CN115270997B - Rocket target attitude stability discrimination method based on transfer learning and related device - Google Patents
Rocket target attitude stability discrimination method based on transfer learning and related device Download PDFInfo
- Publication number
- CN115270997B CN115270997B CN202211143632.4A CN202211143632A CN115270997B CN 115270997 B CN115270997 B CN 115270997B CN 202211143632 A CN202211143632 A CN 202211143632A CN 115270997 B CN115270997 B CN 115270997B
- Authority
- CN
- China
- Prior art keywords
- target
- rocket
- classification
- network
- data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
- G01S7/02—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00
- G01S7/41—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section
- G01S7/417—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section involving the use of neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- General Physics & Mathematics (AREA)
- Software Systems (AREA)
- Medical Informatics (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Computer Networks & Wireless Communication (AREA)
- Radar, Positioning & Navigation (AREA)
- Remote Sensing (AREA)
- Radar Systems Or Details Thereof (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to a rocket target attitude stability discrimination method based on transfer learning and a related device, wherein the method comprises the following steps: acquiring input data to be classified; carrying out segmentation processing on the input data to be classified to obtain a plurality of short-time segmentation data; inputting the short-time segmentation data into a trained target classification network with capsules to obtain a network classification result; based on the obtained continuitiesPAnd judging the secondary network classification result according to a DS evidence theory algorithm to obtain a judgment result. The discrimination method provided by the invention has higher classification accuracy.
Description
Technical Field
The invention belongs to the technical field of target identification, and relates to a rocket target attitude stability discrimination method based on transfer learning and a related device.
Background
When the rocket is launched, if the telemetry data is abnormal or interrupted, the flying state of the rocket needs to be judged by using radar measurement data, and the radar measurement data comprises a rocket body, rocket engine debris and a large amount of rocket body fragments, and the target identification needs to be carried out based on a classifier, the target identity of the rocket body is confirmed, and the attitude stability judgment is carried out by using the measurement data. Limited by the performance of Radar equipment, the data types which can be acquired and calculated in real time are limited, and the RCS (Radar Cross Section) amplitude sequence value of the rocket target measured by the Radar is more typical characteristic data which can be applied in real time. Therefore, the method for extracting the features by using the target RCS amplitude sequence value and further identifying and classifying the targets is a scheme with high feasibility.
The existing rocket target RCS sequence identification technology widely applied to engineering mainly comprises methods such as template matching, statistical identification and the like. The template matching is one of the simplest classification methods used in the field of target identification, and the method mainly comprises the steps of obtaining a sample to be identified, matching the sample to be identified with each template by using certain similarity measurement, calculating and judging the type of the sample to be identified, and finishing the classification identification process. To obtain good recognition performance by a template matching method, a relatively complete template library needs to be established, and both the computational time complexity and the storage space complexity are relatively high. For rocket targets, accurate template acquisition is relatively difficult, and simple template matching is difficult to adapt to rocket main body target characteristics, namely the generalization capability of the method is poor. The statistical identification mainly utilizes the distribution characteristics of various targets, and directly or indirectly utilizes class probability density functions, posterior probability density functions and the like to carry out classification identification. The method mainly comprises a K nearest neighbor method, a linear discrimination method, a Bayesian classification method and the like. The main problems of such methods are that the classification features need to be extracted manually, and the effect is not ideal under the condition that the target sample size is insufficient.
Therefore, how to provide a method for identifying and discriminating a rocket target that can improve the identification stability of the rocket target is a problem to be solved urgently.
Disclosure of Invention
In order to solve the problems in the prior art, the invention provides a method for judging the stability of rocket target postures based on transfer learning and a related device. The technical problem to be solved by the invention is realized by the following technical scheme:
the embodiment of the invention provides a method for judging the attitude stability of a rocket target based on transfer learning, which comprises the following steps:
step 1, obtaining input data to be classified, wherein the input data to be classified comprises RCS amplitude sequence values, mean values, range differences, mean line lengths and time information of rocket targets to be classified;
step 2, carrying out segmentation processing on the input data to be classified to obtain a plurality of short-time segment data, wherein partially repeated data are arranged between two adjacent short-time segment data;
step 3, inputting the short-time segment data into a trained target classification network with capsules to obtain a network classification result, wherein the network classification result is the confidence coefficient that the target is rocket body, rocket engine remains and rocket body fragments, and the trained target classification network is obtained through a transfer learning training method;
step 4, based on the continuous obtained in step 3PAnd judging the secondary network classification result according to a DS evidence theory algorithm to obtain a judgment result.
In one embodiment of the present invention, the step 2 comprises:
taking m as a step, and taking a fixed number of data points from the RCS amplitude sequence value, the average value, the range, the average line length and the time information each time as short-time subsection data, wherein the fixed number n is greater than m, and m and n are integers greater than zero.
In an embodiment of the present invention, the target classification network includes a first convolution layer, a second convolution layer, a third convolution layer, a first capsule layer and a second capsule layer, which are connected in sequence, wherein the first capsule layer includes M capsules, and the second capsule layer includes 3 capsules.
In an embodiment of the present invention, the method for training the target classification network includes:
training the target classification network by using spatial target training data with classification marks to obtain an initially trained target classification network, wherein the spatial target training data comprises RCS amplitude sequence values, mean values, range differences, mean line lengths and time information of spatial targets for training;
and training the initially trained target classification network by using rocket target training data with classification marks to obtain a finally trained target classification network, wherein the rocket target training data comprises RCS amplitude sequence values, mean values, range differences, mean line lengths and time information of rocket targets for training.
