STGCN-based space target attitude anomaly detection method and system
Through the spatial target pose anomaly detection method based on STGCN, the problem of difficult to identify non-cooperative target dynamic determination in the prior art without prior knowledge is solved, and higher model generalization ability and abnormal dynamic recognition accuracy are achieved.
Patent Information
- Application Number
- CN202510395193.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-31
- Publication Date
- 2025-06-06
AI Technical Summary
Existing dynamic monitoring methods are difficult to effectively identify dynamic determination problems of non-cooperative goals without prior knowledge, which limits the generalization ability of the model and the reliability of dynamic identification of abnormal abnormalities in unknown spatial targets.
The spatial target pose abnormal detection method based on STGCN is adopted to capture the deep spatial target information by obtaining spatial target key point information, graph structure modeling, and constructing a spatiotemporal graph convolutional neural network module.
It improves the generalization ability of the model and the accuracy of abnormal movement recognition, can effectively capture the deep-level spatial and temporal information of the target, and improves the reliability of radar target abnormal movement recognition technology.
Smart Images
Figure CN120103339A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of image processing technology, and in particular to a method and system for detecting abnormal posture of space targets based on STGCN. Background Art
[0002] Many satellite dynamic monitoring methods are based on high-resolution ground radar data with prior information. In the absence of prior knowledge, the existing dynamic monitoring methods often have difficulty solving the problem of dynamic determination of non-cooperative targets. This not only limits the generalization ability of the model, but also affects the reliability and effectiveness of abnormal dynamic identification of unknown space targets. Summary of the invention
[0003] The main purpose of the embodiments of the present application is to provide a method and system for detecting anomalies in the posture of space targets based on STGCN.
[0004] The technical solution adopted by the present invention is:
[0005] On the one hand, an embodiment of the present invention provides a method for detecting anomalies in a space target posture based on STGCN, and the method for detecting anomalies in a space target posture based on STGCN comprises the following steps:
[0006] Obtain key point information of space targets;
[0007] Perform graph structure modeling according to the key point information of the space target to obtain graph structure data;
[0008] Build a spatiotemporal graph convolutional neural network module;
[0009] Acquire spatial information and temporal information between key points of spatial targets according to the graph structure data and the spatiotemporal graph convolutional neural network module;
[0010] According to the spatial information and time information between the key points of the space target, detection information of abnormal posture of the space target is obtained.
[0011] Furthermore, the obtaining of key point information of the space target comprises the following steps:
[0012] Acquire inverse synthetic aperture radar images;
[0013] A plurality of key point information of the space target is extracted from the inverse synthetic aperture radar image to obtain the key point information of the space target.
[0014] Furthermore, the graph structure modeling is performed according to the key point information of the space target to obtain graph structure data, including the following steps:
[0015] According to the key point information of the space target, each key point is used as a graph node; the graph node includes the two-dimensional coordinates of the key point in the range-Doppler coordinate system;
[0016] defining graph edges between the graph nodes according to the shape and structure of the space object;
[0017] According to the graph nodes and the graph edges, graph structure modeling is completed to obtain graph structure data.
[0018] Furthermore, the construction of the spatiotemporal graph convolutional neural network module includes the following steps:
[0019] Constructing a graph convolutional network submodule; the graph convolutional network submodule is used to extract spatial features of the target;
[0020] Constructing a time domain convolutional network submodule; the time domain convolutional network submodule is used to aggregate the time features of the target;
[0021] The graph convolutional network submodule and the time domain convolutional network submodule are cascaded to obtain a spatiotemporal graph convolutional neural network module.
[0022] Furthermore, the obtaining of spatial information and time information between key points of spatial targets according to the graph structure data and the spatiotemporal graph convolutional neural network module comprises the following steps:
[0023] According to the graph structure data and the spatiotemporal graph convolutional neural network module, using a graph convolutional network submodule to capture spatial features between graph nodes;
[0024] According to the graph structure data and the spatiotemporal graph convolutional neural network module, aggregating the time features between graph nodes using a time domain convolutional neural network submodule;
[0025] According to the spatial features between the graph nodes and the temporal features between the graph nodes, the spatial information and the temporal information between the key points of the spatial object are obtained.
