constraints provided in the embodiments of this application.

[0056] Figure 5 This is a schematic diagram of the cross-fusion processing provided in an embodiment of this application.

[0057] Figure 6 This is a schematic diagram of the deep fusion process provided in an embodiment of this application. Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0059] Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0060] like Figure 1 As shown in the figure, this application provides an interference source identification system based on vehicle-mounted mobile monitoring data analysis, which may include a memory and a processor.

[0061] In detail, the memory and the processor are electrically connected directly or indirectly to enable data transmission or interaction. For example, the memory and the processor can be electrically connected via one or more communication buses or signal lines. The processor is used to execute executable computer programs stored in the memory to implement the interference source identification method based on vehicle-mounted mobile monitoring data analysis provided in this application embodiment.

[0062] Optionally, the memory may be, but is not limited to, random access memory (RAM), read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), etc.

[0063] Optionally, the processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), a system on chip (SoC), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0064] Understandable. Figure 1 The structure shown is for illustrative purposes only. The interference source identification system based on vehicle-mounted mobile monitoring data analysis may also include... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown may include, for example, a communication unit for exchanging information with other devices, such as sensors.

[0065] Combination Figure 2 This application also provides an interference source identification method based on vehicle-mounted mobile monitoring data analysis, applicable to the aforementioned interference source identification system based on vehicle-mounted mobile monitoring data analysis. The method steps defined in the relevant process of the interference source identification method based on vehicle-mounted mobile monitoring data analysis can be implemented by the interference source identification system based on vehicle-mounted mobile monitoring data analysis (hereinafter referred to as the interference source identification system).

[0066] The following will be about Figure 2 The specific process shown will be explained in detail.

[0067] Step S110: Extract the first monitoring time domain data and the second monitoring time domain data from the vehicle mobile monitoring data obtained from the vehicle monitoring equipment.

[0068] In this embodiment, the interference source identification system can extract first monitoring time-domain data and second monitoring time-domain data from vehicle-mounted motion monitoring data acquired by the vehicle-mounted monitoring device. The first and second monitoring time-domain data reflect the time-domain information of the target signal at two different locations. That is, the vehicle-mounted monitoring device can be used to collect electromagnetic signals, and can collect electromagnetic signals at at least two locations during vehicle movement to obtain vehicle-mounted motion monitoring data. Thus, the interference source identification system can acquire the vehicle-mounted motion monitoring data from the vehicle-mounted monitoring device in real-time or non-real-time, and then extract the electromagnetic signals from the vehicle-mounted motion monitoring data to obtain the first and second monitoring time-domain data belonging to the time domain.

[0069] Step S120: Determine the first monitoring frequency domain data corresponding to the first monitoring time domain data and the second monitoring frequency domain data corresponding to the second monitoring time domain data, respectively.

[0070] In this embodiment, after obtaining the first monitoring time-domain data and the second monitoring time-domain data, the interference source identification system can determine the first monitoring frequency-domain data corresponding to the first monitoring time-domain data and the second monitoring frequency-domain data corresponding to the second monitoring time-domain data, respectively. For example, the first monitoring time-domain data can be time-frequency converted to form the first monitoring frequency-domain data. Furthermore, the second monitoring time-domain data can be time-frequency converted to form the second monitoring frequency-domain data.

[0071] Step S130: Based on the encoded semantic information in the first monitoring frequency domain data, semantic constraints are applied to the encoded semantic information in the first monitoring time domain data to form a first monitoring constraint feature.

[0072] In this embodiment, after obtaining the first monitoring frequency domain data, the interference source identification system can apply semantic constraints to the coded semantic information in the first monitoring time domain data based on the coded semantic information in the first monitoring frequency domain data, forming a first monitoring constraint feature. That is, the latent semantic information in the first monitoring frequency domain data and the first monitoring time domain data can be obtained through semantic encoding, and then corresponding semantic constraints can be applied based on the obtained latent semantic information, thereby forming a more accurate first monitoring constraint feature.

[0073] Step S140: Based on the encoded semantic information in the second monitoring frequency domain data, semantic constraints are applied to the encoded semantic information in the second monitoring time domain data to form a second monitoring constraint feature.

[0074] In this embodiment, after obtaining the second monitoring frequency domain data, the interference source identification system can apply semantic constraints to the coded semantic information in the second monitoring time domain data based on the coded semantic information in the second monitoring frequency domain data, thereby forming a second monitoring constraint feature. In other words, latent semantic information in the second monitoring frequency domain data and the second monitoring time domain data can be obtained through semantic encoding, and then corresponding semantic constraints can be applied based on the obtained latent semantic information, resulting in a more accurate second monitoring constraint feature.

[0075] Step S150: Semantic decoding is performed on the fused features of the first monitoring constraint feature and the second monitoring constraint feature to form the target interference source identification result.

[0076] In this embodiment, after obtaining the first monitoring constraint feature and the second monitoring constraint feature, the interference source identification system can perform semantic decoding on the fused features of the first monitoring constraint feature and the second monitoring constraint feature to form a target interference source identification result. That is, the first monitoring constraint feature and the second monitoring constraint feature can be fused first to achieve feature complementarity, and then semantic decoding can be performed based on the obtained fused features to obtain the corresponding target interference source identification result. The target interference source identification result reflects whether the target signal belongs to an interference source.

