Disclosure of Invention
In order to solve the problems in the prior art, the invention provides a radar high-resolution range profile target recognition method based on width learning. The technical problems to be solved by the invention are realized by the following technical scheme:
the invention provides a radar high-resolution range profile target recognition method based on width learning, which comprises the following steps:
s1: acquiring high-resolution range profile data and generating a training set;
S2: constructing a width learning model;
s3: training the width learning model by utilizing the training set to obtain a trained width learning model;
s4: and identifying the original high-resolution range profile data to be identified by using the trained width learning model, and obtaining an identification result.
In one embodiment of the present invention, the S1 includes:
s1a: extracting high-resolution range profile data in the broadband radar echo signals and preprocessing the high-resolution range profile data to obtain high-resolution range profile time domain feature data;
S1b: establishing a radar target database according to the high-resolution range profile time domain feature data subjected to the modulo two norm normalization processing, and setting a tag value for each target class in the radar target database;
s1c: and selecting a preset number of samples from the radar target database to form a training set.
In one embodiment of the invention, the preprocessing comprises:
Performing modulo-two norm normalization processing on the high-resolution range profile data:
wherein X represents the high-resolution range profile time domain feature data after the modulo-two norm normalization processing, X represents original high resolution range profile data, |·|2 represents a modulo two norm operation.
In one embodiment of the present invention, the S1 further includes:
And selecting a preset number of samples from the radar target database to form a test set, wherein the high-resolution range profile data in the test set and the high-resolution range profile data in the training set have arbitrary pitching angles.
In one embodiment of the present invention, the S2 includes:
S2a: performing power transformation operation on the processed high-resolution range profile time domain feature data:
U＝Xk
wherein U is data after power transformation, X is high-resolution range profile time domain characteristic data after modulo two norms normalization processing, and k is a power transformation coefficient;
s2b: setting an ith mapping characteristic node output Zi of the width learning model;
Wherein Zi is the i-th mapping feature node output, phii is the mapping function,Is a weight matrix,/>N is the number of feature nodes for the bias matrix;
S2c: splicing the mapping characteristic node outputs into a whole:
Zn＝[Z1,Z2…,Zn]
Wherein Zn represents the set of all mapping feature node outputs;
s2d: transforming the mapping characteristic node set Zn to obtain a j-th enhancement node output Ej in the width learning model;
Wherein Ej is the j-th enhancement node output, ζ is a nonlinear transformation function,And/>For a weight matrix and a bias matrix which are randomly generated, m represents the number of the enhancement nodes;
s2e: splicing the enhancement node outputs into one whole:
Ej＝[E1,E2…,Ej]
Wherein Ej represents the set of all enhancement node outputs;
s2f: setting an output of the width learning model:
wherein W is a connection weight value, and H is a set of mapping feature node outputs and enhancement node outputs.
In one embodiment of the present invention, the S3 includes:
inputting all samples in the training set into the width learning model to obtain the connection weight value:
W＝H+Y
H+ is calculated from the following formula:
Where λ represents a regularization coefficient and I represents an identity matrix.
In one embodiment of the present invention, after the step S3, the method further includes:
The trained width learning model is tested using a test set.
Another aspect of the present invention provides a storage medium having stored therein a computer program for executing the steps of the radar high-resolution range profile object recognition method of any one of the above embodiments.
In a further aspect, the present invention provides an electronic device, comprising a memory and a processor, the memory having stored therein a computer program, the processor implementing the steps of the radar high resolution range profile object recognition method according to any of the embodiments described above when invoking the computer program in the memory.
Compared with the prior art, the invention has the beneficial effects that:
According to the radar high-resolution range profile target recognition method based on the width learning, as the width learning model is used, the high-dimensional characteristics of radar high-resolution range profile data are extracted, and the radar target can be rapidly recognized on the premise of ensuring the recognition accuracy. Compared with a depth network, the method has obvious advantages in the aspects of prediction accuracy and recognition time.
The present invention will be described in further detail with reference to the accompanying drawings and examples.
Detailed Description
In order to further explain the technical means and effects adopted by the invention to achieve the preset aim, the following describes a radar high-resolution range profile object recognition method based on width learning according to the invention in detail with reference to the attached drawings and the specific embodiments.
The foregoing and other features, aspects, and advantages of the present invention will become more apparent from the following detailed description of the preferred embodiments when taken in conjunction with the accompanying drawings. The technical means and effects adopted by the present invention to achieve the intended purpose can be more deeply and specifically understood through the description of the specific embodiments, however, the attached drawings are provided for reference and description only, and are not intended to limit the technical scheme of the present invention.
It should be noted that in this document relational terms such as first and second, and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that an article or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in an article or device comprising the element.
Example 1
Referring to fig. 1, fig. 1 is a flowchart of a method for identifying a radar high-resolution range profile object based on width learning according to an embodiment of the present invention. The method comprises the following steps:
S1: and acquiring high-resolution range profile data and generating a training set.
Specifically, step S1 includes:
S1a: and extracting high-resolution range profile data in the broadband radar echo signals, and preprocessing to obtain high-resolution range profile time domain feature data.