In one embodiment of the present invention, the step 4 comprises:
step 4.1 continuous from step 3PObtaining a conflict factor from the secondary network classification result;
and 4.2, obtaining the judgment result according to the network classification result obtained in the step 3 for continuous P times and the conflict factor based on a combined mass function of the DS evidence theory algorithm.
In one embodiment of the invention, the collision factor is expressed as:
wherein, the first and the second end of the pipe are connected with each other,
a factor of the conflict is represented by,
which is indicative of the type of the object,
the target is shown as the body of the rocket,
the representation is targeted to rocket engine debris,
the target is represented as a fragment of an arrow body,
representing a target classification network
The target type of the secondary network classification result is
The degree of confidence of (a) is,
，
which means that the sum is given,
indicating a continuous multiplication.
In one embodiment of the invention, the combined mass function is represented as:
wherein the content of the first and second substances,
to represent
The mass function of (1).
An embodiment of the present invention further provides a device for determining rocket target attitude stabilization based on transfer learning, where the device for determining rocket target attitude stabilization includes:
the system comprises an acquisition module, a classification module and a classification module, wherein the acquisition module is used for acquiring input data to be classified, and the input data to be classified comprises an RCS amplitude sequence value, an average value, a range, an average line length and time information of a rocket target to be classified;
the processing module is used for carrying out segmentation processing on the input data to be classified to obtain a plurality of short-time segmentation data, wherein partially repeated data exists between every two adjacent short-time segmentation data;
the classification module is used for inputting the short-time section data into a trained target classification network with capsules to obtain a network classification result, wherein the network classification result is the confidence coefficient that the target is rocket body, rocket engine remains and rocket body fragments, and the trained target classification network is obtained through the training of a transfer learning training method;
a discrimination module for obtaining the continuity based on the classification modulePAnd judging the secondary network classification result according to a DS evidence theory algorithm to obtain a judgment result.
An embodiment of the present invention further provides an electronic device, including a processor and a memory coupled to each other, wherein:
the memory is for storing program instructions for implementing a method as described in any one of the embodiments above;
the processor is configured to execute the program instructions stored by the memory.
An embodiment of the present invention further provides a computer-readable storage medium storing a program file, the program file being executable to implement the method according to any one of the above embodiments.
Compared with the prior art, the invention has the following beneficial effects:
the invention provides a discrimination method which utilizes RCS amplitude sequence value, mean value, range and average line length measured by a radar as characteristic data, carries out classification recognition on rocket targets based on a target classification network trained by using a transfer learning method, outputs confidence degrees of three types of target recognition of rocket main body targets, rocket engine debris targets and rocket body debris targets, and utilizes DS (Dempster-Shafer, information fusion) evidence theory algorithm to improve recognition stability. The discrimination method provided by the invention has higher classification accuracy.
Other aspects and features of the present invention will become apparent from the following detailed description, which proceeds with reference to the accompanying drawings. It is to be understood, however, that the drawings are designed solely for purposes of illustration and not as a definition of the limits of the invention, for which reference should be made to the appended claims. It should be further understood that the drawings are not necessarily drawn to scale and that, unless otherwise indicated, they are merely intended to conceptually illustrate the structures and procedures described herein.
Drawings
Fig. 1 is a schematic flowchart of a method for judging rocket target attitude stability based on transfer learning according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a capsule network layer according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of a target classification network according to an embodiment of the present invention;
fig. 4 is a schematic structural diagram of a rocket target attitude stabilization decision apparatus based on transfer learning according to an embodiment of the present invention;
fig. 5 is a schematic structural diagram of an electronic device according to an embodiment of the present invention;
fig. 6 is a schematic structural diagram of a computer-readable storage medium according to an embodiment of the present invention.
Detailed Description
The present invention will be described in further detail with reference to specific examples, but the embodiments of the present invention are not limited thereto.
Example one
Referring to fig. 1, fig. 1 is a schematic flow chart of a method for judging the attitude stability of a rocket target based on transfer learning according to an embodiment of the present invention, and the present invention provides a method for judging the attitude stability of a rocket target based on transfer learning, where the method for judging the attitude stability of a rocket target includes steps 1-4, where:
step 1, obtaining input data to be classified, wherein the input data to be classified comprises RCS amplitude sequence values, mean values, range differences, mean line lengths and time information of rocket targets to be classified.
Specifically, the neural network can extract effective and reliable recognition features from the self-learning of the RCS amplitude sequence values, but to enable the neural network to autonomously extract features, more training data volume is required, and the structure of the neural network is relatively complex, so that more training time is consumed, and the calculation process is slower in the model application of the neural network. Considering that the neural network needs to better meet practical application, the embodiment uses a mode that the RCS amplitude sequence value is combined with feature information extracted from the sequence to participate in classification together, reduces the requirement of the neural network on the training data volume, and reduces the structural complexity of the neural network. Wherein the RCS amplitude sequence value of the target a relative to the time t is defined as
Wherein, in the step (A),
the time instant at which the target RCS is measured for the radar,
is a sequence number of a time of day of the data,
for the total time of day of the data,
is as follows
RCS amplitude sequence values of time instants.
Among all data features, the stable features that can effectively distinguish rocket body targets from engine debris targets include: mean value, range, standard deviation, average line length, frequency point corresponding to maximum frequency spectrum value and frequency spectrum mean value; the stable characteristics that can effectively distinguish rocket body target and rocket body fragment target are: mean, maximum, range, mean line length; the stable characteristics that can effectively distinguish the engine debris target from the rocket body debris target are as follows: and the average value, the maximum value, the range, the average line length and the frequency point corresponding to the maximum frequency spectrum value. In summary, the input data is added with the average value, range, average line length, and time information as the identification feature data (i.e. the input data to be classified).