[0026] Furthermore, the method of aggregating the time features between graph nodes using the time domain convolutional neural network submodule according to the graph structure data and the spatiotemporal graph convolutional neural network module includes the following steps:
[0027] According to the graph structure data and the spatiotemporal graph convolutional neural network module, using a time domain convolutional neural network submodule to capture graph node information in different time frames;
[0028] According to the graph node information, the temporal features between the graph nodes are aggregated by applying multiple layers of convolution kernels.
[0029] Furthermore, the STGCN-based space target posture anomaly detection method further includes the following steps:
[0030] According to the spatial information and time information between the key points of the space target, the spatiotemporal graph convolutional neural network module performs binary classification judgment on the normal and abnormal states of the space target.
[0031] On the other hand, an embodiment of the present invention further provides a space target posture anomaly detection system based on STGCN, which is used to implement the space target posture anomaly detection method based on STGCN as described above, and the space target posture anomaly detection system based on STGCN includes:
[0032] The first module is used to obtain key point information of space targets;
[0033] The second module is used to perform graph structure modeling according to the key point information of the space target and obtain graph structure data;
[0034] The third module is used to build a spatiotemporal graph convolutional neural network module;
[0035] A fourth module is used to obtain spatial information and temporal information between key points of spatial targets according to the graph structure data and the spatiotemporal graph convolutional neural network module;
[0036] The fifth module is used to obtain detection information of abnormal posture of the space target based on the spatial information and time information between the key points of the space target.
[0037] On the other hand, an embodiment of the present invention also provides a space target posture anomaly detection device based on STGCN, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the space target posture anomaly detection method based on STGCN as described above.
[0038] On the other hand, an embodiment of the present invention further provides a computer-readable storage medium, which stores computer-executable instructions, and the computer-executable instructions are used to enable a computer to execute the STGCN-based spatial target posture anomaly detection method as described above.
[0039] The embodiments of the present application include at least the following beneficial effects: The present application provides a method and system for detecting abnormal posture of space targets based on STGCN. The present invention can obtain key point information of space targets; perform graph structure modeling based on key point information of space targets to obtain graph structure data; construct a spatiotemporal graph convolutional neural network module; obtain spatial information and temporal information between key points of space targets based on graph structure data and spatiotemporal graph convolutional neural network module; obtain detection information of abnormal posture of space targets based on spatial information and temporal information between key points of space targets. The present invention can capture deep spatiotemporal information in targets and improve the reliability of radar target movement recognition technology. BRIEF DESCRIPTION OF THE DRAWINGS
[0040] Figure 1 is a schematic diagram of a method for detecting abnormal posture of a space target based on STGCN provided by an embodiment of the present invention;
[0041] Figure 2 It is a schematic diagram of the STGCN network architecture provided by an embodiment of the present invention;
[0042] Figure 3 is a schematic diagram of a space satellite target and a graph topology structure provided by an embodiment of the present invention;
[0043] Figure 4 is a schematic diagram of a TCN network structure provided by an embodiment of the present invention;
[0044] Figure 5 is a schematic diagram of the Aura satellite image topology structure provided by an embodiment of the present invention;
[0045] Figure 6 It is a schematic diagram of the accuracy of spatial target movement recognition of the STGCN model after adding noise to the data set provided by an embodiment of the present invention;
[0046] Figure 7 It is a schematic diagram of the spatial target movement recognition accuracy of the STGCN model when the number of data frames in each group of data in the data set is reduced to 3 frames, provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0047] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application is further described in detail below in conjunction with the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the embodiments of the present application. They are only examples of devices and methods consistent with some aspects of the embodiments of the present application as detailed in the attached claims.
[0048] It is understood that the terms "first", "second", etc. used in this application can be used to describe various concepts in this article, but unless otherwise specified, these concepts are not limited by these terms. These terms are only used to distinguish one concept from another concept. For example, without departing from the scope of the embodiment of the present application, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the words "if" and "if" as used herein can be interpreted as "at the time of" or "when" or "in response to determination".
[0049] The terms "at least one", "multiple", "each", "any", etc. used in this application, at least one includes one, two or more, multiple includes two or more, each refers to each of the corresponding multiple, and any refers to any one of the multiple.
[0050] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.
[0051] Before describing the embodiments of the present application in detail, some nouns and terms involved in the embodiments of the present application are first described. The nouns and terms involved in the embodiments of the present application are subject to the following explanations.