[0077] Based on the above, on the one hand, due to the different external influences at different locations, the accuracy of the collected data is relatively low. Therefore, vehicle-mounted monitoring equipment can acquire time-domain information from two different locations, and the data from the two locations can complement each other, reducing errors caused by incomplete or inaccurate data from a single location. For example, a signal at one location may be distorted due to environmental factors (such as obstruction), but by fusing data from another location, the robustness of signal identification can be improved. Furthermore, data from different locations may reflect different interference source characteristics or signal propagation characteristics. By fusing data from these two locations, spatial changes in the signal can be better captured, thereby improving the accuracy of identification. On the other hand, since the time-domain encoded semantic information is also based on the frequency-domain encoded semantic information, the semantic representation accuracy of the formed monitoring constraint features can be higher, making full and effective use of the potential semantic information in the acquired data. Therefore, the reliability of interference source identification can be effectively improved, and reliable target interference source identification results can be obtained, thereby addressing the problem of relatively low reliability in interference source identification in existing technologies.

[0078] Firstly, regarding step S110, it should be noted that the specific method for extracting the first monitoring time domain data and the second monitoring time domain data is not limited and can be selected according to actual needs.

[0079] For example, in one alternative implementation, monitoring time-domain data for two adjacent locations can be extracted from the vehicle-mounted mobile monitoring data to obtain first monitoring time-domain data and second monitoring time-domain data. As another alternative implementation, monitoring time-domain data for two non-adjacent locations can be extracted from the vehicle-mounted mobile monitoring data to obtain first monitoring time-domain data and second monitoring time-domain data.

[0080] Secondly, regarding step S120, it should be noted that the specific methods for determining the first monitoring frequency domain data corresponding to the first monitoring time domain data and the second monitoring frequency domain data corresponding to the second monitoring time domain data are not limited and can be selected according to actual needs.

[0081] For example, in an alternative implementation, the first monitoring time-domain data can be subjected to a Fourier transform to form first monitoring frequency-domain data. This first monitoring frequency-domain data may include either an amplitude spectrum (representing the signal intensity at each frequency) or a phase spectrum (representing the signal phase at each frequency), or it may include both amplitude and phase spectra. Furthermore, the second monitoring time-domain data can be subjected to a Fourier transform to form second monitoring frequency-domain data. This second monitoring frequency-domain data may include either an amplitude spectrum or a phase spectrum, or it may include both amplitude and phase spectra.

[0082] Thirdly, regarding step S130, it should be noted that the specific method for forming the first monitoring constraint feature is not limited and can be selected according to actual needs.

[0083] For example, in an alternative implementation, in order to ensure the reliability of semantic constraints and make the formed first monitoring constraint feature have high semantic representation accuracy, the above steps S131, S132 and S133 are described in detail below.

[0084] Step S131: Perform convolution mining on the first monitoring frequency domain data to form the first monitoring frequency domain feature, and perform convolution mining on the first monitoring time domain data to form the first monitoring time domain feature.

[0085] In this embodiment, convolution mining can be performed on the first monitoring frequency domain data to form first monitoring frequency domain features, and convolution mining can be performed on the first monitoring time domain data to form first monitoring time domain features. For example, two-dimensional convolution can be performed on the first monitoring amplitude spectrum and the first monitoring phase spectrum included in the first monitoring frequency domain data to obtain the corresponding first monitoring amplitude features and first monitoring phase features. That is, the first monitoring frequency domain features can include first monitoring amplitude features and first monitoring phase features. Alternatively, one-dimensional convolution can be performed on the first monitoring time domain data to obtain the first monitoring time domain features. It should be noted that each obtained feature can be represented by a vector after unfolding.

[0086] Step S132: Enhance the first monitoring frequency domain feature to form the first monitoring enhancement feature.

[0087] In this embodiment, after obtaining the first monitoring frequency domain feature, feature enhancement can be performed on the first monitoring frequency domain feature to form a first monitoring enhancement feature. This allows for the full extraction of detailed information in the frequency domain, resulting in better semantic representation capabilities.

[0088] Step S133: Based on the first monitoring enhancement feature, semantic constraints are applied to the first monitoring time-domain feature to form the first monitoring constraint feature.

[0089] In this embodiment, after obtaining the first monitoring enhancement feature, semantic constraints can be applied to the first monitoring time-domain feature based on the first monitoring enhancement feature to form a first monitoring constraint feature. That is, since the first monitoring enhancement feature is formed through feature enhancement, it has better semantic representation accuracy. Therefore, when applying semantic constraints to the first monitoring time-domain feature, the accuracy of the semantic constraints can be effectively guaranteed.

[0090] It is understood that in step S132 above, the specific method of feature enhancement for the first monitoring frequency domain feature is not limited. For example, in an alternative implementation, in order to ensure that the formed first monitoring enhancement feature can fully represent the semantic information related to interference, so that in the subsequent semantic constraint step, the formed first monitoring constraint feature tends to represent the semantic information related to interference, step S132 above may further include steps S132a, S132b, S132c and S132d, the specific contents of which are as follows.

[0091] Step S132a: For each frequency point in the first monitoring amplitude spectrum included in the first monitoring frequency domain data, determine whether the amplitude of the frequency point jumps, and perform convolution mining on the data representing whether the amplitude of each frequency point jumps to form amplitude jump features.