In this embodiment, the mean is the mean of the RCS amplitude sequence values, and is expressed as:
wherein, the first and the second end of the pipe are connected with each other,
is the mean value.
The range is the difference between the maximum value and the minimum value of the RCS amplitude sequence value, and the range is expressed as:
wherein the content of the first and second substances,
is the maximum value of the RCS amplitude sequence values,
is the minimum value of the RCS amplitude sequence values,
is extremely poor.
The average line length is the average length of the distance between each RCS amplitude sequence value, and is expressed as:
wherein the content of the first and second substances,
is as follows
The RCS amplitude sequence value for the time instance,
is the average line length.
The radar measurement timing is typically not at equal intervals, so the time information is taken as the time difference, i.e.:
wherein the content of the first and second substances,
the time at which the target RCS is measured for the radar,
is the time difference.
And 2, carrying out segmentation processing on the input data to be classified to obtain a plurality of short-time segment data, wherein partially repeated data exists between two adjacent short-time segment data.
Specifically, when the radar tracks dense targets, a target jumping and batch phenomenon may occur, so that short-time segmented data is adopted for input data to be classified, and a judgment result is prevented from being influenced by data confusion.
In this embodiment, step 2 may specifically include:
taking m as step, and taking a fixed number of data points from the RCS amplitude sequence value, the mean value, the range, the average line length and the time information each time as short-time segmentation data, wherein the fixed number n is more than m, and m and n are integers more than zero.
That is, the short-time segment data is formed by respectively taking a fixed number of data combinations from the RCS amplitude sequence value, the average value, the range, the average line length and the time information each time, then stepping a data point with the size of m, and extracting the next short-time segment data, thereby obtaining a plurality of short-time segment data, and because the fixed number n is more than m, partial repeated data can be ensured between two adjacent short-time segment data. Because the length of the short-time segment data is fixed, when the equipment measurement frequency is higher, the RCS amplitude sequence value needs to be reasonably sampled, and the short-time segment data with enough total time span is ensured.
It should be noted that, the length of the short time segment data is a fixed value according to the requirement, and considering that the data amount and the time length need to be considered, the fixed number of the invention takes 20 continuous points as a group for input, that is, n takes 20, and the matrix dimension of the short time segment data formed by the method is
。
And 3, inputting the short-time segment data into the trained target classification network with capsules to obtain a network classification result, wherein the network classification result is the confidence coefficient of the rocket body, the rocket engine remains and the rocket body fragments, and the trained target classification network is obtained through the transfer learning training method.
Further, the target classification network comprises a first coiling layer, a second coiling layer, a third coiling layer, a first capsule layer and a second capsule layer which are sequentially connected, wherein the first capsule layer comprises M capsules, and the second capsule layer comprises 3 capsules.
Specifically, the capsule network is a neural network composed of capsules. A capsule is a set of neurons whose output vectors represent instantiation parameters of a particular entity type. A capsule is a group of neurons that represent a vector composed of a set of scalar neurons, the modular length of which represents the probability that a particular entity or feature exists for the vector value output by a capsule, and the orientation of which represents an instantiation parameter (e.g., position, size, attitude, etc.). A schematic diagram of the capsule network layer is shown in FIG. 2, FIG. 2
And
each represents a vector in which, among other things,
、
、
、
a vector representing the output of the capsule of the previous layer,
、
、
、
、
a vector representing the output of the next layer of capsules.
In order to constrain the die length of the capsule output vector within a certain range, a squash (squeeze) function is generally used as the activation function of the capsule layer, namely:
wherein the content of the first and second substances,
is the value of the vector output by the capsule,
as a vector
Die length of (2).
A capsule receives the output vectors of several capsules in the previous layer as input, and passes through a conversion matrix with parameters capable of being trained
And mapping the output vector of the capsule of the previous layer to the vector space of the capsule of the current layer, and updating the output vector value of the capsule of the current layer through a dynamic routing algorithm.
In addition, because the present embodiment adds the feature information except the RCS amplitude sequence value, the target classification network only consists of three convolutional layers and two capsule layers, and the structure of the target classification network is as shown in fig. 3, wherein for the data output after the third convolutional layer, the data format is adjusted to convert the data into a grouped vector form, and the grouped vector form is used as the input of the first capsule layer, the number M of the capsules of the first capsule layer is preferably 10-20, the last layer of the second capsule layer has 3 capsules, and the module lengths of the 3 capsules respectively represent the confidence degrees that the target classification network judges the input target to be a rocket body, a rocket engine debris, and a rocket body fragment.
In addition, the embodiment is trained by the transfer learning training method to obtain a trained target classification network, because the transfer learning is a machine learning method, which reuses a pre-trained model in another similar task. Transfer learning can transfer learned knowledge from one scene to another, typically by transferring models of cat and dog classification to similar tasks, such as resolving eagles and cuckoos (as real image data, extracting features in the same way, and to some extent, having common features).
In this embodiment, in order to solve the problem that a rocket target data set is a small sample and is difficult to directly perform target classification network training, it is noted that there is similarity between an orbiting satellite, a rocket body, and space debris in a space target and a rocket body, rocket engine debris, and rocket body debris in the rocket target in part of physical properties. Namely, the on-orbit satellite and the rocket body have the characteristics of relatively stable attitude and regular physical shape; rocket bodies and rocket engine remains have the characteristics of relatively regular physical shapes and larger sizes, but have the characteristic of spin tumbling generally because the postures of the rocket bodies are not controlled; the space debris has the same characteristics of irregular physical shape and small size as the rocket body debris, and the attitude change is usually severe because the space debris is a projectile during rocket separation.
