Spatial non-cooperative target track classification method, system and device based on improved deep belief network and medium

Abstract

The invention discloses a space non-cooperative target orbit classification method, system and device based on an improved deep belief network and a medium, and the method comprises the steps: generating orbit data according to corresponding orbit motion parameters of a space target, and carrying out the arrangement and marking of the orbit data to make an orbit classification data set; a track classification network model based on an improved deep belief network is constructed through a long short-term memory network, a multi-feature self-attention module and the deep belief network, and the track classification network model is trained by using a self-made track classification data set; finally, the trained track classification network model is used for recognizing the track type of the space non-cooperative target according to the input relative position of the space non-cooperative target, and the final track type is obtained. According to the method, the problems of low timeliness and low accuracy of manual identification in an existing space non-cooperative target track classification method are solved, and compared with the prior art, the method can realize rapid, autonomous and accurate classification of target track types, and can be applied to space debris collision early warning and space threat early warning task scenes.

Description

Technical Field

[0001] The present invention relates to the technical field of space debris collision warning and space threat warning, and in particular to a space non-cooperative target orbit classification method, system, device and medium based on an improved deep belief network. Background Art

[0002] With the continuous development of aerospace technology, more and more countries have the ability to develop space resources. In particular, with the continuous advancement and implementation of giant constellation projects such as Starlink in recent years, the number of space objects in space has increased sharply, posing a serious threat to the safety of on-orbit operation of spacecraft.

[0003] In order to ensure the safe operation of spacecraft, it is necessary to conduct a comprehensive safety analysis of the non-cooperative space targets around it, and timely detect and warn any non-cooperative space targets that may pose a collision threat to the spacecraft to ensure the safety of its own personnel and property. In order to achieve this goal, it is first necessary to determine the orbit type based on the motion information of the non-cooperative space targets near the spacecraft, so as to better understand the motion law and behavior characteristics of the target, and then provide data support for subsequent intention identification and threat assessment. Existing technologies still rely on expert experience and knowledge for manual orbit recursion and classification, and cannot achieve rapid and autonomous classification of non-cooperative target orbits, and have low timeliness and accuracy. Summary of the invention

[0004] In view of the deficiencies in the prior art, the present invention provides a spatial non-cooperative target orbit classification method, system, device and medium based on an improved deep belief network to achieve rapid and autonomous classification of target orbit types.

[0005] To achieve the above object, the present invention adopts the following technical solutions:

[0006] A space non-cooperative target track classification method based on an improved deep belief network comprises the following steps:

[0007] Create track classification datasets;

[0008] Construct a track classification network model based on an improved deep belief network;

[0009] Use the track classification data set to train and test the track classification network model, continuously adjust the training parameters according to the test results, optimize the structure and parameters of the track classification network model, and save the optimal track classification network model;

[0010] The non-cooperative space target is measured to obtain the position of the non-cooperative space target relative to its own spacecraft at each moment, and the positions of two adjacent moments are input into the optimal orbit classification network model as a group to obtain the orbit classification results of the non-cooperative space target at each moment;

[0011] Multi-frame correlation filtering is performed on the orbit classification results at each moment to obtain the final orbit type of the space non-cooperative target.

[0012] To optimize the above technical solutions, the specific measures taken also include:

[0013] Furthermore, the production of the track classification data set is specifically as follows:

[0014] Based on the CW equation, orbital data are generated by simulation software according to the corresponding orbital motion parameters of the space target. The orbital data include closed flyby orbit, flyby orbit, companion orbit and water drop orbit data. Each orbital data is sampled at a fixed time interval to obtain the position of the space non-cooperative target relative to its own spacecraft at each moment. For each orbital data, the relative position data of two adjacent moments are arbitrarily selected as a group of samples, and all samples are labeled to produce an orbit classification data set.

[0015] Furthermore, the track classification network model based on the improved deep belief network specifically includes a long short-term memory network, a multi-feature self-attention module, a deep belief network and a Softmax classifier connected in sequence;

[0016] The long short-term memory network is constructed based on a gated recurrent unit and is used to adaptively extract position change information of a spatial non-cooperative target between two moments;

[0017] The multi-feature self-attention module includes 1 convolutional layer, 1 global average pooling layer and 2 fully connected layers; it is used to dynamically learn and assign weights of different features;

[0018] The deep belief network is constructed by stacking multiple layers of restricted Boltzmann machines.

[0019] Furthermore, the method for measuring non-cooperative targets in space is to use optical, infrared or radar detection payloads for measurement.

[0020] Furthermore, the position of the space non-cooperative target relative to its own spacecraft is the position in the LVLH coordinate system, the origin O of the LVLH coordinate system is the center of mass of its own spacecraft, the direction of the x-axis is the vector direction from the center of the earth to the center of mass of its own spacecraft, the direction of the z-axis is the normal direction of the plane of motion of its own spacecraft relative to the earth, and the y-axis is perpendicular to the x-axis and points to the forward direction of its own spacecraft.

