CN109143199A - Sea clutter small target detecting method based on transfer learning - Google Patents
Sea clutter small target detecting method based on transfer learning Download PDFInfo
- Publication number
- CN109143199A CN109143199A CN201811330236.6A CN201811330236A CN109143199A CN 109143199 A CN109143199 A CN 109143199A CN 201811330236 A CN201811330236 A CN 201811330236A CN 109143199 A CN109143199 A CN 109143199A
- Authority
- CN
- China
- Prior art keywords
- data set
- data
- classifier
- sample
- target
- 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 27
- 238000013526 transfer learning Methods 0.000 title claims abstract description 12
- 238000012549 training Methods 0.000 claims abstract description 57
- 238000012360 testing method Methods 0.000 claims description 11
- 238000001514 detection method Methods 0.000 abstract description 20
- 238000013508 migration Methods 0.000 abstract description 5
- 230000005012 migration Effects 0.000 abstract description 4
- 238000002474 experimental method Methods 0.000 abstract description 2
- 239000003550 marker Substances 0.000 abstract description 2
- 238000005259 measurement Methods 0.000 abstract description 2
- 230000006870 function Effects 0.000 description 6
- 239000004698 Polyethylene Substances 0.000 description 5
- 230000033228 biological regulation Effects 0.000 description 5
- -1 polyethylene Polymers 0.000 description 5
- 229920000573 polyethylene Polymers 0.000 description 5
- 239000013598 vector Substances 0.000 description 5
- 238000004422 calculation algorithm Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 241001269238 Data Species 0.000 description 1
- 239000005030 aluminium foil Substances 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 230000009977 dual effect Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000010287 polarization Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000004088 simulation Methods 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
- G01S7/414—Discriminating targets with respect to background clutter
Landscapes
- Engineering & Computer Science (AREA)
- Computer Networks & Wireless Communication (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Radar, Positioning & Navigation (AREA)
- Remote Sensing (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
The invention proposes small target detecting methods under a kind of sea clutter based on transfer learning.The moving method that this method is combined using TrAdaBoost and SVM classifier screens effective approximate sample from source domain data, the detection of target is realized by Data Migration.By the experiment of the IPIX sea clutter data of actual measurement, when confirmed respectively by historical data and different target Data Migration, which can effectively realize the detection of Small object, and accuracy rate reaches 70% or more.If a small amount of marker samples in aiming field are directly used as training dataset, due to lacking for data volume, then detection accuracy is about 40%.Therefore, the problem of data volume lacks can be made up using the method for transfer learning, improves the accuracy rate of detection.
Description
Technical field
The present invention relates to sea clutter detection technique field more particularly to a kind of sea clutter Small object inspections based on transfer learning
Survey method.
Background technique
The detection of naval target all has critically important answer in terms of marine safety, SAR at Sea, monitoring
With value.Sea clutter is the backscattering echo that radar emission signal is irradiated to that local sea level generates, can be according to sea clutter
In include target echo detect and identify target.But the influence of the various factors such as radar return wind-engaging, wave has echo letter
Miscellaneous relatively low feature, especially Small object, such as: small floating material, the small naval vessel under high sea situation and floating ice etc., then effectively
The marine Weak target of detection there is certain challenge.
But that there are measured datas is few for sea clutter data sample at present, affects the accuracy rate of sea clutter small target deteection.
Summary of the invention
The present invention provides a kind of sea clutter small target detecting method based on transfer learning, to overcome above-mentioned technical problem.
The present invention is based on the sea clutter small target detecting methods of transfer learning, comprising:
It determines that training set includes sample data set and target data set to be detected, and initializes the sample data set and institute
State weight of the target data set to be detected in the training set；
Classification based training is carried out to the training set using SVM classifier and obtains the first classifier, and uses described first point
Class device classifies to the sample data set or target data set to be detected；
Judge first classifier to the classification results of the sample data set or target data set to be detected whether
Correctly, if so, reducing the weight of the sample data set or target data set to be detected, if it is not, then increasing the sample
The weight of data set or target data set to be detected；
Classification based training is carried out to weight training set adjusted using SVM classifier and obtains the second classifier, and is adopted
The sample number is adjusted with classification results of second classifier to the sample data set or target data set to be detected
According to collection or target data set to be detected weight, and so on iteration n times, obtain n-th classifier；
It is final classification device by first classifier to N classifiers combination；
The accuracy rate of the final classification device is calculated according to real-time testing data sample, and whether judges the accuracy rate
Greater than threshold value, if so, being detected using the final classification device to target data set to be measured, if it is not, then to described final
Classifier continues iterated revision.
