CN116500546A - Radar signal sorting method based on point cloud segmentation network - Google Patents
Radar signal sorting method based on point cloud segmentation network Download PDFInfo
- Publication number
- CN116500546A CN116500546A CN202310512899.4A CN202310512899A CN116500546A CN 116500546 A CN116500546 A CN 116500546A CN 202310512899 A CN202310512899 A CN 202310512899A CN 116500546 A CN116500546 A CN 116500546A
- Authority
- CN
- China
- Prior art keywords
- data
- unit
- pdw
- point cloud
- layer
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 75
- 230000011218 segmentation Effects 0.000 title claims abstract description 24
- 230000005855 radiation Effects 0.000 claims description 23
- 238000010606 normalization Methods 0.000 claims description 19
- 238000012549 training Methods 0.000 claims description 19
- 238000001208 nuclear magnetic resonance pulse sequence Methods 0.000 claims description 17
- 230000008569 process Effects 0.000 claims description 16
- 238000012545 processing Methods 0.000 claims description 16
- 230000000694 effects Effects 0.000 claims description 7
- 238000005070 sampling Methods 0.000 claims description 7
- 238000012795 verification Methods 0.000 claims description 6
- 238000012360 testing method Methods 0.000 claims description 4
- 238000002372 labelling Methods 0.000 claims description 2
- 238000013528 artificial neural network Methods 0.000 abstract description 4
- 238000004891 communication Methods 0.000 abstract description 2
- 238000013136 deep learning model Methods 0.000 abstract description 2
- 238000004422 calculation algorithm Methods 0.000 description 12
- 238000003064 k means clustering Methods 0.000 description 6
- 238000005516 engineering process Methods 0.000 description 5
- 230000002776 aggregation Effects 0.000 description 4
- 238000004220 aggregation Methods 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 230000001186 cumulative effect Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000003709 image segmentation Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000005192 partition Methods 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
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
- G01S7/021—Auxiliary means for detecting or identifying radar signals or the like, e.g. radar jamming signals
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Abstract
A radar signal sorting method based on a point cloud segmentation network belongs to the field of electronic reconnaissance of information and communication engineering. The method solves the problems that the parameter setting of the traditional radar signal sorting method needs to depend on manual experience, the sorting flow is complex, the sorting capability of unknown signals is poor, and the radar signal sorting requirement under the current complex electromagnetic environment is difficult to meet. The invention combines the deep learning model with radar signal sorting, realizes an end-to-end sorting scheme without manual intervention from the original PDW to the EDW, and simplifies the sorting realization flow. And because the neural network can deeply mine the characteristic association between the PDWs of different types of radars, the neural network still has better sorting capability for unknown radar signals in a non-cooperative scene. The method can be applied to radar signal sorting.
Description
Technical Field
The invention belongs to the field of electronic reconnaissance of information and communication engineering, and particularly relates to a radar signal sorting method based on a point cloud segmentation network.
Background
The radar signal sorting technology has the main functions of de-interlacing and de-aliasing mixed pulse signals received by a receiver, distinguishing different radiation source types according to the difference of radar characteristic parameters and providing decision information for subsequent electronic reconnaissance and electronic countermeasure. Therefore, radar signal sorting technology is one of the key technologies for electronic countermeasure, around which many scholars have intensively studied and proposed many methods. Current radar signal sorting algorithms are mostly based on pulse descriptors (Pulse Description Word, PDW) for analysis and implementation. The PDW includes radar characteristic parameter information such as an Arrival angle (Direction of Arrival, DOA), carrier frequency (Carrier Frequency, CF), pulse Width (PW), pulse Amplitude (PA), and Pulse Arrival Time (TOA). The radar signal sorting is achieved by analyzing the differences of the PDWs of different pulses, and the sorting result is described by using radiation source pulse description words (Emitter Description Word, EDW).