[0021] Furthermore, the multi-frame correlation filtering of the track classification results at each moment is specifically as follows:

[0022] Randomly select N consecutive orbital classification results. When the N consecutive orbital classification results remain consistent, the orbital classification result is determined to be the final orbital type of the space non-cooperative target.

[0023] The present invention also proposes a space non-cooperative target orbit classification system based on an improved deep belief network, comprising:

[0024] Dataset making module, used to make track classification dataset;

[0025] Network building module, used to build a track classification network model based on improved deep belief network;

[0026] The training module is used to train and test the track classification network model using the track classification data set, continuously adjust the training parameters according to the test results, optimize the structure and parameters of the track classification network model, and save the optimal track classification network model;

[0027] The classification module is used to measure the non-cooperative space target and obtain the position of the non-cooperative space target relative to its own spacecraft at each moment. The positions of two adjacent moments are input into the optimal orbit classification network model as a group to obtain the orbit classification results of the non-cooperative space target at each moment.

[0028] The filtering module is used to perform multi-frame correlation filtering on the orbit classification results at each moment to obtain the final orbit type of the space non-cooperative target.

[0029] The present invention also proposes an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the spatial non-cooperative target orbit classification method based on the improved deep belief network as described above is implemented.

[0030] The present invention also proposes a computer-readable storage medium storing a computer program, wherein the computer program enables a computer to execute the spatial non-cooperative target trajectory classification method based on the improved deep belief network as described above.

[0031] The beneficial effects of the present invention are as follows: the present invention solves the problems of low timeliness and accuracy of manual identification in existing space non-cooperative target orbit classification methods. Compared with the prior art, the present invention can realize rapid, autonomous and accurate classification of target orbit types, and can be better applied to mission scenarios such as space debris collision warning and space threat warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 It is the overall flow chart of the method of the present invention;

[0033] Figure 2 This is a structural diagram of a track classification network model based on an improved deep belief network of the present invention;

[0034] Figure 3 This is a structural diagram of the multi-feature self-attention module of the present invention;

[0035] Figure 4 It is a schematic diagram of four types of tracks in the LVLH coordinate system of the present invention;

[0036] Figure 5 Schematic diagram of four types of orbits in the J2000 coordinate system of the present invention;

[0037] Table 1 shows the classification results of the embodiments of the present invention on the self-made data set. DETAILED DESCRIPTION

[0038] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0039] Embodiment 1

[0040] The present invention proposes a space non-cooperative target trajectory classification method based on an improved deep belief network. The overall process of the method is as follows: Figure 1 As shown, the following steps are included:

[0041] Create a track classification dataset; specifically:

[0042] Based on the CW equation (Clohessy-Wiltshire equation), orbital data are generated through simulation software according to the corresponding orbital motion parameters of the space target. The orbital data include closed flyby orbit, flyby orbit, companion orbit and water drop orbit data; each orbital data is sampled at a fixed time interval (1 second in this embodiment) to obtain the position of the space non-cooperative target relative to its own spacecraft at each moment; for each orbital data, the relative position data of two adjacent moments are arbitrarily selected as a group of samples, and all samples are labeled to produce an orbit classification data set.

[0043] Construct a track classification network model based on an improved deep belief network; the specific structure of the network model is as follows Figure 2 The track classification network model based on the improved deep belief network specifically includes a long short-term memory network, a multi-feature self-attention module, a deep belief network and a Softmax classifier connected in sequence;

[0044] The long short-term memory network is constructed based on a gated recurrent unit (GRU); it is used to adaptively extract position change information of a spatial non-cooperative target between two moments; so as to improve the feature extraction capability of the network model, thereby enhancing the classification effect of the network model.

[0045] The specific structure of the multi-feature self-attention module is as follows Figure 3 As shown in the figure, the multi-feature self-attention module includes 1 convolutional layer, 1 global average pooling layer and 2 fully connected layers; it is used to dynamically learn and assign weights of different features; it enables the network model to pay more attention to important features, thereby improving the accuracy and robustness of the network model.

[0046] The deep belief network is constructed by stacking multiple layers of restricted Boltzmann machines.

