CN113655457B - Self-evolution method and device for radar target detection capability based on sample mining - Google Patents
Self-evolution method and device for radar target detection capability based on sample mining Download PDFInfo
- Publication number
- CN113655457B CN113655457B CN202110978334.6A CN202110978334A CN113655457B CN 113655457 B CN113655457 B CN 113655457B CN 202110978334 A CN202110978334 A CN 202110978334A CN 113655457 B CN113655457 B CN 113655457B
- Authority
- CN
- China
- Prior art keywords
- radar
- target
- time
- sample
- track
- 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
- 238000001514 detection method Methods 0.000 title claims abstract description 36
- 238000000034 method Methods 0.000 title claims abstract description 33
- 238000005065 mining Methods 0.000 title claims abstract description 18
- 238000012549 training Methods 0.000 claims abstract description 42
- 238000010587 phase diagram Methods 0.000 claims description 25
- 238000002372 labelling Methods 0.000 claims description 17
- 239000003550 marker Substances 0.000 claims description 6
- 238000007781 pre-processing Methods 0.000 claims description 5
- 238000004364 calculation method Methods 0.000 claims description 3
- 238000013473 artificial intelligence Methods 0.000 abstract description 8
- 238000004422 calculation algorithm Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 5
- 238000013213 extrapolation Methods 0.000 description 5
- 238000012545 processing Methods 0.000 description 5
- 238000013528 artificial neural network Methods 0.000 description 4
- 230000015572 biosynthetic process Effects 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 238000009795 derivation Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000035485 pulse pressure Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000010187 selection method Methods 0.000 description 1
- 238000007711 solidification Methods 0.000 description 1
- 230000008023 solidification Effects 0.000 description 1
- 230000007704 transition 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
- G01S7/41—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
- G01S7/411—Identification of targets based on measurements of radar reflectivity
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Radar, Positioning & Navigation (AREA)
- Remote Sensing (AREA)
- Computer Networks & Wireless Communication (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
The invention discloses a self-evolution method of radar target detection capability based on sample mining, which comprises the following steps: generating a training sample set; training the model by using a training sample set to generate a training model; detecting a target; digging a new sample; and (5) online learning. The self-evolution method of the radar target detection capability based on sample mining provided by the invention applies tracking logic, does not need manual access, and autonomously mines new samples, so that the self-evolution of the radar detection capability is realized, and the goal of changing from weak artificial intelligence to strong artificial intelligence is reached.
Description
Technical Field
The invention relates to the field of radar target detection, in particular to a self-evolution method and device of radar target detection capacity based on sample mining.
Background
The radar target detection technology is one of the most critical technologies in radar, and is mainly divided into a traditional detection method and an artificial intelligence method at present. The traditional radar target algorithm is mainly characterized in that: engineering technicians use forward derivation methods to strive to derive a physical model that best approximates the objective world for target detection, and the model's optimal inverse depends on the designer's own knowledge. The artificial intelligence method adopts a reverse method. The radar detection model obtains detection capability through learning of a large number of samples, and the superiority and the inverse of the model depend on the scale and coverage rate of training samples. Both algorithms have in common that once the model is cured, the detection capability of the radar is fixed.
Expert and scholars also use deep learning and neural networks to conduct self-evolution research on radar detection capability.
A self-evolution radar target detection algorithm (patent publication number: CN 109239669A) based on deep learning provides a method for realizing model self-evolution by applying a double-view collaborative training algorithm. The training algorithm mainly applies two different initial basic models, and the interaction sample is enhanced to realize the evolution of the respective models. From a global perspective, the new sample generated is still the result of the underlying model. Thus, global detectability fails to get true evolution.
A training sample mining method and device for a deep neural network (patent publication No. CN 109344873A) provides a sample selection method, by which the training speed of the network is improved, and the improvement of the detection capability of a network model cannot be realized.
Disclosure of Invention
In order to solve the problems, the invention provides a self-evolution method of radar target detection capability based on sample mining, which comprises the following steps:
step one: generating a training sample set: preprocessing a plurality of single-frame radar echo original data to generate a radar channel amplitude-phase diagram, and performing target labeling on the radar channel amplitude-phase diagram to form a training sample set;
step two: training the model by using a training sample set to generate a training model;
step three: and (3) target detection: detecting a radar channel amplitude-phase diagram in a training sample set through a training model, and calculating a radar point trace, wherein the radar point trace comprises a target azimuth, a target distance and radar time; tracking the point trace of the radar to form a track of the radar;
step four: new sample mining:
searching for a missing sample according to the radar track, and performing target marking on the missing sample to generate a new sample;
step five: on-line learning:
and (3) merging the new sample into a training sample set, and repeating the second to fifth steps.