The traditional radar signal sorting method combining the pre-sorting process and the main sorting process is widely applied to actual engineering and is also a traditional radar signal sorting technology. The main function of the pre-sorting process is to dilute the original radar pulse stream density, and clustering algorithms such as a k-means clustering algorithm and an improved algorithm thereof, a density-based clustering algorithm DBSCAN and the like are commonly adopted. The main sorting process is mainly to count pulse repetition intervals (Pulse Repetition Interval, PRI) of each group of pulses after clustering, and complete main sorting by matching with pulse searching. The main methods of this process are PRI conversion, sequence difference histogram (Cumulative Difference Histogram, CDIF) and cumulative difference histogram (Sequential Difference Histogram, SDIF). The conventional radar signal sorting method has the following limitations:
(1) Parameters such as a clustering center, the number of clusters, a parameter threshold value, tolerance and the like need to be set in advance, the parameters are unknown in a non-cooperative scene, the setting of the parameters depends on manual experience, and the complex and variable sorting environment cannot be adapted.
(2) The sorting process is complex, and the flexibility is poor due to mutual restriction between the front structure and the rear structure.
The rapid development of modern electronic countermeasure technology and radar has made the current electromagnetic environment more complex to perform in the time, frequency and space domains than ever before. Firstly, the pulse flow density of radar signals reaches the millions, and the signal overlapping is serious; secondly, the common application of radars with various novel complex systems makes sorting overlapping radar pulse sequences more difficult. Therefore, the conventional radar signal sorting algorithm is difficult to meet the radar signal sorting requirement under the current complex electromagnetic environment, and the sorting capability of the conventional radar signal sorting algorithm on unknown signals is poor.
Disclosure of Invention
The invention aims to solve the problems that parameter setting of a traditional radar signal sorting method needs to depend on manual experience, a sorting flow is complex, the sorting capability of unknown signals is poor, and the radar signal sorting requirement under the current complex electromagnetic environment is difficult to meet.
The technical scheme adopted by the invention for solving the technical problems is as follows: a radar signal sorting method based on a point cloud segmentation network specifically comprises the following steps:
step one, intercepting PDW data of continuous N pulses from an original radar pulse sequence received by a receiver according to time sequence to form PDW point cloud data;
the PDW data of each pulse comprises TOA, PW, CF and PA data, namely the dimension of the PDW point cloud data is N multiplied by 4;
step two, carrying out normalization processing on the PDW point cloud data;
step three, inputting the normalized PDW point cloud data into a trained point cloud segmentation network, predicting the radar signal type of the PDW data of each pulse through the trained point cloud segmentation network, and outputting the PDW point cloud data with the dimension of N multiplied by 5 and label information, namely marking the radar signal type of the PDW data of each pulse after normalization;
step four, performing inverse normalization on the PDW point cloud data with the label information output in the step three to obtain original PDW point cloud data with the label;
step five, classifying the PDW data with the same label into the PDW data of the same type of radar signal pulse in the original PDW point cloud data with the label obtained in the step four;
for PDW data of any radar signal pulse, processing the PDW data by adopting a clustering method to distinguish different radar radiation sources corresponding to the radar signal pulse and extract EDW information of each radar radiation source;
similarly, PDW data for each type of radar signal pulse is processed.
The beneficial effects of the invention are as follows:
the invention provides a radar signal sorting method based on a point cloud segmentation network, which has good sorting effect on PDWs with serious overlapping of time domain, frequency domain and space domain and large pulse flow density, can still keep higher sorting accuracy under the condition that pulses have a certain loss rate, and can adapt to the current complex electromagnetic environment; the deep learning model is combined with radar signal sorting, so that an end-to-end sorting scheme without manual intervention from the original PDW to the EDW is realized, and a sorting realization flow is simplified. And because the neural network can deeply mine the characteristic association between the PDWs of different types of radars, the neural network still has better sorting capability for unknown radar signals in a non-cooperative scene.
Experimental results show that the sorting accuracy can reach 97.46% under the condition that the pulse loss rate is 20%, and meanwhile, the method also has the advantage of high sorting efficiency.
Drawings
FIG. 1 is a flow chart of a method for sorting radar signals based on a point cloud segmentation network according to the present invention;
FIG. 2 is a schematic diagram of a PointNet++ network architecture;
FIG. 3 (a) is a schematic diagram of an original PDW pulse sequence;
FIG. 3 (b) is a side view of a PDW point cloud prediction;
FIG. 3 (c) is a top view of the PDW point cloud prediction results;
FIG. 3 (d) is a front view of the PDW point cloud prediction results;
FIG. 4 is a graph comparing sorting effects of different sorting methods.
Detailed Description
The invention will be described in further detail below with reference to the drawings by means of specific embodiments. It will be apparent that the described embodiments are only some, but not all, of the embodiments of the invention. Based on the embodiments of the present invention, other embodiments obtained by those skilled in the art without making any inventive effort are within the scope of the present invention.
