Object hierarchical identification method and identification device based on confusion matrix

Abstract

The invention relates to the technical field of radar signal recognition, and discloses a target hierarchical recognition method and recognition device based on a confusion matrix, and the method comprises the steps: determining the confusion matrix of a to-be-recognized radar signal; generating a directed initial graph with a self-loop based on the confusion matrix; performing bidirectional edge processing on the directed initial graph with the self-loop to generate a processed graph model; generating an initial correct classification list based on the processed graph model; carrying out self-loop removal processing on the processed graph model to generate a self-loop-free undirected graph; and based on the self-loop-free undirected graph and the initial correct classification list, multi-target secondary identification operation is performed on the radar signal to be identified, and a corresponding identification result is generated. A traditional target identification method is improved, and a directed graph with a self-loop is generated by using a confusion matrix to perform confusion distinguishing and identification between target labels, so that the accuracy of secondary identification is effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of radar signal recognition, and in particular to a target hierarchical recognition method based on a confusion matrix and a target hierarchical recognition device based on a confusion matrix. Background Art

[0002] With the rapid development of machine learning and artificial intelligence, classification models have become an important tool for solving multi-target recognition tasks. The confusion matrix is ​​used to evaluate the accuracy of the classification model, display the classification results of the model in a matrix, and compare the model's prediction results with the actual labels.

[0003] Target recognition confusion refers to the situation that during the target classification process, due to the similarity between targets, target 1 may be identified as target 2, or some targets that should belong to different categories may be mistakenly classified into the same category. There are many reasons for target recognition confusion. In order to solve this technical problem, the technical problem of target recognition confusion is solved by analyzing the confusion matrix and improving the model. However, in the actual application process, the technicians found that the existing technology has at least the following technical problems:

[0004] In the existing classification model, when there are fewer classification categories, the target classification effect is better, but when the target category is uncertain, the features are difficult to extract, and there is a lot of noise data, the recognition effect is poor. For example, for radar signals, it has the characteristics of low signal-to-noise ratio, difficult data collection, large interference in echo information, and difficulty in extracting echo information features, which greatly increases the difficulty of multi-target recognition. The target confusion problem cannot be solved by conventional model optimization methods, so it cannot meet actual needs. Summary of the invention

[0005] In order to overcome the above-mentioned technical problems existing in the prior art, an embodiment of the present invention provides a target hierarchical recognition method and a recognition device based on a confusion matrix, which improves the traditional target recognition method and uses the confusion matrix to generate a directed graph with self-loops to perform confusion distinction and recognition between target labels, thereby effectively improving the accuracy of secondary recognition of radar signals.

[0006] In order to achieve the above-mentioned purpose, an embodiment of the present invention provides a target hierarchical recognition method based on a confusion matrix, the method comprising: determining a confusion matrix of a radar signal to be identified; generating a directed initial graph with self-loops based on the confusion matrix; performing bidirectional edge processing on the directed initial graph with self-loops to generate a processed graph model; generating an initial correct classification list based on the processed graph model; performing self-loop removal processing on the processed graph model to generate an undirected graph without self-loops; performing a multi-target secondary recognition operation on the radar signal to be identified based on the undirected graph without self-loops and the initial correct classification list to generate a corresponding recognition result.

[0007] Preferably, generating a directed initial graph with self-loops based on a confusion matrix comprises: obtaining a target label of each radar signal to be identified from the confusion matrix, and generating a vertex set based on the target label; obtaining confused target pairs with confusion from the confusion matrix, and determining a directed edge set based on the confused target pairs; determining a corresponding confusion weight set based on the confused target pairs; and generating a corresponding directed initial graph with self-loops based on the vertex set, the directed edge set, and the confusion weight set.

[0008] Preferably, the bidirectional edge processing of the directed initial graph with self-loops to generate a processed graph model includes: performing unidirectional edge mapping processing on the bidirectional edges in the directed initial graph with self-loops to generate a first processed graph; and optimizing the first processed graph based on the confusion weight set to generate a processed graph model.