Further, the radar time for missing samples includes:
the radar time of the starting point trace of the track-the time of the radar scanning one circle;
the radar time of the ending trace of the track + the time of the radar scan one turn.
Further, if the time difference between two adjacent tracks in the track is greater than the time of one circle of radar scanning, the radar time of missing samples further comprises T k+1 -T r T is as follows k +T r The method comprises the steps of carrying out a first treatment on the surface of the Wherein T is k Representing radar time, T, of a kth trace in a track k+1 Representing radar time, T, of the (k+1) th trace in the track r Indicating the time for the radar to scan one revolution.
Further, labeling the radar channel map includes target locations and target categories, the target locations being identified by coordinates (x min ,y tmin ,x max ,y tmax ) Representation, where x min And y tmin Respectively represent the target to be T in radar time t The abscissa and ordinate, x of the upper left corner of the label frame max And y tmax Respectively represent the target to be T in radar time t The abscissa and the ordinate of the lower right corner of the marking frame; the abscissa of the radar channel phase diagram is the channel pulse, and the ordinate is the range gate.
Further, labeling the missing samples is specifically as follows:
extracting the radar time of the missing sample from the timeThe radar echo data of the frame is used for generating a corresponding radar channel amplitude-phase diagram, and the radar time of the missing sample is assumed to be T l T is then l The target position of the marker frame of (2) is (x) lmin ,y lmin ,x lmax ,y lmax ) Wherein x is lmin And y lmin Respectively represent the target to be T in radar time l The abscissa and the ordinate of the upper left corner of the marking frame; x is x lmax And y lmax Respectively represent the target to be T in radar time l The abscissa and the ordinate of the right lower corner of the marking frame are respectivelyy lmin And y lmax The calculation mode of (2) is as follows:
let T be l The marker frame of the target that can be correctly detected at the previous time is (x min ,y kmin ,x max ,y kmax )，R l According to T l T calculated by radar track before moment l The target distance at the moment is as follows:
。
the device is realized by applying any method.
Compared with the prior art, the invention has the following beneficial effects:
1. the invention adopts a track tracking method, and discovers the sample with missing reasoning through extrapolation, thereby realizing independent autonomous sample mining without relying on manpower;
2. the invention uses the track tracking method, can infer which frame of radar echo data is extrapolated, and on that distance, a real target exists, thereby realizing the automatic labeling problem of new samples.
3. The invention realizes stronger and stronger detection capability through the self-evolution of the model, thereby solving the problems of more false alarms and weak detection capability of the radar.
4. According to the invention, through autonomous sample mining, more and more new samples can be mined under the condition of small samples, so that the problem of model training under the condition of small samples is solved.
5. The invention applies tracking logic, does not need manual access, and autonomously excavates new samples, thereby realizing the self-evolution of radar tracking capability and achieving the goal of changing weak artificial intelligence into strong artificial intelligence.
Drawings
FIG. 1 is a flow chart of the method of the present invention.
Fig. 2 is a radar channel chart according to a first embodiment.
Fig. 3 is a diagram of the result of the calculation in the first embodiment.
Fig. 4 is a diagram of a radar tracking track according to the first embodiment.
Fig. 5 is a schematic diagram of a new sample with labels according to the first embodiment.
Fig. 6 is a tracking trace plot of the model formation after online learning according to the first embodiment.
Detailed Description
The following describes a specific embodiment of a self-evolution method of radar target detection capability based on sample mining in detail with reference to the accompanying drawings.
Example 1
The processing flow chart of the self-evolution method of radar detection capability based on sample mining in this embodiment can be seen in fig. 1.
The processing procedure consists of 4 procedures of radar data preprocessing, target detection, new sample formation and online learning.