Detailed description of the inventionin the first embodiment, this embodiment will be described with reference to fig. 1. The radar signal sorting method based on the point cloud segmentation network specifically comprises the following steps:
step one, intercepting PDW data of continuous N pulses from an original radar pulse sequence received by a receiver according to time sequence to form PDW point cloud data;
the PDW data of each pulse comprises TOA, PW, CF and PA data, namely the dimension of the PDW point cloud data is N multiplied by 4;
according to the method, radar signal sorting can be achieved only by adopting TOA, PW, CF and PA data in PDW data, compared with the existing method, the data quantity to be processed is reduced, the selected data is combined with the PointNet++ model, and the accuracy of radar signal sorting is remarkably improved;
step two, carrying out normalization processing on the PDW point cloud data;
step three, inputting the normalized PDW point cloud data into a trained point cloud segmentation network, predicting the radar signal type of the PDW data of each pulse through the trained point cloud segmentation network, and outputting the PDW point cloud data with the dimension of N multiplied by 5 and label information, namely marking the radar signal type of the PDW data of each pulse after normalization; the last dimension stores the tag information of the radar signal type to which the PDW data of each pulse belongs;
step four, performing inverse normalization on the PDW point cloud data with the label information output in the step three to obtain original PDW point cloud data with the label;
step five, classifying the PDW data with the same label into the PDW data of the same type of radar signal pulse in the original PDW point cloud data with the label obtained in the step four;
for PDW data of any radar signal pulse, processing the PDW data by adopting a clustering method to distinguish different radar radiation sources corresponding to the radar signal pulse and extract EDW information of each radar radiation source;
similarly, PDW data for each type of radar signal pulse is processed.
And fifthly, classifying the PDW data with the same label into a group, further dividing radar radiation sources according to the PDW information of the group, and forming EDW information of the radar radiation sources.
The EDW information extraction method of the radar radiation sources of different types comprises the following steps:
1. PRI is a conventional radar. The PRI of the PRI conventional radar is almost fixed or the jitter amount is not more than 1%, and the CF and PW of the same PRI conventional radar are kept unchanged or the jitter amount is not more than 1%. Therefore, for PRI conventional radars, CF and PW can be subjected to two-stage clustering in sequence by using a dynamic k-means clustering algorithm to obtain a preliminary grouping result, then PRI values of each group of results are counted, and PRI conventional radars are further divided according to the difference of the PRI values.
2. PRI dithered radar. The PRI dither radar is a radar in which PRI dithers at a certain central PRI value and the dithering rate is generally not more than 30%, and both CF and PW of the same PRI dither radar remain unchanged or the dithering amount is not more than 1%. Therefore, for PRI (pulse-rate indicator) dithering radars, CF and PW can be clustered in two stages in sequence by using a dynamic k-means clustering algorithm to obtain a preliminary grouping result, then the central PRI value and the dithering rate of each group of results are counted and calculated, and PRI dithering radars are further divided according to the difference.
3. PRI spread radar. The PRI spread radar is composed of a plurality of sub-periods PRI and is repeated in sequence within the frame period PRI. Both CF and PW of the same PRI spread radar remain unchanged or the amount of jitter does not exceed 1%. And for PRI spread radars, a dynamic k-means clustering algorithm is utilized to sequentially perform two-stage clustering on CF and PW to obtain a preliminary grouping result, then the sub-period PRI value, the frame period PRI value and the spread number of each group of results are counted and calculated, and different PRI spread radars are continuously distinguished according to the difference of the values.
4. Frequency agile radar. The PRI and PW values of frequency agile radar are typically fixed values, with CF constantly changing over different pulses. The CF of the frequency agility radar makes random jump in a agility range which is 15% -20% of the working center frequency of the frequency agility radar. For the radiation sources, the dynamic k-means clustering algorithm can be utilized to cluster PW to finish preliminary grouping, PRI of each group of pulse sequences is calculated, radar radiation source individuals are further distinguished according to the difference of PRI, the center frequency and agility range of each radiation source individual are calculated to finish final grouping, and EDW information of each radiation source individual is formed.