[0047] Use the track classification data set to train and test the track classification network model, continuously adjust the training parameters according to the test results, optimize the structure and parameters of the track classification network model, and save the optimal track classification network model;

[0048] Use optical, infrared or radar detection payloads to measure non-cooperative space targets to obtain the position of non-cooperative space targets relative to their own spacecraft at each moment. The position of non-cooperative space targets relative to their own spacecraft is the position in the LVLH coordinate system. The origin O of the LVLH coordinate system is the center of mass of their own spacecraft, the direction of the x-axis is the vector direction from the center of the earth to the center of mass of their own spacecraft, the direction of the z-axis is the normal direction of the plane of motion of their own spacecraft relative to the earth, and the y-axis is perpendicular to the x-axis and points to the forward direction of their own spacecraft. The positions of two adjacent moments are input as a group into the optimal orbit classification network model to obtain the orbit classification results of non-cooperative space targets at each moment, that is, the probability of the target belonging to the orbit of flyby, flyby, companion and water droplet at each moment.

[0049] The orbit classification results at each moment are subjected to multi-frame correlation filtering to obtain the final orbit type of the non-cooperative target in space. The multi-frame correlation filtering is specifically as follows:

[0050] Randomly select N consecutive orbital classification results. When the N consecutive orbital classification results remain consistent, the orbital classification result is determined to be the final orbital type of the space non-cooperative target.

[0051] Embodiment 2

[0052] The present invention proposes a space non-cooperative target orbit classification system based on an improved deep belief network corresponding to the method of embodiment 1, comprising:

[0053] Dataset making module, used to make track classification dataset;

[0054] Network building module, used to build a track classification network model based on improved deep belief network;

[0055] The training module is used to train and test the track classification network model using the track classification data set, continuously adjust the training parameters according to the test results, optimize the structure and parameters of the track classification network model, and save the optimal track classification network model;

[0056] The classification module is used to measure the non-cooperative space target and obtain the position of the non-cooperative space target relative to its own spacecraft at each moment. The positions of two adjacent moments are input into the optimal orbit classification network model as a group to obtain the orbit classification results of the non-cooperative space target at each moment.

[0057] The filtering module is used to perform multi-frame correlation filtering on the orbit classification results at each moment to obtain the final orbit type of the space non-cooperative target.

[0058] The implementation methods of each module and module function in the system are completely consistent with the steps of the method in Example 1, so they will not be repeated here.

[0059] Embodiment 3

[0060] The present invention proposes an electronic device, comprising: 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 spatial non-cooperative target trajectory classification method based on an improved deep belief network as described in Example 1 is implemented.

[0061] Embodiment 4

[0062] The present invention proposes a computer-readable storage medium storing a computer program, wherein the computer program enables a computer to execute the spatial non-cooperative target trajectory classification method based on an improved deep belief network as described in Example 1.

[0063] In order to illustrate the effectiveness of the present invention, a space non-cooperative orbit classification experiment is completed as follows:

[0064] (1) Experimental setup

[0065] Using the Tensorflow neural network learning framework and Anaconda4.7.10's Spyder (Python3.9) as the simulation platform, the computer is configured with an I9-12900H 2.5GHz CPU, 32G memory, and an NVIDIA RTX3060 GPU. Using Numpy, OpenCV-Python, Pytorch and other libraries, an independently designed space non-cooperative target orbit classification network model is built.

[0066] The network model of the present invention is trained offline and tested online using a self-made orbit classification dataset. The self-made orbit classification dataset contains four types of orbits: closed flyby orbit, flyby orbit, companion orbit and water drop orbit. The forms of each orbit in the LVLH coordinate system and the J2000 coordinate system are as follows: Figure 4 and Figure 5As shown. Each type of space target orbit contains 250 orbit data, a total of 1000 orbit data. The training set contains 800 orbit data, the validation set contains 100 orbit data, and the test set contains 100 orbit data. The sampling time interval of each orbit data is 1 second. 1000 sets of sample data are randomly extracted from each orbit data to construct the data set. The data set contains a total of 1000000 sets of sample data. According to the common data set division rules, the data set is divided into training set, validation set and test set. The final training set contains 800000 sets of sample data, the validation set contains 100000 sets of sample data, and the test set contains 100000 sets of sample data.

[0067] (2) Experimental results analysis

[0068] In order to verify the effectiveness of the method of the present invention, the accuracy and F1 score are used as evaluation criteria to quantitatively analyze the track classification results. Accuracy is the ratio of the number of correctly predicted samples to the total number of samples, and F1 score is the harmonic mean of precision and recall. The closer the value is to 1, the better the classification effect of the network model. The calculation method of each evaluation standard is as follows: For the evaluation standard mAP of target detection, its calculation formula is shown in formulas (1)-(4):

[0069]

[0070] Among them, TP represents the number of correctly predicted positive samples, TN represents the number of correctly predicted negative samples, FP represents the number of incorrectly predicted negative samples, and FN represents the number of incorrectly predicted positive samples.

[0071] The method of the present invention is used to test the test set of the self-made track classification data set, and the test results are shown in Table 1.

