Radar target identification method and device based on high-resolution one-dimensional range profile views

Abstract

The invention discloses a radar target recognition method and device based on a high-resolution one-dimensional range profile viewer. The method comprises the following steps: extracting a target time sequence in a preprocessed radar high-resolution one-dimensional range profile by using a noise threshold; obtaining a node set according to the amplitude value in the target time sequence; traversing slopes of a node i and a node j in the node set to obtain an edge set corresponding to the nodes in the target time sequence; wherein the range of the node i is 0 to L-1, the range of the node j is i + 1 to L-1, and L is the number of nodes in the node set; updating the node set for multiple times according to the edge set to obtain a plurality of node sets with different updating times; and performing graph classification according to the plurality of node sets with different updating times to obtain a classification result so as to realize target identification of the radar high-resolution one-dimensional range profile. The method has higher recognition accuracy, and improves the target recognition performance of a high-resolution radar system.

Description

Technical Field

[0001] The invention belongs to the field of radar target recognition, and in particular relates to a radar target recognition method and device based on a high-resolution one-dimensional range image visual graph. Background Art

[0002] With the continuous development of radar technology, the ability to accurately identify targets is of vital importance in many fields such as military and civilian use. The radar high-resolution one-dimensional range image reflects the distribution of scattering points of the target in the direction of the radar line of sight, contains rich target structure information, and has become an important research direction in the field of target recognition. In the modern complex electromagnetic environment and diverse target scenes, traditional low-resolution radars are difficult to meet the needs of high-precision target recognition. High-resolution radars can obtain more detailed target features, and high-resolution one-dimensional range images have attracted much attention in target recognition applications due to their relatively simple data acquisition and high processing efficiency.

[0003] However, the traditional method of directly extracting the structural features of the target's one-dimensional range image using a convolutional network or a temporal network can only obtain the target's spatial structure from a local or sequential relationship. When using a convolutional network to process a one-dimensional range image, the convolution window determines that it can only capture part of the input data. As the number of network layers increases, local features may be gradually transmitted, but the integration of global features may become inadequate; when using a temporal network to process high-resolution one-dimensional range images, the structure of the target may not be a simple temporal sequence relationship, but a nonlinear and spatial structure involving multiple feature dimensions.

[0004] From the perspective of exploring the structural information in the target range image as much as possible, the traditional method of directly extracting features from time series has a fixed relative position of the time series, and the convolutional network extraction will result in a single window scanning range during extraction. The sequential scanning of the time series network is difficult to capture the nonlinear relationship between different parts of the target. For example, there is still room for improvement in extracting features from each structural position point and adjacent position points of the aircraft. Summary of the invention

[0005] In order to solve the above problems existing in the prior art, the present invention provides a radar target recognition method and device based on a high-resolution one-dimensional range image visual graph.

[0006] The technical problem to be solved by the present invention is achieved through the following technical solutions:

[0007] In a first aspect, the present invention provides a radar target recognition method based on a high-resolution one-dimensional range image visual graph, the method comprising:

[0008] The target time series in the preprocessed radar high-resolution one-dimensional range image is extracted using noise threshold;

[0009] Obtain a node set based on the amplitude values in the target time series;

[0010] Traverse the slopes of nodes i and j in the node set to obtain an edge set corresponding to the nodes in the target time series; wherein, the range of the node i is from 0 to L - 1, the range of the node j is from i + 1 to L - 1, and L is the number of nodes in the node set;

[0011] Update the node set multiple times according to the edge set to obtain multiple node sets with different update times;

[0012] Perform graph classification according to multiple node sets with different update times to obtain a classification result, so as to realize target recognition of the radar high-resolution one-dimensional range profile.

[0013] Optionally, the method for using a noise threshold to extract the target time series in the preprocessed radar high-resolution one-dimensional range profile includes:

[0014] Perform amplitude normalization on the radar high-resolution one-dimensional range profile to obtain the preprocessed radar high-resolution one-dimensional range profile;

[0015] Obtain the noise threshold according to the preprocessed radar high-resolution one-dimensional range profile;

[0016] Use the noise threshold to search the preprocessed radar high-resolution one-dimensional range profile from front to back for the first range cell greater than the noise threshold as the starting cell;

[0017] Use the noise threshold to search the preprocessed radar high-resolution one-dimensional range profile from back to front for the first range cell greater than the noise threshold as the ending cell;

[0018] Extract the target time series according to the starting cell and the ending cell.