(1) Generating training sample sets
Preprocessing a plurality of single-frame radar echo original data to form a radar channel phase diagram capable of being calculated by a neural network, wherein a typical radar channel phase diagram can be seen in fig. 2;
preprocessing the part of work mainly to finish pulse pressure, channel correction and moving target detection; the operation is standard partial radar signal processing, and in actual products, the partial radar signal processing is mostly finished by adopting an FPGA or a DSP; and the finished product is mostly transmitted to a subsequent processing module through rapidIO.
After the radar channel phase diagram is formed, the radar channel phase diagram needs to be marked to generate a training sample set. The method for generating the radar channel amplitude and phase diagram refers to a patent document of a radar target detection method (patent application number: 202011641433.7).
The specific method of the labeling is to draw the position of the target in the radar channel amplitude-phase diagram, and the labeling content has two points: target position (x) min ,y tmin ,x max ,y tmax ) And a target class, wherein x min And y tmin Respectively represent the target to be T in radar time t The abscissa and ordinate, x of the upper left corner of the label frame max And y tmax Respectively represent the target to be T in radar time t The abscissa and the ordinate of the lower right corner of the marking frame; the abscissa of the radar channel phase diagram is the channel pulse, and the ordinate is the range gate. Since the amplitude of the radar channel amplitude-phase diagram is unchanged, x is min And x max Remain unchanged.
To improve the accuracy of labeling, there are generally three methods: a) Labeling the target by referring to GPS data; b) Labeling the target by referring to the CFAR detection result; c) And a skilled labeling person labels the target through experience accumulated for a long time. The manually labeled schematic and the automatically labeled schematic are the same, see fig. 5.
(2) Training a model by using a training sample set to generate a training model
The present embodiment trains the identification network using the Retinanet model. The identification category is a class (valid target). After the labeling of the training samples is completed, training is performed on Nvidia V100 for 8 days, and the solidification of the model is completed, so that a detection model is generated. The deep neural network may be selected by self-designed network, or by using public network (including YoloV3, faster RCNN or CenterNet).
(3) Target detection
Deducing a radar channel amplitude-phase diagram in a training sample set through a detection model, calculating a target distance in the frame radar echo and a direction of the frame radar echo to form a target position (polar coordinate), and combining radar time of the frame radar echo to form radar points (direction, distance and time); the result of the solution can be seen in FIG. 3; the black frame in fig. 3 indicates the target position, and the center point of the frame is taken as the distance gate where the target is located.
Tracking: and tracking the point trace of the radar to form a track of the radar. The radar track map of this embodiment can be seen in fig. 4, where the abscissa is the azimuth and the ordinate is the distance.
(5) New sample mining: by extrapolation of the track break points, it is inferred which frame and which distance the target cannot be detected. Then, firstly selecting radar echo according to the frame number, and generating a radar channel amplitude-phase diagram. Then, marking targets on the radar channel amplitude and phase diagram according to the reasoning result to form a new training sample, wherein the sample is a new sample for the initial training model; FIG. 5 can be seen as a schematic diagram of the generation of a new sample with labels; the method specifically comprises the following steps:
1) Missing sample search
When a batch of radar echo data is processed, a plurality of tracks are formed, and then missing samples are searched. Let a certain track be L, which is the set of all points of a certain target. Each trace includes at least a target azimuth, a target distance, and a radar time. The radar time difference between two adjacent points in a certain track is the time of the radar scanning one circle under the normal condition, and the radar time difference is set as T r 。
If T k+1 -T k >T r Indicating that there are missing samples between these two points. Therefore, the time corresponding to the missing sample to be found should be T k+1 -T r T is as follows k +T r Wherein T is k Representing radar time, T, of a kth trace in a track k+1 Representing radar time, T, of the (k+1) th trace in the track r Representing the time of the radar scan for one turn, i.e., the radar time difference for two adjacent spots.
There are two cases, namely the starting point and the ending point of the track, respectively, by extrapolation of the trackPushing out the front point of the starting point and the rear point of the ending point; setting the track starting point moment as T 0 The time of the initial point extrapolation is T 0 -T r The method comprises the steps of carrying out a first treatment on the surface of the Let the ending point of the track be T N The time of the end point extrapolation is T N +T r . These two extrapolated points also act as missing samples.