[0092] In the embodiments of this application, combined with Figure 3 For each frequency point in the first monitoring amplitude spectrum included in the first monitoring frequency domain data, it is determined whether the amplitude of that frequency point has jumped. Furthermore, the data representing whether the amplitude of each frequency point has jumped is subjected to convolution mining to form amplitude jump features. For example, an amplitude threshold can be set. Then, the difference between the amplitude of each frequency point and the amplitude of the previous frequency point can be calculated. If the absolute value of the difference is greater than the amplitude threshold, a jump can be considered to have occurred; if the absolute value of the difference is not greater than the amplitude threshold, a jump can be considered not to have occurred. A jump can be represented by "1", and no jump can be represented by "0". Thus, a one-dimensional array (1*n, where n is the number of frequency points) can be formed based on multiple frequency points. Then, a one-dimensional convolution can be performed on this one-dimensional array to obtain the amplitude jump features. For example, in wireless communication, narrowband interference may manifest as a sudden increase in intensity at a certain frequency point in the amplitude spectrum. Therefore, by mining amplitude jump features, it is possible to characterize whether something belongs to an interference source to a certain extent.

[0093] Step S132b: For each frequency point in the first monitoring phase spectrum included in the first monitoring frequency domain data, determine whether the phase of that frequency point has jumped, and perform convolution mining on the data characterizing whether the phase of each frequency point has jumped to form a phase jump feature.

[0094] In this embodiment, for each frequency point in the first monitoring phase spectrum included in the first monitoring frequency domain data, it is determined whether the phase of that frequency point has jumped. Furthermore, the data characterizing whether the phase of each frequency point has jumped is convolved to mine phase jump features. For example, a phase threshold can be set. Then, the difference between the phase of each frequency point and the phase of the previous frequency point can be calculated. If the absolute value of the difference is greater than the phase threshold, a jump can be considered to have occurred; if the absolute value of the difference is not greater than the phase threshold, no jump can be considered to have occurred. A jump can be represented by "1", and no jump by "0". Thus, a one-dimensional array (1*n, where n is the number of frequency points) can be formed based on multiple frequency points. Then, a one-dimensional convolution can be performed on this one-dimensional array to obtain the phase jump features. For example, interference signals typically exhibit unstable or random phase changes, while normal signals usually maintain a consistent or regular phase change. Interference signals, on the other hand, may exhibit drastic fluctuations or irregular phase changes. Therefore, by mining phase jump features, it is possible to characterize whether a signal belongs to an interference source to a certain extent.

[0095] Step S132c: Based on the amplitude jump feature and the phase jump feature, focus mining is performed on the first monitoring amplitude feature and the first monitoring phase feature included in the first monitoring frequency domain feature to form the first amplitude focus feature and the first phase focus feature.

[0096] In this embodiment, after obtaining the amplitude jump feature and the phase jump feature, the first monitoring amplitude feature and the first monitoring phase feature included in the first monitoring frequency domain feature can be focused and mined based on the amplitude jump feature and the phase jump feature, respectively, to form a first amplitude focused feature and a first phase focused feature. That is, the first monitoring amplitude feature can be focused and mined based on the amplitude jump feature to obtain the first amplitude focused feature. And, the second monitoring amplitude feature can be focused and mined based on the phase jump feature to obtain the second amplitude focused feature. The first monitoring amplitude feature is formed by convolution mining of the first monitoring amplitude spectrum, and the first monitoring phase feature is formed by convolution mining of the first monitoring phase spectrum. In this way, the goal of focusing global semantic information based on local semantic information related to interference can be achieved. This allows the formed first amplitude focused feature and first phase focused feature to represent global semantic information while also focusing on semantic information related to interference. Therefore, when used for interference source identification, it can have a higher accuracy in semantic representation. Furthermore, it should be noted that the focused mining can be implemented based on a cross-attention mechanism.

[0097] Step S132d: Based on the first amplitude focusing feature and the first phase focusing feature, determine the first monitoring enhancement feature.

[0098] In this embodiment of the application, after obtaining the first amplitude focusing feature and the first phase focusing feature, a first monitoring enhancement feature can be determined based on the first amplitude focusing feature and the first phase focusing feature. For example, the first amplitude focusing feature and the first phase focusing feature can be fused by means of averaging or weighted averaging to form the first monitoring enhancement feature.

[0099] It is understood that in step S133 above, the specific method of semantically constraining the first monitoring time-domain feature is not limited. For example, in an alternative implementation, in order to ensure the accuracy of the semantic constraint and make the reliability of the formed first monitoring constraint feature high, step S133 above may further include steps S133a, S133b and S133c, wherein the specific contents of each step are as follows.

[0100] Step S133a: Perform feature space transformation on the first monitoring enhancement feature to form a first monitoring transformation feature, and perform self-attention processing on the first monitoring transformation feature to form a monitoring attention feature.

[0101] In the embodiments of this application, combined with Figure 4 The first monitoring enhancement feature can be transformed into a first monitoring transformed feature by performing feature space transformation, and then self-attention processing can be applied to the first monitoring transformed feature to form a monitoring attention feature. It should be noted that since the first monitoring enhancement feature belongs to the frequency domain and the first monitoring time domain feature belongs to the time domain, feature space transformation can be performed first to ensure the reliability of semantic constraints. This allows semantic constraints to be performed in a similar semantic space. Furthermore, self-attention processing can focus important semantic information, resulting in higher accuracy during semantic constraints. The feature space transformation can be achieved using a linear function, such as Y=AX+B, where A is the weight matrix and B is the bias parameter.