[0019] Optionally, the noise threshold is expressed as follows:

[0020]

[0021] wherein, threshold represents the noise threshold, K is a preset coefficient, x norm = [x norm,1 , x norm,2 ,..., x norm,n ,..., x norm,N represents the preprocessed radar high-resolution one-dimensional range profile, a, b ∈ [0, N], a < b, and N is the number of range cells in the preprocessed radar high-resolution one-dimensional range profile.

[0022] Optionally, traversing the slopes of nodes i and j in the node set to obtain an edge set corresponding to the nodes in the target time series includes:

[0023] When traversing, if the current slope value of node i and node j in the node set corresponding to the current moment is greater than the slope threshold, an undirected edge is added between node i and node j at the current moment, and the slope threshold is updated to the current slope value, and the traversal continues;

[0024] After the traversal is completed, the edge set consisting of the undirected edges between the node i and the node j at each time is obtained.

[0025] Optionally, the updating the node set multiple times according to the edge set to obtain multiple updated node sets includes:

[0026] Performing word embedding processing on the node set to obtain the node set after word embedding;

[0027] Get the neighbor node v of any node v in the node set after word embedding neighbor The node importance of the node v; wherein the neighbor node v of the node v neighbor Obtained according to the node v and the edge set;

[0028] Obtaining a normalized attention coefficient corresponding to the node v according to the node importance and the softmax function;

[0029] Update the node v according to the attention coefficient to obtain the updated node v;

[0030] The updated node v is updated again until a preset update number is reached, so as to obtain a plurality of nodes v with different update numbers, and further obtain a plurality of node sets with different update numbers.

[0031] Optionally, the neighbor node v neighbor The node importance of the node v is expressed as follows:

[0032] e neighbor =LeakyReLU(a T [Wv||Wv neighbor ]);

[0033] Among them, e neighbor Represents the neighbor node v neighbor The node importance of the node v, LeakyReLU is the activation function, a T is the first learnable parameter, and W is the second learnable parameter.

[0034] Optionally, the normalized attention coefficient corresponding to the node v is expressed as follows:

[0035]

[0036] Among them, α neighbor represents the normalized attention coefficient corresponding to the node v, exp is an exponential function, M neighbor Represents the neighbor node set of the node v, v neighbor ' are other neighbor nodes of the node v.

[0037] Optionally, the performing graph classification according to the node sets with multiple different update times includes:

[0038] Inputting the node sets with multiple different update times into the graph pooling layer respectively to obtain the node sets with multiple pooled node sets with different update times;

[0039] The node sets after multiple pooling with different update times are input into a fully connected layer to obtain the classification result.

[0040] In a second aspect, the present invention provides a radar target recognition device based on a high-resolution one-dimensional range image visual map, the device comprising:

[0041] A sequence extraction module is used to extract the target time series in the preprocessed radar high-resolution one-dimensional range image by using a noise threshold;

[0042] A first set acquisition module, used to obtain a node set according to the amplitude value in the target time series;

[0043] A second set acquisition module is used to traverse the slopes of nodes i and j in the node set to obtain the edge set corresponding to the nodes in the target time series; wherein the range of node i is 0 to L-1, the range of node j is i+1 to L-1, and L is the number of nodes in the node set;

[0044] An updating module, used for updating the node set multiple times according to the edge set to obtain the node sets with multiple different update times;

[0045] The recognition module is used to perform graph classification according to the node sets with different update times to obtain classification results, so as to realize target recognition of the radar high-resolution one-dimensional range image.

[0046] The technical solution provided by the embodiments of the present invention may have the following beneficial effects:

[0047] In the above technical solution, by constructing a node set and an edge set according to the target time series, the obtained visible graph structure is classified, so as to extract features at a deeper level. Compared with the traditional method of directly extracting features from time series through convolutional networks or temporal networks, it has a higher recognition accuracy and improves the target recognition performance of the high-resolution radar system.

