Action recognition method and device for space target

Abstract

The invention discloses an action recognition method and device for a space target, and belongs to the field of deep learning. The method comprises the following steps: acquiring a to-be-recognized high-resolution one-dimensional range profile sequence of a space target; extracting a time-frequency spectrum of the high-resolution one-dimensional range profile sequence by adopting continuous wavelet transform to obtain a wavelet frequency spectrum sequence; and performing spatial-temporal feature extraction and recognition on the wavelet spectrum sequence by using a pre-trained three-dimensional convolution model to obtain a motion state of the space target. According to the method, the HRRP sequence is adopted as input, the space and time characteristics in the HRRP sequence can be effectively analyzed in a mode of combining the continuous wavelet transform and the three-dimensional convolutional neural network, and the space target action recognition accuracy can be effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of deep learning technology, and in particular to a method and device for identifying a motion of a space target. Background Art

[0002] In recent years, with the rapid increase in the number of targets on orbit, the space traffic environment has become increasingly crowded, and the probability of various dangerous collision accidents has increased significantly. In order to ensure the safety of space activities, it is necessary to continuously track and monitor the trajectory of space targets, and at the same time identify and analyze the operating status of the targets. The behavioral interpretation of out-of-control, out-of-communication and non-cooperative targets not only helps to understand the real working status of the targets, thereby reducing the risk of collision and other adverse factors, but also helps to analyze the root causes of orbital anomalies, thereby providing support for the design of orbital debris removal missions.

[0003] The task of analyzing the motion state of space targets still faces many challenges. Due to the complex structure of space targets, the existing neural network recognition algorithms are difficult to achieve high accuracy in the motion recognition of space targets. Therefore, there is an urgent need to provide a new method that can achieve space target motion recognition with high accuracy. Summary of the invention

[0004] In order to solve the problem that the existing neural network recognition algorithm is difficult to have a high accuracy rate in the motion recognition of space targets, the embodiments of the present invention provide a method and device for the motion recognition of space targets.

[0005] On the one hand, a method for identifying a space target's motion is provided, the method comprising:

[0006] Obtain a high-resolution one-dimensional range image sequence of the space target to be identified;

[0007] Using continuous wavelet transform to extract the time-frequency spectrum of the high-resolution one-dimensional range image sequence to obtain a wavelet spectrum sequence;

[0008] The pre-trained three-dimensional convolution model is used to extract and identify the spatiotemporal features of the wavelet spectrum sequence to obtain the motion state of the space target.

[0009] On the other hand, a space target motion recognition device is provided, which is used to implement the steps described in any method embodiment of the specification, and the device includes:

[0010] An acquisition unit, used for acquiring a high-resolution one-dimensional range image sequence of a space target to be identified;

[0011] An extraction unit, used for extracting the time-frequency spectrum of the high-resolution one-dimensional range image sequence by using continuous wavelet transform to obtain a wavelet spectrum sequence;

[0012] The recognition unit is used to extract and recognize the spatiotemporal features of the wavelet spectrum sequence using a pre-trained three-dimensional convolution model to obtain the motion state of the space target.

[0013] On the other hand, a computer device is provided, which includes a memory and a processor, wherein the memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory to implement the steps of the above-mentioned method.

[0014] On the other hand, a computer-readable storage medium is provided, wherein a computer program is stored in the storage medium, and when the computer program is executed by a processor, the steps of the above-mentioned method are implemented.

[0015] On the other hand, a computer program product is provided, comprising a computer program, wherein the computer program implements the steps of the above method when executed by a processor.

[0016] The technical solution provided by the present invention can at least bring the following beneficial effects:

[0017] The use of HRRP sequence (High Resolution Range Profile, high-resolution one-dimensional range profile) can effectively and accurately reflect the motion information of space targets and has good applicability in target dynamic analysis tasks. Under different motion states, the position and intensity of the scattering points in the HRRP sequence will change regularly over time. Therefore, the mapping relationship between this change feature and the target action pattern can be mined through deep learning and other technologies, and the motion state of the space target can be effectively identified. Continuous wavelet transform can provide frequency information of the signal at different time points, which makes it very effective in analyzing non-stationary signals. Combined with the characteristics of three-dimensional convolutional neural network that can simultaneously integrate spatial and temporal features in image sequences, it can achieve high-accuracy HRRP sequence action recognition of space targets, which can effectively improve the accuracy of space target action recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0019] Figure 1 This is a flow chart of a method for identifying a space target's motion provided by an embodiment of the present invention;

