Method for constructing space target recognition model, recognition method, device and equipment

Abstract

The embodiments of the present invention relate to the field of image processing technology, and in particular to a method for constructing a space target recognition model, a recognition method, a device and equipment. The method comprises: taking a number of training samples; the training samples are space target ISAR images containing target type labels; using a number of training samples to extract and train the features of the recognition neural network for the space target, so as to construct a space target recognition model; the recognition neural network comprises a plurality of sparse autoencoder network modules. The technical solution provided by the present invention can solve the problem that there is currently no space target ISAR image recognition method based on unsupervised learning sparse autoencoder network, and can perform efficient data compression and classification recognition on space target ISAR images.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of image processing technology, and in particular to a method for constructing a space target recognition model, a recognition method, a device and equipment. Background Art

[0002] Space target recognition technology is a military research focus of major powers in the world. Today, it has become an important indicator for measuring a country's overall defense system. ISAR (Inverse Synthetic Aperture Radar) images of space targets contain very rich target information, from which features with good separability can be extracted, and are insensitive to posture, which has important research value in the field of space target recognition.

[0003] There are two main types of existing space target ISAR image recognition methods: one is the traditional recognition method based on local discriminants and Trace transforms; the other is supervised deep learning methods based on convolutional neural networks (CNNs) and inverse space transform networks.

[0004] However, there is currently no method for space target ISAR image recognition based on unsupervised learning sparse autoencoder networks. Summary of the invention

[0005] Since there is currently no method for space target ISAR image recognition based on unsupervised learning sparse autoencoder networks, the embodiments of the present invention provide a method for constructing a space target recognition model, a recognition method, an apparatus and a device.

[0006] In a first aspect, an embodiment of the present invention provides a method for constructing a space target recognition model, comprising:

[0007] Acquire a number of training samples; the training samples are space target ISAR images containing target type labels;

[0008] A plurality of training samples are used to extract and train the features of space targets on a recognition neural network, so as to construct a space target recognition model; the recognition neural network includes a plurality of sparse autoencoder network modules.

[0009] In a possible design, the recognition neural network includes: an input layer, a compression layer, a recognition layer and an output layer;

[0010] The method of extracting and training the features of the space target by using a plurality of training samples to identify the neural network, so as to construct a space target recognition model, includes:

[0011] Inputting a plurality of space target ISAR images into the first sparse autoencoder network module of the compression layer, performing feature extraction and data compression on each space target ISAR image, and obtaining a plurality of compressed data matrices;

[0012] Inputting a plurality of compressed data matrices into the second sparse autoencoder network module and the classifier of the recognition layer, classifying and identifying the space targets in each space target ISAR image, and outputting the corresponding recognition results;

[0013] According to the outputted recognition results and the target type label corresponding to each space target ISAR image in the training sample, the network parameters of the recognition neural network are adjusted until a space target recognition model that meets the expectations is obtained.

[0014] In one possible design, the first sparse autoencoder network module includes a first sparse autoencoder network submodule and a second sparse autoencoder network submodule;

[0015] The step of inputting a plurality of space target ISAR images into the first sparse autoencoder network module of the compression layer, and performing feature extraction and data compression on each space target ISAR image comprises:

[0016] Inputting the plurality of space target ISAR images into the first sparse autoencoder network submodule, and performing row compression on the two-dimensional data matrix corresponding to each space target ISAR image;

[0017] Each row-compressed two-dimensional data matrix is ​​input into the second sparse autoencoder network submodule, and the column compression is performed on the two-dimensional data matrix corresponding to each space target ISAR image.

[0018] In a possible design, after obtaining the plurality of compressed data matrices and before inputting the plurality of compressed data matrices into the second sparse autoencoder network module and the classifier of the recognition layer, the method further includes:

[0019] Each compressed data matrix is ​​expanded in one dimension, and the corresponding one-dimensional data matrix is ​​used as the compressed data matrix.

[0020] In one possible design, the first sparse autoencoder network submodule, the second sparse autoencoder network submodule and the second sparse autoencoder network module all include three intermediate hidden layers.

[0021] In a second aspect, an embodiment of the present invention further provides an embodiment of the present invention further provides an identification method for space target identification using a space target identification model constructed by any of the methods for constructing a space target identification model described in this specification, including:

[0022] Acquire ISAR images of space targets to be identified;

[0023] Inputting the space target ISAR image into the space target recognition model;

[0024] Receive the recognition result output by the space target recognition model.