[0048] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 It is a flow chart of a radar target recognition method based on a high-resolution one-dimensional range image visual graph provided by an embodiment of the present invention;

[0050] Figure 2a is a schematic diagram of a high-resolution one-dimensional range image before extraction provided by an embodiment of the present invention;

[0051] Figure 2b is a schematic diagram of a high-resolution one-dimensional range image after extraction provided by an embodiment of the present invention;

[0052] Figure 3 is a flow chart of constructing an edge set proposed by an embodiment of the present invention;

[0053] Figure 4 It is a flow chart of graph classification proposed in an embodiment of the present invention;

[0054] Figure 5 is a comparison chart of classification accuracy of different models proposed in an embodiment of the present invention;

[0055] Figure 6 This is a comparison chart of classification accuracy of different graph structures proposed in an embodiment of the present invention;

[0056] Figure 7 The invention discloses a block diagram of a radar target recognition device based on a high-resolution one-dimensional range image visualization diagram. DETAILED DESCRIPTION

[0057] In order to facilitate the understanding of the scheme of the present invention, the relevant state of the prior art and the inventive concept of the present invention are first briefly described.

[0058] The traditional method of directly extracting the structural features of the target's high-resolution one-dimensional range image using convolutional networks or temporal networks can only obtain the target's spatial structure from local or sequential relationships. When using convolutional networks to process high-resolution one-dimensional range images, the convolution window determines that it can only capture part of the information of the input data. As the number of network layers increases, local features may be gradually transmitted, but the integration of global features may become stretched; when using temporal networks to process high-resolution one-dimensional range images, the structure of the target may not be a simple temporal order relationship, but a nonlinear and spatial structure involving multiple feature dimensions. By constructing the high-resolution one-dimensional range image as a visible graph structure, each scattering point of the target can be regarded as a node attribute in the graph, and the connection edge between each two nodes can represent the physical or spatial relationship between them. The scattering points of some targets may appear as a regular structure in the range image, such as a specific shape or symmetry. These local regularities can be captured and transmitted to the entire graph through the connection relationship of the graph structure, thereby providing rich target structure information for target recognition. And through the graph structure, the graph neural network can integrate global information and propagate information through the connection between nodes, thereby capturing the global scattering characteristics of the target. The connection relationship between different scattering points can reflect the relationship between the various parts of the target structure. In this way, the graph neural network can more comprehensively understand the global features of the target such as structure and shape, thereby improving the recognition accuracy. Therefore, according to the above-mentioned inventive concept, the present invention proposes a radar target recognition method based on a high-resolution one-dimensional range image visual graph.

[0059] Figure 1 is a flow chart of a radar target recognition method based on a high-resolution one-dimensional range image visual graph provided by an embodiment of the present invention, such as Figure 1 As shown, the method may include the following steps:

[0060] S101, extracting the target time series in the preprocessed radar high-resolution one-dimensional range image by using a noise threshold.

[0061] Optionally, S101 may include:

[0062] The radar high-resolution one-dimensional range image is amplitude normalized to obtain the preprocessed radar high-resolution one-dimensional range image;

[0063] The noise threshold is obtained according to the preprocessed radar high-resolution one-dimensional range image;

[0064] The noise threshold is used to search the preprocessed radar high-resolution one-dimensional range image from front to back for the first range unit that is larger than the noise threshold, which is used as the starting unit.

[0065] The noise threshold is used to search the preprocessed radar high-resolution one-dimensional range image from back to front for the first range unit that is larger than the noise threshold, which is taken as the end unit.

[0066] The target time series is extracted based on the starting unit and the ending unit.

[0067] It is understandable that before performing target recognition on the radar high-resolution one-dimensional range image, a preprocessing operation is required because the one-dimensional range image has amplitude sensitivity, translation sensitivity and azimuth sensitivity. Amplitude sensitivity refers to the degree of dependence of the one-dimensional range image on the amplitude change of the target reflection signal. Even for the same radar and the same target, the echo amplitude may have certain differences. For a high-resolution one-dimensional range image sample x with a length of N s =[x 1 ,x 2 ,...,x N ], use the two-norm to normalize the amplitude to eliminate this difference, and get the high-resolution one-dimensional distance image after preprocessing:

[0068]