2) Automatic labeling of missing samples
When the radar time of the missing sample is found, the time is set as T l (T l The radar time of the missing sample is represented, and the trace is the trace corresponding to the newly mined missing sample). By T l For searching, extracting radar echo data of the frame at the time from radar echo data original data to form a radar channel amplitude-phase diagram, wherein the abscissa of the radar channel amplitude-phase diagram is a channel pulse, and the ordinate is a range gate. After the radar channel phase diagram is formed, T is calculated l The mark frame of the object is set as (x lmin ,y lmin ,x lmax ,y lmax )，x lmin And y lmin Respectively represent the target to be T in radar time l The abscissa and the ordinate of the upper left corner of the marking frame; x is x lmax And y lmax Respectively represent the target to be T in radar time l The abscissa and the ordinate of the lower right corner of the marking frame.
Since the target is detected on the single frame echo datax max -x min The amplitude of the radar channel amplitude-phase diagram is shown in fig. 3 and 5.
Estimating y lmin ，y lmax The method comprises the following steps: first according to T l Radar trace before time (or T l Radar trace after time), using least square method to calculate T l Target distance R at time l ，R l Is the ordinate of the center point of the mark frame, R l ＝(y lmin +y lmax )/2. Let T be l The marker frame of the target that can be correctly detected at the previous time is (x min ,y kmin ,x max ,y kmax ). Since it is the same target, there is y lmax -y lmin ＝y kmax -y kmin Thus y is lmin ＝R l -(y kmax -y kmin )/2，y lmax ＝R l +(y kmax -y kmin )/2. To this end, a marker frame of the missing sample is formed.
Finally, online learning is finished, a plurality of new samples formed by different track missing samples are fused into a training sample set, the training model is trained again, a new training model is formed, and model capacity is evolved accordingly. And detecting the radar echo data of the batch again by the evolved model, and forming a track diagram shown in fig. 6. From the figure, the amplified track in the box is newly detected by a new training model, the detection capability is enhanced, and the radar power is increased.
The new sample formation and online learning are the most central parts of the invention, and realize the self-evolution of radar tracking capability.
The invention applies tracking logic to find radar echo samples which can not detect targets, and the samples are brand new samples for an original model; the invention applies the tracking logic to realize the automatic labeling of the new sample, so that the new sample can automatically enter the training set; the invention can realize sample searching, sample marking and new model training automatic running operation, can realize self-evolution of radar detection capability without manual intervention, and realizes the transition from weak artificial intelligence to strong artificial intelligence.
The foregoing description of the preferred embodiments of the invention is not intended to limit the invention to the precise form disclosed, and any such modifications, equivalents, and alternatives falling within the spirit and scope of the invention are intended to be included within the scope of the invention.
Claims (4)
1. The self-evolution method of radar target detection capability based on sample mining is characterized by comprising the following steps:
step one: generating a training sample set: preprocessing a plurality of single-frame radar echo original data to generate a radar channel amplitude-phase diagram, and performing target labeling on the radar channel amplitude-phase diagram to form a training sample set;
step two: training the model by using a training sample set to generate a training model;
step three: and (3) target detection: detecting a radar channel amplitude-phase diagram in a training sample set through a training model, and calculating a radar point trace, wherein the radar point trace comprises a target azimuth, a target distance and radar time; tracking the point trace of the radar to form a track of the radar;
step four: new sample mining:
searching for a missing sample according to the radar track, and carrying out automatic target labeling on the missing sample to generate a new sample;
labeling the missing samples specifically comprises the following steps:
extracting radar echo data of the frame at the missing sample according to the radar time of the missing sample, generating a corresponding radar channel amplitude-phase diagram, and assuming the radar time of the missing sample to be T l T is then l The target position of the marker frame of (2) is (x) lmin ，y lmin ，x lmax ，y lmax ) Wherein x is lmin And y lmin Respectively represent the target to be T in radar time l The abscissa and the ordinate of the upper left corner of the marking frame; x is x lmax And y lmax Respectively represent the target to be T in radar time l The abscissa and the ordinate of the right lower corner of the marking frame are respectivelyy lmin And y lmax The calculation mode of (2) is as follows:
let T be l The marker frame of the target that can be correctly detected at the previous time is (x min ，y kmin ，x max ，y kmax )，R l According to T l T calculated by radar track before moment l The target distance at the moment is as follows:
the radar time for missing samples includes:
the radar time of the starting point trace of the track-the time of the radar scanning one circle;
the radar time of the ending point trace of the track+the time of one circle of radar scanning;
step five: on-line learning:
and (3) merging the new sample into a training sample set, and repeating the second to fifth steps.