[0052] 1) STGCN (Spatio-Temporal Graph Convolutional Network), spatio-temporal graph convolutional network;
[0053] 2) ISAR (Inverse Synthetic Aperture Radar), a radar imaging technology;
[0054] 3) STGNN (Spatio-Temporal Graph Neural Network), spatio-temporal graph neural network;
[0055] 4) GCN (Graph Convolutional Network), graph convolutional network;
[0056] 5)TCN (Temporal Convolutional Network), temporal convolutional network;
[0057] 6) BN (Batch Normalization), batch normalization;
[0058] 7) ReLU (Rectified Linear Unit), activation function;
[0059] 8) Dropout, random dropout;
[0060] 9) DeepSpaceI, an unmanned space probe;
[0061] 10) Dragon, a reusable cargo / passenger spacecraft;
[0062] 11) Soyuz, a series of manned spacecraft;
[0063] 12) TianGongI, Tiangong-1;
[0064] 13) Aura, an Earth observation satellite;
[0065] 14) GNN (Graph Neural Network), graph neural network.
[0066] Many satellite dynamic monitoring methods are based on high-resolution ground radar data with prior information. In the absence of prior knowledge, the existing dynamic monitoring methods often have difficulty solving the problem of dynamic determination of non-cooperative targets. This not only limits the generalization ability of the model, but also affects the reliability and effectiveness of abnormal dynamic identification of unknown space targets.
[0067] In order to solve this problem, the present invention proposes an innovative solution: a method and system for detecting anomalies in the posture of space targets based on STGCN. The space-time graph (STGNN) is an advanced graph model that can capture the connections between objects contained in the image by modeling the image as a graph. Compared with directly using ISAR images for anomaly recognition, this method can not only flexibly model ISAR space targets, but also capture the temporal and spatial information of the targets through the temporal edges and spatial edges in the space-time graph, respectively. The graph convolutional neural network (GCN) designed by the present invention learns the information representation of nodes by propagating the intrinsic relationship of the graph structure, and can learn the relationship between nodes according to the topological structure of the graph to obtain feature representation.
[0068] Through this method, the present invention not only solves the problem of difficulty in identifying abnormal dynamics of unknown targets in the absence of prior knowledge, but also greatly improves the generalization ability of the model and the accuracy of abnormal motion recognition. The STGCN model can capture deep temporal and spatial information in the target and improve the reliability of radar target abnormal motion recognition technology.
[0069] The embodiments of the present invention are further described below in conjunction with the accompanying drawings.
[0070] On the one hand, the embodiment of the present invention provides a method for detecting abnormal posture of a space target based on STGCN, referring to Figure 1 ,The space target posture anomaly detection method based on STGCN includes the following steps:
[0071] S100, obtaining key point information of space targets;
[0072] S200, performing graph structure modeling according to key point information of the space target to obtain graph structure data;
[0073] S300, constructing a spatiotemporal graph convolutional neural network module;
[0074] S400, acquiring spatial information and temporal information between key points of spatial targets according to the graph structure data and the spatiotemporal graph convolutional neural network module;
[0075] S500: Acquire detection information of abnormal posture of the space target according to the spatial information and time information between key points of the space target.
[0076] S100 disclosed in the embodiment of the present invention obtains key point information of a space target, including the following steps:
[0077] S110, acquiring an inverse synthetic aperture radar image;
[0078] S120, extracting a plurality of key point information of the space target from the inverse synthetic aperture radar image to obtain the key point information of the space target.
[0079] As an optional implementation, the embodiment of the present invention acquires continuous frame images of the space target through an inverse synthetic aperture radar (ISAR), and these images contain multi-polarization and multi-viewing angle information of the target.
[0080] S200 disclosed in the embodiment of the present invention performs graph structure modeling according to key point information of the space target to obtain graph structure data, including the following steps:
[0081] S210, according to the key point information of the space target, taking each key point as a graph node; the graph node includes the two-dimensional coordinates of the key point in the range-Doppler coordinate system;
[0082] S220, defining graph edges between graph nodes according to the shape and structure of the space object;
[0083] S230: Complete graph structure modeling based on graph nodes and graph edges to obtain graph structure data.
[0084] As an optional implementation, the embodiment of the present invention models the key point information of the space target as graph structure data. Each key point is a node of the graph, and the connection (edge) between the nodes is defined according to the shape and structure of the target. The nodes in the graph structure contain the two-dimensional coordinates of the key point in the range-Doppler coordinate system.
