A bird-detecting radar data processing system and method optimized for small targets

Abstract

The present invention relates to the technical field of bird-detecting radar data processing, and specifically to a bird-detecting radar data processing system and method optimized for small targets. The present invention includes a data acquisition module, an adaptive filtering module and a graph neural network processing module, wherein: the data acquisition module acquires radar data and expected data; the adaptive filtering module dynamically adjusts the step size and weight based on the current error during the filtering process by setting an initial step size and weight to generate a radar feature vector; the graph neural network processing module generates a target node according to the radar feature vector, defines edge weights through adjacent errors, calculates attention weights, and finally updates the radar feature vector of the target node. This scheme improves the accuracy and stability of small target signal extraction and solves the signal noise and parameter fixation problems in traditional methods.

Description

Technical Field

[0001] The present invention relates to the technical field of bird-detecting radar data processing, and in particular to a bird-detecting radar data processing system and method optimized for small targets. Background Art

[0002] Traditional bird-detecting radar data processing systems often face difficulties in signal extraction due to large signal clutter when processing small target signals. Since the echo signals of small targets such as birds are relatively weak, traditional systems usually use fixed filter parameters and weights when processing these signals, which causes the filter to be unstable under different signal conditions. If the step size of the filter is set too large, it will cause oscillation and unstable output, while if it is set too small, it will result in slow processing speed and difficulty in convergence.

[0003] In addition, traditional processing methods are highly dependent on initial parameters and are difficult to cope with the variable characteristics of radar signals, resulting in large errors between the filtering results and the expected data, affecting the accuracy and stability of signal extraction. In terms of graph neural network processing, traditional systems often fail to fully utilize the relationship between each radar feature vector, relying only on the information of a single node for processing, ignoring the contribution of adjacent nodes to the target node characteristics, resulting in the lack of sufficient accuracy and representativeness of the final generated feature vector.

[0004] In summary, when dealing with small target signals, the traditional bird-detecting radar data processing system has problems such as fixed step size affecting the convergence speed, unstable filtering, low feature extraction accuracy, and insufficient estimation of neighboring node contributions. A more adaptive and effective optimization method is urgently needed to improve the overall system performance. Summary of the invention

[0005] The purpose of the present invention is to provide a bird-detecting radar data processing system and method optimized for small targets to solve the problems raised in the above background technology.

[0006] To achieve the above-mentioned purpose, one of the purposes of the present invention is to provide a bird detection radar data processing system optimized for small targets, comprising a data acquisition module, an adaptive filtering module and a graph neural network processing module, wherein:

[0007] The data acquisition module obtains the expected data of the bird detection radar and radar data, of which Refers to the number of radar data obtained from the radar equipment;

[0008] The adaptive filtering module sets the initial step size and initial weight, and uses the filter to The radar data is filtered and output as the filter output result; the current error is determined according to the filter output result and the expected data of the bird detection radar, and the initial step length is adjusted according to the current error to generate the current step length, and the difference between the initial step length and the current step length is recorded as the adjacent error; and the initial weight is adjusted according to the current step length to generate the current weight, and the radar feature vector is obtained according to the current weight and the filter output result; it is repeated times until the generation radar feature vectors;

[0009] The graph neural network processing module is based on radar feature vectors, generating target nodes, and define the edge weights of adjacent target nodes based on the adjacent errors. Each target node calculates the attention weight according to the radar feature vectors and edge weights of its adjacent nodes to determine the contribution of neighboring nodes to the update of its own radar feature vector. According to the contribution, the radar feature vector of each target node is updated. The radar feature vector is used as the final extracted target feature.

[0010] As a further improvement of the technical solution, the adaptive filtering module includes an initialization unit, which sets an initial step size and initial weights , where the initial step size determines the speed of each weight adjustment; the initial weight determines the state of the filter when processing the initial input.