5. Pulse set frequency agile radar. The CF variation of pulse group agile radar has a frame period comprising several sub-periods, a special form of agile radar. For the radiation sources, a dynamic k-means clustering algorithm is also utilized to cluster PW to form a preliminary grouping result, PRI of each group of pulse sequences is calculated, the PRI is further divided according to PRI differences, and finally agile frequency points and agile range of each radiation source individual are counted to form EDW information of each radiation source individual.
The existing sorting method using the image segmentation network needs to map the original PDW data into images, the data processing amount is large, and the problem of point overlapping exists in the mapping caused by the limitation of the image resolution, so that the processing efficiency and the performance are limited. According to the method, firstly, original PDW data is directly divided into a plurality of sub clusters according to different radar types through a PointNet++ point cloud segmentation network, so that PDW pulse streams with high density and serious overlap are grouped according to the radar types, meanwhile, the density of the original PDW pulse streams can be diluted, and then different radar radiation sources are divided for each group of pulses by using a clustering method. No manual intervention exists in the sorting process, no manual parameter setting is relied on, and no restriction exists between structures, so that the method is applicable to the current complex electromagnetic environment. The method does not need to carry out complex transformation on the original PDW data, so that the data processing amount is small, the sorting efficiency is obviously improved compared with the prior machine learning method due to the light-weight structure of the model, the situation that the PDW data is lost or distorted is avoided, the original PDW information can be completely reserved, and the method is an end-to-end intelligent sorting scheme.
The second embodiment is as follows: this embodiment differs from the specific embodiment in that the value of N is not less than 1024.
In this embodiment, the value of the parameter N is set to be not less than 1024, so that the throughput of data can be reduced as much as possible while the sorting accuracy is ensured.
Other steps and parameters are the same as in the first embodiment.
And a third specific embodiment: the difference between this embodiment and the first or second embodiment is that the specific process of the second step is:
for TOA data in PDW data, the process of normalizing the TOA data is as follows:
wherein t is i TOA data, t, in the intercepted ith pulse PDW data i,nor TOA data, t, in the normalized ith pulse PDW data means The mean value of TOA data in PDW data of N pulses, and sigma is the variance of TOA data in PDW data of N pulses;
for CF, PW, and PA data in PDW data, the normalization method is the same as for TOA data.
The size of the parameter is limited to [ -1,1 by normalization]The dimension influence among different characteristic parameters can be eliminated, the statistical distribution of the unified sample can be induced, and the problem that the accuracy of the follow-up model prediction is reduced due to singular point cloud data is avoided. Normalization processing is carried out while normalization parameters t of the PDW point cloud data of the batch are required to be reserved means And sigma, so as to extract the original PDW point cloud data when the inverse normalization is performed subsequently.
Other steps and parameters are the same as in the first or second embodiment.
The specific embodiment IV is as follows: this embodiment will be described with reference to fig. 2. The difference between this embodiment and one to three embodiments is that the point cloud partition network is a pointnet++ model, and the pointnet++ model specifically includes: a first SA unit (Set Abstraction Layer, SA), a second SA unit, a third SA unit, a first FP unit (Feature Propagation, FP), a second FP unit, a third FP unit, and a full connection layer.
As shown in fig. 2, the left half part is an encoding part of the PointNet++ network, and the encoding part gradually downsamples the point cloud to mainly realize the local feature space encoding of the point cloud; the right half part is a decoding part of the PointNet++ network, each point in the point cloud is up-sampled, the original point cloud structure is gradually restored, and meanwhile, the characteristic information aggregation of each point in the point cloud is realized; and finally, carrying out semantic discrimination on each point through the full connection layer.
In the encoding section, the purpose of the SA unit structure is to learn the structure and geometry information from point to point, encode it onto part of the points by downsampling, and then aggregate each point its neighborhood information by subsequent decoding operations. The normalized PDW point cloud data is characterized in that 256 points with the farthest distance from the normalized PDW point cloud data are randomly sampled through a first SA unit, then a local space is constructed by searching neighborhood points, each space contains 16 points, and finally feature information of each space is obtained through a PointNet layer. The PDW point cloud data passes through the first SA unit to obtain 256 sampling points and global features of local spaces of the sampling points, and the global features of the local spaces are used in subsequent decoding layers after being obtained. These 256 samples continue through the second SA unit and the third SA unit to build deeper features for the decoding section.