[0102] Step S133b: Nonlinear activation is performed on the monitoring attention feature to form a monitoring activation feature.

[0103] In this embodiment of the application, after obtaining the monitoring attention features, the monitoring attention features can be nonlinearly activated to form monitoring activation features. For example, they can be mapped using functions such as sigmoid, so that each parameter in the obtained monitoring activation features belongs to 0-1, which can be used as a basis for characterizing the importance of the corresponding position.

[0104] Step S133c: Multiply the monitoring activation feature and the first monitoring time domain feature to multiply the two feature parameters at corresponding positions to form the first monitoring constraint feature.

[0105] In this embodiment, after obtaining the monitoring activation feature, the monitoring activation feature and the first monitoring time-domain feature can be multiplied together, such that the two feature parameters at corresponding positions are multiplied, i.e., bitwise multiplication is achieved to form the first monitoring constraint feature. In this way, constraints on each parameter in the first monitoring time-domain feature can be implemented based on the importance of each parameter in the first monitoring enhancement feature.

[0106] Fourthly, regarding step S140, it should be noted that the specific method for forming the second monitoring constraint feature is not limited and can be selected according to actual needs.

[0107] For example, in an alternative implementation, in order to ensure the reliability of semantic constraints and make the formed second monitoring constraint features have high semantic representation accuracy, the above steps S141, S142 and S143 are described in detail below.

[0108] Step S141: Perform convolution mining on the second monitoring frequency domain data to form the second monitoring frequency domain features, and perform convolution mining on the second monitoring time domain data to form the second monitoring time domain features.

[0109] In this embodiment, convolution mining can be performed on the second monitoring frequency domain data to form second monitoring frequency domain features, and convolution mining can be performed on the second monitoring time domain data to form second monitoring time domain features. For example, two-dimensional convolution can be performed on the second monitoring amplitude spectrum and the second monitoring phase spectrum included in the second monitoring frequency domain data to obtain the corresponding second monitoring amplitude features and second monitoring phase features. That is, the second monitoring frequency domain features can include second monitoring amplitude features and second monitoring phase features. Alternatively, one-dimensional convolution can be performed on the second monitoring time domain data to obtain the second monitoring time domain features. It should be noted that each obtained feature can be represented by a vector after unfolding.

[0110] Step S142: Enhance the second monitoring frequency domain feature to form a second monitoring enhancement feature.

[0111] In this embodiment, after obtaining the second monitoring frequency domain feature, feature enhancement can be performed on the second monitoring frequency domain feature to form a second monitoring enhanced feature. This allows for the full extraction of detailed information in the frequency domain, resulting in better semantic representation capabilities.

[0112] Step S143: Based on the second monitoring enhancement feature, semantic constraints are applied to the second monitoring time domain feature to form the second monitoring constraint feature.

[0113] In this embodiment, after obtaining the second monitoring enhancement feature, semantic constraints can be applied to the second monitoring time-domain feature based on the second monitoring enhancement feature to form a second monitoring constraint feature. That is, since the second monitoring enhancement feature is formed through feature enhancement, it has better semantic representation accuracy. Therefore, when applying semantic constraints to the second monitoring time-domain feature, the accuracy of the semantic constraints can be effectively guaranteed.

[0114] Fifthly, regarding step S150, it should be noted that the specific method for forming the target interference source identification result is not limited and can be selected according to actual needs.

[0115] For example, in an alternative implementation, in order to ensure that the identified target interference source has high reliability, the above step S150 may further include steps S151, S152, S153 and S154, wherein the specific contents of each step are as follows.

[0116] Step S151: Perform cross-fusion processing on the first monitoring constraint feature and the second monitoring constraint feature to form a cross-fusion monitoring feature.

[0117] In this embodiment of the application, the first monitoring constraint feature and the second monitoring constraint feature can be cross-fused to form a monitoring cross-fusion feature. That is, the first monitoring constraint feature and the second monitoring constraint feature can be fused together so that the formed monitoring cross-fusion feature can represent the two features.

[0118] Step S152: During the semantic forward fusion process, based on the first monitoring constraint feature, the monitoring cross-fusion feature is subjected to deep fusion processing to form a monitoring deep fusion feature.

[0119] In this embodiment of the application, after obtaining the monitoring cross-fusion features, the monitoring cross-fusion features can be deeply fused based on the first monitoring constraint features during the semantic forward fusion process to form monitoring deep fusion features.

[0120] Step S153: During the semantic backward fusion process, based on the second monitoring constraint features, the monitoring deep fusion features are subjected to deep fusion processing to form monitoring target fusion features.

[0121] In this embodiment, after obtaining the monitoring deep fusion features, deep fusion processing can be performed on the monitoring deep fusion features based on the second monitoring constraint features during the semantic backward fusion process to form monitoring target fusion features. That is, after achieving initial fusion through cross-fusion processing, to further improve the fusion effect, further deep fusion processing can be performed on the first and second monitoring constraint features during the semantic forward fusion and semantic backward fusion processes, respectively, so that the fusion of semantic information can be more complete. Furthermore, it should be noted that forward and backward can refer to two inverse processes. For example, in the forward processing, features are compressed, and in the backward processing, features are expanded. This ensures that the final monitoring target fusion features have the same size as the original features (such as the first and second monitoring constraint features). Moreover, since the size of the features in the processing is smaller than the original features, the computational load can be reduced, lowering the computational cost.

[0122] Step S154: Semantically decode the fused features of the monitored target to form the target interference source identification result.

[0123] In this embodiment, after obtaining the fused features of the monitored target, semantic decoding can be performed on the fused features to form a target interference source identification result. Semantic decoding can be implemented through fully connected processing and activation processing. For example, the fused features of the monitored target can be fully connected to obtain a 1*2 fully connected feature. Then, a classification function such as softmax can be used to activate this fully connected feature, resulting in a 1*2 probability distribution. This probability distribution includes two parameters representing the probability of belonging to an interference source and the probability of not belonging to an interference source, respectively. The type corresponding to the larger probability can then be determined as the target interference source identification result.

[0124] It is understood that in step S151 above, the specific method of cross-fusion processing of the first monitoring constraint feature and the second monitoring constraint feature is not limited. For example, in an alternative embodiment, in order to make the formed cross-fusion monitoring feature take into account both the first monitoring constraint feature and the second monitoring constraint feature through cross-fusion processing, step S151 above may further include steps S151a, S151b and S151c, wherein the specific contents of each step are as follows.

[0125] Step S151a: Based on the first monitoring constraint feature, perform cross-attention processing on the second monitoring constraint feature to form a first cross-fusion feature, and determine the first weight parameter of the first cross-fusion feature based on the attention parameters formed during the cross-attention processing.

[0126] In the embodiments of this application, combined with Figure 5 Based on the first monitoring constraint feature, the second monitoring constraint feature can be subjected to cross-attention processing to form a first cross-fusion feature. Based on the attention parameters formed during the cross-attention processing, a first weight parameter of the first cross-fusion feature can be determined. For example, the first weight parameter can be determined based on the dispersion of the attention parameter. There can be a positive correlation between the dispersion and the first weight parameter; that is, higher dispersion indicates uneven parameter distribution and higher significance of the attention processing, while lower dispersion indicates uniform parameter distribution and generally lower significance of the attention processing, meaning that effective mining of important semantic information is not achieved. The attention parameter can be determined by mapping the first monitoring constraint feature to a query feature and the second monitoring constraint feature to a key feature. Then, the transposes of the query feature and the key feature can be multiplied to obtain the corresponding attention parameter.

[0127] Step S151b: Based on the second monitoring constraint feature, perform cross-attention processing on the first monitoring constraint feature to form a second cross-fusion feature, and determine the second weight parameter of the second cross-fusion feature based on the attention parameters formed during the cross-attention processing.

[0128] In this embodiment of the application, based on the second monitoring constraint feature, the first monitoring constraint feature is subjected to cross-attention processing to form a second cross-fusion feature, and based on the attention parameters formed during the cross-attention processing, the second weight parameter of the second cross-fusion feature is determined. The specific processing procedure and principle are as described above.

[0129] Step S151c: Based on the first weight parameter and the second weight parameter, the first cross-fusion feature and the second cross-fusion feature are weighted and averaged to form the monitoring cross-fusion feature.

[0130] In this embodiment of the application, after obtaining the first weight parameter and the second weight parameter, the first cross-fusion feature and the second cross-fusion feature can be weighted and averaged based on the first weight parameter and the second weight parameter to form a monitoring cross-fusion feature, such as the first weight parameter * the first cross-fusion feature + the second weight parameter * the second cross-fusion feature. In this way, the fusion of weighted averages can be achieved through the effect of attention processing, which can make the fusion more reliable.

[0131] It is understood that the specific method of performing deep fusion processing on the monitored cross-fusion features in step S152 above is not limited. For example, in an alternative implementation, in order to ensure the sufficiency of deep fusion, step S152 above may include steps S152a, S152b, S152c and S152d, wherein the specific contents of each step are as follows.

[0132] Step S152a: In the first stage, the first monitoring constraint feature is subjected to self-attention processing to form the first attention feature of the first stage; and the first attention feature of the first stage is mapped to form the first gating adjustment feature of the first stage; and the monitoring cross-fusion feature is adjusted based on the first gating adjustment feature of the first stage to form the first monitoring adjustment feature of the first stage.

[0133] In the embodiments of this application, combined with Figure 6 In the first stage, the first monitoring constraint feature can be processed with self-attention to form the first attention feature of the first stage. Then, based on the first attention feature, mapping (e.g., using functions like sigmoid) is performed to form the first gating adjustment feature of the first stage. Finally, based on the first gating adjustment feature, the monitoring cross-fusion feature is adjusted (e.g., by performing bitwise multiplication) to form the first monitoring adjustment feature of the first stage. In other words, performing self-attention processing before gating mapping allows important semantic information to be extracted first. Then, mapping based on the first attention feature allows the parameters in the resulting first gating adjustment feature to more accurately represent the importance of the corresponding position, thereby ensuring the reliability of the gating adjustment.