[0069] In addition, Figure 2a is a schematic diagram of a high-resolution one-dimensional range image before extraction provided by an embodiment of the present invention, Figure 2b is a schematic diagram of a high-resolution one-dimensional range image after extraction provided by an embodiment of the present invention, such as Figure 2a and Figure 2b As shown in the figure, translation sensitivity refers to the impact of the target moving along the distance in the physical space on the one-dimensional range image. When the target displacement is greater than the distance resolution unit, the target peak will jump from one unit to another, resulting in uneven range images of the same target. Considering the influence of the noise area, if traditional methods such as centroid alignment are used to align the one-dimensional range image without removing the noise, the noise will also be constructed as a node in the construction of the graph structure, which will cause interference from useless signals and a huge amount of calculation. The noise threshold refers to setting a suitable threshold by the amplitude of the noise. When the signal amplitude of the range unit is greater than the threshold, the unit is determined to be the target range image, thereby extracting the unit where the target is located, eliminating translation sensitivity, and removing the noise area.

[0070] Optionally, the noise threshold is expressed as follows:

[0071]

[0072] Where threshold represents the noise threshold, K is the preset coefficient, and x norm =[x norm,1 ,x norm,2 ,...,x norm,n ,...,x norm,Nrepresents the preprocessed radar high-resolution one-dimensional range profile. Let \(a,b\in[0,N]\) where \(a < b\), and \(N\) is the number of range bins in the preprocessed radar high-resolution one-dimensional range profile.

[0073] It is worth mentioning that azimuth sensitivity refers to the sensitivity of the high-resolution one-dimensional range profile to the change in the relative azimuth angle between the target and the radar. There are significant differences in the high-resolution one-dimensional range profiles obtained by observing the target from multiple angles. Therefore, the present invention uses data that covers as many azimuth angles as possible to observe the target for training, enabling the model to learn the target azimuth information at different angles and enhancing the azimuth generalization ability of the model.

[0074] S102. Obtain a node set according to the amplitude values in the target time series.

[0075] It can be understood that a target time series corresponds to a graph sample, and a graph sample is composed of a node set and an edge set. In this section, the target time series is to be transformed into a graph sample. Among them, the node set \(y = [y 1 ,y 2 ,...,y L is the amplitude value of all data points in the target time series.

[0076] S103. Traverse the slopes of nodes \(i\) and \(j\) in the node set to obtain the edge set corresponding to the nodes in the target time series. Among them, the range of node \(i\) is from 0 to \(L - 1\), and the range of node \(j\) is from \(i + 1\) to \(L - 1\), where \(L\) is the number of nodes in the node set.

[0077] Optionally, S103 may include:

[0078] When traversing, if the current slope values of nodes \(i\) and \(j\) in the node set corresponding to the current moment are greater than the slope threshold, an undirected edge is added between nodes \(i\) and \(j\) at the current moment, and the slope threshold is updated to the current slope value, and the traversal continues;

[0079] After the traversal ends, an edge set composed of undirected edges between nodes \(i\) and \(j\) at each moment is obtained.

[0080] It can be understood that Figure 3 is a flowchart for constructing an edge set proposed in an embodiment of the present invention. As shown in Figure 3As shown, num is the target time series, node i and node j are two pointers, len(num) is the length of the target time series, slope is the slope threshold, and snowSlope is the current slope value; the traditional method first calculates the straight line of two data points on the time series, and then traverses the middle point between the two points. If there is a middle point with an amplitude higher than the straight line, the two points are determined to be invisible. The present invention traverses and calculates the slope of node i and node j, and saves the current slope value of the two points as the current maximum slope value slope. As node j continues to traverse backward, it may encounter a larger snowSlope. When snowSlope is greater than slope, it is determined that node i and node j are visible, and slope is updated. An undirected edge is connected to node i and node j in the edge set of the graph; otherwise, node i and node j are determined to be invisible.

[0081] S104. Update the node set multiple times according to the edge set to obtain multiple node sets with different update times.

[0082] Optionally, S104 may include:

[0083] Perform word embedding processing on the node set to obtain the word embedded node set;

[0084] Get the neighbor node v of any node v in the node set after word embedding neighbor The node importance to node v; among them, the neighbor node v of node v neighbor Obtained according to the node v and edge set;

[0085] According to the node importance and softmax function, the normalized attention coefficient corresponding to node v is obtained;

[0086] Update node v according to the attention coefficient to obtain the updated node v;

[0087] The updated node v is updated again until the preset update times are reached, thereby obtaining multiple nodes v with different update times, and further obtaining multiple node sets with different update times.