[0009] Preferably, performing unidirectional edge mapping processing on the bidirectional edges in the directed initial graph with self-loops to generate a first processed graph includes: obtaining a weight pair corresponding to the confused target pair based on the confusion weight set; judging whether the confused target pair meets the preset unidirectional edge requirement based on the weight pair and a preset empirical threshold; if so, generating a unidirectional edge based on the confused target pair, generating a unidirectional weight of the unidirectional edge based on the weight pair, adjusting the directed initial graph with self-loops based on the unidirectional edge and the unidirectional weight to generate a first adjusted graph; otherwise, deleting the bidirectional edges corresponding to the confused target pair from the directed initial graph with self-loops to generate a second adjusted graph; and using the first adjusted graph or the second adjusted graph as the first processed graph.

[0010] Preferably, the performing a multi-target secondary recognition operation on the radar signal to be identified based on the self-loop-free undirected graph and the initial correct classification list to generate a corresponding recognition result includes: determining the adjacent points of any vertex in the self-loop-free undirected graph; determining the degrees of all vertices based on the adjacent points of each vertex, and generating a vertex set arranged in descending order of degree according to the degrees of all vertices; performing vertex analysis processing on each vertex in the vertex set in turn based on the initial correct classification list to generate a processing result; and updating the initial correct classification list based on the processing result to generate a recognition result for the radar signal to be identified.

[0011] Preferably, the vertex analysis processing is performed on each vertex in the vertex set in turn based on the initial correct classification list to generate a processing result, including: obtaining the adjacent points of the current vertex in turn; obtaining the indirect adjacent points of the adjacent points; judging whether the current vertex is adjacent to the indirect adjacent points; if so, generating a first correct classification value of the current vertex; otherwise, obtaining the edge weights between all points adjacent to the adjacent points, judging whether the edge weights between the current vertex and the adjacent points are greater than the remaining edge weights; if so, generating a second correct classification value of the current vertex; otherwise, generating a third correct classification value of the current vertex; and determining the first correct classification value or the second correct classification value or the third correct classification value as the processing result.

[0012] Correspondingly, the present invention also provides a target hierarchical identification device based on a confusion matrix, the device comprising: a matrix determination unit, used to determine the confusion matrix of a radar signal to be identified; a graph generation unit, used to generate a directed initial graph with self-loops based on the confusion matrix; a first processing unit, used to perform bidirectional edge processing on the directed initial graph with self-loops to generate a processed graph model; a table generation unit, used to generate an initial correct classification list based on the processed graph model; a second processing unit, used to perform self-loop removal processing on the processed graph model to generate an undirected graph without self-loops; a secondary identification unit, used to perform a multi-target secondary identification operation on the radar signal to be identified based on the undirected graph without self-loops and the initial correct classification list to generate a corresponding identification result.

[0013] Preferably, the graph generating unit is specifically used to: obtain a target label of each radar signal to be identified from the confusion matrix, and generate a vertex set based on the target label; obtain confused target pairs with confusion from the confusion matrix, and determine a directed edge set based on the confused target pairs; determine a corresponding confusion weight set based on the confused target pairs; and generate a corresponding directed initial graph with self-loops based on the vertex set, the directed edge set, and the confusion weight set.

[0014] Preferably, the first processing unit is specifically used to: perform unidirectional edge mapping processing on the bidirectional edges in the directed initial graph with self-loops to generate a first processed graph; optimize the first processed graph based on the confusion weight set to generate a processed graph model.

[0015] Preferably, the secondary identification unit is specifically used to: determine the adjacent points of any vertex in the self-loop-free undirected graph; determine the degrees of all vertices based on the adjacent points of each vertex, and generate a vertex set arranged in descending order of degree according to the degrees of all vertices; perform vertex analysis processing on each vertex in the vertex set in turn based on the initial correct classification list to generate a processing result; update the initial correct classification list based on the processing result to generate an identification result for the radar signal to be identified.

[0016] Through the technical solution provided by the present invention, the present invention has at least the following technical effects:

[0017] By improving the traditional method, on the basis of the confusion matrix used in the existing technology, further graph processing is performed on it, a self-loop and directed graph model is established, and the relationship between target labels is optimized, so as to achieve better label distinction effect, effectively improve the secondary recognition accuracy of the target, and meet the actual needs.