Further, data analysis shows that the target classification network can be trained by a migration learning method on the basis that the orbits and rocket bodies, rocket bodies and rocket engine remains, and space fragments and rocket body fragments have high similarity on RCS sequence characteristics in combination with actual conditions that space target data is convenient to obtain in large quantities, namely, the target classification network is trained by a large quantity of space target data, and then the obtained target classification network is migrated to small sample data of rocket targets for further training and optimization, so that the target classification network with better performance is obtained.
Based on the above reasons, the method for training the target classification network may specifically include:
s1, training a target classification network by using spatial target training data with classification marks to obtain an initially trained target classification network, wherein the spatial target training data comprises RCS amplitude sequence values, mean values, range, average line lengths and time information of a spatial target for training;
and S2, training the initially trained target classification network by utilizing rocket target training data with classification marks to obtain the finally trained target classification network, wherein the rocket target training data comprises RCS amplitude sequence values, mean values, range differences, mean line lengths and time information of rocket targets for training.
In the specific training process, a model of the target classification network is constructed first, and model parameters of the target classification network are initialized. The method comprises the steps of training a model of a target classification network initialized by parameters by using space target training data with classification marks until the accuracy of the model on a space target verification set is not increased (namely the testing accuracy of the target classification network on the space target verification set is changed from continuous increase to decrease and is stopped), thus obtaining the initially trained target classification network, pre-training the model by using the space target training data, continuing training by using rocket target training data until the accuracy of the rocket target training data on the rocket target verification set is not increased, stopping training at the moment to obtain model parameters of the finally trained target classification network, and testing the performance of the target classification network after transfer learning on a test set. It should be noted that the training data used in this embodiment may also be segmented data, and the specific segmentation method is described in step 2, and is not described herein again.
In summary, the design of the target classification network is the core of the classification recognition algorithm, and there are many actually available network structures, such as a support vector machine classifier, a K nearest neighbor classifier, a BP (Back Propagation) neural network classifier, and the like. The capsule network is a neural network model newly proposed in the field of machine learning, and as a novel classifier, the performance of the capsule network is fully verified in the fields of image recognition and the like. When the image is two-dimensional matrix data and the RCS amplitude sequence value is one-dimensional matrix data, and the RCS amplitude sequence value and the statistical characteristic value thereof are organized into a two-dimensional matrix as input data, the RCS amplitude sequence value of the object can be identified by the capsule network.
The actual conditions are limited, the number of targets of the real rocket measured by the radar is small, and therefore effective actual measurement rocket target data which can be directly used for target classification network training is limited. If the rocket target data is directly used for training the target classification network, the performance of the target classification network is limited by small data size and is not ideal. Considering that the measurement data acquisition difficulty of the space target is lower than that of the rocket target, the available data amount is more, and the space target class and the rocket target class have stronger corresponding relation on some physical attributes, a transfer learning method can be adopted, the space target data is used for pre-training the target classification network, and the rocket target data is used for further training and optimizing the target classification network on the basis.
Step 4, based on the continuous obtained in step 3PAnd judging the secondary network classification result according to a DS (Dempster-Shafer, information fusion) evidence theory algorithm to obtain a judgment result.
Specifically, because the target classification network uses the short-time segment data as input, the input and output of the short-time segment data bring disadvantages in the stability of output results, and therefore, a discrimination layer is added on the output results of the target classification network to correlate the front and rear classification results, thereby improving the stability of classification confidence.
The DS evidence theory algorithm is an uncertain reasoning method, is popularized to a Bayesian reasoning method, does not need prior probability, and has the advantage of simple reasoning form. The embodiment utilizes the theory to design a discrimination algorithm suitable for the calculation result of the target classification network.
According to the definition of DS evidence theory algorithm, since there are 3 classification results in the present embodiment, the set of discriminant types in the present embodiment is set as
Wherein, in the step (A),
the target is shown as the body of the rocket,
the representation is targeted to rocket engine debris,
representing the target as an arrow body fragment; classifying result output by network for discriminating evidence as target classification
Wherein, in the step (A),
representing a target classification network
The secondary network classification result is
The degree of confidence of (a) is,
representing a target classification network
The secondary network classification result is
The degree of confidence of (a) is,
representing a target classification network
The secondary network classification result is
Of (2) and
the secondary classification results require:
。
based on the above, step 4 may specifically include steps 4.1 to 4.2, where:
step 4.1 continuation obtained according to step 3PSecondary network classification nodeAnd obtaining the conflict factor.
In particular, continuouslyPThe secondary network classification result is a continuous network classification result obtained by inputting continuous P short-time segment data into the target classification networkPAs a result, the collision factor is expressed as:
wherein, the first and the second end of the pipe are connected with each other,
a factor of the conflict is represented by,
the type of the object is represented by,
the target is shown as the body of the rocket,
the representation is targeted to rocket engine debris,
the target is represented as a fragment of an arrow body,
representing a target classification network
The target type of the secondary network classification result is
The degree of confidence of (a) is,
，
which means that the sum is given,
indicating a continuous multiplication.
Closer to 1, the greater the conflict between evidences, i.e., the greater the divergence between features; on the contrary, the method can be used for carrying out the following steps,
closer to 0, the less the divergence between features, the higher the consistency.
And 4.2, obtaining a judgment result according to the network classification result obtained in the step 3 for continuous P times and the conflict factor based on a combined mass function of the DS evidence theory algorithm.
Specifically, the Dempster combination rule (i.e., DS evidence theory) based on P network classification results can record the combined mass function as:
wherein the content of the first and second substances,
to represent
The mass function of (2).