[0011] As a further improvement of the technical solution, the adaptive filtering module includes a feature vector generating unit, the feature vector generating unit generates The process of generating a radar feature vector specifically includes:

[0012] Use filters to The radar data is filtered and output as the filter output result ,in , , yes Radar data in radar data;

[0013] Determine the current error based on the filter output and the expected data of the bird detection radar ,in , Indicates the expected data of bird detection radar; error Represents the difference between the filter output result and the expected data of the bird detection radar. The goal of adaptive filtering is to improve the accuracy of the output signal by minimizing this error;

[0014] According to the current error , adjust the initial step size to generate the current step size ,but ,in and is the adjustment parameter, is the signal-to-noise ratio;

[0015] Record the difference between the initial step length and the current step length as the adjacent error ,in ;

[0016] Based on the current step length , for the initial weight Make adjustments to generate the current weight ,in ;According to the current weight and filter output results, the radar feature vector is obtained ,in ;

[0017] Repeat the above process times, until The radar data is iterated and generated radar feature vectors.

[0018] As a further improvement of the technical solution, the graph neural network processing module includes a graph construction unit, and the process of defining the edge of the target node based on the adjacent error by the graph construction unit specifically includes:

[0019] definition The radar feature vector is ,in ;

[0020] Each radar feature vector Each of them is a node in the graph neural network ,generate Target Nodes ;

[0021] Defining two adjacent target nodes based on neighboring errors and The edge weight between them is given by the formula , after defining edge weights based on adjacent errors, the edge set is generated , where each edge Indicates the target node and the target node The edge between them has a weight equal to the adjacent error ,but .

[0022] As a further improvement of the technical solution, the graph neural network processing module includes a graph processing unit, and the process of updating the radar feature vector of each target node by the graph processing unit is as follows:

[0023] To determine each target node The radar characteristic vector The contribution size of each adjacent node during update, the edge set The adjacent error in Normalize and get the attention weight between the target node and its adjacent nodes ;

[0024] According to the calculated attention weight , determine the contribution of neighboring nodes to the update of the radar feature vector of the target node, and perform weighted summation of the feature vectors of all neighboring nodes of each target node. The formula is as follows:

[0025] ,in Indicates the target node The set of adjacent node indices, Is the target node For the target node The attention weight, is the updated target node The radar feature vector of

[0026] According to the calculated attention weight and the feature vectors of the adjacent nodes, the radar feature vector of each target node is updated. The updated radar feature vector is the new radar feature vector of each node after integrating the information of its neighboring nodes. The radar feature vector is used as the final extracted target feature.

[0027] The second object of the present invention is to provide a method for using a bird detection radar data processing system optimized for small targets, comprising the following method steps:

[0028] S1. Obtain the expected data of bird detection radar and radar data, of which Refers to the number of radar data obtained from the radar equipment;

[0029] S2, set the initial step size and initial weight, use the filter to The radar data is filtered and output as the filter output result; the current error is determined according to the filter output result and the expected data of the bird detection radar, and the initial step length is adjusted according to the current error to generate the current step length, and the difference between the initial step length and the current step length is recorded as the adjacent error; and the initial weight is adjusted according to the current step length to generate the current weight, and the radar feature vector is obtained according to the current weight and the filter output result; it is repeated times until the generation radar feature vectors;

[0030] S3. According to radar feature vectors, generating target nodes, and define the edge weights of adjacent target nodes based on the adjacent errors. Each target node calculates the attention weight according to the radar feature vectors and edge weights of its adjacent nodes to determine the contribution of neighboring nodes to the update of its own radar feature vector. According to the contribution, the radar feature vector of each target node is updated. The radar feature vector is used as the final extracted target feature.