[0025] In a third aspect, an embodiment of the present invention further provides a device for constructing a space target recognition model, including:

[0026] An acquisition unit, used for acquiring a plurality of training samples; the training samples are ISAR images of space targets containing target type labels;

[0027] The construction unit is used to extract and train the features of the space target on the recognition neural network using a number of training samples to construct a space target recognition model; the recognition neural network includes a plurality of sparse autoencoder network modules.

[0028] In a fourth aspect, an embodiment of the present invention further provides a space target recognition device, including:

[0029] An image acquisition unit, used for acquiring an ISAR image of a space target to be identified;

[0030] An input unit, used for inputting the space target ISAR image into the space target recognition model;

[0031] The recognition unit is used to receive the recognition result output by the space target recognition model.

[0032] In a fifth aspect, an embodiment of the present invention further provides an electronic device, comprising a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the method described in any embodiment of this specification is implemented.

[0033] In a sixth aspect, an embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed in a computer, enables the computer to execute the method described in any embodiment of this specification.

[0034] The embodiment of the present invention provides a method for constructing a space target recognition model, a recognition method, an apparatus and a device. First, a plurality of space target ISAR images containing target type labels are obtained. By utilizing a recognition neural network including a plurality of sparse autoencoder network modules, feature extraction is performed on the space target ISAR images to train a space target recognition model. The space target ISAR images can be efficiently compressed and classified for recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] 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.

[0036] Figure 1 It is a flow chart of a method for constructing a space target recognition model provided by one embodiment of the present invention;

[0037] Figure 2 is a hardware architecture diagram of an electronic device provided by an embodiment of the present invention;

[0038] Figure 3 It is a structural diagram of a device for constructing a space target recognition model provided by an embodiment of the present invention;

[0039] Figure 4 is a hardware architecture diagram of another electronic device provided by an embodiment of the present invention;

[0040] Figure 5 It is a structural diagram of a space target recognition device provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0041] 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.

[0042] As mentioned above, there are two main types of existing space target ISAR image recognition methods: one is the traditional recognition method based on local discriminants and trace transforms, and the other is supervised deep learning methods based on convolutional neural networks (CNNs) and inverse space transform networks. Currently, there is no space target ISAR image recognition method based on unsupervised learning sparse autoencoder networks.

[0043] In order to solve the above technical problems, the inventors may consider obtaining several space target ISAR images containing target type labels as training samples, and inputting several training samples into a recognition neural network including multiple sparse autoencoder network modules to perform feature extraction on the space targets in each space target ISAR image, so as to train a space target recognition model, which can perform efficient data compression and classification recognition on the space target ISAR images.

[0044] The specific implementation of the above concept is described below.

[0045] Please refer to Figure 1 , an embodiment of the present invention provides a method for constructing a space target recognition model, the method comprising:

[0046] Step 100: obtaining a number of training samples; the training samples are space target ISAR images containing target type labels;

[0047] Step 102: extracting and training the features of the space target for the recognition neural network using a number of training samples to construct a space target recognition model; the recognition neural network includes a plurality of sparse autoencoder network modules.

[0048] In an embodiment of the present invention, a plurality of space target ISAR images containing target type labels are first acquired, and features of the space target ISAR images are extracted by utilizing a recognition neural network including a plurality of sparse autoencoder network modules to train a space target recognition model, thereby enabling efficient data compression and classification recognition of the space target ISAR images.

[0049] Described below Figure 1 How the various steps are performed.

[0050] For step 100:

[0051] It should be noted that each space target ISAR image contains a label that characterizes the type of target in the image, which is used to compare with the recognition result to adjust the network parameters in the recognition neural network.

[0052] Regarding step 102:

[0053] In the embodiment of the present invention, the recognition neural network includes: an input layer, a compression layer, a recognition layer and an output layer.

[0054] In some implementations, step 102 may include the following steps S1-S3:

[0055] Step S1, inputting a plurality of space target ISAR images into the first sparse autoencoder network module of the compression layer, performing feature extraction and data compression on each space target ISAR image, and obtaining a plurality of compressed data matrices.

[0056] Since the sparse autoencoder network can only process one-dimensional data, when compressing the two-dimensional ISAR image of space targets, it is necessary to compress the "rows" and "columns" separately.

[0057] Therefore, in some embodiments, the first sparse autoencoder network module includes a first sparse autoencoder network submodule and a second sparse autoencoder network submodule.

[0058] In this embodiment, the step of “inputting a plurality of space target ISAR images into the first sparse autoencoder network module of the compression layer, and performing feature extraction and data compression on each space target ISAR image” may include:

[0059] Inputting a plurality of space target ISAR images into the first sparse autoencoder network submodule, and performing row compression on the two-dimensional data matrix corresponding to each space target ISAR image;

[0060] Each row-compressed two-dimensional data matrix is ​​input into the second sparse autoencoder network submodule, and the two-dimensional data matrix corresponding to each space target ISAR image is column compressed.