[0018] Other features and advantages of the embodiments of the present invention will be described in detail in the subsequent detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The accompanying drawings are used to provide a further understanding of the embodiments of the present invention and constitute a part of the specification. Together with the following specific implementations, they are used to explain the embodiments of the present invention, but do not constitute a limitation on the embodiments of the present invention. In the accompanying drawings:

[0020] Figure 1 It is a specific implementation flow chart of the target hierarchical recognition method based on confusion matrix provided by an embodiment of the present invention;

[0021] Figure 2 This is a specific implementation flow chart of generating a directed initial graph with self-loops provided by an embodiment of the present invention;

[0022] Figure 3 is a schematic diagram of a directed initial graph with a self-loop provided by an embodiment of the present invention;

[0023] Figure 4 The embodiment of the present invention provides Schematic diagram of

[0024] Figure 5 is an undirected graph without self-loops provided by an embodiment of the present invention Schematic diagram of

[0025] Figure 6 It is a specific implementation flow chart of performing multi-target secondary recognition to generate recognition results provided by an embodiment of the present invention;

[0026] Figure 7It is a structural diagram of a target hierarchical recognition device based on a confusion matrix provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0027] The specific implementation of the embodiment of the present invention is described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation described here is only used to illustrate and explain the embodiment of the present invention, and is not used to limit the embodiment of the present invention.

[0028] The terms "system" and "network" in the embodiments of the present invention can be used interchangeably. "Multiple" means two or more than two. In view of this, "multiple" can also be understood as "at least two" in the embodiments of the present invention. "And / or" describes the association relationship of associated objects, indicating that three relationships can exist. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / ", unless otherwise specified, generally indicates that the previous and subsequent associated objects are in an "or" relationship. In addition, it should be understood that in the description of the embodiments of the present invention, the words "first", "second", etc. are only used to distinguish the purpose of description, and cannot be understood as indicating or implying relative importance, nor can they be understood as indicating or implying order.

[0029] See also Figure 1 , an embodiment of the present invention provides a target hierarchical recognition method based on a confusion matrix, the method comprising:

[0030] S10) determining a confusion matrix of the radar signal to be identified;

[0031] S20) generating a directed initial graph with self-loops based on the confusion matrix;

[0032] S30) performing bidirectional edge processing on the directed initial graph with self-loops to generate a processed graph model;

[0033] S40) generating an initial correct classification list based on the processed graph model;

[0034] S50) performing self-loop removal processing on the processed graph model to generate an undirected graph without self-loops;

[0035] S60) performing a multi-target secondary recognition operation on the radar signal to be recognized based on the self-loop-free undirected graph and the initial correct classification list to generate a corresponding recognition result.

[0036] In a possible implementation, a confusion matrix of the radar signal to be identified is first generated. For example, in one embodiment, 16 radar signals need to be identified, and the 16 radar signals have a certain degree of similarity, which is easy to cause confusion. Therefore, a confusion matrix of the above 16 radar signals is first created. Then, a directed initial graph with self-loops is generated based on the confusion matrix. See Figure 2 In an embodiment of the present invention, the step of generating a directed initial graph with self-loops based on a confusion matrix includes:

[0037] S21) obtaining a target label of each radar signal to be identified from the confusion matrix, and generating a vertex set based on the target label;

[0038] S22) determining at least one confused signal group with confusion based on the confusion matrix, and determining a directed edge set based on the vertex set and the confused signal group;

[0039] S23) determining a corresponding confusion weight set based on the confusion signal group;

[0040] S24) generating a corresponding directed initial graph with self-loops based on the vertex set, the directed edge set and the confusion weight set.