[0031] Compared with the prior art, the present invention has the following beneficial effects:

[0032] The bird-detecting radar data processing system and method optimized for small targets generates the current step size by adjusting the initial step size according to the current error. The system can flexibly adapt to different signal conditions, thereby improving the efficiency and stability of the filter. Specifically, the initial step size determines the speed of each weight adjustment. If it is set too large, it will cause oscillation, and if it is set too small, it will lead to slow convergence. By adjusting the step size in real time, the system can better cope with the changing radar data and ensure that the filter gradually minimizes the error between the output result and the expected data of the bird-detecting radar. This dynamic adjustment mechanism enables the filter to extract small target signals more accurately and generate a set of accurate radar feature vectors after multiple iterations. These feature vectors are the basis for subsequent graph neural network processing, which improves the target recognition ability of the entire system.

[0033] In addition, the adjustment of the current step size is also associated with the adjacent error. By recording the difference between the initial step size and the current step size as part of the adjacent error, it is used to define the edge weights of adjacent target nodes. In the graph neural network processing, based on these edge weights, each target node calculates the attention weight according to the radar feature vectors and edge weights of its adjacent nodes, and then determines the contribution of the neighboring nodes to the update of its own radar feature vector; in this way, through reasonable step size adjustment, the system can not only effectively optimize the weights to make the filter output closer to the real signal, but also generate more accurate target features in the graph neural network based on these optimized feature vectors; finally, the updated radar feature vector integrates the information of adjacent nodes after multiple iterations and optimizations, and has better comprehensive signal feature representation capabilities, which improves the recognition effect of the entire system when detecting small targets (such as birds). BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 It is a schematic diagram of the overall method flow of the present invention;

[0035] Figure 2 It is a schematic diagram of the overall module unit of the present invention.

[0036] In the figure: 100, data acquisition module; 200, adaptive filtering module; 201, initialization unit; 202, feature vector generation unit; 300, graph neural network processing module; 301, graph construction unit; 302, graph processing unit. DETAILED DESCRIPTION

[0037] The following will be combined with the accompanying drawings in the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0038] Unless otherwise specified, the terms (including technical terms) used herein have the commonly understood meanings to those skilled in the art. In addition, it is understood that the terms defined in commonly used dictionaries should be understood to have the same meanings as those in the context of the relevant fields, and should not be understood as idealized or overly formal meanings.

[0039] According to the document, the following are some of the definitions of terms:

[0040] Filter output result: The adaptive filter module uses filters to filter the radar data in turn. The output data obtained is called the filter output result. This is the signal data after filtering and is the basis for subsequent error calculation.

[0041] Current error: This refers to the difference between the filter output and the expected data of the bird detection radar. The current error is used to measure the deviation between the processed signal and the ideal signal, and provides feedback information for the adjustment of the step size and weight;

[0042] Current step size: The step size generated after adjusting the initial step size according to the current error is called the current step size. The current step size is used to control the speed of weight update to adapt to different signal conditions and improve the efficiency and stability of the filter;

[0043] Adjacent error: The difference between the initial step size and the current step size is called the adjacent error. The adjacent error is used to define the edge weights of adjacent target nodes and is an important parameter for calculating attention weights in graph neural network processing.

[0044] Current weight: The weight generated after adjusting the initial weight according to the current step size is called the current weight. The current weight is used to update the state of the filter so that the output signal gradually approaches the ideal signal;

[0045] Radar feature vector: The feature vector generated by the adaptive filtering module based on the current weights and filter output results is called the radar feature vector. These vectors are the basic information of each node in the subsequent graph neural network processing;

[0046] Edge weights of adjacent target nodes: This refers to the weights of the edges defined between adjacent target nodes in the graph neural network based on the neighboring errors. The edge weights determine the contribution of the adjacent nodes to the update of the target node feature vector.

[0047] Contribution size: In graph neural network processing, the contribution size of each adjacent node to the update of the target node feature vector is determined based on the calculated attention weights and the feature vectors of the adjacent nodes. The contribution size determines the influence of each adjacent node in the update process of the target node feature vector.

[0048] Next, see Figure 1-Figure 2 One of the purposes of this embodiment is to provide a bird-detecting radar data processing system optimized for small targets, including a data acquisition module 100, an adaptive filtering module 200 and a graph neural network processing module 300.