Specifically, the pretreatment includes: the high-resolution range profile data is subjected to the following modulo-two norm normalization processing:
Wherein X represents the high-resolution range profile time domain feature data after the modulo-two norm normalization processing, X represents the acquired original high-resolution range profile data, |·|2 represents a modulo two norm operation.
S1b: establishing a radar target database according to the high-resolution range profile time domain feature data subjected to the modulo two norm normalization processing, and setting a tag value for each target class in the radar target database;
in particular, the target categories described herein may be different aircraft types to be identified, such as data for a B-737 aircraft, data for an A-330 aircraft, and so forth.
Further, setting tag values for respective target categories within the radar target database, including:
marking the label of the data of the first category as d1,d1 and taking the value as 1;
Marking the label of the data of the second category as d2,d2 and the value as 2;
Analogize … … in turn
And marking the label of the data of the M-th class as dM,dM and taking the value as M, wherein M is the total number of the target classes.
For example, assuming that different aircraft types in the high-resolution range profile data are to be identified, the target class is the type of the aircraft, and assuming that the B-737 aircraft is the first class, all tags containing the high-resolution range profile time domain feature data of the B-737 aircraft in the radar target database are marked as d1; assuming that the a-330 aircraft is in the second category, all tags in the radar target database containing high-resolution range profile time domain feature data of the a-330 aircraft are denoted as d2, and so on.
During data acquisition, a radar is generally aimed at an airplane of a certain model (such as B-747) to acquire a large amount of data, wherein the data comprise various pitching angles; the next model (e.g., a-330) is then collected, again including various pitch angles, and so on. These different models of aircraft target data together constitute the radar target database.
It should be noted that the target class described herein may be other objects, such as ships, etc.
S1c: and selecting a preset number of samples from the radar target database to form a training set, and selecting a preset number of samples from the radar target database to form a test set, wherein high-resolution range profile data in the test set and the high-resolution range profile data in the training set have different pitching angles.
The radar target database contains data of various target types, such as data of the B-737 aircraft and data of the A-330 aircraft. The data of each target class is usually obtained from different pitch angles, that is to say, comprises different pitch angles, for example, the data of the B-737 aircraft have pitch angles with different degrees of 3 degrees, 5 degrees and the like, and the data of the A-330 aircraft also have pitch angles with different degrees of 3 degrees, 5 degrees and the like. In order to verify the popularization capability of the method of the embodiment, pitch angles with different degrees are adopted when a training set and a testing set are selected.
Specifically, selecting samples with the pitch angle of each target being 3 degrees from a radar target database to form a training sample set; and selecting samples with the pitch angle of each target of 5 degrees from the radar target database to form a test sample set. However, in other embodiments, any pitch angle samples may be selected to form the training set, as well as any pitch angle samples to form the test set.
It should be noted that in the training sample set and the test sample set, each of the target types to be identified described above, for example, different aircraft types, are included.
S2: construction of a Width learning model
The width learning model generates feature nodes by mapping the input data through features, and then generates enhancement nodes by nonlinear transformation of the feature nodes. The feature nodes and the enhancement nodes are spliced together to form a hidden layer. The input data is output finally through the hidden layer. Referring to fig. 2, fig. 2 is a schematic diagram of a width learning model according to an embodiment of the invention.
Specifically, step S2 includes:
S2a: performing power transformation operation on the processed high-resolution range profile time domain feature data
U＝Xk
Wherein U is data after power transformation, X is high-resolution range profile time domain characteristic data after module two-norm normalization processing, k is a power transformation coefficient, and the value of k can be obtained according to practical experience and can be 0.1, 0.3, 0.5, 0.7 and 0.9, preferably 0.3.
S2b: setting an ith mapping characteristic node output Zi of the width learning model;
Wherein Zi is the i-th mapping feature node output, phii is the mapping function,Is a weight matrix,/>For the bias matrix, n is the number of feature nodes. The weight matrix and the bias matrix are randomly generated and the mapping function is a linear transformation function.
S2c: splicing the mapping characteristic node outputs into a whole:
Zn＝[Z1,Z2…,Zn]
where Zn represents the set of all map feature node outputs.
S2d: transforming the mapping characteristic node set Zn through the following method to obtain an ith enhancement node output Ej in the width learning model;
Wherein Ej is the j-th enhancement node output, ζ is a nonlinear transformation function,And/>For the randomly generated weight matrix and bias matrix, m represents the number of enhancement nodes.
S2e: the enhanced nodes are aggregated and connected into a whole by the following formula:
Ej＝[E1,E2…,Ej]
where Ej represents the set of all enhancement node outputs.
S2f: from the feature nodes and enhancement nodes set forth above, the output of the width learning model is expressed as:
wherein Y is the output of the width learning model, W is a connection weight value, and is a set of mapping feature node output and enhancement node output. According to the above equation, the connection weight values are solved as follows:
W＝H+Y，
Wherein H+ can be calculated from the following formula:
Where λ represents a regularization coefficient and I represents an identity matrix.
S3: and training the width learning model by utilizing the training set to obtain a trained width learning model.
And (3) inputting the high-resolution range profile data in the training set obtained in the step (S1) into the width learning model, and obtaining and updating the connection weight value to obtain a trained width learning model.
Specifically, all samples in the training set are input into the width learning model, and the connection weight value is obtained:
W＝H+Y
H+ is calculated from the following formula:
Where λ represents a regularization coefficient and I represents an identity matrix.
Further, after S3, the method further includes:
The trained width learning model is tested using a test set.
Specifically, the test set is used as high-resolution range profile data to be identified, the data in the test set is input into a trained width learning model for identification, and a classification label of an object to be identified is obtained so as to test the identification effect of the width learning model. It should be noted that, in order to test the effect of the model, the test set of this embodiment selects high-resolution range profile data with a pitch angle of 5 °.
S4: and processing the original high-resolution range profile data to be identified by using the trained width learning model to obtain an identification result.
Specifically, the high-resolution range profile data to be identified is input into a trained width learning model, and then the class label of the target in the high-resolution range profile data can be obtained. For example, the recognition result of the width learning model on the airplane type can be obtained by inputting the high-resolution range profile of the airplane into the width learning model.
In practical application, the trained width learning model can identify the target in the high-resolution range profile data of any pitch angle.
Effects of the embodiments of the present invention are described in further detail below in connection with simulation experiments.
1. Simulation conditions:
The hardware platform of the simulation experiment is as follows: the processor is an Intel (R) Core (TM) i7-8700k CPU, the main frequency is 3.2GHz, and the memory is 16GB.
The software platform of the simulation experiment is as follows: windows 10 operating system and python 3.6.
The data used in the simulation experiment are high-resolution range profile simulation data of 10 types of airplanes, the number of training samples of various airplane targets is 9000, and the total number of training samples is 90000; the number of the target test samples of each type of airplane is 1000, the total number of the test samples is 10000, and the pitch angles of each type of airplane training set and test set are 3 degrees and 5 degrees respectively.
2. Simulation content and result analysis:
The method of the embodiment of the invention and the existing convolutional neural network method are used for identifying the high-resolution range profile, the classification accuracy results under different bandwidths are shown in figure 3, wherein the abscissa represents the bandwidths, and the abscissa represents the accurate identification rates, namely 400MHz, 500MHz, 600MHz, 700MHz, 800MHz, 900MHz, 1000MHz, 1100MHz, 1200MHz and 1300 MHz. In fig. 3, the solid square node line represents the relationship curve between the accurate recognition rate obtained by the method and different bandwidths, and the solid circular node line represents the relationship curve between the accurate recognition rate obtained by the existing convolutional neural network and different bandwidths.
As can be seen from fig. 3, the accuracy of the identification method of the present invention is superior to that of the existing convolutional neural network method. When the bandwidth is 400MHz to 1000MHz, the recognition rate of the width learning model is higher than that of the conventional convolutional neural network by 3 percentage points on average, which shows that the characteristic extraction capability of the width learning model is better than that of the convolutional neural network. When the bandwidth is 1100MHz to 1300MHz, the recognition rates of the two methods are similar, because the large bandwidth contains more information, the two methods can better extract the characteristics of the high-resolution range profile, and the recognition rates are similar.
FIG. 3 shows the recognition rate curves of the two methods, and Table 1 shows a comparison of the calculation time of the width learning model method of the present invention and the existing convolutional neural network method.
Table 1 calculation time contrast table of breadth learning model and convolutional neural network
As can be seen from table 1, the training time of the width learning model was about 9.5s, and the training time of the convolutional neural network was about 420 s. The computation time of the convolutional neural network is about 45 times that of the width learning model. This is because the iterative updating of the convolutional neural network is very time consuming, which results in computation time much higher than that of the breadth learning, which does not have the complex structure of the convolutional neural network, and thus computation time is greatly reduced.
In summary, the method for identifying the radar high-resolution range profile target according to the embodiment uses the width learning model to extract the high-dimensional features of the radar high-resolution range profile data, so that the radar target can be identified rapidly on the premise of ensuring the identification accuracy. Compared with the neural network, the method has obvious advantages in the aspects of prediction accuracy and recognition time.
Another embodiment of the present invention provides a storage medium having stored therein a computer program for executing the steps of the radar high-resolution range profile object recognition method in the above embodiment. In a further aspect, the present invention provides an electronic device, including a memory and a processor, where the memory stores a computer program, and the processor, when calling the computer program in the memory, implements the steps of the radar high-resolution range profile object recognition method according to the above embodiment. In particular, the integrated modules described above, implemented in the form of software functional modules, may be stored in a computer readable storage medium. The software functional module is stored in a storage medium and includes instructions for causing an electronic device (which may be a personal computer, a server, or a network device, etc.) or a processor (processor) to perform part of the steps of the methods described in the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (Random Access Memory, RAM), a magnetic disk, or an optical disk, or other various media capable of storing program codes.
The foregoing is a further detailed description of the invention in connection with the preferred embodiments, and it is not intended that the invention be limited to the specific embodiments described. It will be apparent to those skilled in the art that several simple deductions or substitutions may be made without departing from the spirit of the invention, and these should be considered to be within the scope of the invention.