[0041] In the specific drawing process, first obtain the target label of each radar signal to be identified from the confusion matrix. For example, the 16 radar signals mentioned above are numbered 1-16, and the numbers are the target labels of the corresponding radar signals. At this time, the target labels are mapped to the vertices on the graph, and then a directed relationship between the target labels is established. Specifically, if there is confusion between two radar signals, the group of confused signals is connected with a directed edge, and the probability of target v being identified as target u is determined as the weight of the directed edge, thereby generating a directed edge set. For example, see Figure 3 , which is a schematic diagram of a directed initial graph with a self-loop provided by an embodiment of the present invention. In the schematic diagram, the directed initial graph with a self-loop is characterized as , where V is the graph The vertex set of the target label is mapped into a directed graph The vertex set of If there is confusion between two targets, they are connected by directed edges. The edge vu indicates that v is the starting point and u is the end point. L is the weight set on the edge, and the weight L(vu) on the edge vu is equal to the probability that the target v is recognized as the target u.

[0042] In an embodiment of the present invention, by converting an abstract confusion matrix into an intuitive directed initial graph with self-loops, the abstract mathematical problem is converted into a visual image problem, so that the confusion between targets can be better distinguished and processed from another dimension, thereby achieving a better recognition effect.

[0043] At this time, the above directed initial graph with self-loops is further processed.

[0044] In an embodiment of the present invention, the bidirectional edge processing of the directed initial graph with self-loops to generate a processed graph model includes: performing unidirectional edge mapping processing on the bidirectional edges in the directed initial graph with self-loops to generate a first processed graph; and optimizing the first processed graph based on the confusion weight set to generate a processed graph model.

[0045] In one possible implementation, if There are bidirectional edges uv and vu between the two target vertices in >0 and , then the two bidirectional edges are processed, specifically, the bidirectional edges are mapped into unidirectional edges. Specifically, in an embodiment of the present invention, the bidirectional edges in the directed initial graph with self-loops are mapped into unidirectional edges to generate a first processed graph, including: obtaining a weight pair corresponding to the confusion signal group based on the confusion weight set; judging whether the confusion signal group meets the preset unidirectional requirement based on the weight pair and a preset empirical threshold; if so, generating a unidirectional edge based on the confusion signal group, generating a unidirectional weight of the unidirectional edge based on the weight pair, adjusting the directed initial graph with self-loops based on the unidirectional edge and the unidirectional weight to generate a first adjusted graph; otherwise, deleting the bidirectional edge corresponding to the confusion signal group from the directed initial graph with self-loops to generate a second adjusted graph; and using the first adjusted graph or the second adjusted graph as the first processed graph.

[0046] Specifically, = , and then obtain the preset experience threshold , and further verify whether the above obfuscation target pair meets the preset unilateral requirements. Specifically, if , then the bidirectional edge is mapped to a unidirectional edge , whose edge weight is ; Otherwise, delete uv and vu, and the generated graph model is recorded as .

[0047] In the embodiment of the present invention, on the one hand, since the confusion matrix can provide some data feature information of target similarity, on the other hand, human experience has a stronger guiding role for the characteristics of radar signals such as high confidentiality, strong coupling and non-cooperation, so by using the work experience of technicians combined with the similar data characteristics of the confusion matrix, the confusion relationship between different radar signals with confusion characteristics is screened, so that different radar signals with weak similarity can be effectively distinguished, and reliable distinction and identification of some similar radar signals can be achieved. On this basis, hierarchical identification is performed on the remaining radar signals with extreme similarity.

[0048] Further, the first processed graph is optimized to generate a processed graph model. Specifically, if the graph side The weight on If yes, keep it; otherwise delete it. , see Figure 4 , which is a diagram provided by an embodiment of the present invention Schematic diagram of .

[0049] In an embodiment of the present invention, by processing the bidirectional edges in the directed initial graph with self-loops, the association relationship between the target labels is greatly optimized, and at the same time, the distinguishing characteristics between the target labels are more prominent, which greatly reduces the complexity of subsequent analysis and processing and improves the accuracy and efficiency of analysis and identification.

[0050] At this time, the generated graph model After the above processing, it can be seen that the relationship between each target label has been greatly simplified and effectively distinguished, so the graph model can be used to Generate an initial correct classification list. Specifically, combined with the embodiment of the present invention, observe the figure If the vertex still retains a self-loop, put its vertex number into the list Obviously, all vertices in DM have self-loops, and their edge weights are , determine the target Identify The result is correct. Identified as correctly classified targets. That is, the vertices in DM0 are correctly classified targets judged by the confusion matrix.