[0049] The data acquisition module 100 obtains the expected data of the bird detection radar and radar data, of which Refers to the number of radar data obtained from the radar device. Bird detection radar expected data refers to the real label data obtained in the experiment, which is the ideal echo signal of small targets (i.e. birds). It is used as a reference in the filtering process to measure the difference between radar data and expected data. The original observation data obtained from the radar device is the data within a specific area. The detection results of each target reflect the specific characteristics of the target and are used for adaptive filtering and graph neural network processing.

[0050] The initialization unit 201 in the adaptive filtering module 200 sets the initial step size and the initial weight, specifically including:

[0051] Set the initial step size and initial weights , where the initial step size determines the speed of each weight adjustment. The appropriateness of the initial step size setting directly affects the convergence speed of the filter; the initial weight determines the state of the filter when processing the initial input and is the starting point for calculation and adjustment;

[0052] The feature vector generation unit 202 in the adaptive filtering module 200 uses a filter to sequentially The radar data is filtered and output as the filter output result ,in , , yes Radar data in radar data;

[0053] Determine the current error based on the filter output and the expected data of the bird detection radar ,in , Indicates the expected data of bird detection radar; error Represents the difference between the filter output and the expected data of the bird detection radar. The goal of adaptive filtering is to improve the accuracy of the output signal by minimizing this error. The error provides feedback information for subsequent steps (such as step size adjustment and weight update), so that the filter can gradually optimize the weight vector and reduce the error.

[0054] According to the current error , adjust the initial step size to generate the current step size ,but ,in and is the adjustment parameter, is the signal-to-noise ratio; by adjusting the step size in real time, the filter can respond to different signal conditions more flexibly and adapt to changes in radar data. Appropriate step size adjustment can accelerate convergence and improve stability;

[0055] Since the initial step size determines the speed of each weight adjustment, if it is set too large, it will cause oscillation, and if it is set too small, it will lead to slow convergence. Therefore, by adjusting the initial step size in real time, the system can better cope with the changing radar data and ensure that the filter gradually minimizes the error between the output result and the expected data of the bird detection radar.

[0056] Record the difference between the initial step length and the current step length as the adjacent error ,in ;

[0057] Based on the current step length , for the initial weight Make adjustments to generate the current weight ,in ;According to the current weight and filter output results, the radar feature vector is obtained ,in ; Weight adjustment is the core of the adaptive filtering module 200. Through error feedback and step size adjustment, the filter continuously optimizes the weight vector according to the input signal to make the output signal close to the target signal. As the weight is continuously updated, the output of the filter gradually approaches the expected output signal, thereby achieving higher signal extraction accuracy. This step directly affects the final signal processing effect;

[0058] Repeat the above process times, until The radar data is iterated and generated radar feature vectors.

[0059] This dynamic adjustment mechanism enables the filter to extract small target signals more accurately and generate a set of accurate radar feature vectors after multiple iterations; the radar feature vectors are the basis for subsequent graph neural network processing, improving the target recognition capability of the entire system.

[0060] The graph construction unit 301 in the graph neural network processing module 300 is based on radar feature vectors, generating target nodes, and define the edges of the target nodes based on the adjacent errors, including:

[0061] definition The radar feature vector is ,in ;

[0062] Each radar feature vector Each of them is a node in the graph neural network ,generate Target Nodes ;

[0063] Defining two adjacent target nodes based on neighboring errors and The edge weight between them is given by the formula , after defining edge weights based on adjacent errors, the edge set is generated , where each edge Indicates the target node and the target node The edge between them has a weight equal to the adjacent error ,but .