[0020] Figure 2is a structural diagram of a three-dimensional convolutional network provided by an embodiment of the present invention;

[0021] Figure 3 is a two-dimensional image of HRRP amplitude provided by an embodiment of the present invention;

[0022] Figure 4 is a three-dimensional representation of HRRP provided by an embodiment of the present invention;

[0023] Figure 5 It is a wavelet spectrum sequence diagram provided by an embodiment of the present invention;

[0024] Figure 6 It is a structural diagram of a space target motion recognition device provided by one embodiment of the present invention;

[0025] Figure 7 It is a hardware architecture diagram of a computer device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0026] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0027] The specific implementation of the above concept is described below.

[0028] Please refer to Figure 1 , an embodiment of the present invention provides a method for identifying a space target action, the method comprising:

[0029] Step 100: Acquire a high-resolution one-dimensional range image sequence of a space target to be identified;

[0030] Step 102: extracting the time-frequency spectrum of the high-resolution one-dimensional range image sequence by continuous wavelet transform to obtain a wavelet spectrum sequence;

[0031] Step 104: Use the pre-trained three-dimensional convolution model to extract and identify the spatiotemporal features of the wavelet spectrum sequence to obtain the motion state of the space target.

[0032] In the embodiment of the present invention, the use of HRRP sequence (High Resolution Range Profile, high-resolution one-dimensional range profile) can effectively and accurately reflect the motion information of the space target, and has good applicability in the target dynamic analysis task. Under different motion states, the position and intensity of the scattering points in the HRRP sequence will change regularly over time. Therefore, the mapping relationship between this change feature and the target action pattern can be mined through deep learning and other technologies, and the motion state of the space target can be effectively identified. Continuous wavelet transform can provide frequency information of the signal at different time points, which makes it very effective in analyzing non-stationary signals. Combined with the characteristics of three-dimensional convolutional neural network that can simultaneously integrate spatial and temporal features in image sequences, it can achieve high-accuracy HRRP sequence action recognition of space targets, which can effectively improve the accuracy of space target action recognition.

[0033] Described below Figure 1 How the various steps are performed.

[0034] For step 100:

[0035] In the embodiment of the present invention, since HRRP is a high-resolution representation of the target in the direction of the radar beam, it can reflect the distribution of the target scattering center along the distance direction and can be regarded as a non-stationary signal. At present, HRRP is mostly used for target recognition and classification, and there are few studies reported on its use in target motion recognition. In fact, the HRRP sequence can effectively and accurately reflect the motion information of the target and has good applicability in target dynamic analysis tasks. Under different target motion modes, the position and intensity of the scattering points in the HRRP sequence will change regularly over time. Therefore, the mapping relationship between this change feature and the target motion mode can be mined through deep learning and other technologies, thereby effectively identifying the target motion state. Therefore, this embodiment uses a high-resolution one-dimensional range image sequence of a space target as an input item for motion recognition to improve the accuracy of motion recognition.

[0036] Regarding step 102:

[0037] In this embodiment, continuous wavelet transform is used to extract the time-frequency spectrum of each HRRP in the sequence. HRRP characterizes the distribution of scattering centers in the distance dimension. These scattering centers have different distances relative to the radar. They will appear at different time points in the radar receiver. Therefore, each HRRP can be regarded as an independent time series. In other words, the scattering points of HRRP can be regarded as instantaneous changes in non-stationary time series signals. Through the multi-scale wavelet basis of wavelet transform, the characteristics of these instantaneous changes can be captured, and then the one-dimensional HRRP time-frequency spectrum can be extracted. It can be understood that a frame of HRRP image can extract a time-frequency spectrum, and the time-frequency spectrum of a high-resolution one-dimensional range image sequence forms a three-dimensional wavelet spectrum sequence. The extracted wavelet spectrum sequence is as follows. Figure 5 As shown, only the first 6 frames of the frequency spectrum in the sequence are shown here.