[0088] It can be understood that at this point, the amplitude values ​​of all data points of a target time series are taken as a node set, and the visible graph is constructed to obtain an edge set. With the node set and the edge set, a high-resolution one-dimensional distance image is constructed into a graph sample, and the graph classification task can be used to extract and classify its features. The obtained graph sample is sent to the graph neural network for message propagation and aggregation, and finally the extracted features are sent to the classifier for graph classification. The present invention uses a graph attention network to implement the graph classification task. The node set y of the target time series = [y 1 ,y 2 ,...,y L]After word embedding, each point amplitude value will be mapped into a high-dimensional vector, and the node set V after word embedding is obtained. 0 ,v 1 ,v 2 ,...,v M ] T , where M is the number of nodes, and the vector of the amplitude value of each point after word embedding is m=1,2,...,M, their dimension is the size F of word embedding dimension.

[0089] The attention mechanism is an unsupervised self-learning weight coefficient calculation mechanism. It can be used to explain message propagation and aggregation by taking a certain amplitude node in the target time series as an example. For a certain node v and its neighboring nodes on a target time series, the node v is first linearly transformed to convert it into a higher-level feature, and then the attention mechanism a is used to calculate the attention coefficient. The neighboring node v neighbor The node importance of node v is expressed as follows:

[0090] e neighbor =LeakyReLU(a T [Wv||Wv neighbor ]);

[0091] Among them, e neighbor Represents the neighbor node v neighbor The node importance of node v, LeakyReLU is the activation function, a T is the first learnable parameter, and W is the second learnable parameter.

[0092] Optionally, the normalized attention coefficient corresponding to node v is expressed as follows:

[0093]

[0094] Among them, α neighbor represents the normalized attention coefficient corresponding to node v, exp is the exponential function, M neighbor Represents the set of neighbor nodes of node v, v neighbor ' are other neighbor nodes of node v.

[0095] The updated node v can be expressed as:

[0096]

[0097] It is worth mentioning that v' represents the updated node v. At this time, the node v completes a feature extraction and updates the value of the vector. This node v' contains the neighbor node information and becomes the new feature value. This operation must be performed on all nodes in the graph sample.

[0098] S105, performing graph classification according to a plurality of node sets with different update times to obtain classification results, so as to realize target recognition of radar high-resolution one-dimensional range image.

[0099] Optionally, S105 may include:

[0100] Input multiple node sets with different update times into the graph pooling layer respectively to obtain multiple pooled node sets with different update times;

[0101] Multiple pooled node sets with different update times are input into the fully connected layer to obtain the classification results.

[0102] Understandably, Figure 4 It is a flowchart of a graph classification proposed in an embodiment of the present invention. Through the attention mechanism, the attention is focused as much as possible on the information useful for extracting target structural features in each distance unit information, so that the model converges in the right direction. After the message propagation and aggregation are completed, the nodes in the graph sample save the characteristic values, and it is necessary to extract and integrate a global feature. Since the present invention adopts a graph classification task, it is also necessary to add a pooling layer and an activation function layer to the graph sample after message propagation and aggregation. The pooling layer can downsample the nodes of a graph sample, and the number of nodes will become smaller, but the connection relationship between the edges of the nodes can retain the original structure. The present invention introduces the TopKPooling layer as a pooling layer to reduce the scale of the graph and extract the global information of the graph, such as Figure 4 As shown in the figure. After each convolution operation, the TopKPooling layer selects the nodes with the top ratio and their corresponding edge information for retention based on the importance score of the node features, thus realizing the downsampling operation of the graph data. This not only helps to reduce the amount of calculation, but also can retain the key information in the graph to a certain extent, so that the model can focus on more representative nodes and structures for subsequent learning and feature extraction processes.

[0103] After multiple layers of convolution, pooling and readout operations, the model integrates the features extracted from the graph data at different levels through a global average pooling operation to obtain multiple fixed-dimensional feature vectors, and sums these vectors to fuse multi-scale information. Subsequently, these fused features are further transformed and mapped through two fully connected layers in turn. Between the fully connected layers, the introduced ReLU activation function introduces nonlinear factors to the model, enhances the model's expressiveness, and enables it to learn more complex data patterns. The last layer of the model is a fully connected layer, which maps the input features to the final output dimension and normalizes them through a softmax function to obtain the predicted probability distribution of each category. This design enables the model to predict multi-classification tasks based on the input graph data and output the logarithmic probability value of each sample belonging to a different category, so that the cross entropy loss function indicator can be used for optimization and performance evaluation in the subsequent training and evaluation process.