The combined mass function value is the discrimination confidence coefficient, and the above formula shows that when the confidence coefficient of a certain type (rocket body target, rocket engine debris target or rocket body fragment target) in the single network classification result is maximum, the recognition tendency that the target is the type is maximum. However, in a single network classification result, the confidence of the target in other types is also high, and after calculation by a DS evidence theory algorithm, the identification tendency is further strengthened, namely, the discrimination confidence of the target in the type is increased, and the discrimination confidence of the target in the other types is reduced. In the same principle, after P times of network classification results are calculated by a DS evidence theory algorithm, the discrimination confidence degree aiming at the type is continuously and stably kept to be maximum, and a dominance tendency is formed, so that discrimination is formed. That is, in this embodiment, after one of three discrimination confidences obtained by calculating the P-times network classification result through the combined mass function is greater than a set threshold, it may be determined that the target is the type corresponding to the discrimination confidence, and in this embodiment, when the discrimination confidence of the certain type is greater than 0.8, the discrimination target is the type.
Preferably, in order to enable the target classification network to quickly respond to the situation that the radar tracks the target jump, the times for calculating the identification result of the combined mass function are not too large, and the times are takenPAnd (5) = 3. In actual measurement, the local fluctuation of the RCS amplitude sequence value obtained by measurement is possibly unstable due to the influence of precession and nutation of the target motion state and the influence of target jump tracking of a radar, so that the judgment confidence coefficient output by a target classification network fluctuates to a certain degree, and the DS evidence theory algorithm is used for comprehensively judging the recognition results of multiple times before and after the network, so that the final output result is kept relatively stable.
For example: the 1 st, 2 nd and 3 rd network classification results output by the target classification network are shown in table 1.
TABLE 1 output results of the target Classification network
| 1 st output of the result | 2 nd output of the result | 3 rd output result | Value of the collision factor K | Combined Mass function values | |
| Confidence of rocket body | 0.1 | 0.15 | 0.1 | 0.4505 | 0.003 |
| Confidence of rocket engine debris | 0.7 | 0.8 | 0.8 | 0.4505 | 0.994 |
| Confidence of arrow body fragment | 0.2 | 0.05 | 0.1 | 0.4505 | 0.003 |
Wherein, the conflict factor K value is:
the combined Mass function values are:
as can be seen from table 1, the output results of the target classification network for 3 times are that the confidence of the rocket engine debris is high, and the output results are 0.7 and 0.8. After the calculation of the DS evidence theory algorithm, the Mass function value of rocket engine remains reaches 0.99, the confidence coefficient of other types of targets becomes lower, and the 3 Mass function values are used as the final judgment result of three types of targets, so that the output result is more stable.
The invention not only designs a target classification network based on a capsule network, which takes rocket target RCS original sequence data and characteristic quantity thereof as input; the spatial target RCS sequence data with large sample size is used as pre-training data, a transfer learning method is used for training a target classification network, and the difficulty that the rocket target RCS sequence data is small in sample size is solved; and judging a plurality of continuous network classification results output by the target classification network through a DS evidence theory algorithm to obtain stable and definite classification results.
In summary, the invention uses the object classification network with capsules as a classifier model, and applies the object classification network and the discrimination algorithm to the classification of rocket objects at the same time. The target classification network with capsules provided by the invention is one of the latest artificial neural network models in the field of machine learning, and the actual inspection shows that the target classification network with capsules has good performance on solving the multi-classification problem. Considering that few available training samples are obtained through actual measurement, the method utilizes a transfer learning method, firstly uses spatial target training data to pre-train a classifier model of a target classification network, then transfers the pre-trained target classification network to rocket target training data for further training and optimization, and finally adopts a DS evidence theory algorithm to judge the classification result of the target classification network, so that the output of the judgment confidence coefficient is more stable. Simulation experiments show that the method provided by the invention has higher classification accuracy, can overcome the problems existing in actual measurement such as tracking mixed batch and the like to a certain extent, can effectively solve the problem that artificial neural network training is difficult to directly carry out due to less rocket target samples, exerts the advantages of autonomous learning and memory and reliable identification characteristics of the neural network, and finally obtains a more stable identification and judgment result through DS evidence judgment.
Example two
Referring to fig. 4, fig. 4 is a schematic structural diagram of an embodiment of a rocket target attitude stabilization decision device based on transfer learning, which includes:
the system comprises an acquisition module, a classification module and a classification module, wherein the acquisition module is used for acquiring input data to be classified, and the input data to be classified comprises an RCS amplitude sequence value, an average value, a range, an average line length and time information of a rocket target to be classified;
the processing module is used for carrying out segmentation processing on the input data to be classified to obtain a plurality of short-time segment data, wherein partially repeated data are arranged between every two adjacent short-time segment data;
the classification module is used for inputting the short-time section data into a trained target classification network with capsules to obtain a network classification result, wherein the network classification result is the confidence coefficient that the target is rocket body, rocket engine remains and rocket body fragments, and the trained target classification network is obtained through the training of a transfer learning training method;
a discrimination module for obtaining the continuity based on the classification modulePAnd judging the secondary network classification result according to a DS evidence theory algorithm to obtain a judgment result.
EXAMPLE III
Referring to fig. 5, fig. 5 is a schematic structural diagram of an electronic device provided in the present invention. The electronic device comprises a processor 51 and a memory 52 connected to each other.
The memory 52 is used to store program instructions implementing the method of any one of the above.
The processor 51 is operative to execute program instructions stored in the memory 52.