2. The method for self-evolution of radar target detection capability based on sample mining according to claim 1, wherein if the time difference between two adjacent points in the track is greater than the time of one radar scan, the radar time for missing the sample further comprises T k+1 -T r T is as follows k +T r The method comprises the steps of carrying out a first treatment on the surface of the Wherein T is k Representing radar time, T, of a kth trace in a track k+1 Representing radar time, T, of the (k+1) th trace in the track r Indicating the time for the radar to scan one revolution.
3. The self-evolution method of radar target detection capability based on sample mining according to claim 2, wherein labeling the radar channel graph comprises target location and target class, the target location being defined by coordinates (x min ，y tmin ，x max ，y tmax ) Representation, where x min And y tmin Respectively represent the target to be T in radar time t The abscissa and ordinate, x of the upper left corner of the label frame max And y tmax Respectively represent the target to be T in radar time t The abscissa and the ordinate of the lower right corner of the marking frame; the abscissa of the radar channel phase diagram is the channel pulse, and the ordinate is the range gate.
4. A self-evolving device of radar target detection capability based on sample mining, characterized in that the device is implemented by applying the method of any of claims 1-3.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110978334.6A CN113655457B (en) | 2021-08-24 | 2021-08-24 | Self-evolution method and device for radar target detection capability based on sample mining |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110978334.6A CN113655457B (en) | 2021-08-24 | 2021-08-24 | Self-evolution method and device for radar target detection capability based on sample mining |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113655457A CN113655457A (en) | 2021-11-16 |
| CN113655457B true CN113655457B (en) | 2023-11-24 |
Family
ID=78492784
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110978334.6A Active CN113655457B (en) | 2021-08-24 | 2021-08-24 | Self-evolution method and device for radar target detection capability based on sample mining |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113655457B (en) |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107506444A (en) * | 2017-08-25 | 2017-12-22 | 中国人民解放军海军航空工程学院 | Interruption flight path, which continues, associates machine learning system |
| CN109117793A (en) * | 2018-08-16 | 2019-01-01 | 厦门大学 | Direct-push high Resolution Range Profile Identification of Radar method based on depth migration study |
| CN109239669A (en) * | 2018-08-16 | 2019-01-18 | 厦门大学 | A kind of self-evolution Radar Targets'Detection algorithm based on deep learning |
| CN109344873A (en) * | 2018-08-31 | 2019-02-15 | 北京智芯原动科技有限公司 | A kind of the training sample method for digging and device of deep neural network |
| CN111175719A (en) * | 2020-01-08 | 2020-05-19 | 中国船舶重工集团公司第七二四研究所 | Intelligent track starting method based on BP neural network |
| CN111694830A (en) * | 2020-06-12 | 2020-09-22 | 复旦大学 | Missing data completion method based on deep ensemble learning |
| CN111814582A (en) * | 2020-06-15 | 2020-10-23 | 开易（北京）科技有限公司 | Method and device for processing driver behavior monitoring image |
| CN112085125A (en) * | 2020-09-29 | 2020-12-15 | 西安交通大学 | Missing value filling method based on linear self-learning network, storage medium and system |
| CN112816982A (en) * | 2020-12-31 | 2021-05-18 | 中国电子科技集团公司第十四研究所 | Radar target detection method |
| CN112986950A (en) * | 2020-12-25 | 2021-06-18 | 南京理工大学 | Single-pulse laser radar echo feature extraction method based on deep learning |
| CN113011568A (en) * | 2021-03-31 | 2021-06-22 | 华为技术有限公司 | Model training method, data processing method and equipment |
-
2021
- 2021-08-24 CN CN202110978334.6A patent/CN113655457B/en active Active
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107506444A (en) * | 2017-08-25 | 2017-12-22 | 中国人民解放军海军航空工程学院 | Interruption flight path, which continues, associates machine learning system |
| CN109117793A (en) * | 2018-08-16 | 2019-01-01 | 厦门大学 | Direct-push high Resolution Range Profile Identification of Radar method based on depth migration study |