[0061] For example, assuming that the two-dimensional data matrix of each space target ISAR image is S=M*N, the S matrix is ​​first compressed by the first sparse autoencoder network submodule to obtain a two-dimensional data matrix A=M1*N, where M1<M; then the two-dimensional data matrix A is compressed by the second sparse autoencoder network submodule to obtain a two-dimensional data matrix B=M1*N1, where N1<N, which is the compressed data matrix of the space target ISAR image.

[0062] In some embodiments, after obtaining the plurality of compressed data matrices, before inputting the plurality of compressed data matrices into the second sparse autoencoder network module and the classifier of the recognition layer, the method further includes:

[0063] Each compressed data matrix is ​​expanded in one dimension, and the corresponding one-dimensional data matrix is ​​used as the compressed data matrix.

[0064] For example, the compressed data matrix B=M1*N1 of the space target ISAR image is expanded in one dimension, and the one-dimensional data matrix C=1*(M1*N1) is obtained by connecting the columns end to end. At this time, the length of the one-dimensional data is the multiplication of the rows and columns of the matrix B (M1*N1).

[0065] It should be noted that, since in the embodiment of the present invention, the second sparse self-encoding network module is a sparse self-encoding network module, it is necessary to expand the compressed data matrix in one dimension. Of course, it is also possible not to expand the compressed data matrix in one dimension, then the second sparse self-encoding network module needs to include two sparse self-encoding network sub-modules to decode the rows and columns of the two-dimensional compressed data matrix respectively.

[0066] Step S2, inputting several compressed data matrices into the second sparse autoencoder network module and classifier of the recognition layer, classifying and identifying the space targets in each space target ISAR image, and outputting the corresponding recognition results.

[0067] In the embodiment of the present invention, the first sparse autoencoding network submodule, the second sparse autoencoding network submodule and the second sparse autoencoding network module are all three-layer sparse autoencoding networks, that is, they include three intermediate hidden layers.

[0068] It should be noted that the sparse autoencoding networks of the first sparse autoencoding network submodule, the second sparse autoencoding network submodule and the second sparse autoencoding network module may have the same number of layers or different numbers of layers, and the number of layers is not specifically limited.

[0069] For example, the compressed data matrices corresponding to several space target ISAR images are input into the second sparse autoencoder network module and classifier for unsupervised deep learning, the initial network parameters can be trained, and the corresponding recognition results can be output, where each compressed data matrix is ​​a one-dimensional data matrix C=1*(M1*N1).

[0070] Step S3, according to the outputted recognition results and the target type label corresponding to each space target ISAR image in the training sample, the network parameters of the recognition neural network are adjusted until a space target recognition model that meets the expectations is obtained.

[0071] In this step, the initial network parameters are adjusted according to the residual between the recognition result of each space target ISAR image output by the classifier and the corresponding target type label to obtain a space target recognition model that meets the expectations.

[0072] This embodiment also provides a method for identifying a space target using a space target recognition model constructed based on any of the methods for constructing a space target recognition model described in the specification, including:

[0073] Acquire ISAR images of space targets to be identified;

[0074] Inputting the space target ISAR image into the space target recognition model;

[0075] Receive the recognition result output by the space target recognition model.

[0076] In an embodiment of the present invention, a plurality of ISAR images of space targets to be identified can be directly input into a space target recognition model in sequence, wherein each space target ISAR image does not contain a target type label, and the recognition result of each space target can be quickly obtained.

[0077] like Figure 2 , Figure 3 As shown, an embodiment of the present invention provides a device for constructing a space target recognition model. The device embodiment can be implemented by software, or by hardware or a combination of software and hardware. From the hardware level, Figure 2 As shown, it is a hardware architecture diagram of an electronic device where a device for constructing a space target recognition model provided by an embodiment of the present invention is located, except Figure 3 In addition to the processor, memory, network interface, and non-volatile memory shown, the electronic device in the embodiment may also include other hardware, such as a forwarding chip responsible for processing messages, etc. Taking software implementation as an example, Figure 3 As shown, as a device in a logical sense, the CPU of the electronic device in which it is located reads the corresponding computer program in the non-volatile memory into the internal memory and runs it.

[0078] like Figure 3 As shown, this embodiment provides a device for constructing a space target recognition model, including:

[0079] The acquisition unit 301 is used to acquire a plurality of training samples; the training samples are ISAR images of space targets containing target type labels;

[0080] The construction unit 302 is used to extract and train the features of the space target for the recognition neural network using a number of training samples to construct a space target recognition model; the recognition neural network includes a plurality of sparse autoencoder network modules.