[0085] S300 disclosed in the embodiment of the present invention constructs a spatiotemporal graph convolutional neural network module, comprising the following steps:
[0086] S310, constructing a graph convolutional network submodule; the graph convolutional network submodule is used to extract spatial features of the target;
[0087] S320, constructing a time domain convolutional network submodule; the time domain convolutional network submodule is used to aggregate the time features of the target;
[0088] S330, cascade the graph convolutional network submodule and the time domain convolutional network submodule to obtain a spatiotemporal graph convolutional neural network module.
[0089] As an optional implementation, the embodiment of the present invention uses a graph topology structure. GCN (graph convolutional network) captures the spatial information between key points through convolution kernels. GCN extracts the spatial features of the target by learning the relationship between nodes. TCN (time domain convolutional network) is used to capture the time information of the same key point in different time frames. Through the application of multi-layer convolution kernels, TCN can effectively aggregate time features and capture the time dynamic changes of the target.
[0090] S400 disclosed in the embodiment of the present invention obtains spatial information and time information between key points of spatial targets according to graph structure data and a spatiotemporal graph convolutional neural network module, including the following steps:
[0091] S410, capturing spatial features between graph nodes using a graph convolutional network submodule according to the graph structure data and the spatiotemporal graph convolutional neural network module;
[0092] S420, according to the graph structure data and the spatiotemporal graph convolutional neural network module, using the time domain convolutional neural network submodule to aggregate the time features between the graph nodes;
[0093] S430. Obtain spatial information and temporal information between key points of the spatial target according to the spatial features between the graph nodes and the temporal features between the graph nodes.
[0094] S420 disclosed in the embodiment of the present invention aggregates the time features between graph nodes using the time domain convolutional neural network submodule according to the graph structure data and the spatiotemporal graph convolutional neural network module, including the following steps:
[0095] S421, according to the graph structure data and the spatiotemporal graph convolutional neural network module, using the time domain convolutional neural network submodule to capture graph node information in different time frames;
[0096] S422. According to the graph node information, the time features between the graph nodes are aggregated by applying multi-layer convolution kernels.
[0097] The method for detecting abnormal posture of a space target based on STGCN disclosed in an embodiment of the present invention further includes the following steps:
[0098] S600: Based on the spatial information and time information between the key points of the space target, the spatiotemporal graph convolutional neural network module performs binary classification judgment on the normal and abnormal states of the space target.
[0099] On the other hand, an embodiment of the present invention further provides a space target posture anomaly detection system based on STGCN, which is used to implement the space target posture anomaly detection method based on STGCN as described above. The space target posture anomaly detection system based on STGCN includes:
[0100] The first module is used to obtain key point information of space targets;
[0101] The second module is used to perform graph structure modeling based on key point information of space targets and obtain graph structure data;
[0102] The third module is used to build a spatiotemporal graph convolutional neural network module;
[0103] The fourth module is used to obtain the spatial information and time information between the key points of the spatial target according to the graph structure data and the spatiotemporal graph convolutional neural network module;
[0104] The fifth module is used to obtain detection information of abnormal posture of space targets based on spatial information and time information between key points of space targets.
[0105] Inverse synthetic aperture radar (ISAR) imaging technology can obtain high-resolution scattering images of space targets. Combined with the optimized design of the system and working mode, ISAR has the ability to continuously measure space targets with high resolution in multi-polarization and multi-view. On-orbit status monitoring of space targets is a key technology for situational awareness of space targets and a prerequisite for ensuring the safe operation of our space targets. Through status monitoring, the "three-state" information of the target's dynamics, attitude and morphology can be obtained, and then the target's category, action intention and prediction can be obtained. Using ISAR imaging, the perception of abnormal dynamic states of space targets can be realized.
[0106] Since the Doppler resolution of normal moving targets shows a low-high-low trend, while the Doppler resolution of abnormal moving space targets is mainly determined by spin motion and remains basically unchanged in different orbital segments. Since the normal and abnormal motion of space targets show different Doppler resolution trends on the orbit, the spatiotemporal graph convolutional neural network (STGCN) can be used to solve the time series classification problem and realize abnormal motion recognition. Facing the urgent need for active intelligence and information acquisition of unknown space targets, the existing space target state monitoring methods have weak active information acquisition capabilities and basically no research on unknown target morphology, dynamics and posture analysis. A space target motion projection imaging model is established, combined with STGCN, to achieve active acquisition of dynamic information of non-cooperative targets.