[0051] On the other hand, the processed graph model is processed to remove self-loops to generate an undirected graph without self-loops. Specifically, the graph is first deleted. Then, all directed edges are converted to undirected edges, and the weight of each undirected edge remains unchanged, that is, , and finally generate the corresponding undirected graph without self-loops as the base graph for subsequent secondary recognition. For example, see Figure 5 , which is an undirected graph without self-loops provided by an embodiment of the present invention Schematic diagram of .

[0052] See also Figure 6 In an embodiment of the present invention, the performing of a multi-target secondary recognition operation on the radar signal to be recognized based on the self-loop-free undirected graph and the initial correct classification list to generate a corresponding recognition result includes:

[0053] S61) determining adjacent points of any vertex in the self-loop-free undirected graph;

[0054] S62) determining the degrees of all vertices based on the adjacent points of each vertex, and generating a vertex set arranged in descending order of degree according to the degrees of all vertices;

[0055] S63) performing vertex analysis processing on each vertex in the vertex set in turn based on the initial correct classification list to generate a processing result;

[0056] S64) updating the initial correct classification list based on the processing result to generate a recognition result for the radar signal to be recognized.

[0057] In a possible implementation, firstly, the neighboring points of any vertex in the undirected graph without self-loops are determined. For example, in an undirected graph without self-loops, if there is an edge connecting two vertices, then the two vertices are determined to be adjacent. For any vertex v, the set of neighboring nodes of v is recorded as Based on this, the total number of neighbors of a vertex is determined as the degree of the vertex , sort the vertex set from large to small according to the size of the degree, and the formed vertex set is denoted as D, and then further vertex analysis is performed.

[0058] In an embodiment of the present invention, vertex analysis processing is performed on each vertex in the vertex set in turn based on the initial correct classification list to generate a processing result, including: obtaining the adjacent points of the current vertex in turn; obtaining the indirect adjacent points of the adjacent points; judging whether the current vertex is adjacent to the indirect adjacent points; if so, generating a first correct classification value of the current vertex; otherwise, obtaining the edge weights between all points adjacent to the adjacent points, judging whether the edge weights between the current vertex and the adjacent points are greater than the remaining edge weights; if so, generating a second correct classification value of the current vertex; otherwise, generating a third correct classification value of the current vertex; processing the first correct classification value or the second correct classification value or the third correct classification value based on the initial correct classification list to generate a processing result.

[0059] In a specific embodiment, first take the first vertex in D, assuming it is v, and let .consider ,like If v has only one neighbor, then ; If u has other neighbors o besides neighbor v, that is, determine the indirect neighbor points. If o is also adjacent to v, then generate the first correct classification value, that is, ; If o and v are not adjacent, then judge Is the edge weight of The largest of the associated edges, if so, then the second correct classification value is generated, that is, , otherwise the third correct classification value is generated, that is, , according to the above method traverse Then update all the vertices in and D, delete the classified vertices from D, and then repeat the above process until D is an empty set, that is, the analysis and processing of all vertices in D is completed, and the final secondary recognition result is output. For example, in one embodiment, The results of the secondary identification are represented as: {2, 3, 4, 8, 11, 12, 13, 14}; {9, 16}; {7, 10}; {5}; {6}; {1}; {15}, thereby achieving accurate classification and identification of the above 16 radar signals.

[0060] In an embodiment of the present invention, by adopting a multi-target recognition technology based on a confusion graph, the self-loops in a directed graph with self-loops are used to distinguish which results in the target classification results are correct and which results are wrong, and optimization processing is performed on them. On this basis, the confusion graph is used to accurately classify targets with less similarity into categories with greater similarity, thereby achieving better confusion distinction and greatly improving the accuracy of target recognition.

[0061] The following describes the target hierarchical recognition device based on confusion matrix provided by the embodiment of the present invention in conjunction with the accompanying drawings.