| CN109239669A (en) * | 2018-08-16 | 2019-01-18 | 厦门大学 | A kind of self-evolution Radar Targets'Detection algorithm based on deep learning |
| CN109344873A (en) * | 2018-08-31 | 2019-02-15 | 北京智芯原动科技有限公司 | A kind of the training sample method for digging and device of deep neural network |
| CN111175719A (en) * | 2020-01-08 | 2020-05-19 | 中国船舶重工集团公司第七二四研究所 | Intelligent track starting method based on BP neural network |
| CN111694830A (en) * | 2020-06-12 | 2020-09-22 | 复旦大学 | Missing data completion method based on deep ensemble learning |
| CN111814582A (en) * | 2020-06-15 | 2020-10-23 | 开易（北京）科技有限公司 | Method and device for processing driver behavior monitoring image |
| CN112085125A (en) * | 2020-09-29 | 2020-12-15 | 西安交通大学 | Missing value filling method based on linear self-learning network, storage medium and system |
| CN112986950A (en) * | 2020-12-25 | 2021-06-18 | 南京理工大学 | Single-pulse laser radar echo feature extraction method based on deep learning |
| CN112816982A (en) * | 2020-12-31 | 2021-05-18 | 中国电子科技集团公司第十四研究所 | Radar target detection method |
| CN113011568A (en) * | 2021-03-31 | 2021-06-22 | 华为技术有限公司 | Model training method, data processing method and equipment |
Non-Patent Citations (5)
| Title |
|---|
| 基于动态行为学习的空中目标识别方法;成磊峰 等;计算机与数字工程(第03期);全文 * |
| 基于数据挖掘的雷达探测目标误差测量技术;张彤;计算机测量与控制(第010期);全文 * |
| 基于深度学习的雷达自动目标识别架构研究;李士国 等;《现代雷达》;全文 * |
| 基于神经网络的多功能雷达行为辨识方法;方旖;毕大平;潘继飞;陈秋菊;;空军工程大学学报(自然科学版)(第03期);全文 * |
| 数据库样本缺失下的雷达辐射源识别;李蒙;朱卫纲;;电讯技术(第07期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113655457A (en) | 2021-11-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111126152B (en) | Multi-target pedestrian detection and tracking method based on video | |
| CN101364259B (en) | Method for extracting road various information of multi-level knowledge driven panchromatic remote sensing image | |
| CN109708638B (en) | Ship track point extraction method | |
| CN111368633A (en) | AUV-based side-scan sonar image identification method | |
| CN114782626B (en) | Transformer substation scene map building and positioning optimization method based on laser and vision fusion | |
| CN108664930A (en) | A kind of intelligent multi-target detection tracking | |
| CN110569843B (en) | Intelligent detection and identification method for mine target | |
| CN107688170A (en) | A kind of Radar Target Track initial mode based on random forest | |
| CN104880160B (en) | Two-dimensional-laser real-time detection method of workpiece surface profile | |
| CN109753853A (en) | One kind being completed at the same time pedestrian detection and pedestrian knows method for distinguishing again | |
| CN110533720A (en) | Semantic SLAM system and method based on joint constraint | |
| CN104299243A (en) | Target tracking method based on Hough forests | |
| CN105427342A (en) | Method and system for detecting and tracking underwater small-target sonar image target | |
| CN103759732A (en) | Angle information assisted centralized multi-sensor multi-hypothesis tracking method | |
| CN107133966A (en) | A kind of three-dimensional sonar image background dividing method based on sampling consistency algorithm | |
| CN109213204A (en) | AUV sub-sea floor targets based on data-driven search navigation system and method | |
| CN114998276B (en) | Robot dynamic obstacle real-time detection method based on three-dimensional point cloud | |
| CN112985263A (en) | Method, device and equipment for detecting geometrical parameters of bow net | |
| CN114092517A (en) | Multi-target tracking method based on traditional and deep learning algorithm | |
| CN116277025A (en) | Object sorting control method and system of intelligent manufacturing robot | |
| CN109712171A (en) | A kind of Target Tracking System and method for tracking target based on correlation filter | |
| Martirena et al. | Automated annotation of lane markings using lidar and odometry | |
| Wang et al. | Regression forest based RGB-D visual relocalization using coarse-to-fine strategy | |
| CN113655457B (en) | Self-evolution method and device for radar target detection capability based on sample mining | |
| CN113724293A (en) | Vision-based intelligent internet public transport scene target tracking method and system |
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 |