[0107] As an optional implementation, the embodiment of the present invention proposes a method for identifying abnormalities based on STGCN, referring to the network architecture as Figure 2 As shown in the figure. The core architecture of the network, the spatiotemporal graph convolutional neural network (STGCN), consists of a cascade of graph convolutional neural networks (GCN) and time domain convolutional neural networks (TCN). The input of the STGCN network architecture is the information of ten key points of spatial targets in continuous frame ISAR images. First, the GCN network uses a custom topological graph structure and a convolution kernel to capture the spatial information between key points; secondly, the TCN uses three convolution kernels to integrate the features of the same node at different times to capture the temporal information of the same key point. The STGCN network architecture uses the cascade of the GCN network and the TCN network to achieve the binary classification task of judging the normal and abnormal spatial targets.
[0108] Space targets have unique structures, that is, they have stable shapes and components, so the overall structure of space targets can be represented by a graph. For space satellite targets, the graph topology structure customized by the embodiment of the present invention is as follows: Figure 3 As shown. The embodiment of the present invention defines ten key points of the space target as graph nodes, and defines graph edges according to the target shape and the position of the key points. In this way, the embodiment of the present invention constructs a graph structure data composed of graph nodes and graph edges to express space satellite targets of regular shapes, and converts the image into a graph. The embodiment of the present invention models the space satellite target in the ISAR image as a skeleton, and the skeleton framework is an array containing 10 tuples. The skeleton learns the information representation of the node by propagating the intrinsic relationship of the graph structure data, and can effectively capture the spatial information between the key points of the satellite target.
[0109] The graph nodes in the graph structure data include the two-dimensional coordinate (r, d) tuples of the ten key points of the space target in the range-Doppler coordinate system, where r represents the coordinate of the key point in the range dimension and d represents the coordinate of the key point in the Doppler dimension. ; Spatial edge E S It is the connection information of key points in the same frame.
[0110] E S = {v ti v tj |(i,j)∈G}
[0111] G=[(2,3),(3,4),(4,1),(1,2),(1,9),(4,9),(9,5),(9,8),(5,8),(5,6),
[0112] (8,7),(7,6),(9,10),(9,10),(1,1),(2,2),(3,3),(4,4),(5,5),
[0113] (6,6),(7,7),(8,8),(9,9),(10,10)]
[0114] Among them, E S represents the set of spatial edges; G represents the topological structure of the graph, defining the connection relationship between key points (graph nodes); v ti represents the i-th graph node in the t-th frame; v tj represents the j-th graph node in the t-th frame; (i, j) indicates that there is a spatial edge between the i-th graph node and the j-th graph node.
[0115] refer to Figure 4 The TCN network structure diagram includes: using BN (Batch Normalization) to batch normalize the input data; using ReLU (Rectified Linear Unit) activation function to enhance nonlinearity; extracting local features of the time dimension through the convolution kernel; BN again to normalize the convolution output twice; using Dropout to randomly mask some features to enhance generalization ability.
[0116] The embodiment of the present invention uses the TCN network structure to capture the node information of the same key point in different frames for the graph structure of the continuous multi-frame ISAR images obtained above. The application of the three-layer convolution kernel effectively aggregates the time features, can effectively capture the time information between the key points of the satellite target, and provides strong support for abnormal movement recognition.
[0117] Time Edge f Connection information for the same key point in different frames:
[0118] E f = {v ti v (t+1)i}
[0119] Among them, E f Represents a set of time edges, describing the temporal connections of graph nodes; v tirepresents the i-th graph node in the t-th frame; v (t+1)i The i-th graph node in the t+1th frame; v ti v (t+1)i Represents the node v of the tth frame ti and node v in frame t+1 (t+1)i The time between the edges.
[0120] As an optional implementation, the experimental results generated by the embodiment of the present invention are displayed as follows:
[0121] In order to verify the performance of STGCN, the embodiment of the present invention tested it on a data set containing key point data of DeepSpaceI, Dragon, Soyuz, TianGongI and Aura satellites. Each set of data in the data set contains 36 consecutive frames of ISAR images and the label (normal / abnormal) of each frame. The evaluation process of the embodiment of the present invention includes the accuracy of determining the normal and abnormal motion of space targets, the performance of the model under certain noise conditions, the performance of the model when some key point information is missing, and the accuracy of normal and abnormal motion determination when the number of input frames decreases. These extensive analyses will provide the embodiment of the present invention with a comprehensive evaluation of the performance of the model in abnormal motion recognition.