[0038] Regarding step 104:

[0039] In some implementations, step 104 may include:

[0040] Use three-dimensional convolutional network to extract time dimension features and space dimension features from wavelet spectrum sequence;

[0041] The Softmax classifier is used to calculate the category probability distribution based on the time dimension features and space dimension features of the wavelet spectrum sequence extracted by the three-dimensional convolutional network, and the motion state of the space target is identified.

[0042] In this embodiment, the 3D convolutional neural network uses a 3D convolution kernel, which can simultaneously integrate the spatial and temporal features in the image sequence and has excellent performance in the image sequence analysis task. Therefore, the 3D convolutional network can be used to extract temporal and spatial features from the wavelet spectrum sequence, thereby improving the recognition accuracy.

[0043] In some embodiments, the 3D convolutional model is trained as follows:

[0044] Construct a high-resolution one-dimensional range image sequence dataset, and use the motion state corresponding to each sequence sample as the label of the sequence sample;

[0045] Divide the dataset into training, validation and test sets;

[0046] Perform wavelet transform on the sequence samples in the training set to obtain the wavelet spectrum sequence corresponding to each sequence sample;

[0047] The wavelet spectrum sequences in the training set are input into the pre-built three-dimensional convolutional network in batches, and the network parameters of the three-dimensional convolutional network are trained and adjusted using the label of each sequence sample until a three-dimensional convolutional model that meets the requirements is obtained.

[0048] In this embodiment, the extracted wavelet spectrum sequence is input into a three-dimensional convolutional network to extract time domain and space domain features. The structure of the three-dimensional convolutional network is as follows: Figure 2 As shown in the figure. Traditional convolutional neural networks use two-dimensional convolution kernels, which only slide in the width and height dimensions of the image, so they can only extract spatial features. The three-dimensional convolution kernel used in the three-dimensional convolutional network not only slides in the spatial dimension, but also slides along the time dimension of the input image sequence, which can effectively realize the information integration of the time dimension and space dimension channels of the wavelet spectrum sequence, and extract the motion characteristics of the target with excellent performance.

[0049] In some embodiments, a high-resolution one-dimensional range image sequence dataset is constructed as follows:

[0050] The action image sequences of the test target at different spin axes and different spin periods are collected as out-of-control samples;

[0051] Collect the action image sequence of the test target under three-axis stabilization as normal samples;

[0052] For each action image sequence, a high-resolution one-dimensional range image is generated for each image in the sequence, and a one-dimensional range image sequence sample corresponding to each action image sequence is obtained to generate a high-resolution one-dimensional range image sequence data set.

[0053] In this embodiment, the study of space targets shows that: during normal operation, the target maintains a three-axis posture stability, and the orientation of the components remains unchanged in the target body coordinate system; after losing control, the target rotates around a spin axis and gradually loses rotational energy, and the spin period and spin axis are always fixed, or there are some slight oscillations. Based on this, the three-axis stable action of the target is simulated and a series of out-of-control spin actions are designed. During modeling, the X-axis of the target model is aligned with the target body and points forward, the Z-axis points directly downward, and the Y-axis is perpendicular to the other two axes, following the right-hand rule. For out-of-control spin actions, set the target spin axis to the X-axis, Y-axis or Z-axis, and set the spin period. For three-axis stable actions, the spin axis and spin period are not set. Each sample consists of 16 consecutive frames of HRRP.

[0054] In some embodiments, the spin axis is the X axis, the Y axis or the Z axis; and the spin period includes 10 s, 15 s and 30 s.

[0055] In this embodiment, the out-of-control sample sets the target spin axis to the X-axis, Y-axis or Z-axis, and the spin period is 10s, 15s or 30s. The naming rule of the sample shows the spin axis and rotation period of the target, for example, X-spin 30s-T means that the target spins around the X-axis with a spin period of 30s; Stable means that the target is stable in three axes. In order to better display the sample characteristics, samples are selected from the categories X-spin15s-T, Y-spin 30s-T, Z-spin10s-T, and Stable, and two forms of sample images are drawn. Figure 3 A two-dimensional image of the amplitude of each HRRP in the sample is shown; Figure 4 A 3D representation of the HRRP is shown, with the scattering centers marked with red dots. To prevent visual confusion, only the first 6 HRRPs in the sequence are shown here as an illustration.