[0104] In one embodiment, Figure 5 This is a comparison chart of the classification accuracy of different models proposed in an embodiment of the present invention. This example uses high-resolution one-dimensional range image data of three types of aircraft to compare with the traditional deep learning network model. Under the same measured data, the accuracy of CNN, RNN, LSTM and the method proposed in the present invention is shown in Figure 2. Figure 5 As shown; Figure 6 This is a comparison chart of the classification accuracy of different graph structures proposed in an embodiment of the present invention. This example uses the same graph neural network model and different graph structures to construct graphs for three types of aircraft to obtain a comparison of classification performance. Under the same measured data, the comparison results of the classification performance between different graph structures are as follows: Figure 6 shown.

[0105] Figure 7 is a block diagram of a radar target recognition device based on a high-resolution one-dimensional range image visual graph proposed in an embodiment of the present invention, such as Figure 7 As shown, the apparatus 700 may include:

[0106] A sequence extraction module 701 is used to extract the target time sequence in the preprocessed radar high-resolution one-dimensional range image by using a noise threshold;

[0107] A first set acquisition module 702, configured to obtain a node set according to the amplitude value in the target time series;

[0108] The second set acquisition module 703 is used to traverse the slopes of nodes i and j in the node set to obtain the edge set corresponding to the nodes in the target time series; wherein the range of node i is 0 to L-1, the range of node j is i+1 to L-1, and L is the number of nodes in the node set;

[0109] An updating module 704 is used to update the node set multiple times according to the edge set to obtain multiple node sets with different update times;

[0110] The identification module 705 is used to perform graph classification according to a plurality of node sets with different update times to obtain classification results, so as to realize target recognition of radar high-resolution one-dimensional range image.

[0111] It should be noted that the terms "first", "second", etc. are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Instead, they are merely examples of devices and methods consistent with some aspects of the present invention.

[0112] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification.

[0113] Although the present invention is described herein in conjunction with various embodiments, in the process of implementing the claimed invention, those skilled in the art can understand and implement other changes to the disclosed embodiments by viewing the drawings and the disclosed content. In the description of the present invention, the term "comprising" does not exclude other components or steps, "one" or "an" does not exclude multiple situations, and "multiple" means two or more, unless otherwise clearly and specifically limited. In addition, certain measures are recorded in different embodiments, but this does not mean that these measures cannot be combined to produce good results.

[0114] As for the device / electronic device / storage medium embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0115] It should be noted that the device of the embodiment of the present invention is a device that applies the above-mentioned radar target recognition method based on a high-resolution one-dimensional range image visible graph, so all embodiments of the above-mentioned radar target recognition method based on a high-resolution one-dimensional range image visible graph are applicable to the device and can achieve the same or similar beneficial effects.

[0116] The above contents are further detailed descriptions of the present invention in combination with specific preferred embodiments, and it cannot be determined that the specific implementation of the present invention is limited to these descriptions. For ordinary technicians in the technical field to which the present invention belongs, several simple deductions or substitutions can be made without departing from the concept of the present invention, which should be regarded as falling within the protection scope of the present invention.

Claims

1. A radar target recognition method based on a high-resolution one-dimensional range image visualization diagram, characterized in that: The method comprises: The target time series in the preprocessed radar high-resolution one-dimensional range image is extracted using noise threshold; Obtaining a node set according to the amplitude value in the target time series; Traversing the slopes of nodes i and j in the node set, obtaining the edge set corresponding to the nodes in the target time series; wherein the range of node i is 0 to L-1, the range of node j is i+1 to L-1, and L is the number of nodes in the node set; The node set is updated multiple times according to the edge set to obtain the node sets with multiple different update times; Graph classification is performed according to the node sets with multiple different update times to obtain classification results, so as to realize target recognition of the radar high-resolution one-dimensional range image.