[0122] (1) Theoretical value accuracy
[0123] Table 1
[0124]
[0125] Referring to Table 1, in the dataset containing DeepSpaceI, Dragon, Soyuz, and TianGongI satellites, the STGCN model has a 100% accuracy rate in identifying anomalies.
[0126] (2) Aura Satellite
[0127] The Aura satellite contains only six key points. Figure 5 The key number of Aura satellites is shown in Figure 1. This modeling method does not increase the amount of information.
[0128] Table 2
[0129]
[0130] Referring to Table 2, in the datasets containing DeepSpaceI, Dragon, Soyuz, TianGongI and Aura satellites, the STGCN model still has a 100% accuracy rate in identifying abnormal movements, which indicates that the STGCN model has a strong ability to learn the spatial and temporal information of satellite target key points.
[0131] (3) Noise
[0132] refer to Figure 6 In the case where a certain amount of noise is added to the data set in the embodiment of the present invention, when the position of the key point deviates within 3 pixels, the accuracy of the model in identifying the movement of the space target is still 100%, indicating that the model has a certain tolerance rate to noise.
[0133] (4) Frame rate
[0134] refer to Figure 7 When the number of frames of each data group in the data set is reduced to 3 frames in the embodiment of the present invention, the accuracy of the model in identifying the movement of space targets is still 100%.
[0135] The key points of the present invention are:
[0136] 1. Compared with directly using ISAR images for anomaly recognition, the graph model proposed in this invention can model ISAR spatial targets more flexibly. GCN can learn the information representation of nodes by propagating the intrinsic relationship of the graph structure, and can learn the relationship between nodes according to the topological structure of the graph to obtain feature representation. The customized topological graph structure of this model can effectively capture the spatial information between key points.
[0137] 2. TCN is one of the key technologies for capturing temporal information of spatial targets. By temporally aggregating features through three layers of convolution kernels, the TCN model can effectively capture local temporal relationships in images. TCN strengthens the features of the temporal dimension in key points, making it easier to capture the temporal information contained in key points.
[0138] The STGCN model of the present invention is more accurate in describing the relationship between ISAR target image data. First, GNN can customize spatial edges according to the shape of the space target. GCN has a strong processing capability for graph structure data and has a strong topological relationship modeling capability. STGCN can capture the deep spatiotemporal relationship between target key points.
[0139] In addition, the application of convolution kernels in GCN and TCN networks can effectively capture and strengthen the local spatiotemporal relationship and key features of spatial targets in images. The present invention can still achieve good results when the data set contains noise, the number of key points in the data is missing, or there are a small number of image frames, and can realize the recognition of abnormal movement of spatial targets, effectively capture spatiotemporal information, and have a certain fault tolerance for the input key point data.
[0140] On the other hand, an embodiment of the present invention also provides a space target posture anomaly detection device based on STGCN, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the space target posture anomaly detection method based on STGCN as described above.
[0141] The processor and the memory can be connected via a bus or other means. The memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may optionally include a memory remotely arranged relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.
[0142] On the other hand, an embodiment of the present invention further provides a computer-readable storage medium, which stores computer-executable instructions, and the computer-executable instructions are used to enable a computer to execute the above-mentioned STGCN-based space target posture anomaly detection method.
[0143] It will be appreciated by those skilled in the art that all or some of the steps and systems in the disclosed method above may be implemented as software, firmware, hardware and appropriate combinations thereof. Some physical components or all physical components may be implemented as software executed by a processor, such as a central processing unit, a digital signal processor or a microprocessor, or may be implemented as hardware, or may be implemented as an integrated circuit, such as an application specific integrated circuit. Such software may be distributed on a computer-readable medium, which may include a computer storage medium (or a non-transitory medium) and a communication medium (or a temporary medium). As known to those skilled in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tapes, disk storage or other magnetic storage devices, or any other medium that may be used to store desired information and may be accessed by a computer. Furthermore, it is well known to those skilled in the art that communication media typically embodies computer readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and may include any information delivery media.
[0144] The preferred embodiments of the present invention are described above with reference to the accompanying drawings, but the scope of the rights of the present invention is not limited thereto. Any modification, equivalent substitution and improvement made by a person skilled in the art without departing from the scope and essence of the present invention should be within the scope of the rights of the present invention.