[0134] In step S152b, in the second stage, the first attention feature and the first monitoring and regulation feature of the first stage are downsampled to form the first attention sampling feature and the first monitoring sampling feature of the second stage, and the first attention sampling feature of the second stage is mapped to form the first gating regulation feature of the second stage, and the first monitoring sampling feature of the second stage is regulated based on the first gating regulation feature of the second stage to form the first monitoring and regulation feature of the second stage.

[0135] In this embodiment, in the second stage, the first attention feature and the first monitoring and adjustment feature of the first stage can be downsampled (downsampling can be achieved by convolution and / or pooling, etc., to reduce the size of the features) to form the first attention sampling feature and the first monitoring sampling feature of the second stage. Based on the first attention sampling feature of the second stage, mapping is performed to form the first gating and adjustment feature of the second stage. Based on the first gating and adjustment feature of the second stage, the first monitoring sampling feature of the second stage is adjusted to form the first monitoring and adjustment feature of the second stage.

[0136] In step S152c, in each subsequent stage, the first attention sampling feature and the first monitoring regulation feature of the previous stage are downsampled to form the first attention sampling feature and the first monitoring sampling feature of the current stage, and the first attention sampling feature of the current stage is mapped to form the first gating regulation feature of the current stage, and the first monitoring sampling feature of the current stage is regulated based on the first gating regulation feature of the current stage to form the first monitoring regulation feature of the current stage.

[0137] In the embodiments of this application, in each subsequent stage (such as the third stage, the fourth stage, etc.), the first attention sampling feature and the first monitoring and adjustment feature in the previous stage can be downsampled to form the first attention sampling feature and the first monitoring and adjustment feature of the current stage. Furthermore, the first attention sampling feature of the current stage can be mapped to form the first gating and adjustment feature of the current stage. Finally, the first monitoring and adjustment feature of the current stage can be adjusted based on the first gating and adjustment feature of the current stage to form the first monitoring and adjustment feature of the current stage.

[0138] In step S152d, the first monitoring regulation feature of the last stage is used as the monitoring deep fusion feature.

[0139] In this embodiment of the application, after obtaining the first monitoring and adjustment feature of the last stage, the first monitoring and adjustment feature of the last stage can be used as the monitoring deep fusion feature.

[0140] It is understood that in step S153 above, the specific method of performing deep fusion processing on the monitoring deep fusion features is not limited. For example, in an alternative implementation, in order to ensure the sufficiency of deep fusion, step S153 above may include steps S153a, S153b, S153c and S153d, wherein the specific contents of each step are as follows.

[0141] Step S153a: In the first stage, the compressed features of the second monitoring constraint features are subjected to self-attention processing to form the second attention features of the first stage; and the second attention features of the first stage are mapped to form the second gating adjustment features of the first stage; and the monitoring deep fusion features are adjusted based on the second gating adjustment features of the first stage to form the second monitoring adjustment features of the first stage.

[0142] In this embodiment, in the first stage, the compressed feature of the second monitoring constraint feature (which can be compressed by convolution and / or pooling to reduce its size to the same size as the monitoring deep fusion feature, so as to facilitate subsequent gating adjustment) can be subjected to self-attention processing to form the second attention feature of the first stage; and the second attention feature of the first stage can be mapped to form the second gating adjustment feature of the first stage; and the monitoring deep fusion feature can be adjusted based on the second gating adjustment feature of the first stage to form the second monitoring adjustment feature of the first stage.

[0143] In step S153b, in the second stage, the second attention feature and the second monitoring regulation feature of the first stage are upsampled to form the second attention sampling feature and the second monitoring sampling feature of the second stage, and the second attention sampling feature of the second stage is mapped to form the second gating regulation feature of the second stage, and the second monitoring sampling feature of the second stage is regulated based on the second gating regulation feature of the second stage to form the second monitoring regulation feature of the second stage.

[0144] In the embodiments of this application, in the second stage, the second attention feature and the second monitoring adjustment feature in the first stage can be upsampled respectively (for example, upsampling can be achieved by transposing convolution, etc., so that the feature size can be expanded) to form the second attention sampling feature and the second monitoring sampling feature of the second stage. Based on the second attention sampling feature of the second stage, mapping is performed to form the second gating adjustment feature of the second stage. Based on the second gating adjustment feature of the second stage, the second monitoring sampling feature of the second stage is adjusted to form the second monitoring adjustment feature of the second stage.

[0145] In step S153c, in each subsequent stage, the second attention sampling feature and the second monitoring regulation feature in the previous stage are upsampled to form the second attention sampling feature and the second monitoring sampling feature of the current stage, and the second attention sampling feature of the current stage is mapped to form the second gating regulation feature of the current stage, and the second monitoring sampling feature of the current stage is regulated based on the second gating regulation feature of the current stage to form the second monitoring regulation feature of the current stage.

[0146] In the embodiments of this application, in each subsequent stage (such as the third stage, the fourth stage, etc.), the second attention sampling feature and the second monitoring and adjustment feature in the previous stage can be upsampled to form the second attention sampling feature and the second monitoring and adjustment feature of the current stage. Furthermore, the second attention sampling feature of the current stage can be mapped to form the second gating and adjustment feature of the current stage. Finally, the second monitoring and adjustment feature of the current stage can be adjusted based on the second gating and adjustment feature of the current stage to form the second monitoring and adjustment feature of the current stage.