The main purpose of the decoding part is to map the local space information obtained by coding aggregation to all point clouds of the point cloud set to obtain the point characteristics of each original PDW point cloud in the PDW point clouds. The specific implementation process is that the FP unit is utilized to transmit the characteristics of the points from the previous unit to the next unit until the original PDW point cloud data is restored.
Other steps and parameters are the same as in one to three embodiments.
Fifth embodiment: this embodiment is different from one to four of the embodiments in that the first, second, and third SA units have the same structure, and each SA unit includes a sampling layer, a packet layer, and a PointNet layer.
Each SA unit is composed of three associated network structures, namely a Sampling Layer (Sampling Layer) for point cloud Sampling point selection, a Grouping Layer (Grouping Layer) for constructing a local space, and a PointNet Layer for feature aggregation.
Other steps and parameters are the same as in one to four embodiments.
Specific embodiment six: this embodiment is different from one of the first to fifth embodiments in that the first FP unit, the second FP unit, and the third FP unit have the same structure, and each FP unit includes an IDW layer, a Concate layer, and a PointNet layer;
taking the normalized PDW point cloud data as the input of the first SA unit to obtain the output of the first SA unit;
taking the output of the first SA unit as the input of the second SA unit to obtain the output of the second SA unit;
taking the output of the second SA unit as the input of the third SA unit to obtain the output of the third SA unit;
the output of the third SA unit is used as the input of the IDW layer of the first FP unit, the output of the IDW layer of the first FP unit is spliced with the output of the third SA unit, and the splicing result is used as the input of the Concate layer of the first FP unit;
taking the output of the PointNet layer of the first FP unit as the input of the IDW layer of the second FP unit, splicing the output of the IDW layer of the second FP unit with the output of the second SA unit, and taking the splicing result as the input of the Concate layer of the second FP unit;
taking the output of the PointNet layer of the second FP unit as the input of the IDW layer of the third FP unit, splicing the output of the IDW layer of the third FP unit with the output of the first SA unit, and taking the splicing result as the input of the Concate layer of the third FP unit;
and then taking the output of the PointNet layer of the third FP unit as the input of a full connection layer, and outputting a radar signal type prediction result of the PDW data of each pulse in the normalized PDW point cloud data through the full connection layer.
The output of the IDW layer is used as the input of the PointNet layer in the same FP unit, the IDW layer is used for calculating interpolation features of each point of the shallow network, the Concate layer is used for feature stitching, and the PointNet layer is used for feature aggregation.
Each hierarchy contains PointNet layers, and the composition of the PointNet layers in the different hierarchies is shown in Table 1:
table 1 PointNet structure for each level
Other steps and parameters are the same as in one of the first to fifth embodiments.
Seventh embodiment: the difference between this embodiment and one to six embodiments is that the interpolation method of the IDW layer is:
wherein f (j) (x) The j-th characteristic value of unknown point, f l (j) For the j-th feature value of the first known point selected, C is a feature number, l=1, 2, … L, L represents the number of known points selected from the set of known points, w l (x) Is the weight of the first known point and the unknown point, which is inversely proportional to the distance between the unknown point and the known point, the closer the distance, the greater the effect, d (x, x l ) Is the distance between the unknown point and the first known point, and p represents the influence factor of the distance on the weight.
The invention adopts tri-linear interpolation, namely, p=2 and k=3 are taken, 3 points closest to an unknown point are selected from a known point set to carry out interpolation calculation, interpolation characteristics of 64 points are obtained after interpolation is completed, characteristics of 64 points obtained in a second SA unit are spliced with the interpolation characteristics to obtain new characteristics, and the characteristics are aggregated through a PointNet layer. And sequentially passing the point features acquired by the first FP unit through the second FP unit and the third FP unit to acquire new point features. And finally, processing the characteristics obtained by the third FP unit through the full connection layer to obtain a prediction result of the PDW data of each pulse, wherein the output of the network is PDW point cloud data with a prediction label.
Other steps and parameters are the same as in one of the first to sixth embodiments.