Claims
1. A method for detecting abnormal posture of space targets based on STGCN, characterized in that: The STGCN-based space target posture anomaly detection method comprises the following steps: Obtain key point information of space targets; Perform graph structure modeling according to the key point information of the space target to obtain graph structure data; Build a spatiotemporal graph convolutional neural network module; Acquire spatial information and temporal information between key points of spatial targets according to the graph structure data and the spatiotemporal graph convolutional neural network module; According to the spatial information and time information between the key points of the space target, detection information of abnormal posture of the space target is obtained.
2. The method for detecting abnormal posture of space targets based on STGCN according to claim 1 is characterized in that: The step of obtaining key point information of a space target comprises the following steps: Acquire inverse synthetic aperture radar images; A plurality of key point information of the space target is extracted from the inverse synthetic aperture radar image to obtain the key point information of the space target.
3. The method for detecting abnormal posture of space targets based on STGCN according to claim 1 is characterized in that: The step of performing graph structure modeling according to the key point information of the space target and obtaining graph structure data comprises the following steps: According to the key point information of the space target, each key point is used as a graph node; the graph node includes the two-dimensional coordinates of the key point in the range-Doppler coordinate system; defining graph edges between the graph nodes according to the shape and structure of the space object; According to the graph nodes and the graph edges, graph structure modeling is completed to obtain graph structure data.
4. The method for detecting abnormal posture of space targets based on STGCN according to claim 1 is characterized in that: The construction of the spatiotemporal graph convolutional neural network module includes the following steps: Constructing a graph convolutional network submodule; the graph convolutional network submodule is used to extract spatial features of the target; Constructing a time domain convolutional network submodule; the time domain convolutional network submodule is used to aggregate the time features of the target; The graph convolutional network submodule and the time domain convolutional network submodule are cascaded to obtain a spatiotemporal graph convolutional neural network module.
5. The method for detecting abnormal posture of space targets based on STGCN according to claim 1, characterized in that: The step of obtaining spatial information and temporal information between key points of a spatial target according to the graph structure data and the spatiotemporal graph convolutional neural network module comprises the following steps: According to the graph structure data and the spatiotemporal graph convolutional neural network module, using a graph convolutional network submodule to capture spatial features between graph nodes; According to the graph structure data and the spatiotemporal graph convolutional neural network module, aggregating the time features between graph nodes using a time domain convolutional neural network submodule; According to the spatial features between the graph nodes and the temporal features between the graph nodes, the spatial information and the temporal information between the key points of the spatial object are obtained.
6. The method for detecting abnormal posture of space targets based on STGCN according to claim 5 is characterized in that: The method of aggregating the time features between graph nodes using the time domain convolutional neural network submodule according to the graph structure data and the spatiotemporal graph convolutional neural network module comprises the following steps: According to the graph structure data and the spatiotemporal graph convolutional neural network module, using a time domain convolutional neural network submodule to capture graph node information in different time frames; According to the graph node information, the temporal features between the graph nodes are aggregated by applying multiple layers of convolution kernels.
7. The method for detecting abnormal posture of space targets based on STGCN according to claim 1, characterized in that: The STGCN-based space target posture anomaly detection method further includes the following steps: According to the spatial information and time information between the key points of the space target, the spatiotemporal graph convolutional neural network module performs binary classification judgment on the normal and abnormal states of the space target.
8. A space target posture anomaly detection system based on STGCN, used to implement the space target posture anomaly detection method based on STGCN as claimed in any one of claims 1 to 7, characterized in that: The STGCN-based space target posture anomaly detection system includes: The first module is used to obtain key point information of space targets; The second module is used to perform graph structure modeling according to the key point information of the space target and obtain graph structure data; The third module is used to build a spatiotemporal graph convolutional neural network module; A fourth module is used to obtain spatial information and temporal information between key points of spatial targets according to the graph structure data and the spatiotemporal graph convolutional neural network module; The fifth module is used to obtain detection information of abnormal posture of the space target based on the spatial information and time information between the key points of the space target.
9. A space target posture anomaly detection device based on STGCN, characterized in that: It includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the STGCN-based space target posture anomaly detection method as described in any one of claims 1 to 7 is implemented.
10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are used to enable a computer to execute the STGCN-based space target posture anomaly detection method as described in any one of claims 1 to 7.