The invention proposes small target detecting methods under a kind of sea clutter based on transfer learning.This method uses
The moving method that TrAdaBoost and SVM classifier combine screens effective approximate sample from source domain data, is moved by data
The detection of in-migration realization target.By the experiment of the IPIX sea clutter data of actual measurement, confirmed respectively by historical data and difference
When target data migrates, which can effectively realize the detection of Small object, and accuracy rate reaches 70% or more.If in aiming field
A small amount of marker samples be directly used as training dataset, due to lacking for data volume, then detection accuracy is about 40%.Therefore, make
The problem of data volume missing can be made up with the method for transfer learning, improve the accuracy rate of detection.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is this hair
Bright some embodiments for those of ordinary skill in the art without any creative labor, can be with
It obtains other drawings based on these drawings.
Fig. 1 is that the present invention is based on the sea clutter small target detecting method flow charts of transfer learning；
Fig. 2 is the classifier accuracy rate curve graph of present invention target polyethylene ball to be detected；
Fig. 3 is the classifier accuracy rate curve graph that present invention target to be detected is maritime buoyage 14# and long 4.57m canoe.
Specific embodiment
In order to make the object, technical scheme and advantages of the embodiment of the invention clearer, below in conjunction with the embodiment of the present invention
In attached drawing, technical scheme in the embodiment of the invention is clearly and completely described, it is clear that described embodiment is
A part of the embodiment of the present invention, instead of all the embodiments.Based on the embodiments of the present invention, those of ordinary skill in the art
Every other embodiment obtained without creative efforts, shall fall within the protection scope of the present invention.
Fig. 1 is that the present invention is based on the sea clutter small target detecting method flow charts of transfer learning, as shown in Figure 1, this example
Method may include:
Step 101 determines that training set includes sample data set and target data set to be detected, and initializes the sample number
According to collection and weight of the target data set to be detected in the training set；
Specifically, the present embodiment sample data set is the sample data set for largely having marked Small object, it is denoted as Ta.To
Detection target data set is the sample data set comprising having marked Small object on a small quantity, is denoted as Tb.The training dataset is two
The union of a data set, is denoted as T=Ta∪Tb.The data of the training dataset mainly include following 3 kinds of situations:
(1) using the historical data and real time data on same sea area radar acquisition sea level；
(2) using the analogue data and real time data on target sea area radar acquisition sea level；
(3) same sea area is marked in the data on certain sea area radar acquisition sea level and the second classification using first kind target
Data.
N the first data set samples, m the second data set samples are randomly selected in three cases above.
The parameter of the training dataset is initialized, the parameter includes: sample data set and target data set to be detected
Weighted value in training set, is denoted asInitialize weighted value are as follows:
The weight distribution P on training set T after mergingt, following expression.
Wherein, wtFor the weight vectors of the t times iteration, n is the first data set number of samples, and m is the second data set sample
Number,For the weight of the t times iteration, i-th of sample.
The echo-signal that radar illumination sea level receives, referred to as sea clutter.There may be reality for the irradiation in same sea area
When data and history data, may have the data of different target type, may there is the data of simulation, may there is identical region
Data.The data that collection capacity is big and mark is more can be defined as source domain, and the target and collection capacity for solution and mark are few
Data definition is aiming field, completes the target identification in aiming field by the migration of more source domain data.According to discussed above,
Can be using analogue data as source domain, measured data is as aiming field；Using historical data as source domain, real time data is as target
Domain；Using a kind of target data as source domain, using another target data as aiming field.In short, by the energy of existing mass data
Enough learn the conduct source domain of reliable model out, only the conduct aiming field for being not enough to train reliable model of low volume data.This
The data in two domains are different from different data sources, their data distribution.Our target is to utilize source domain data
It goes that target numeric field data is helped to construct a more reliable disaggregated model, so that classification of this model in test data is accurate
It spends as high as possible.The present invention uses McMaster university of Canada IPIX (Intelligent Pixel processing X-
Band) radar data, it is that current lot of domestic and international scholar generally use when research in terms of sea clutter and approved
Measured data.Data in 1993 are considered source data domain, i.e. historical data.It is target data domain data in 1998,
The detection for attempting to realize target is migrated by historical data.Target is to be wrapped in the polyethylene ball that the diameter of aluminium foil is 1 meter.
IPIX radar is the relevant dual polarization radar of X-band, is the data of acquisition in 1993 and 1998 respectively, for convenience of wide
Big domestic and foreign scholars' research, is announced in official website in June, 2001.The database includes 392 acquired under different sea situations
Group data, every group of echo data for having 14 distance unit, the data of each distance unit are sampled by continuous 131072
Point data composition, each data include the data of I (same to phase) and Q (orthogonal) channel.Using the average value of I and Q channel data as
The feature vector of classifier (present invention uses SVM classifier), selection markers are good from the data of history source domain in 1993
200, the sample training datasets as source domain are defined as Ta；The good sample of selection markers from the data of aiming field in 1998
This 40 training datasets as aiming field are defined as Tb；Merge the training set T=T mergeda∪Tb.From 1998
200, the sample good test sets as aiming field of selection markers, are defined as S in the data of the aiming field in year.
The relevant parameter of initialization algorithm.
1. whole process needs to train better classifier, successive ignition is carried out, first initialization the number of iterations N, this
N=40 in invention example.
2. the main thought of algorithm is the weight of the training data of multiple adjustment misclassification, to obtain mould of preferably classifying
Type.Therefore need to initialize the weight vectors of each sample in training set TWherein, n and m are respectively
Data set TaAnd TbSize, n=200, m=40 here.As shown in formula 1.
The source domain training dataset T of historical data in 1993aThe weight of each sample adjusted by a regulation coefficient β
It is whole.Initialization
Aiming field training dataset T in 1998bEach sample weight pass through a regulation coefficient βtTo adjust.βt
Definition see below step.
Step 102 obtains the first classifier to training set progress classification based training using SVM classifier, and uses institute
The first classifier is stated to classify to the sample data set or target data set to be detected
Specifically, according to the weight distribution P on training dataset T and Tt, use SVM classifier, Selection of kernel function
Radial basis function obtains the first classifier h by trainingt。
Pass through htTo the target data set T to be detected of training setbClassify, obtains classification error rate lt, ltCalculating side
Method following expression.
Wherein, ht(xi) it is to pass through htObtained classification category, c (xi) it is true classification category, xiFor i-th sample
Characteristic value.
N times loop iteration is carried out, following several operations are completed, cyclic variable t=1,2,3 ..., N are set.
1. for the weight distribution P on the training set T and T after mergingt(as shown in formula 2) and test data S is called
SVM classifier, the Selection of kernel function radial basis function of SVM classifier used in present example obtain one by training
Weak Classifier ht。
2. passing through htTo aiming field training dataset T in 1998bClassify, obtains classification error rate lt, ltCalculating
Method is as shown in formula 3.
In formula, ht(xi) it is to pass through htObtained classification category, c (xi) it is true classification category.
3. regulation coefficient β is arrangedt=lt/(1-lt), to avoid algorithm from stopping, if ltGreater than 0.5, it is set as 0.5.
4. the training sample of the aiming field for 1998, if by misclassification, it is believed that it this be the difficult instruction of a comparison
Practice sample.Then, increase the weight of this training sample, for emphasizing this sample.Next time when classification based training, this
Sample will be reduced by the probability of misclassification.For the source domain training sample of historical data in 1993, if by misclassification, it is believed that it
This be it is very different with target data, then, reduce the weight of these data to reduce their influences in classification based training.Tool
Body updates weight vectorsMethod it is as shown in formula 4.
Step 103 judges classification of first classifier to the sample data set or target data set to be detected
As a result whether correct, if so, reducing the weight of the sample data set or target data set to be detected, if it is not, then increasing
The weight of the sample data set or target data set to be detected；
Specifically, setting regulation coefficient βt=lt/(1-lt), to avoid training process from stopping, if ltGreater than 0.5, if
It is set to 0.5.
For the target data set T to be detected of training setbClassify, if some sample by misclassification, determines the sample
This is the difficult training sample of a comparison.Increase the weight of the training sample, then for emphasizing the sample.Classification based training next time
When, this sample will be reduced by the probability of misclassification.For sample data set TaClassify, if recognized by misclassification
It is very different for the sample and target data, then reduce the weight of sample data set to reduce their influences in classification based training.
It is specific to update weight vectorsMethod following expression.
By sample weightsIt is normalized.
Step 104 obtains the second classification to weight training set adjusted progress classification based training using SVM classifier
Device, and using described in classification results adjustment of second classifier to the sample data set or target data set to be detected
The weight of sample data set or target data set to be detected, and so on iteration N number, obtain n-th classifier；
Specifically, carrying out classification instruction to the sample data set or target data set to be detected using SVM classifier
It gets to the first classifier, is denoted as hi.If some training sample is correctly classified by the first classifier, reduce corresponding weight；
On the contrary, increasing its weight if some training sample is classified by mistake.Weight is by calculating classification error rate, setting
Regulation coefficient is completed.The sample set that right value update is crossed be used to train next classifier, and entire training process carries out multiple
Iteration.
As shown in Fig. 2, the sample data set T of the present embodimentaFor the polyethylene ball historical data of acquisition in 1993, TbFor
The data of the polyethylene ball of target to be detected in 1998, in the corresponding curve of accuracy rate of iteration 40 times corresponding 40 classifiers
The accuracy rate highest of 3rd classifier.Sample data set TaFor the polyethylene ball historical data of acquisition in 1993, TbIt is respectively to be checked
The 229# data that target is maritime buoyage 14# and long 4.57m canoe are surveyed, this method iteration 40 times corresponding 40 classifiers are used
The corresponding curve of accuracy rate it is as shown in Figure 3.
Step 105, by first classifier to N classifiers combination be final classification device；
Specifically, each classifiers combination that each training is obtained is at final classification device.Each classifier was trained
After journey, the weight of the small classifier of error in classification rate is increased, it is made to play biggish decision in final classification function
Effect, and the weight of the big classifier of error in classification rate is reduced, so that it is played lesser decision in final classification function and makees
With final classification device are as follows:
By sample weightsIt is normalized.
Available final strong classifier hf(x), as shown in formula 5.
Classify for test set S, compared with actual classification results, the detection for obtaining current iteration is accurate
Rate.
Final result exports the Detection accuracy of each iteration.
Step 106, the accuracy rate that the final classification device is calculated according to real-time testing data sample, and judge the standard
Whether true rate is greater than threshold value, if so, being detected using the final classification device to target data set to be measured, if it is not, then right
The final classification device continues iterated revision.
Specifically, using the part sample of real-time testing data as test set, using obtained strong classifier, for
Test data carries out detection classification, and compares with actual classification category, obtains using detection classification after the strong classifier
Accuracy rate, be denoted as R.
The threshold calculations process of the present embodiment are as follows: not by the migration of sample data set in training set, be only used only to be checked
Target data set is surveyed as training set, directly obtains a classifier using SVM algorithm, and detected for test data,
The accuracy rate for calculating detection, is denoted as R*.If RtsLess than 1, the threshold value that Detection accuracy is arranged is Rts=1.5 R*, if Rts
More than or equal to 1, then threshold value is set as 0.8.Need the final classification device of iterated revision that cycle-index s=1,2 can be set,
3,...,M.It reselects training set to be trained, repeats step 101 to step 106, until R > RtsOr cycle-index > M knot
Beam.Final classification device is obtained finally by training, the detection for this sea area sea clutter Small object.
Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention., rather than its limitations；To the greatest extent
Pipe present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that: its according to
So be possible to modify the technical solutions described in the foregoing embodiments, or to some or all of the technical features into
Row equivalent replacement；And these are modified or replaceed, various embodiments of the present invention technology that it does not separate the essence of the corresponding technical solution
The range of scheme.
Claims (1)
1. a kind of sea clutter small target detecting method based on transfer learning characterized by comprising
Determine that training set includes sample data set and target data set to be detected, and initialize the sample data set and it is described to
Detect weight of the target data set in the training set；
Classification based training is carried out to the training set using SVM classifier and obtains the first classifier, and uses first classifier
Classify to the sample data set or target data set to be detected；
Judge whether first classifier is correct to the classification results of the sample data set or target data set to be detected,
If so, reducing the weight of the sample data set or target data set to be detected, if it is not, then increasing the sample data set
Or the weight of target data set to be detected；
Classification based training is carried out to weight training set adjusted using SVM classifier and obtains the second classifier, and uses institute
It states the second classifier and the sample data set is adjusted to the classification results of the sample data set or target data set to be detected
Or the weight of target data set to be detected, and so on iteration n times, obtain n-th classifier；
It is final classification device by first classifier to N classifiers combination；
The accuracy rate of the final classification device is calculated according to real-time testing data sample, and judges whether the accuracy rate is greater than
Threshold value, if so, being detected using the final classification device to target data set to be measured, if it is not, then to the final classification
Device continues iterated revision.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811330236.6A CN109143199A (en) | 2018-11-09 | 2018-11-09 | Sea clutter small target detecting method based on transfer learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811330236.6A CN109143199A (en) | 2018-11-09 | 2018-11-09 | Sea clutter small target detecting method based on transfer learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109143199A true CN109143199A (en) | 2019-01-04 |
Family
ID=64807950
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811330236.6A Pending CN109143199A (en) | 2018-11-09 | 2018-11-09 | Sea clutter small target detecting method based on transfer learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109143199A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109902742A (en) * | 2019-02-28 | 2019-06-18 | 深圳前海微众银行股份有限公司 | Sample complementing method, terminal, system and medium based on encryption transfer learning |
| CN111736127A (en) * | 2020-06-11 | 2020-10-02 | 北京理工大学 | Source domain generation and distribution parameter generalization method for unknown sea clutter |
| CN112765141A (en) * | 2021-01-13 | 2021-05-07 | 杭州电子科技大学 | Continuous large-scale water quality missing data filling method based on transfer learning |
| CN114220026A (en) * | 2021-12-30 | 2022-03-22 | 杭州电子科技大学 | Sea surface small target detection method based on multi-classification idea |
| CN116959696A (en) * | 2023-09-20 | 2023-10-27 | 武汉光盾科技有限公司 | Data processing method and device based on laser therapeutic instrument |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102521656A (en) * | 2011-12-29 | 2012-06-27 | 北京工商大学 | Integrated transfer learning method for classification of unbalance samples |
| US20160078359A1 (en) * | 2014-09-12 | 2016-03-17 | Xerox Corporation | System for domain adaptation with a domain-specific class means classifier |
| CN107292246A (en) * | 2017-06-05 | 2017-10-24 | 河海大学 | Infrared human body target identification method based on HOG PCA and transfer learning |
-
2018
- 2018-11-09 CN CN201811330236.6A patent/CN109143199A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102521656A (en) * | 2011-12-29 | 2012-06-27 | 北京工商大学 | Integrated transfer learning method for classification of unbalance samples |