The processor 51 may also be referred to as a CPU (Central Processing Unit). The processor 51 may be an integrated circuit chip having signal processing capabilities. The processor 51 may also be a general purpose processor, a digital signal processor, an application specific integrated circuit, a field programmable gate array or other programmable logic device, discrete gate or transistor logic, discrete hardware components. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.
The storage 52 may be a memory bank, a TF card, etc., and may store all information in the electronic device, including the raw data input, the computer program, the intermediate operation results, and the final operation results. It stores and retrieves information based on the location specified by the controller. With the memory, the electronic device can only guarantee normal operation if it has memory function. The memory of the electronic device is classified into a main memory (internal memory) and an auxiliary memory (external memory) according to the use, and also classified into an external memory and an internal memory. The external memory is usually a magnetic medium, an optical disk, or the like, and can store information for a long period of time. The memory refers to a storage component on the main board, which is used for storing data and programs currently being executed, but is only used for temporarily storing the programs and the data, and the data is lost when the power is turned off or the power is cut off.
In the embodiments provided in the present invention, it should be understood that the disclosed method and apparatus can be implemented by other methods. For example, the above-described apparatus implementation methods are merely illustrative, e.g., the division of modules or units into only one logical functional division, and additional division methods may be implemented in practice, e.g., units or components may be combined or integrated into another system, or some features may be omitted, or not implemented. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection through some interfaces, devices or units, and may be in an electrical, mechanical or other form.
The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one position, or may be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the embodiment of the method.
In addition, functional units in the embodiments of the present application may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit. The integrated unit can be realized in a form of hardware, and can also be realized in a form of a software functional unit.
The integrated unit, if implemented in the form of a software functional unit and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present application may be substantially implemented or contributed to by the prior art, or all or part of the technical solution may be embodied in a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a system server, a network device, or the like) or a processor (processor) to execute all or part of the steps of the implementation method of the present application.
Example four
Referring to fig. 6, fig. 6 is a schematic structural diagram of a computer-readable storage medium according to the present invention. The storage medium of the present invention stores a program file 61 capable of implementing all the methods, wherein the program file 61 may be stored in the storage medium in the form of a software product, and includes several instructions to enable a computer device (which may be a personal computer, a server, or a network device) or a processor to execute all or part of the steps of each implementation method of the present application. The foregoing storage device includes: various media capable of storing program codes, such as a U disk, a removable hard disk, a read-only memory, a random access memory, a magnetic disk or an optical disk, or terminal equipment, such as a computer, a server, a mobile phone, a tablet and the like.
In the description of the invention, the terms "first", "second" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implying any number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of the present invention, "a plurality" means two or more unless specifically defined otherwise.
In the description herein, references to the description of the term "one embodiment," "some embodiments," "an example," "a specific example," or "some examples," etc., mean that a particular feature, structure, material, or characteristic data point described in connection with the embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the terms used above are not necessarily intended to refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristic data points described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, various embodiments or examples described in this specification can be combined and combined by those skilled in the art.
The foregoing is a more detailed description of the invention in connection with specific preferred embodiments and it is not intended that the invention be limited to these specific details. For those skilled in the art to which the invention pertains, several simple deductions or substitutions can be made without departing from the spirit of the invention, and all shall be considered as belonging to the protection scope of the invention.
Claims (8)
1. A rocket target attitude stability discrimination method based on transfer learning is characterized by comprising the following steps:
step 1, obtaining input data to be classified, wherein the input data to be classified comprises RCS amplitude sequence values, mean values, range differences, mean line lengths and time information of rocket targets to be classified;
step 2, carrying out segmentation processing on the input data to be classified to obtain a plurality of short-time segment data, wherein partially repeated data are arranged between two adjacent short-time segment data;
step 3, inputting the short-time segment data into a trained target classification network with capsules to obtain a network classification result, wherein the network classification result is the confidence coefficient that the target is rocket body, rocket engine remains and rocket body fragments, and the trained target classification network is obtained through a transfer learning training method;
step 4, based on the continuous obtained in step 3PJudging the secondary network classification result according to a DS evidence theory algorithm to obtain a judgment result;
the step 4 comprises the following steps:
step 4.1 continuation obtained according to said step 3PObtaining a conflict factor from the secondary network classification result;
step 4.2, based on a combined mass function of the DS evidence theory algorithm, obtaining the discrimination result according to the network classification result obtained in the step 3 for continuous P times and the conflict factor;
the collision factor is expressed as:
wherein, the first and the second end of the pipe are connected with each other,
a factor of the conflict is represented by,
the type of the object is represented by,
the target is shown as the body of the rocket,
the representation is targeted to rocket engine debris,
the target is represented as a fragment of an arrow body,
representing a target classification network
The secondary network classification result is of the target type
The degree of confidence of (a) is,
，
which means that the sum is given,
indicating a continuous multiplication.
2. The method for judging attitude stabilization of rocket targets based on transfer learning according to claim 1, wherein the step 2 comprises:
taking m as a step, and taking a fixed number of data points from the RCS amplitude sequence value, the average value, the range, the average line length and the time information each time as short-time subsection data, wherein the fixed number n is greater than m, and m and n are integers greater than zero.
3. A rocket target attitude stabilization discrimination method based on transfer learning according to claim 1, wherein the target classification network comprises a first convolution layer, a second convolution layer, a third convolution layer, a first capsule layer and a second capsule layer which are connected in sequence, wherein the first capsule layer comprises M capsules, and the second capsule layer comprises 3 capsules.