[0081] In one embodiment of the present invention, in the construction unit 302, the recognition neural network includes: an input layer, a compression layer, a recognition layer, and an output layer;

[0082] In one embodiment of the present invention, the construction unit 302 is used to perform the following operations:

[0083] Inputting a plurality of space target ISAR images into the first sparse autoencoder network module of the compression layer, performing feature extraction and data compression on each space target ISAR image, and obtaining a plurality of compressed data matrices;

[0084] Input a number of compressed data matrices into the second sparse autoencoder network module and the classifier of the recognition layer, classify and recognize each space target in the ISAR image of the space target, and output the corresponding recognition result;

[0085] According to the output recognition results and the target type label corresponding to each space target ISAR image in the training sample, the network parameters of the recognition neural network are adjusted until a space target recognition model that meets the expectations is obtained.

[0086] In one embodiment of the present invention, in the construction unit 302, the first sparse autoencoder network module includes a first sparse autoencoder network submodule and a second sparse autoencoder network submodule;

[0087] In the first sparse autoencoder network module for inputting a plurality of space target ISAR images into the compression layer, when feature extraction and data compression are performed on each space target ISAR image, the following operations are performed:

[0088] Inputting a plurality of space target ISAR images into the first sparse autoencoder network submodule, and performing row compression on the two-dimensional data matrix corresponding to each space target ISAR image;

[0089] Each row-compressed two-dimensional data matrix is ​​input into the second sparse autoencoder network submodule, and the two-dimensional data matrix corresponding to each space target ISAR image is column compressed.

[0090] In one embodiment of the present invention, the construction unit 302, after obtaining the plurality of compressed data matrices and before inputting the plurality of compressed data matrices into the second sparse autoencoder network module and the classifier of the recognition layer, is further configured to perform the following operations:

[0091] Each column-compressed two-dimensional data matrix is ​​expanded one-dimensionally to obtain the corresponding one-dimensional data matrix.

[0092] In one embodiment of the present invention, in the construction unit 302, the first sparse autoencoder network submodule, the second sparse autoencoder network submodule and the second sparse autoencoder network module all include three intermediate hidden layers.

[0093] like Figure 4 , Figure 5 As shown, an embodiment of the present invention provides a space target recognition device. The device embodiment can be implemented by software, or by hardware or a combination of software and hardware. From the hardware level, Figure 4 As shown, it is a hardware architecture diagram of an electronic device in which a space target identification device provided by an embodiment of the present invention is located. Figure 4In addition to the processor, memory, network interface, and non-volatile memory shown, the electronic device in the embodiment may also include other hardware, such as a forwarding chip responsible for processing messages, etc. Taking software implementation as an example, Figure 5 As shown, as a device in a logical sense, the CPU of the electronic device in which it is located reads the corresponding computer program in the non-volatile memory into the internal memory and runs it.

[0094] like Figure 5 As shown, this embodiment provides a space target recognition device, including:

[0095] An image acquisition unit 501 is used to acquire an ISAR image of a space target to be identified;

[0096] An input unit 502 is used to input the space target ISAR image into the space target recognition model;

[0097] The recognition unit 503 is used to receive the recognition result output by the space target recognition model.

[0098] It is to be understood that the structure illustrated in the embodiment of the present invention does not constitute a specific limitation on a device for constructing a space target recognition model / a space target recognition device. In other embodiments of the present invention, a device for constructing a space target recognition model / a space target recognition device may include more or fewer components than shown in the figure, or combine some components, or split some components, or arrange the components differently. The components shown in the figure may be implemented in hardware, software, or a combination of software and hardware.

[0099] The information interaction, execution process and other contents between the modules in the above-mentioned device are based on the same concept as the embodiment of the method of the present invention. For the specific contents, please refer to the description in the embodiment of the method of the present invention, and no further description is given here.

[0100] An embodiment of the present invention also provides an electronic device, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, a device for constructing a space target recognition model in any embodiment of the present invention and a method for identifying space targets using a space target recognition model constructed using any of the methods described in this specification are implemented.

[0101] An embodiment of the present invention also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the processor executes a device for constructing a space target recognition model in any embodiment of the present invention and a method for identifying space targets using a space target recognition model constructed using any of the methods described in this specification.

[0102] Specifically, a system or device equipped with a storage medium can be provided, on which software program code that implements the functions of any of the above-mentioned embodiments is stored, and a computer (or CPU or MPU) of the system or device can be enabled to read and execute the program code stored in the storage medium.