2. The radar target recognition method based on high-resolution one-dimensional range image visualization according to claim 1, characterized in that: The method of extracting the target time series in the preprocessed radar high-resolution one-dimensional range image by using the noise threshold comprises: Normalizing the amplitude of the radar high-resolution one-dimensional range image to obtain the preprocessed radar high-resolution one-dimensional range image; Obtaining the noise threshold according to the preprocessed radar high-resolution one-dimensional range image; Using the noise threshold, searching the preprocessed radar high-resolution one-dimensional range image from front to back for the first range unit that is greater than the noise threshold as a starting unit; Using the noise threshold, searching the preprocessed radar high-resolution one-dimensional range image from back to front for the first range unit that is greater than the noise threshold as the end unit; The target time series is extracted according to the starting unit and the ending unit.

3. The radar target recognition method based on high-resolution one-dimensional range image visualization according to claim 2 is characterized in that: The noise threshold is expressed as follows: wherein, threshold represents the noise threshold, K is a preset coefficient, and x norm = [x norm,1 , x norm,2 ,..., x norm,n ,..., x norm,N represents the preprocessed one-dimensional high-resolution range profile of the radar, a, b ∈ [0, N], a < b, and N is the number of range cells in the preprocessed one-dimensional high-resolution range profile of the radar.

4. The radar target recognition method based on high-resolution one-dimensional range image visualization according to claim 1, characterized in that: The traversing the slopes of nodes i and j in the node set to obtain the edge set corresponding to the nodes in the target time series includes: When traversing, if the current slope value of node i and node j in the node set corresponding to the current moment is greater than the slope threshold, an undirected edge is added between node i and node j at the current moment, and the slope threshold is updated to the current slope value, and the traversal continues; After the traversal is completed, the edge set consisting of the undirected edges between the node i and the node j at each time is obtained.

5. The radar target recognition method based on high-resolution one-dimensional range image visualization according to claim 1, characterized in that: The step of updating the node set multiple times according to the edge set to obtain multiple updated node sets includes: Performing word embedding processing on the node set to obtain the node set after word embedding; Get the neighbor node v of any node v in the node set after word embedding neighbor The node importance of the node v; wherein the neighbor node v of the node v neighbor Obtained according to the node v and the edge set; Obtaining a normalized attention coefficient corresponding to the node v according to the node importance and the softmax function; Update the node v according to the attention coefficient to obtain the updated node v; The updated node v is updated again until a preset update number is reached, so as to obtain a plurality of nodes v with different update numbers, and further obtain a plurality of node sets with different update numbers.

6. The radar target recognition method based on high-resolution one-dimensional range image visualization according to claim 5, characterized in that: The neighbor node v neighbor The node importance of the node v is expressed as follows: e neighbor =LeakyReLU(a T [Wv||Wv neighbor ]); Among them, e neighbor Represents the neighbor node v neighbor The node importance of the node v, LeakyReLU is the activation function, a T is the first learnable parameter, and W is the second learnable parameter.

7. The radar target recognition method based on high-resolution one-dimensional range image visual graph according to claim 6 is characterized in that: The normalized attention coefficient corresponding to the node v is expressed as follows: Among them, α neighbor represents the normalized attention coefficient corresponding to the node v, exp is an exponential function, M neighbor Represents the neighbor node set of the node v, v neighbor ' are other neighbor nodes of the node v.

8. The radar target recognition method based on high-resolution one-dimensional range image visualization according to claim 6 is characterized in that: The graph classification according to the node sets with multiple different update times includes: Inputting the node sets with multiple different update times into the graph pooling layer respectively to obtain the node sets with multiple pooled node sets with different update times; The node sets after multiple pooling with different update times are input into a fully connected layer to obtain the classification result.

9. A radar target recognition device based on a high-resolution one-dimensional range image visualization diagram, characterized in that: The device comprises: A sequence extraction module is used to extract the target time series in the preprocessed radar high-resolution one-dimensional range image by using a noise threshold; A first set acquisition module, used to obtain a node set according to the amplitude value in the target time series; A second set acquisition module is used to traverse the slopes of nodes i and j in the node set to obtain the edge set corresponding to the nodes in the target time series; wherein the range of node i is 0 to L-1, the range of node j is i+1 to L-1, and L is the number of nodes in the node set; An updating module, used for updating the node set multiple times according to the edge set to obtain the node sets with multiple different update times; The recognition module is used to perform graph classification according to the node sets with different update times to obtain classification results, so as to realize target recognition of the radar high-resolution one-dimensional range image.

Images