[0062] See also Figure 7 Based on the same inventive concept, an embodiment of the present invention provides a target hierarchical identification device based on a confusion matrix, the device comprising: a matrix determination unit, used to determine the confusion matrix of a radar signal to be identified; a graph generation unit, used to generate a directed initial graph with self-loops based on the confusion matrix; a first processing unit, used to perform bidirectional edge processing on the directed initial graph with self-loops to generate a processed graph model; a table generation unit, used to generate an initial correct classification list based on the processed graph model; a second processing unit, used to perform self-loop removal processing on the processed graph model to generate an undirected graph without self-loops; a secondary identification unit, used to perform a multi-target secondary identification operation on the radar signal to be identified based on the undirected graph without self-loops and the initial correct classification list to generate a corresponding identification result.

[0063] In an embodiment of the present invention, the graph generating unit is specifically used to: obtain a target label of each radar signal to be identified from the confusion matrix, and generate a vertex set based on the target label; determine at least one group of confused signal groups with confusion based on the confusion matrix, and determine a directed edge set based on the vertex set and the confused signal group; determine a corresponding confusion weight set based on the confused signal group; and generate a corresponding directed initial graph with self-loops based on the vertex set, the directed edge set, and the confusion weight set.

[0064] In an embodiment of the present invention, the first processing unit is specifically used to: perform unidirectional edge mapping processing on the bidirectional edges in the directed initial graph with self-loops to generate a first processed graph; and optimize the first processed graph based on the confusion weight set to generate a processed graph model.

[0065] In an embodiment of the present invention, the secondary identification unit is specifically used to: determine the adjacent points of any vertex in the self-loop-free undirected graph; determine the degrees of all vertices based on the adjacent points of each vertex, and generate a vertex set arranged in descending order of degree according to the degree of all vertices; perform vertex analysis processing on each vertex in the vertex set in turn based on the initial correct classification list to generate a processing result; update the initial correct classification list based on the processing result to generate an identification result for the radar signal to be identified.

[0066] The optional implementation modes of the embodiments of the present invention are described in detail above in conjunction with the accompanying drawings. However, the embodiments of the present invention are not limited to the specific details in the above implementation modes. Within the technical concept of the embodiments of the present invention, various simple modifications can be made to the technical scheme of the embodiments of the present invention, and these simple modifications all belong to the protection scope of the embodiments of the present invention.

[0067] It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. To avoid unnecessary repetition, the embodiments of the present invention will not further describe various possible combinations.

[0068] Those skilled in the art can understand that all or part of the steps in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a program, and the program is stored in a storage medium, including a number of instructions to enable a single-chip microcomputer, a chip or a processor to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0069] In addition, various implementation modes of the embodiments of the present invention may be arbitrarily combined, and as long as they do not violate the concept of the embodiments of the present invention, they should also be regarded as the contents disclosed by the embodiments of the present invention.

Claims

1. A target hierarchical recognition method based on confusion matrix, characterized in that: The method comprises: Determine a confusion matrix of the radar signal to be identified; Generate an initial directed graph with self-loops based on the confusion matrix; Performing bidirectional edge processing on the directed initial graph with self-loops to generate a processed graph model; generating an initial correct classification list based on the processed graph model; Performing self-loop removal processing on the processed graph model to generate an undirected graph without self-loops; Based on the self-loop-free undirected graph and the initial correct classification list, a multi-target secondary recognition operation is performed on the radar signal to be recognized to generate a corresponding recognition result.

2. The method according to claim 1, characterized in that The generating of a directed initial graph with self-loops based on a confusion matrix includes: Obtaining a target label of each radar signal to be identified from the confusion matrix, and generating a vertex set based on the target label; Determine at least one confused signal group with confusion based on the confusion matrix, and determine a directed edge set based on the vertex set and the confused signal group; Determining a corresponding confusion weight set based on the confusion signal group; A corresponding directed initial graph with self-loops is generated based on the vertex set, the directed edge set and the confusion weight set.

3. The method according to claim 2, characterized in that The step of performing bidirectional edge processing on the directed initial graph with self-loops to generate a processed graph model includes: Performing unidirectional edge mapping processing on the bidirectional edges in the directed initial graph with self-loops to generate a first processed graph; The first processed graph is optimized based on the confusion weight set to generate a processed graph model.