4. The method for discriminating rocket target attitude stabilization based on transfer learning according to claim 3, wherein the training method of the target classification network comprises:
training the target classification network by using spatial target training data with classification marks to obtain an initially trained target classification network, wherein the spatial target training data comprises RCS amplitude sequence values, mean values, range differences, mean line lengths and time information of spatial targets for training;
and training the initially trained target classification network by using rocket target training data with classification marks to obtain a finally trained target classification network, wherein the rocket target training data comprises RCS amplitude sequence values, mean values, range differences, mean line lengths and time information of rocket targets for training.
5. A method for judging rocket target attitude stabilization based on transfer learning according to claim 1, characterized in that said combined mass function is expressed as:
wherein, the first and the second end of the pipe are connected with each other,
to represent
The mass function of (2).
6. A rocket target attitude stability discrimination device based on transfer learning is characterized by comprising the following components:
the system comprises an acquisition module, a classification module and a classification module, wherein the acquisition module is used for acquiring input data to be classified, and the input data to be classified comprises an RCS amplitude sequence value, an average value, a range, an average line length and time information of a rocket target to be classified;
the processing module is used for carrying out segmentation processing on the input data to be classified to obtain a plurality of short-time segment data, wherein partially repeated data are arranged between every two adjacent short-time segment data;
the classification module is used for inputting the short-time section data into a trained target classification network with capsules to obtain a network classification result, wherein the network classification result is the confidence coefficient that the target is rocket body, rocket engine remains and rocket body fragments, and the trained target classification network is obtained through the training of a transfer learning training method;
a discrimination module for obtaining continuity based on the classification modulePJudging the secondary network classification result according to a DS evidence theory algorithm to obtain a judgment result;
the continuity obtained based on the classification modulePAnd judging the secondary network classification result according to a DS evidence theory algorithm to obtain a judgment result, wherein the judgment result comprises the following steps:
continuity obtained according to the classification modulePObtaining a conflict factor from the secondary network classification result;
based on a combined mass function of the DS evidence theory algorithm, obtaining the discrimination result according to the network classification result obtained by the classification module for continuous P times and the conflict factor;
the collision factor is expressed as:
wherein the content of the first and second substances,
a factor of the conflict is represented by,
which is indicative of the type of the object,
the target is shown as the body of the rocket,
the representation is targeted to rocket engine debris,
the target is represented as a fragment of an arrow body,
representing a target classification network
The target type of the secondary network classification result is
The degree of confidence of (a) is,
，
which means that the sum is given,
indicating a continuous multiplication.
7. An electronic device comprising a processor and a memory coupled to each other, wherein:
the memory for storing program instructions for implementing the method of any one of claims 1-5;
the processor is configured to execute the program instructions stored by the memory.
8. A computer-readable storage medium, characterized in that a program file is stored, which program file can be executed to implement the method according to any one of claims 1-5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211143632.4A CN115270997B (en) | 2022-09-20 | 2022-09-20 | Rocket target attitude stability discrimination method based on transfer learning and related device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211143632.4A CN115270997B (en) | 2022-09-20 | 2022-09-20 | Rocket target attitude stability discrimination method based on transfer learning and related device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115270997A CN115270997A (en) | 2022-11-01 |
| CN115270997B true CN115270997B (en) | 2022-12-27 |
Family
ID=83756028
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211143632.4A Active CN115270997B (en) | 2022-09-20 | 2022-09-20 | Rocket target attitude stability discrimination method based on transfer learning and related device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115270997B (en) |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109918752A (en) * | 2019-02-26 | 2019-06-21 | 华南理工大学 | Mechanical failure diagnostic method, equipment and medium based on migration convolutional neural networks |
| CN110210320A (en) * | 2019-05-07 | 2019-09-06 | 南京理工大学 | The unmarked Attitude estimation method of multiple target based on depth convolutional neural networks |
| CN110858277A (en) * | 2018-08-22 | 2020-03-03 | 阿里巴巴集团控股有限公司 | Method and device for obtaining attitude classification model |
| CN111191690A (en) * | 2019-12-16 | 2020-05-22 | 上海航天控制技术研究所 | Spatial target autonomous identification method based on transfer learning, electronic device and storage medium |
| CN111460818A (en) * | 2020-03-31 | 2020-07-28 | 中国测绘科学研究院 | Web page text classification method based on enhanced capsule network and storage medium |
| CN111898652A (en) * | 2020-07-10 | 2020-11-06 | 西北工业大学 | Spatial target posture classification and identification method based on convolutional neural network |
| CN112183601A (en) * | 2020-09-22 | 2021-01-05 | 中国人民解放军63921部队 | Spatial object classification method, device and storage medium |
| CN114399829A (en) * | 2022-03-25 | 2022-04-26 | 浙江壹体科技有限公司 | Posture migration method based on generative countermeasure network, electronic device and medium |
| CN114648052A (en) * | 2020-12-18 | 2022-06-21 | 武汉智行者科技有限公司 | Multi-sensor information fusion algorithm based on DS evidence theory |