Eighth embodiment: this embodiment differs from one of the first to seventh embodiments in that the training process of the PointNet++ model is:
the method comprises the steps of respectively acquiring PDW data of pulse sequences of a single radar radiation source and PDW data of pulse sequences of a plurality of radar radiation sources (2 radars, 3 radars and more radars can be used for meeting the diversity of training data), and respectively intercepting the PDW data of each pulse sequence to form PDW point cloud data; carrying out normalization processing on the PDW point cloud data, and labeling the PDW point cloud data subjected to normalization processing to obtain a data set;
according to 8:1:1 randomly dividing the obtained data set into a training set, a verification set and a test set, training the PointNet++ model by using the training set, evaluating and adjusting the training effect of the PointNet++ model by using the verification set until the prediction accuracy of the PointNet++ model on the test set reaches a threshold value, stopping training and storing the PointNet++ model, wherein the stored PointNet++ model is the trained PointNet++ model.
The super-parameters settings at the time of PointNet++ model training are shown in Table 2:
table 2 model super parameter settings
Adam is selected by the model parameter optimization function to adaptively adjust the learning rate of each parameter. The initial learning rate is 0.001, the total number of training samples is 1800, the number of batch processing samples is 16, the training iteration number is 120, so that the full coverage of the training samples is ensured, training is stopped until the prediction accuracy meets the requirement, and a model with successful training is stored.
Other steps and parameters are the same as those of one of the first to seventh embodiments.
Experimental part
(1) And (5) feasibility verification. A group of PDW pulse sequences are randomly generated, the number of the pulses is 1024, and the pulse sequences comprise PRI conventional radars, PRI shaking radars, PRI raging radars, agile radars and pulse group agile radars, and the number of the radars is ten. The TOA, PW and CF are visualized as coordinates as shown in FIG. 3 (a).
Considering that certain jitter exists in each characteristic parameter of PDW in an actual sorting environment, certain jitter is set for PW, CF and PA in the simulation process. After being predicted by the PointNet++ model, the total pulse number is 1015, the total sorting accuracy is 99.12%, the sorting accuracy of different types of radars is shown in table 3, and the visualized results are shown in fig. 3 (b), 3 (c) and 3 (d). Through the experiment, the feasibility of the method for sorting the multi-radar signals is verified.
TABLE 3 sorting accuracy of various radars
(2) And (5) verifying reliability. In order to further verify the sorting reliability of the method in a complex environment, a group of PDW pulse sequences are randomly generated, the number of the pulses is 1024, the types and the number of radars contained in the feasibility verification process are contained, and certain jitter exists in all radar parameters. Except that the pulses had a 20% loss rate to simulate the PDW received in a complex sorting environment. After being predicted by the PointNet++ model, the total pulse number is 998, the sorting accuracy is 97.46%, and the sorting accuracy of different types of radars is shown in Table 4. It can be seen that the method of the invention can still maintain a higher sorting accuracy under the condition of higher pulse loss rate. For PRI conventional radars, PRI ragged radars and pulse group frequency agile radars, the radar characteristic parameters of the type show a certain regularity of radars, the regularity of the radars is locally destroyed under the condition of high pulse loss rate, and the traditional method is difficult to sort correctly. The method of the invention deeply digs and combines the local features and the global features among the PDWs, so that the local features and the global features are concerned and can be mutually complemented when the prediction is carried out. Therefore, the method can still sort correctly under the environment of high pulse loss rate.
TABLE 4 sorting accuracy of various radars
(3) Sorting method comparison and performance analysis. In order to further explain the sorting performance of the method, the sorting effect of the method is compared with that of other sorting methods, and under the condition of different pulse loss rates, the same PDW pulse sequence is sorted by different methods respectively to obtain a relation curve of the sorting accuracy of the different methods along with the change of the pulse loss rate, as shown in figure 4. As can be seen from fig. 4, the conventional method has a lower sorting accuracy at a high pulse loss rate, and is more sensitive to pulse loss. The sorting effect of the method under the same condition is superior to that of other sorting methods, the method is less sensitive to pulse loss, and the sorting accuracy is still kept high under a complex environment. The sorting performance of the method is superior to that of other methods, the method can adapt to the current complex electromagnetic environment, and the sorting requirement of the current electronic warfare is met.
The above examples of the present invention are only for describing the calculation model and calculation flow of the present invention in detail, and are not limiting of the embodiments of the present invention. Other variations and modifications of the above description will be apparent to those of ordinary skill in the art, and it is not intended to be exhaustive of all embodiments, all of which are within the scope of the invention.
Claims (8)
1. The radar signal sorting method based on the point cloud segmentation network is characterized by comprising the following steps of:
step one, intercepting PDW data of continuous N pulses from an original radar pulse sequence received by a receiver according to time sequence to form PDW point cloud data;
the PDW data of each pulse comprises TOA, PW, CF and PA data, namely the dimension of the PDW point cloud data is N multiplied by 4;
step two, carrying out normalization processing on the PDW point cloud data;
step three, inputting the normalized PDW point cloud data into a trained point cloud segmentation network, predicting the radar signal type of the PDW data of each pulse through the trained point cloud segmentation network, and outputting the PDW point cloud data with the dimension of N multiplied by 5 and label information, namely marking the radar signal type of the PDW data of each pulse after normalization;
step four, performing inverse normalization on the PDW point cloud data with the label information output in the step three to obtain original PDW point cloud data with the label;
step five, classifying the PDW data with the same label into the PDW data of the same type of radar signal pulse in the original PDW point cloud data with the label obtained in the step four;
for PDW data of any radar signal pulse, processing the PDW data by adopting a clustering method to distinguish different radar radiation sources corresponding to the radar signal pulse and extract EDW information of each radar radiation source;
similarly, PDW data for each type of radar signal pulse is processed.
2. The method for sorting radar signals based on a point cloud segmentation network according to claim 1, wherein the value of N is not less than 1024.
3. The radar signal sorting method based on the point cloud segmentation network according to claim 2, wherein the specific process of the second step is as follows:
for TOA data in PDW data, the process of normalizing the TOA data is as follows:
wherein t is i TOA data, t, in the intercepted ith pulse PDW data i,nor TOA data, t, in the normalized ith pulse PDW data means The mean value of TOA data in PDW data of N pulses, and sigma is the variance of TOA data in PDW data of N pulses;
for CF, PW, and PA data in PDW data, the normalization method is the same as for TOA data.
4. The radar signal sorting method based on the point cloud segmentation network according to claim 3, wherein the point cloud segmentation network is a PointNet++ model, and the PointNet++ model specifically comprises: the first SA unit, the second SA unit, the third SA unit, the first FP unit, the second FP unit, the third FP unit and the full connection layer.
5. The method for sorting radar signals based on a point cloud segmentation network according to claim 4, wherein the first SA unit, the second SA unit and the third SA unit have the same structure, and each SA unit includes a sampling layer, a grouping layer and a PointNet layer.
6. The method for sorting radar signals based on a point cloud segmentation network according to claim 5, wherein the first FP unit, the second FP unit, and the third FP unit have the same structure, and each FP unit includes an IDW layer, a Concate layer, and a PointNet layer;
taking the normalized PDW point cloud data as the input of the first SA unit to obtain the output of the first SA unit;
taking the output of the first SA unit as the input of the second SA unit to obtain the output of the second SA unit;
taking the output of the second SA unit as the input of the third SA unit to obtain the output of the third SA unit;
the output of the third SA unit is used as the input of the IDW layer of the first FP unit, the output of the IDW layer of the first FP unit is spliced with the output of the third SA unit, and the splicing result is used as the input of the Concate layer of the first FP unit;
taking the output of the PointNet layer of the first FP unit as the input of the IDW layer of the second FP unit, splicing the output of the IDW layer of the second FP unit with the output of the second SA unit, and taking the splicing result as the input of the Concate layer of the second FP unit;
taking the output of the PointNet layer of the second FP unit as the input of the IDW layer of the third FP unit, splicing the output of the IDW layer of the third FP unit with the output of the first SA unit, and taking the splicing result as the input of the Concate layer of the third FP unit;
and then taking the output of the PointNet layer of the third FP unit as the input of a full connection layer, and outputting a radar signal type prediction result of the PDW data of each pulse in the normalized PDW point cloud data through the full connection layer.
7. The method for sorting radar signals based on the point cloud segmentation network according to claim 6, wherein the interpolation method of the IDW layer is as follows:
wherein f (j) (x) The j-th characteristic value of unknown point, f l (j) For the j-th feature value of the first known point selected, C is a feature number, l=1, 2, … L, L represents the number of known points selected from the set of known points, w l (x) Is the weight of the first known point and the unknown point, d (x, x l ) Is the distance between the unknown point and the first known point, and p represents the influence factor of the distance on the weight.