[0056] In some embodiments, the tag includes three types of information: out-of-control or normal state, spin axis, and spin period; when the tag is in a normal state, the spin axis and spin period information are empty;

[0057] The motion state of a space target includes three types of information: the target's out-of-control or normal state, spin axis, and spin period.

[0058] In this embodiment, the motion information of the space target can be identified, including whether it is out of control or operating normally. When it is out of control, the spin axis and spin period of the spin motion can be identified, and the identification result is precise and accurate.

[0059] Please refer to Figure 6 The embodiment of the present invention provides a space target motion recognition device, which is used to implement the steps of any method embodiment in the specification, and the device includes:

[0060] An acquisition unit 601 is used to acquire a high-resolution one-dimensional range image sequence of a space target to be identified;

[0061] An extraction unit 602 is used to extract the time-frequency spectrum of the high-resolution one-dimensional range image sequence by using continuous wavelet transform to obtain a wavelet spectrum sequence;

[0062] The recognition unit 603 is used to extract and recognize the spatiotemporal features of the wavelet spectrum sequence using a pre-trained three-dimensional convolution model to obtain the motion state of the space target.

[0063] In one embodiment of the present invention, the identification unit 603 is used to perform:

[0064] Use three-dimensional convolutional network to extract time dimension features and space dimension features from wavelet spectrum sequence;

[0065] The Softmax classifier is used to calculate the category probability distribution based on the time dimension features and space dimension features of the wavelet spectrum sequence extracted by the three-dimensional convolutional network, and the motion state of the space target is identified.

[0066] In one embodiment of the present invention, the three-dimensional convolution model in the recognition unit 603 is trained in the following manner:

[0067] Construct a high-resolution one-dimensional range image sequence dataset, and use the motion state corresponding to each sequence sample as the label of the sequence sample;

[0068] Divide the dataset into training, validation and test sets;

[0069] Perform wavelet transform on the sequence samples in the training set to obtain the wavelet spectrum sequence corresponding to each sequence sample;

[0070] The wavelet spectrum sequences in the training set are input into the pre-built three-dimensional convolutional network in batches, and the network parameters of the three-dimensional convolutional network are trained and adjusted using the label of each sequence sample until a three-dimensional convolutional model that meets the requirements is obtained.

[0071] In one embodiment of the present invention, the high-resolution one-dimensional range image sequence data set in the recognition unit 603 is constructed in the following manner:

[0072] The action image sequences of the test target at different spin axes and different spin periods are collected as out-of-control samples;

[0073] Collect the action image sequence of the test target under three-axis stabilization as normal samples;

[0074] For each action image sequence, a high-resolution one-dimensional range image is generated for each image in the sequence, and a one-dimensional range image sequence sample corresponding to each action image sequence is obtained to generate a high-resolution one-dimensional range image sequence data set.

[0075] In one embodiment of the present invention, the spin axes in the identification unit 603 are respectively the X-axis, the Y-axis or the Z-axis; and the spin periods include 10s, 15s and 30s.

[0076] In one embodiment of the present invention, the tag in the identification unit 603 includes three types of information: out-of-control or normal state, spin axis and spin period; when the tag is in a normal state, the spin axis and spin period information are empty;

[0077] The motion state of a space target includes three types of information: the target's out-of-control or normal state, spin axis, and spin period.

[0078] It should be noted that the motion recognition device for space targets provided in the above embodiment is only illustrated by the division of the above functional units. In practical applications, the above functions can be assigned to different functional units as needed, that is, the internal structure of the device can be divided into different functional units to complete all or part of the functions described above. In addition, the above device embodiment and the method embodiment belong to the same concept, and the specific implementation process is detailed in the method embodiment, which will not be repeated here.

[0079] The embodiment of the present application also provides a computer device, please refer to Figure 7 The computer device includes a processor and a memory, in which at least one instruction, at least one program, a code set or an instruction set is stored, and the at least one instruction, at least one program, a code set or an instruction set is loaded and executed by the processor to implement the motion recognition method of the space target provided by the above-mentioned method embodiments.