[0064] By recording the difference between the initial step size and the current step size as part of the adjacent error, it is used to define the edge weights of adjacent target nodes; based on these edge weights, each target node calculates the attention weight according to the radar feature vectors and edge weights of its adjacent nodes, and then determines the contribution of the neighboring nodes to the update of its own radar feature vector. In this way, through reasonable step size adjustment, the system can not only effectively optimize the weights to make the filter output closer to the real signal, but also generate more accurate target features in the graph neural network based on these optimized feature vectors.

[0065] The graph processing unit 302 in the graph neural network processing module 300 calculates the attention weight for each target node based on the radar feature vectors and edges of its neighboring nodes, so as to determine the contribution of the neighboring nodes to the update of its own radar feature vector. According to the contribution, the radar feature vector of each target node is updated. The radar feature vectors are used as the final extracted target features, including:

[0066] To determine each target node The radar characteristic vector The contribution size of each adjacent node during update, the edge set The adjacent error in Normalize and get the attention weight between the target node and its adjacent nodes ;

[0067] According to the calculated attention weight , determine the contribution of neighboring nodes to the update of the radar feature vector of the target node, and perform weighted summation of the feature vectors of all neighboring nodes of each target node. The formula is as follows:

[0068] ,in Indicates the target node The set of adjacent node indices, Is the target node For the target node The attention weight, is the updated target node The radar feature vector of

[0069] According to the calculated attention weight and the feature vectors of the adjacent nodes, the radar feature vector of each target node is updated. The updated radar feature vector is the new radar feature vector of each node after integrating the information of its neighboring nodes. The radar feature vectors are used as the final extracted target features. These final extracted feature vectors represent the comprehensive features of the target node after considering the information of the adjacent nodes, which is a better signal representation.

[0070] The second purpose of this embodiment is to provide a method for using a bird detection radar data processing system optimized for small targets, including the following method steps:

[0071] S1. Obtain the expected data of bird detection radar and radar data, of which Refers to the number of radar data obtained from the radar equipment;

[0072] S2, set the initial step size and initial weight, use the filter to The radar data is filtered and output as the filter output result; the current error is determined according to the filter output result and the expected data of the bird detection radar, and the initial step length is adjusted according to the current error to generate the current step length, and the difference between the initial step length and the current step length is recorded as the adjacent error; and the initial weight is adjusted according to the current step length to generate the current weight, and the radar feature vector is obtained according to the current weight and the filter output result; it is repeated times until the generation radar feature vectors;

[0073] S3. According to radar feature vectors, generating target nodes, and define the edge weights of adjacent target nodes based on the adjacent errors. Each target node calculates the attention weight according to the radar feature vectors and edge weights of its adjacent nodes to determine the contribution of neighboring nodes to the update of its own radar feature vector. According to the contribution, the radar feature vector of each target node is updated. The radar feature vector is used as the final extracted target feature.

[0074] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and descriptions are only preferred examples of the present invention and are not intended to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A bird-detecting radar data processing system optimized for small targets, characterized in that: It comprises a data acquisition module (100), an adaptive filtering module (200) and a graph neural network processing module (300), wherein: The data acquisition module (100) obtains the expected data of the bird detection radar and radar data, of which Refers to the number of radar data obtained from the radar equipment; The adaptive filtering module (200) sets an initial step size and an initial weight, and uses a filter to sequentially The radar data is filtered and output as the filter output result; the current error is determined according to the filter output result and the expected data of the bird detection radar, and the initial step length is adjusted according to the current error to generate the current step length, and the difference between the initial step length and the current step length is recorded as the adjacent error; and the initial weight is adjusted according to the current step length to generate the current weight, and the radar feature vector is obtained according to the current weight and the filter output result; it is repeated times until the generation radar feature vectors; among them: The adaptive filtering module (200) comprises an initialization unit (201), wherein the initialization unit (201) sets an initial step length. and initial weights , where the initial step size determines the speed of each weight adjustment; the initial weight determines the state of the filter when processing the initial input; The adaptive filtering module (200) comprises a feature vector generating unit (202), wherein the feature vector generating unit (202) generates The process of generating a radar feature vector specifically includes: Use filters to The radar data is filtered and output as the filter output result ,in , , yes Radar data in radar data; Determine the current error based on the filter output and the expected data of the bird detection radar ,in , Indicates the expected data of bird detection radar; error Represents the difference between the filter output result and the expected data of the bird detection radar. The goal of adaptive filtering is to improve the accuracy of the output signal by minimizing this error; According to the current error , adjust the initial step size to generate the current step size ,but ,in and is the adjustment parameter, is the signal-to-noise ratio; Record the difference between the initial step length and the current step length as the adjacent error ,in ; Based on the current step length , for the initial weight Make adjustments to generate the current weight ,in ;According to the current weight and filter output results, the radar feature vector is obtained ,in ; Repeat the above process times, until The radar data is iterated and generated radar feature vectors; The graph neural network processing module (300) is based on radar feature vectors, generating target nodes, and define the edge weights of adjacent target nodes based on the adjacent errors. Each target node calculates the attention weight according to the radar feature vectors and edge weights of its adjacent nodes to determine the contribution of neighboring nodes to the update of its own radar feature vector. According to the contribution, the radar feature vector of each target node is updated. The radar feature vector is used as the final extracted target feature.