| US20160078359A1 (en) * | 2014-09-12 | 2016-03-17 | Xerox Corporation | System for domain adaptation with a domain-specific class means classifier |
| CN107292246A (en) * | 2017-06-05 | 2017-10-24 | 河海大学 | Infrared human body target identification method based on HOG PCA and transfer learning |
Non-Patent Citations (2)
| Title |
|---|
| 林永锋: "基于迁移学习的计算机辅助肺结节检测研究", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
| 范剑超 等: "高分三号SAR影像在国家海域使用动态监测中的应用", 《雷达学报》 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109902742A (en) * | 2019-02-28 | 2019-06-18 | 深圳前海微众银行股份有限公司 | Sample complementing method, terminal, system and medium based on encryption transfer learning |
| CN109902742B (en) * | 2019-02-28 | 2021-07-16 | 深圳前海微众银行股份有限公司 | Sample completion method, terminal, system and medium based on encryption migration learning |
| CN111736127A (en) * | 2020-06-11 | 2020-10-02 | 北京理工大学 | Source domain generation and distribution parameter generalization method for unknown sea clutter |
| CN112765141A (en) * | 2021-01-13 | 2021-05-07 | 杭州电子科技大学 | Continuous large-scale water quality missing data filling method based on transfer learning |
| CN114220026A (en) * | 2021-12-30 | 2022-03-22 | 杭州电子科技大学 | Sea surface small target detection method based on multi-classification idea |
| CN116959696A (en) * | 2023-09-20 | 2023-10-27 | 武汉光盾科技有限公司 | Data processing method and device based on laser therapeutic instrument |
| CN116959696B (en) * | 2023-09-20 | 2023-12-08 | 武汉光盾科技有限公司 | Data processing method and device based on laser therapeutic instrument |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109143199A (en) | Sea clutter small target detecting method based on transfer learning | |
| CN113269073B (en) | Ship multi-target tracking method based on YOLO V5 algorithm | |
| Kowalska et al. | Maritime anomaly detection using Gaussian process active learning | |
| US11822007B2 (en) | System and method for identification of an airborne object | |
| CN111722199B (en) | Radar signal detection method based on convolutional neural network | |
| CN105005796B (en) | Satellite-borne SAR image Ship target recognition method based on step analysis | |
| CN109086679A (en) | A kind of millimetre-wave radar safety check instrument foreign matter detecting method | |
| CN105738888B (en) | Bicharacteristic offshore floating small target detecting method based on ocean clutter cancellation | |
| CN110018453A (en) | Intelligent type recognition methods based on aircraft track feature | |
| CN105866758B (en) | Time-frequency bicharacteristic small targets detection in sea clutter method based on block albefaction clutter recognition | |
| CN109753874A (en) | A kind of low slow small classification of radar targets method based on machine learning | |
| CN107064177B (en) | Ultra wide band soil signal water content recognition methods based on adaptive nuero-fuzzy inference system | |
| CN106326938B (en) | SAR image target discrimination method based on Weakly supervised study | |
| CN104391290A (en) | CFAR detector suitable for complex inhomogeneous clutters | |
| CN106569193A (en) | Sea-surface small target detection method based on front-back revenue reference particle filter | |
| CN111505649B (en) | Towed passive array sonar low signal-to-noise ratio ship moving target detection method | |
| CN107492114A (en) | The heavy detecting method used when monocular is long during the tracking failure of visual tracking method | |
| Guo et al. | Fusion of ship perceptual information for electronic navigational chart and radar images based on deep learning | |
| CN113609895A (en) | Road traffic information acquisition method based on improved Yolov3 | |
| CN104268574A (en) | SAR image change detecting method based on genetic kernel fuzzy clustering | |
| Küppers et al. | Bayesian confidence calibration for epistemic uncertainty modelling | |
| Zhang et al. | Space target classification with corrupted HRRP sequences based on temporal–spatial feature aggregation network | |
| CN111274911A (en) | Dense fog monitoring method based on wireless microwave attenuation characteristic transfer learning | |
| CN110824478B (en) | Automatic classification method and device for precipitation cloud types based on diversified 3D radar echo characteristics | |
| CN116542912A (en) | Flexible body bridge vibration detection model with multi-target visual tracking function and application |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190104 |
|
| RJ01 | Rejection of invention patent application after publication |