4. The method according to claim 3, characterized in that The step of performing unidirectional edge mapping processing on the bidirectional edges in the directed initial graph with self-loops to generate a first processed graph includes: Acquire a weight pair corresponding to the confusion signal group based on the confusion weight set; Determining whether the confusion signal group meets the preset unilateral requirement based on the weight pair and the preset empirical threshold; If yes, generate a unidirectional edge based on the confusion signal group, generate a unidirectional weight of the unidirectional edge based on the weight pair, and adjust the directed initial graph with self-loops based on the unidirectional edge and the unidirectional weight to generate a first adjusted graph; Otherwise, deleting the bidirectional edges corresponding to the obfuscated signal group from the directed initial graph with self-loops to generate a second adjusted graph; The first adjustment image or the second adjustment image is used as a first processed image.

5. The method according to claim 1, characterized in that The performing a multi-target secondary recognition operation on the radar signal to be identified based on the self-loop-free undirected graph and the initial correct classification list to generate a corresponding recognition result includes: Determine adjacent points of any vertex in the self-loop-free undirected graph; Determine the degree of all vertices based on the adjacent points of each vertex, and generate a set of vertices arranged in descending order according to the degree of all vertices; Based on the initial correct classification list, perform vertex analysis processing on each vertex in the vertex set in turn to generate a processing result; The initial correct classification list is updated based on the processing result to generate a recognition result for the radar signal to be recognized.

6. The method according to claim 5, characterized in that The step of performing vertex analysis processing on each vertex in the vertex set in turn based on the initial correct classification list to generate a processing result includes: Get the adjacent points of the current vertex in sequence; Obtain indirect neighboring points of the neighboring point; Determine whether the current vertex is adjacent to the indirect adjacent point; If so, generate the first correct classification value of the current vertex; Otherwise, obtain the edge weights between all points adjacent to the adjacent point, and determine whether the edge weight between the current vertex and the adjacent point is greater than the remaining edge weights; If yes, generate the second correct classification value of the current vertex; Otherwise, generate the third correct classification value of the current vertex; The first correct classification value, the second correct classification value, or the third correct classification value is processed based on the initial correct classification list to generate a processing result.

7. A target hierarchical recognition device based on confusion matrix, characterized in that: The device comprises: A matrix determination unit, used for determining a confusion matrix of a radar signal to be identified; A graph generation unit, used for generating a directed initial graph with self-loops based on a confusion matrix; A first processing unit is used to perform bidirectional edge processing on the directed initial graph with self-loops to generate a processed graph model; A table generating unit, configured to generate an initial correct classification list based on the processed graph model; A second processing unit is used to perform self-loop removal processing on the processed graph model to generate an undirected graph without self-loops; The secondary recognition unit is used to perform a multi-target secondary recognition operation on the radar signal to be recognized based on the self-loop-free undirected graph and the initial correct classification list to generate a corresponding recognition result.

8. The device according to claim 7, characterized in that The graph generating unit is specifically used for: Obtaining a target label of each radar signal to be identified from the confusion matrix, and generating a vertex set based on the target label; Obtaining confused target pairs with confusion from the confusion matrix, and determining a directed edge set based on the confused target pairs; Determining a corresponding confusion weight set based on the confusion target pair; A corresponding directed initial graph with self-loops is generated based on the vertex set, the directed edge set and the confusion weight set.

9. The device according to claim 8, characterized in that The first processing unit is specifically configured to: Performing unidirectional edge mapping processing on the bidirectional edges in the directed initial graph with self-loops to generate a first processed graph; The first processed graph is optimized based on the confusion weight set to generate a processed graph model.

10. The device according to claim 7, characterized in that The secondary identification unit is specifically used for: Determine adjacent points of any vertex in the self-loop-free undirected graph; Determine the degree of all vertices based on the adjacent points of each vertex, and generate a set of vertices arranged in descending order according to the degree of all vertices; Based on the initial correct classification list, perform vertex analysis processing on each vertex in the vertex set in turn to generate a processing result; The initial correct classification list is updated based on the processing result to generate a recognition result for the radar signal to be recognized.

Images