2. The bird-detecting radar data processing system optimized for small targets according to claim 1 is characterized in that: The graph neural network processing module (300) comprises a graph construction unit (301), and the process of defining the edge of the target node based on the adjacent error of the graph construction unit (301) specifically comprises: definition The radar feature vector is ,in ; Each radar feature vector Each of them is a node in the graph neural network ,generate Target Nodes ; Defining two adjacent target nodes based on neighboring errors and The edge weight between them is given by the formula , after defining edge weights based on adjacent errors, the edge set is generated , where each edge Indicates the target node and the target node The edge between them has a weight equal to the adjacent error ,but .

3. The bird-detecting radar data processing system optimized for small targets according to claim 2 is characterized in that: The graph neural network processing module (300) comprises a graph processing unit (302), and the process of the graph processing unit (302) updating the radar feature vector of each target node is specifically as follows: To determine each target node The radar characteristic vector The contribution size of each adjacent node during update, the edge set The adjacent error in Normalize and get the attention weight between the target node and its adjacent nodes ; According to the calculated attention weight , determine the contribution of neighboring nodes to the update of the radar feature vector of the target node, and perform weighted summation of the feature vectors of all neighboring nodes of each target node. The formula is as follows: ,in Indicates the target node The set of adjacent node subscripts, Is the target node For the target node The attention weight, is the updated target node The radar feature vector of According to the calculated attention weight and the feature vectors of the adjacent nodes, the radar feature vector of each target node is updated. The updated radar feature vector is the new radar feature vector of each node after integrating the information of its neighboring nodes. The radar feature vector is used as the final extracted target feature.

4. A method for using a bird-detecting radar data processing system optimized for small targets as claimed in any one of claims 1 to 3, characterized in that: The method comprises the following steps: S1. Obtain the expected data of bird detection radar and radar data, of which Refers to the number of radar data obtained from the radar equipment; S2, set the initial step size and initial weight, use the filter to The radar data is filtered and output as the filter output result; the current error is determined according to the filter output result and the expected data of the bird detection radar, and the initial step length is adjusted according to the current error to generate the current step length, and the difference between the initial step length and the current step length is recorded as the adjacent error; And according to the current step size, the initial weight is adjusted to generate the current weight, and the radar feature vector is obtained according to the current weight and the filter output result; iterate repeatedly times until the generation radar feature vectors; S3. According to radar feature vectors, generating target nodes, and define the edge weights of adjacent target nodes based on the adjacent errors. Each target node calculates the attention weight according to the radar feature vectors and edge weights of its adjacent nodes to determine the contribution of neighboring nodes to the update of its own radar feature vector. According to the contribution, the radar feature vector of each target node is updated. The radar feature vector is used as the final extracted target feature.

Images