[0147] In step S153d, the second monitoring regulation feature of the last stage is used as the monitoring target fusion feature.

[0148] In this embodiment of the application, after obtaining the second monitoring and adjustment feature of the last stage, the second monitoring and adjustment feature of the last stage can be used as the monitoring target fusion feature.

[0149] In summary, the interference source identification method and system based on vehicle-mounted mobile monitoring data analysis provided in this application firstly extracts first monitoring time-domain data and second monitoring time-domain data from the vehicle-mounted mobile monitoring data acquired by the vehicle-mounted monitoring device; secondly, it determines the first monitoring frequency-domain data corresponding to the first monitoring time-domain data and the second monitoring frequency-domain data corresponding to the second monitoring time-domain data; then, based on the encoded semantic information in the first monitoring frequency-domain data, it performs semantic constraints on the encoded semantic information in the first monitoring time-domain data to form a first monitoring constraint feature; further, based on the encoded semantic information in the second monitoring frequency-domain data, it performs semantic constraints on the encoded semantic information in the second monitoring time-domain data to form a second monitoring constraint feature; finally, it performs semantic decoding on the fusion feature of the first monitoring constraint feature and the second monitoring constraint feature to form the target interference source identification result. Based on the above, on the one hand, due to the different external influences at different locations, the accuracy of the collected data is relatively low. Therefore, vehicle-mounted monitoring equipment can acquire time-domain information from two different locations, and the data from the two locations can complement each other, reducing errors caused by incomplete or inaccurate data from a single location. For example, a signal at one location may be distorted due to environmental factors (such as obstruction), but by fusing data from another location, the robustness of signal identification can be improved. Furthermore, data from different locations may reflect different interference source characteristics or signal propagation characteristics. By fusing data from these two locations, spatial changes in the signal can be better captured, thereby improving the accuracy of identification. On the other hand, since the time-domain encoded semantic information is also based on the frequency-domain encoded semantic information, the semantic representation accuracy of the formed monitoring constraint features can be higher, making full and effective use of the potential semantic information in the acquired data. Therefore, the reliability of interference source identification can be effectively improved, and reliable target interference source identification results can be obtained, thereby addressing the problem of relatively low reliability in interference source identification in existing technologies.

[0150] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for identifying interference sources based on vehicle-mounted mobile monitoring data analysis, characterized in that, include: First monitoring time domain data and second monitoring time domain data are extracted from the vehicle mobile monitoring data obtained from the vehicle monitoring equipment. The first monitoring time domain data and the second monitoring time domain data are used to reflect the time domain information of the target signal at two different locations. Determine the first monitoring frequency domain data corresponding to the first monitoring time domain data and the second monitoring frequency domain data corresponding to the second monitoring time domain data, respectively; Based on the encoded semantic information in the first monitoring frequency domain data, semantic constraints are applied to the encoded semantic information in the first monitoring time domain data to form a first monitoring constraint feature, including: performing convolutional mining on the first monitoring frequency domain data to form a first monitoring frequency domain feature, and performing convolutional mining on the first monitoring time domain data to form a first monitoring time domain feature; performing feature enhancement on the first monitoring frequency domain feature to form a first monitoring enhancement feature; and applying semantic constraints on the first monitoring time domain feature based on the first monitoring enhancement feature to form a first monitoring constraint feature. Based on the encoded semantic information in the second monitoring frequency domain data, semantic constraints are applied to the encoded semantic information in the second monitoring time domain data to form second monitoring constraint features. This includes: performing convolutional mining on the second monitoring frequency domain data to form second monitoring frequency domain features, and performing convolutional mining on the second monitoring time domain data to form second monitoring time domain features; performing feature enhancement on the second monitoring frequency domain features to form second monitoring enhancement features; and applying semantic constraints on the second monitoring time domain features based on the second monitoring enhancement features to form second monitoring constraint features. The fused features of the first monitoring constraint feature and the second monitoring constraint feature are semantically decoded to form a target interference source identification result, wherein the target interference source identification result is used to reflect whether the target signal belongs to an interference source.

2. The interference source identification method based on vehicle-mounted mobile monitoring data analysis according to claim 1, characterized in that, The step of enhancing the first monitoring frequency domain features to form the first monitoring enhanced features includes: For each frequency point in the first monitoring amplitude spectrum included in the first monitoring frequency domain data, it is determined whether the amplitude of that frequency point jumps, and the data characterizing whether the amplitude of each frequency point jumps is subjected to convolution mining to form amplitude jump features. For each frequency point in the first monitoring phase spectrum included in the first monitoring frequency domain data, it is determined whether the phase of that frequency point has jumped, and the data characterizing whether the phase of each frequency point has jumped is subjected to convolution mining to form phase jump features. Based on the amplitude jump feature and the phase jump feature respectively, the first monitoring amplitude feature and the first monitoring phase feature included in the first monitoring frequency domain feature are focused and mined to form the first amplitude focus feature and the first phase focus feature; Based on the first amplitude focusing feature and the first phase focusing feature, the first monitoring enhancement feature is determined.