[0072] Table 1 Test results of space non-cooperative target orbit classification network

[0073] Accuracy (%) F1 score Run time (ms) Method of the present invention 99.75 0.9974 5.36

[0074] As can be seen from the table, the method of the present invention can quickly and autonomously classify the target track type, with a classification accuracy of 99.75%, an F1 score of 0.9974, and a running time of 5.36ms, with good timeliness and accuracy.

[0075] In the embodiments disclosed in the present application, the computer storage medium can be a tangible medium that can contain or store a program for use by or in conjunction with an instruction execution system, device or equipment. The computer storage medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the above. More specific examples of computer storage media can include electrical connections based on one or more lines, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the above.

[0076] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in this application can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0077] The above are only preferred embodiments of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should be regarded as the protection scope of the present invention.

Claims

1. A spatial non-cooperative target trajectory classification method based on an improved deep belief network, characterized in that: The following steps are involved: Create track classification datasets; Construct a track classification network model based on an improved deep belief network; Use the track classification data set to train and test the track classification network model, continuously adjust the training parameters according to the test results, optimize the structure and parameters of the track classification network model, and save the optimal track classification network model; The non-cooperative space target is measured to obtain the position of the non-cooperative space target relative to its own spacecraft at each moment, and the positions of two adjacent moments are input into the optimal orbit classification network model as a group to obtain the orbit classification results of the non-cooperative space target at each moment; Multi-frame correlation filtering is performed on the orbit classification results at each moment to obtain the final orbit type of the space non-cooperative target.

2. The spatial non-cooperative target trajectory classification method based on the improved deep belief network as claimed in claim 1, characterized in that: The production track classification data set is specifically: Based on the CW equation, orbital data are generated by simulation software according to the corresponding orbital motion parameters of the space target. The orbital data include closed flyby orbit, flyby orbit, companion orbit and water drop orbit data. Each orbital data is sampled at a fixed time interval to obtain the position of the space non-cooperative target relative to its own spacecraft at each moment. For each orbital data, the relative position data of two adjacent moments are arbitrarily selected as a group of samples, and all samples are labeled to produce an orbit classification data set.

3. The spatial non-cooperative target orbit classification method based on improved deep belief network as claimed in claim 1, characterized in that: The track classification network model based on the improved deep belief network specifically includes a long short-term memory network, a multi-feature self-attention module, a deep belief network and a Softmax classifier connected in sequence; The long short-term memory network is constructed based on a gated recurrent unit and is used to adaptively extract position change information of a spatial non-cooperative target between two moments; The multi-feature self-attention module includes 1 convolutional layer, 1 global average pooling layer and 2 fully connected layers; Used to dynamically learn and assign weights to different features; The deep belief network is constructed by stacking multiple layers of restricted Boltzmann machines.

4. The spatial non-cooperative target orbit classification method based on improved deep belief network as claimed in claim 1, characterized in that: The method for measuring non-cooperative targets in space is to use optical, infrared or radar detection payloads for measurement.

5. The spatial non-cooperative target trajectory classification method based on improved deep belief network as claimed in claim 1, characterized in that: The position of the space non-cooperative target relative to its own spacecraft is the position in the LVLH coordinate system, the origin O of the LVLH coordinate system is the center of mass of its own spacecraft, the direction of the x-axis is the vector direction from the center of the earth to the center of mass of its own spacecraft, the direction of the z-axis is the normal direction of the own spacecraft relative to the plane of motion of the earth, and the y-axis is perpendicular to the x-axis and points to the forward direction of its own spacecraft.

6. The spatial non-cooperative target orbit classification method based on improved deep belief network as claimed in claim 1, characterized in that: The multi-frame association filtering of the track classification results at each moment is specifically as follows: Randomly select N consecutive orbital classification results. When the N consecutive orbital classification results remain consistent, the orbital classification result is determined to be the final orbital type of the space non-cooperative target.

7. A space non-cooperative target orbit classification system based on an improved deep belief network, characterized in that: include: Dataset making module, used to make track classification dataset; Network building module, used to build a track classification network model based on improved deep belief network; The training module is used to train and test the track classification network model using the track classification data set, continuously adjust the training parameters according to the test results, optimize the structure and parameters of the track classification network model, and save the optimal track classification network model; The classification module is used to measure the non-cooperative space target and obtain the position of the non-cooperative space target relative to its own spacecraft at each moment. The positions of two adjacent moments are input into the optimal orbit classification network model as a group to obtain the orbit classification results of the non-cooperative space target at each moment. The filtering module is used to perform multi-frame correlation filtering on the orbit classification results at each moment to obtain the final orbit type of the space non-cooperative target.

8. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the spatial non-cooperative target orbit classification method based on the improved deep belief network as described in any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium storing a computer program, characterized in that: The computer program enables the computer to execute the spatial non-cooperative target trajectory classification method based on the improved deep belief network as described in any one of claims 1-6.

Images