[0103] In this case, the program code itself read from the storage medium can realize the function of any one of the above-mentioned embodiments, and thus the program code and the storage medium storing the program code constitute a part of the present invention.

[0104] The storage medium embodiments for providing the program code include a floppy disk, a hard disk, a magneto-optical disk, an optical disk (such as CD-ROM, CD-R, CD-RW, DVD-ROM, DVD-RAM, DVD-RW, DVD+RW), a magnetic tape, a non-volatile memory card, and a ROM. Alternatively, the program code can be downloaded from a server computer by a communication network.

[0105] In addition, it should be clear that the functions of any of the above embodiments can be implemented not only by executing the program code read by the computer, but also by enabling an operating system operating on the computer to complete part or all of the actual operations based on instructions from the program code.

[0106] In addition, it can be understood that the program code read from the storage medium is written to a memory provided in an expansion board inserted into the computer or to a memory provided in an expansion module connected to the computer, and then based on the instructions of the program code, a CPU installed on the expansion board or expansion module is enabled to perform part or all of the actual operations, thereby realizing the functions of any of the above-mentioned embodiments.

[0107] It should be noted that, in this article, relational terms such as first and second 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 statement "comprise a ..." do not exclude the presence of other identical factors in the process, method, article or device including the elements.

[0108] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for constructing a space target recognition model, characterized in that: include: Obtain several training samples; The training samples are space target ISAR images containing target type labels; Using several training samples to extract and train the recognition neural network for space target features, so as to construct a space target recognition model; The recognition neural network includes a plurality of sparse autoencoder network modules; The recognition neural network comprises: an input layer, a compression layer, a recognition layer and an output layer; The method of extracting and training the features of the space target by using a plurality of training samples to identify the neural network, so as to construct a space target recognition model, includes: Inputting a plurality of space target ISAR images into the first sparse autoencoder network module of the compression layer, performing feature extraction and data compression on each space target ISAR image, and obtaining a plurality of compressed data matrices; Inputting a plurality of compressed data matrices into the second sparse autoencoder network module and the classifier of the recognition layer, classifying and identifying the space targets in each space target ISAR image, and outputting the corresponding recognition results; According to the outputted recognition results and the target type label corresponding to each space target ISAR image in the training sample, adjusting the network parameters of the recognition neural network until a space target recognition model that meets expectations is obtained; The first sparse autoencoder network module includes a first sparse autoencoder network submodule and a second sparse autoencoder network submodule; The step of inputting a plurality of space target ISAR images into the first sparse autoencoder network module of the compression layer, and performing feature extraction and data compression on each space target ISAR image comprises: Inputting the plurality of space target ISAR images into the first sparse autoencoder network submodule, and performing row compression on the two-dimensional data matrix corresponding to each space target ISAR image; Input each row-compressed two-dimensional data matrix into the second sparse autoencoder network submodule, and perform column compression on the two-dimensional data matrix corresponding to each space target ISAR image; After obtaining the plurality of compressed data matrices and before inputting the plurality of compressed data matrices into the second sparse autoencoder network module and the classifier of the recognition layer, the method further includes: Each compressed data matrix is ​​expanded in one dimension, and the corresponding one-dimensional data matrix is ​​used as the compressed data matrix.

2. The method according to claim 1, characterized in that The first sparse autoencoder network submodule, the second sparse autoencoder network submodule and the second sparse autoencoder network module all include three intermediate hidden layers.

3. A method for identifying space targets using the space target recognition model constructed by the method of claim 1 or 2, characterized in that: include: Acquire ISAR images of space targets to be identified; Inputting the space target ISAR image into the space target recognition model; Receive the recognition result output by the space target recognition model.

4. A device for constructing a space target recognition model based on the method described in claim 1 or 2, characterized in that: include: An acquisition unit, used for acquiring a number of training samples; The training samples are space target ISAR images containing target type labels; The construction unit is used to extract and train the features of the space target on the recognition neural network using a number of training samples to construct a space target recognition model; the recognition neural network includes a plurality of sparse autoencoder network modules.

5. A space target recognition device, used to implement the method described in claim 1 or 2, characterized in that: include: An image acquisition unit, used for acquiring an ISAR image of a space target to be identified; An input unit, used for inputting the space target ISAR image into the space target recognition model; The recognition unit is used to receive the recognition result output by the space target recognition model.

6. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the method according to claim 1 or 2 is implemented.

7. A computer-readable storage medium having a computer program stored thereon, which, when executed in a computer, causes the computer to execute the method of claim 1 or 2.

Images