3. The interference source identification method based on vehicle-mounted mobile monitoring data analysis according to claim 1, characterized in that, The step of semantically constraining the first monitoring time-domain features based on the first monitoring enhancement features to form the first monitoring constraint features includes: The first monitoring enhancement feature is subjected to feature space transformation to form a first monitoring transformation feature, and the first monitoring transformation feature is subjected to self-attention processing to form a monitoring attention feature; The monitoring attention features are nonlinearly activated to form monitoring activation features; The monitoring activation feature and the first monitoring time domain feature are multiplied together to form the first monitoring constraint feature.

4. The interference source identification method based on vehicle-mounted mobile monitoring data analysis according to claim 1, characterized in that, The step of semantically decoding the fused features of the first monitoring constraint features and the second monitoring constraint features to form the target interference source identification result includes: The first monitoring constraint feature and the second monitoring constraint feature are cross-fused to form a cross-fused monitoring feature. During the semantic forward fusion process, based on the first monitoring constraint features, the monitoring cross-fusion features are subjected to deep fusion processing to form monitoring deep fusion features; During the semantic backward fusion process, based on the second monitoring constraint features, the monitoring deep fusion features are subjected to deep fusion processing to form monitoring target fusion features; The fused features of the monitored target are semantically decoded to form the target interference source identification result.

5. The interference source identification method based on vehicle-mounted mobile monitoring data analysis according to claim 4, characterized in that, The step of performing deep fusion processing on the monitoring cross-fusion features based on the first monitoring constraint features during the semantic forward fusion process to form monitoring deep fusion features includes: In the first stage, the first monitoring constraint feature is subjected to self-attention processing to form the first attention feature of the first stage, and the first attention feature of the first stage is mapped to form the first gating adjustment feature of the first stage, and the monitoring cross-fusion feature is adjusted based on the first gating adjustment feature of the first stage to form the first monitoring adjustment feature of the first stage. In the second stage, the first attention feature and the first monitoring and regulation feature of the first stage are downsampled to form the first attention sampling feature and the first monitoring sampling feature of the second stage. Based on the first attention sampling feature of the second stage, a mapping is performed to form the first gating regulation feature of the second stage. Based on the first gating regulation feature of the second stage, the first monitoring sampling feature of the second stage is regulated to form the first monitoring regulation feature of the second stage. In each subsequent stage, the first attention sampling feature and the first monitoring regulation feature of the previous stage are downsampled to form the first attention sampling feature and the first monitoring sampling feature of the current stage. Based on the first attention sampling feature of the current stage, a mapping is performed to form the first gating regulation feature of the current stage. Based on the first gating regulation feature of the current stage, the first monitoring sampling feature of the current stage is regulated to form the first monitoring regulation feature of the current stage. The first monitoring and regulation feature of the last stage is used as the monitoring deep fusion feature.

6. The interference source identification method based on vehicle-mounted mobile monitoring data analysis according to claim 4, characterized in that, The step of performing deep fusion processing on the monitoring deep fusion features based on the second monitoring constraint features to form monitoring target fusion features during the semantic backward fusion process includes: In the first stage, the compressed features of the second monitoring constraint features are subjected to self-attention processing to form the second attention features of the first stage; and the second attention features of the first stage are mapped to form the second gating adjustment features of the first stage; and the monitoring deep fusion features are adjusted based on the second gating adjustment features of the first stage to form the second monitoring adjustment features of the first stage. In the second stage, the second attention feature and the second monitoring and regulation feature of the first stage are upsampled to form the second attention sampling feature and the second monitoring sampling feature of the second stage. Based on the second attention sampling feature of the second stage, a mapping is performed to form the second gating regulation feature of the second stage. Based on the second gating regulation feature of the second stage, the second monitoring sampling feature of the second stage is regulated to form the second monitoring regulation feature of the second stage. In each subsequent stage, the second attention sampling feature and the second monitoring regulation feature in the previous stage are upsampled to form the second attention sampling feature and the second monitoring sampling feature of the current stage, and the second attention sampling feature of the current stage is mapped to form the second gating regulation feature of the current stage, and the second monitoring sampling feature of the current stage is regulated based on the second gating regulation feature of the current stage to form the second monitoring regulation feature of the current stage. The second monitoring and regulation feature of the last stage is used as the monitoring target fusion feature.

7. The interference source identification method based on vehicle-mounted mobile monitoring data analysis according to claim 4, characterized in that, The step of cross-fusion processing the first monitoring constraint feature and the second monitoring constraint feature to form a cross-fusion monitoring feature includes: Based on the first monitoring constraint feature, the second monitoring constraint feature is subjected to cross-attention processing to form a first cross-fusion feature, and based on the attention parameters formed during the cross-attention processing, the first weight parameter of the first cross-fusion feature is determined. Based on the second monitoring constraint feature, the first monitoring constraint feature is subjected to cross-attention processing to form a second cross-fusion feature, and based on the attention parameters formed during the cross-attention processing, the second weight parameter of the second cross-fusion feature is determined. Based on the first weight parameter and the second weight parameter, the first cross-fusion feature and the second cross-fusion feature are weighted and averaged to form the monitoring cross-fusion feature.

8. An interference source identification system based on vehicle-mounted mobile monitoring data analysis, characterized in that, include: Memory, used to store computer programs; A processor connected to the memory is used to execute the computer program stored in the memory to implement the interference source identification method based on vehicle mobile monitoring data analysis as described in any one of claims 1-7.

Images