[0080] An embodiment of the present application also provides a computer-readable storage medium, on which is stored at least one instruction, at least one program, code set or instruction set, and the at least one instruction, at least one program, code set or instruction set is loaded and executed by a processor to implement the motion recognition method of space targets provided in the above-mentioned method embodiments.

[0081] An embodiment of the present application also provides a computer program product, which includes a computer program. A processor of a computer device reads the computer program from a computer-readable storage medium, and the processor executes the computer program, so that the computer device executes any of the space target motion recognition methods in the above embodiments.

[0082] For the convenience of description, the above system or device is described by dividing it into various modules or units according to its functions. Of course, when implementing the present application, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0083] It can be known from the description of the above implementation methods that those skilled in the art can clearly understand that the present application can be implemented by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product can be stored in a storage medium such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods of each embodiment of the present application or some parts of the embodiments.

[0084] Finally, it should be noted that, in this article, relational terms such as first, second, third and fourth are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0085] The above are only preferred implementations of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A method for identifying a space target's motion, characterized in that: The method comprises: Obtain a high-resolution one-dimensional range image sequence of the space target to be identified; Using continuous wavelet transform to extract the time-frequency spectrum of the high-resolution one-dimensional range image sequence to obtain a wavelet spectrum sequence; The pre-trained three-dimensional convolution model is used to extract and identify the spatiotemporal features of the wavelet spectrum sequence to obtain the motion state of the space target.

2. The method according to claim 1, characterized in that The extracting and identifying the spatiotemporal features of the wavelet spectrum sequence using a pre-trained three-dimensional convolution model to obtain the motion state of the space target includes: Using a three-dimensional convolutional network to extract time dimension features and space dimension features from the wavelet spectrum sequence; The Softmax classifier is used to calculate the category probability distribution based on the time dimension features and space dimension features of the wavelet spectrum sequence extracted by the three-dimensional convolutional network, and the motion state of the space target is identified.

3. The method according to claim 1, characterized in that The 3D convolutional model is trained as follows: Construct a high-resolution one-dimensional range image sequence dataset, and use the motion state corresponding to each sequence sample as the label of the sequence sample; Dividing the data set into a training set, a validation set, and a test set; Performing wavelet transform on the sequence samples in the training set to obtain a wavelet spectrum sequence corresponding to each sequence sample; The wavelet spectrum sequences in the training set are input into a pre-built three-dimensional convolutional network in batches, and the network parameters of the three-dimensional convolutional network are trained and adjusted using the label of each sequence sample until a three-dimensional convolutional model that meets the requirements is obtained.

4. The method according to claim 3, characterized in that The high-resolution one-dimensional range image sequence data set is constructed in the following manner: The action image sequences of the test target at different spin axes and different spin periods are collected as out-of-control samples; Collecting a sequence of action images of the test target under three-axis stabilization as a normal sample; For each action image sequence, a high-resolution one-dimensional range image is generated for each image in the sequence, and a one-dimensional range image sequence sample corresponding to each action image sequence is obtained to generate a high-resolution one-dimensional range image sequence data set.

5. The method according to claim 4, characterized in that The spin axes are respectively the X axis, the Y axis or the Z axis; the spin periods include 10s, 15s and 30s.

6. The method according to claim 3 or 4, characterized in that: The tag includes three types of information: out-of-control or normal state, spin axis and spin period; when the tag is in normal state, the spin axis and spin period information are empty; The motion state of the space target includes three kinds of information: the target's out-of-control or normal state, the spin axis and the spin period.

7. A space target motion recognition device, used to implement the steps of any of the methods described in claims 1 to 6, characterized in that: The device comprises: An acquisition unit, used for acquiring a high-resolution one-dimensional range image sequence of a space target to be identified; An extraction unit, used for extracting the time-frequency spectrum of the high-resolution one-dimensional range image sequence by using continuous wavelet transform to obtain a wavelet spectrum sequence; The recognition unit is used to extract and recognize the spatiotemporal features of the wavelet spectrum sequence using a pre-trained three-dimensional convolution model to obtain the motion state of the space target.

8. A computer device, characterized in that: The computer device includes a memory and a processor, the memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory to implement the steps of any one of the methods described in claims 1-6.

9. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method described in any one of claims 1 to 6 are implemented.

10. A computer program product, characterized in that The method comprises a computer program, wherein when the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images