8. The method for sorting radar signals based on the point cloud segmentation network according to claim 7, wherein the training process of the PointNet++ model is as follows:
the method comprises the steps of respectively acquiring PDW data of pulse sequences of a single radar radiation source and PDW data of pulse sequences of a plurality of radar radiation sources, and respectively intercepting the PDW data of each pulse sequence to form PDW point cloud data; carrying out normalization processing on the PDW point cloud data, and labeling the PDW point cloud data subjected to normalization processing to obtain a data set;
according to 8:1:1 randomly dividing the obtained data set into a training set, a verification set and a test set, training the PointNet++ model by using the training set, evaluating and adjusting the training effect of the PointNet++ model by using the verification set until the prediction accuracy of the PointNet++ model on the test set reaches a threshold value, stopping training and storing the PointNet++ model, wherein the stored PointNet++ model is the trained PointNet++ model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310512899.4A CN116500546A (en) | 2023-05-08 | 2023-05-08 | Radar signal sorting method based on point cloud segmentation network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310512899.4A CN116500546A (en) | 2023-05-08 | 2023-05-08 | Radar signal sorting method based on point cloud segmentation network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116500546A true CN116500546A (en) | 2023-07-28 |
Family
ID=87318100
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310512899.4A Pending CN116500546A (en) | 2023-05-08 | 2023-05-08 | Radar signal sorting method based on point cloud segmentation network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116500546A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117272086A (en) * | 2023-11-22 | 2023-12-22 | 中国电子科技集团公司第二十九研究所 | Radar signal scanning envelope segmentation method based on DBSCAN |
-
2023
- 2023-05-08 CN CN202310512899.4A patent/CN116500546A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117272086A (en) * | 2023-11-22 | 2023-12-22 | 中国电子科技集团公司第二十九研究所 | Radar signal scanning envelope segmentation method based on DBSCAN |
| CN117272086B (en) * | 2023-11-22 | 2024-02-13 | 中国电子科技集团公司第二十九研究所 | Radar signal scanning envelope segmentation method based on DBSCAN |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112364943B (en) | Federal prediction method based on federal learning | |
| Knoblauch et al. | Spatio-temporal Bayesian on-line changepoint detection with model selection | |
| CN111081016B (en) | Urban traffic abnormity identification method based on complex network theory | |
| CN107292097B (en) | Chinese medicine principal symptom selection method based on feature group | |
| CN108564788B (en) | Method for discovering vehicles in same line based on streaming big data | |
| CN110519128B (en) | Random forest based operating system identification method | |
| CN110751121B (en) | Unsupervised radar signal sorting method based on clustering and SOFM | |
| CN106528874B (en) | The CLR multi-tag data classification method of big data platform is calculated based on Spark memory | |
| CN100595782C (en) | Classification method for syncretizing optical spectrum information and multi-point simulation space information | |
| CN113591380B (en) | Traffic flow prediction method, medium and equipment based on graph Gaussian process | |
| CN104166859A (en) | Polarization SAR image classification based on SSAE and FSALS-SVM | |
| CN109117856B (en) | Intelligent edge cloud-based person and object tracking method, device and system | |
| CN116500546A (en) | Radar signal sorting method based on point cloud segmentation network | |
| CN108647691A (en) | A kind of image classification method based on click feature prediction | |
| CN114595732B (en) | Radar radiation source sorting method based on depth clustering | |
| CN111464469B (en) | Hybrid digital modulation mode identification method based on neural network | |
| CN113759323B (en) | Signal sorting method and device based on improved K-Means joint convolution self-encoder | |
| CN112817755A (en) | Edge cloud cooperative deep learning target detection method based on target tracking acceleration | |
| CN111598325A (en) | Traffic speed prediction method based on hierarchical clustering and hierarchical attention mechanism | |
| CN114254141A (en) | End-to-end radar signal sorting method based on depth segmentation | |
| CN110932807A (en) | Spectrum sensing method of MIMO (multiple input multiple output) system under non-Gaussian noise | |
| CN114004336A (en) | Three-dimensional ray reconstruction method based on enhanced variational self-encoder | |
| CN106570104A (en) | Multi-partition clustering preprocessing method of stream data | |
| Xian et al. | A novel intrusion detection method based on clonal selection clustering algorithm | |
| CN113133078B (en) | Light-weight inter-satellite switching device and method for giant low-orbit satellite network |
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 |