| CN115032602A (en) * | 2022-04-14 | 2022-09-09 | 杭州电子科技大学 | Radar target identification method based on multi-scale convolution capsule network |
Family Cites Families (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109117793B (en) * | 2018-08-16 | 2021-10-29 | 厦门大学 | Direct-push type radar high-resolution range profile identification method based on deep migration learning |
| CN109344871A (en) * | 2018-08-30 | 2019-02-15 | 西北工业大学 | A kind of target classification identification method based on multi-source field fusion transfer learning |
| CN109934276B (en) * | 2019-03-05 | 2020-11-17 | 安翰科技(武汉)股份有限公司 | Capsule endoscope image classification system and method based on transfer learning |
| EP3786845A1 (en) * | 2019-08-29 | 2021-03-03 | Robert Bosch GmbH | Difficulty-adaptive training for machine learning modules |
| CN111243066B (en) * | 2020-01-09 | 2022-03-22 | 浙江大学 | Facial expression migration method based on self-supervision learning and confrontation generation mechanism |
| US20210262329A1 (en) * | 2020-02-20 | 2021-08-26 | Exxonmobil Upstream Research Company | Method for Generating Initial Models For Least Squares Migration Using Deep Neural Networks |
| CN111860124B (en) * | 2020-06-04 | 2024-04-02 | 西安电子科技大学 | Remote sensing image classification method based on space spectrum capsule generation countermeasure network |
| CN112818941A (en) * | 2021-03-04 | 2021-05-18 | 西北大学 | Cultural relic fragment microscopic image classification method, system, equipment and storage medium based on transfer learning |
| CN114187506A (en) * | 2021-11-22 | 2022-03-15 | 武汉科技大学 | Remote sensing image scene classification method of viewpoint-aware dynamic routing capsule network |
| CN114819061A (en) * | 2022-04-02 | 2022-07-29 | 南京航空航天大学 | Sparse SAR target classification method and device based on transfer learning |
-
2022
- 2022-09-20 CN CN202211143632.4A patent/CN115270997B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110858277A (en) * | 2018-08-22 | 2020-03-03 | 阿里巴巴集团控股有限公司 | Method and device for obtaining attitude classification model |
| CN109918752A (en) * | 2019-02-26 | 2019-06-21 | 华南理工大学 | Mechanical failure diagnostic method, equipment and medium based on migration convolutional neural networks |
| CN110210320A (en) * | 2019-05-07 | 2019-09-06 | 南京理工大学 | The unmarked Attitude estimation method of multiple target based on depth convolutional neural networks |
| CN111191690A (en) * | 2019-12-16 | 2020-05-22 | 上海航天控制技术研究所 | Spatial target autonomous identification method based on transfer learning, electronic device and storage medium |
| CN111460818A (en) * | 2020-03-31 | 2020-07-28 | 中国测绘科学研究院 | Web page text classification method based on enhanced capsule network and storage medium |
| CN111898652A (en) * | 2020-07-10 | 2020-11-06 | 西北工业大学 | Spatial target posture classification and identification method based on convolutional neural network |
| CN112183601A (en) * | 2020-09-22 | 2021-01-05 | 中国人民解放军63921部队 | Spatial object classification method, device and storage medium |
| CN114648052A (en) * | 2020-12-18 | 2022-06-21 | 武汉智行者科技有限公司 | Multi-sensor information fusion algorithm based on DS evidence theory |
| CN114399829A (en) * | 2022-03-25 | 2022-04-26 | 浙江壹体科技有限公司 | Posture migration method based on generative countermeasure network, electronic device and medium |
| CN115032602A (en) * | 2022-04-14 | 2022-09-09 | 杭州电子科技大学 | Radar target identification method based on multi-scale convolution capsule network |
Non-Patent Citations (3)
| Title |
|---|
| 基于极限学习机参数迁移的域适应算法;许夙晖等;《自动化学报》;20171211(第02期);全文 * |
| 基于激光雷达的火箭垂直起飞段姿态测量技术;师恒 等;《光子学报》;20220430;第51卷(第4期);全文 * |
| 迁移学习下的火箭发动机参数异常检测策略;张晨曦，唐曙，唐珂;《计算机应用》;20200930;第40卷(第9期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115270997A (en) | 2022-11-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Wang et al. | Symmetric cross entropy for robust learning with noisy labels | |
| Pei et al. | SAR automatic target recognition based on multiview deep learning framework | |
| US9811718B2 (en) | Method and a system for face verification | |
| Zhou et al. | Automatic radar waveform recognition based on deep convolutional denoising auto-encoders | |
| CN111950656B (en) | Image recognition model generation method and device, computer equipment and storage medium | |
| CN109063719B (en) | Image classification method combining structure similarity and class information | |
| CN111353463A (en) | Hyperspectral image classification method based on random depth residual error network | |
| CN112529146B (en) | Neural network model training method and device | |
| CN110705489B (en) | Training method and device for target recognition network, computer equipment and storage medium | |
| CN110210625A (en) | Modeling method, device, computer equipment and storage medium based on transfer learning | |
| US11568212B2 (en) | Techniques for understanding how trained neural networks operate | |
| CN111223128A (en) | Target tracking method, device, equipment and storage medium | |
| CN113902944A (en) | Model training and scene recognition method, device, equipment and medium | |
| CN110135428B (en) | Image segmentation processing method and device | |
| CN115392357A (en) | Classification model training and labeled data sample spot inspection method, medium and electronic equipment | |
| Ouf | Leguminous seeds detection based on convolutional neural networks: Comparison of faster R-CNN and YOLOv4 on a small custom dataset | |
| CN114595352A (en) | Image identification method and device, electronic equipment and readable storage medium | |
| Fu et al. | An instance selection approach to multiple instance learning | |
| CN115270997B (en) | Rocket target attitude stability discrimination method based on transfer learning and related device | |
| CN114155388B (en) | Image recognition method and device, computer equipment and storage medium | |
| CN114693997A (en) | Image description generation method, device, equipment and medium based on transfer learning | |
| CN115410250A (en) | Array type human face beauty prediction method, equipment and storage medium | |
| Zou et al. | An intelligent image feature recognition algorithm with hierarchical attribute constraints based on weak supervision and label correlation | |
| Chen et al. | Feature fusion based on convolutional neural network for SAR ATR | |
| CN114065901A (en) | Method and device for training neural network model |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |