CN110222632A - A kind of waterborne target detection method of gray prediction auxiliary area suggestion - Google Patents
A kind of waterborne target detection method of gray prediction auxiliary area suggestion Download PDFInfo
- Publication number
- CN110222632A CN110222632A CN201910480821.2A CN201910480821A CN110222632A CN 110222632 A CN110222632 A CN 110222632A CN 201910480821 A CN201910480821 A CN 201910480821A CN 110222632 A CN110222632 A CN 110222632A
- Authority
- CN
- China
- Prior art keywords
- frame
- waterborne target
- waterborne
- target
- target detection
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/41—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
Landscapes
- Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The present invention discloses a kind of waterborne target detection method of gray prediction auxiliary area suggestion, belongs to Intelligent unattended ship domain.The present invention includes: firstly, using trained model inspection to waterborne target；Then, gray prediction is carried out using the waterborne target location information in successive video frames；Then, suggest being fed back and be guided with region of the result of gray prediction to neural network；Finally, newly entering the waterborne target in video frame by more accurate region advisory result identification obtained in the previous step.The present invention can promote the accuracy rate and speed of waterborne target identification by simplifying neural network structure and proposing more accurate region suggestion.
Description
Technical field
The invention belongs to Intelligent unattended ship domains, and in particular to a kind of waterborne target of gray prediction auxiliary area suggestion
Detection method.
Background technique
In recent years, it is developed rapidly using the technology that deep neural network carries out target detection, the technology is also wide
It is general to be applied to intelligent surface ship object detection and recognition field；However, these classical ways do not account for waterborne target inspection
When survey in successive video frames same target spatial continuity.Using deep neural network carry out target detection technology increasingly at
It is ripe, it is also widely used in Intelligent unattended ship domain, however, merely not using classical target detection network
The specific application scene of Intelligent unattended ship can preferably be adapted to.Firstly, existing method for tracking target is in complicated aquatic environment
Under be easy to appear as target part blocks, target temporarily exceeds the visual field and caused by algorithm fail the problem of；Secondly, classical
Deep neural network object detection method has ignored the inner link of target position between successive frame under aquatic environment；In addition, through
In allusion quotation object detection method, the method operand that anchor point and candidate frame are uniformly distributed in entire image is huge, computing redundancy, no
The special scenes for adapting to waterborne target detection cause and calculate power waste.
Summary of the invention
The range of area to be tested is reduced using gray prediction arrangement anchor point and candidate frame the present invention is to provide a kind of,
The method that auxiliary mark detection network rapidly and accurately identifies waterborne target.This method is detected using the sky-line and gray prediction
Method guidance field suggest network, while simplifying the number of plies of feature extraction network.The object of the present invention is achieved like this:
A kind of waterborne target detection method of gray prediction auxiliary area suggestion comprising the steps of:
Step 1: using classical target detection Network Recognition waterborne target, and the seat of target identification frame central point is obtained
Mark；
Step 2: establishing grey forecasting model, and waterborne target identifies in the next video frame obtained using the model prediction
The central point abscissa of frame；
Step 3: opposite residual test is carried out to the grey forecasting model being calculated and grade compares bias test, it is ensured that mould
The accuracy of type and prediction result；
Step 4: region suggestion is carried out to classical target detection network is improved using gray prediction, and is detected, is obtained
Final detection result；
Step 5: analyzing the testing result of acquisition, if acquired results are correct, updates grey forecasting model, into
The target detection of row next frame；If waterborne target is not detected in continuous three frame, it is re-execute the steps one, calls classical target inspection
Survey Network Recognition waterborne target.
The finite element sequence data of grey forecasting model described in step 2 isWhereinThe central point abscissa u of target identification frame respectively in kth frame sample frame；
To X(0)One-accumulate is done to generate to obtain sequenceWherein
Enable Z(1)For X(1)Close to average generation sequenceWherein
The grey differential equation model for establishing GM (1,1) isWherein a is development coefficient, and b is grey effect
Amount；The least-squares estimation parameter column of grey differential equation meet (a, b)T=(BTB)-1BTYn, wherein
The albefaction equation for establishing Grey Differential Equation, final prediction result can be obtained by asking it to solve and doing regressive reduction；
Waterborne target identification as in the obtained next video frame of gray prediction
The central point abscissa of frame.
Opposite residual test described in step 3 are as follows:
WhereinFor the waterborne target frame midpoint abscissa detected in kth frame,To predict in obtained kth frame
Waterborne target frame midpoint abscissa；If | ε (k) | < 0.2, then it is assumed that model meets residual test；
Ratio bias test described in step 3 are as follows:
WhereinFor the grade ratio of initial data, a is development coefficient；If | ρ (k) | < 0.2, then it is assumed that meet grade
Compare bias test.
Compared with prior art, the beneficial effects of the present invention are:
Compared with classical target detection and track algorithm, this method has the specific of waterborne target identification mission preferably
Scene, robustness is stronger, recognition result more fast and accurately advantage；The present invention can by simplify neural network structure and
It is proposed that more accurate region is suggested to promote the accuracy rate and speed of waterborne target identification.
Detailed description of the invention
Fig. 1 a is that typical waterborne target identifies scene one；
Fig. 1 b is that typical waterborne target identifies scene two；
Fig. 2 is algorithm flow chart of the invention；
Fig. 3 is network structure of the invention.
Specific embodiment
The present invention is explained in detail with reference to the accompanying drawing:
Region suggestion is carried out using gray prediction the present invention is to provide a kind of, neural network is promoted and detects successive video frames
The method of the speed and accuracy rate of middle waterborne target.Firstly, utilizing trained model inspection to waterborne target；Then, sharp
Gray prediction is carried out with the waterborne target location information in successive video frames；Then, with the result of gray prediction to nerve net
Suggest being fed back and being guided in the region of network；Finally, new by more accurate region advisory result identification obtained in the previous step
Enter the waterborne target in video frame.The present invention can be mentioned by simplifying neural network structure and proposing more accurate region suggestion
Rise the accuracy rate and speed of waterborne target identification.The present invention obtains more accurate anchor point position using gray prediction and sea horizon detection
It sets, and then promotes the accuracy and speed of target detection；Simplify feature extraction network, reduces and calculate force request.
The object of the present invention is to provide a kind of waterborne target detection methods of gray prediction auxiliary area suggestion.Waterborne target
Often have near the sky-line in samples pictures, and the position of the same waterborne target of continuous videos interframe in the video frame exists
Inner link, as illustrated in figs. 1A and ib；The present invention utilizes this feature, uses gray prediction and sea horizon detection arrangement anchor point
And candidate frame, FasterRCNN Network Recognition waterborne target is assisted, the rapidity and accuracy of waterborne target identification are promoted.It has
Body implementation method is as follows.
The present invention is further elaborated below with reference to the citing of Fig. 2 algorithm flow chart.
Firstly, identifying waterborne target using classical FasterRCNN method, and records target identification frame central point and exist
Pixel coordinate (u, v) in video frame, is arranged sampling time interval t=0.2s, and sampling frame number is 5 frames.
Grey forecasting model is established using the sample information in initial target frame, finite element sequence data isWhereinThe central point abscissa u of target identification frame respectively in kth frame sample frame；
To X(0)One-accumulate is done to generate to obtain sequenceWhereinEnable Z(1)For X(1)
Close to average generation sequenceWhereinEstablish the grey differential of GM (1,1)
Equation model isWherein a is development coefficient, and b is grey actuating quantity.The least-squares estimation of grey differential equation is joined
Ordered series of numbers meets (a, b)T=(BTB)-1BTYn, wherein
The albefaction equation for establishing Grey Differential Equation, final prediction result can be obtained by asking it to solve and doing regressive reduction
Waterborne target identification as in the obtained next video frame of gray prediction
The central point abscissa of frame.
Opposite residual test is carried out to the grey forecasting model being calculatedWhereinFor in kth frame
Obtained waterborne target frame midpoint abscissa is detected,To predict waterborne target frame midpoint abscissa in obtained kth frame；If |
ε (k) | < 0.2, then it is assumed that model meets residual test.It then carries out grade and compares bias testIts
InFor the grade ratio of initial data, a is development coefficient described above；If | ρ (k) | < 0.2, then it is assumed that meet grade
Compare bias test.After only meeting this two predicted values inspections, gray model just can be used to assist given region suggestion, it is ensured that knot
Otherwise the accuracy of fruit should use newest 5 frame video frame to rebuild grey forecasting model.
Classics FasterRCNN neural network structure is improved, carries out region suggestion using the result of gray prediction.It ties below
Fig. 3 network structure is closed the present invention is described in detail done in FasterRCNN network structure.
The some characteristics that the special scenes of waterborne target detection have its intrinsic；Firstly, waterborne target to be detected is often distributed
Near the sky-line, a possibility that there are targets to be detected above and below video frame, is very low；Secondly, waterborne target detects scene
Middle background characteristics is more single, mostly sky and ocean.Therefore, this method uses sea horizon detection to obtain sea in video frame first
Anchor is then arranged according to Gaussian Profile on sea horizon, near the waterborne target abscissa that gray prediction obtains in the position of antenna
Point still selects totally nine kinds of candidate frames of three kinds of sizes and three kinds of length-width ratios, by these candidate frame input areas at each anchor point
Network (RegionProposalNetwork, RPN) is suggested in domain, so that region suggestion is more accurate.On the other hand, have benefited from essence
True region is suggested and more single background characteristics, simplifies the feature extraction network number of plies, replaces VGG16 net using VGG13 network
Network promotes network operation speed.Pond is then carried out, target classification and bounding box, which return, to be operated, and final testing result is obtained.
Judge whether to detect target, if detecting target, gray model is carried out in real time more using video frame is newly entered
Newly, it prevents model prediction result from dissipating at any time, then re-starts prediction and detection using new model.If not detecting mesh
Mark, illustrates that waterborne target may have been moved off the visual field or forecasting inaccuracy is true, should stop this detection process at this time, will newly enter video
Frame restarts next detection process as target initial frame.
Claims (3)
1. a kind of waterborne target detection method of gray prediction auxiliary area suggestion, which is characterized in that comprise the steps of:
Step 1: using classical target detection Network Recognition waterborne target, and the coordinate of target identification frame central point is obtained；
Step 2: establishing grey forecasting model, waterborne target identification frame in the next video frame obtained using the model prediction
Central point abscissa；
Step 3: opposite residual test is carried out to the grey forecasting model that is calculated and grade compares bias test, it is ensured that model and
The accuracy of prediction result；
Step 4: region suggestion is carried out to classical target detection network is improved using gray prediction, and is detected, is obtained final
Testing result；
Step 5: analyzing the testing result of acquisition, if acquired results are correct, update grey forecasting model, carries out down
The target detection of one frame；If waterborne target is not detected in continuous three frame, it is re-execute the steps one, calls classical target detection net
Network identifies waterborne target.
2. a kind of waterborne target detection method of gray prediction auxiliary area suggestion according to claim 1, feature exist
In the finite element sequence data of grey forecasting model described in step 2 isWhereinPoint
Not Wei in kth frame sample frame target identification frame central point abscissa u；
To X(0)One-accumulate is done to generate to obtain sequenceWherein
Enable Z(1)For X(1)Close to average generation sequenceWherein
The grey differential equation model for establishing GM (1,1) isWherein a is development coefficient, and b is grey actuating quantity；Ash is micro-
The least-squares estimation parameter column of equation are divided to meet (a, b)T=(BTB)-1BTYn, wherein
The albefaction equation for establishing Grey Differential Equation, final prediction result can be obtained by asking it to solve and doing regressive reduction；
Waterborne target identification frame as in the obtained next video frame of gray prediction
Central point abscissa.
3. a kind of waterborne target detection method of gray prediction auxiliary area suggestion according to claim 1, feature exist
In opposite residual test described in step 3 are as follows:
WhereinFor the waterborne target frame midpoint abscissa detected in kth frame,To predict water surface mesh in obtained kth frame
Mark frame midpoint abscissa；If | ε (k) | < 0.2, then it is assumed that model meets residual test；
Ratio bias test described in step 3 are as follows:
WhereinFor the grade ratio of initial data, a is development coefficient；If | ρ (k) | < 0.2, then it is assumed that meet grade than inclined
Difference is examined.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910480821.2A CN110222632A (en) | 2019-06-04 | 2019-06-04 | A kind of waterborne target detection method of gray prediction auxiliary area suggestion |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910480821.2A CN110222632A (en) | 2019-06-04 | 2019-06-04 | A kind of waterborne target detection method of gray prediction auxiliary area suggestion |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110222632A true CN110222632A (en) | 2019-09-10 |
Family
ID=67819226
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910480821.2A Pending CN110222632A (en) | 2019-06-04 | 2019-06-04 | A kind of waterborne target detection method of gray prediction auxiliary area suggestion |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110222632A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110909948A (en) * | 2019-11-29 | 2020-03-24 | 北京信息科技大学 | Soil pollution prediction method and system |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105701352A (en) * | 2016-01-15 | 2016-06-22 | 中国科学院光电研究院 | Space motion object locus prediction method |
| CN106909885A (en) * | 2017-01-19 | 2017-06-30 | 博康智能信息技术有限公司上海分公司 | A kind of method for tracking target and device based on target candidate |
| CN107679657A (en) * | 2017-09-28 | 2018-02-09 | 黑龙江省科学院火山与矿泉研究所 | Water quality prediction method for mineral spring |
| CN107886120A (en) * | 2017-11-03 | 2018-04-06 | 北京清瑞维航技术发展有限公司 | Method and apparatus for target detection tracking |
| CN108052940A (en) * | 2017-12-17 | 2018-05-18 | 南京理工大学 | SAR remote sensing images waterborne target detection methods based on deep learning |
| CN108200534A (en) * | 2018-01-10 | 2018-06-22 | 海信集团有限公司 | A kind of method and apparatus positioned to terminal |
| CN108537826A (en) * | 2018-05-28 | 2018-09-14 | 深圳市芯汉感知技术有限公司 | A kind of Ship Target tracking based on manual intervention |
| CN108828550A (en) * | 2018-08-03 | 2018-11-16 | 国网福建省电力有限公司 | A kind of sea cable regional boundary ship target method for detecting position |
| CN109711295A (en) * | 2018-12-14 | 2019-05-03 | 北京航空航天大学 | A kind of remote sensing image offshore Ship Detection |
-
2019
- 2019-06-04 CN CN201910480821.2A patent/CN110222632A/en active Pending
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105701352A (en) * | 2016-01-15 | 2016-06-22 | 中国科学院光电研究院 | Space motion object locus prediction method |
| CN106909885A (en) * | 2017-01-19 | 2017-06-30 | 博康智能信息技术有限公司上海分公司 | A kind of method for tracking target and device based on target candidate |
| CN107679657A (en) * | 2017-09-28 | 2018-02-09 | 黑龙江省科学院火山与矿泉研究所 | Water quality prediction method for mineral spring |
| CN107886120A (en) * | 2017-11-03 | 2018-04-06 | 北京清瑞维航技术发展有限公司 | Method and apparatus for target detection tracking |
| CN108052940A (en) * | 2017-12-17 | 2018-05-18 | 南京理工大学 | SAR remote sensing images waterborne target detection methods based on deep learning |
| CN108200534A (en) * | 2018-01-10 | 2018-06-22 | 海信集团有限公司 | A kind of method and apparatus positioned to terminal |
| CN108537826A (en) * | 2018-05-28 | 2018-09-14 | 深圳市芯汉感知技术有限公司 | A kind of Ship Target tracking based on manual intervention |
| CN108828550A (en) * | 2018-08-03 | 2018-11-16 | 国网福建省电力有限公司 | A kind of sea cable regional boundary ship target method for detecting position |
| CN109711295A (en) * | 2018-12-14 | 2019-05-03 | 北京航空航天大学 | A kind of remote sensing image offshore Ship Detection |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110909948A (en) * | 2019-11-29 | 2020-03-24 | 北京信息科技大学 | Soil pollution prediction method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110705457B (en) | Remote sensing image building change detection method | |
| CN106952288B (en) | Based on convolution feature and global search detect it is long when block robust tracking method | |
| CN111127493A (en) | Remote sensing image semantic segmentation method based on attention multi-scale feature fusion | |
| CN110119728A (en) | Remote sensing images cloud detection method of optic based on Multiscale Fusion semantic segmentation network | |
| US20120328161A1 (en) | Method and multi-scale attention system for spatiotemporal change determination and object detection | |
| CN110796679B (en) | Target tracking method for aerial image | |
| CN110427807A (en) | A kind of temporal events motion detection method | |
| CN111476159A (en) | Method and device for training and detecting detection model based on double-angle regression | |
| CN107622507B (en) | Air target tracking method based on deep learning | |
| CN103093458B (en) | The detection method of key frame and device | |
| CN106557740B (en) | The recognition methods of oil depot target in a kind of remote sensing images | |
| Sun et al. | Global Mask R-CNN for marine ship instance segmentation | |
| CN114677323A (en) | Semantic vision SLAM positioning method based on target detection in indoor dynamic scene | |
| CN116524062B (en) | Diffusion model-based 2D human body posture estimation method | |
| CN105787950A (en) | Infrared image sea-sky-line detection algorithm based on line gradient accumulation | |
| CN116665095B (en) | Method and system for detecting motion ship, storage medium and electronic equipment | |
| CN109308713A (en) | A kind of improvement core correlation filtering Method for Underwater Target Tracking based on Forward-looking Sonar | |
| CN110852243A (en) | Improved YOLOv 3-based road intersection detection method and device | |
| CN114494870A (en) | Double-time-phase remote sensing image change detection method, model construction method and device | |
| CN110334727A (en) | A kind of tunnel slot intelligent Matching detection method | |
| CN116109812A (en) | Target detection method based on non-maximum suppression threshold optimization | |
| CN106558069A (en) | A kind of method for tracking target and system based under video monitoring | |
| Sun et al. | Image recognition technology in texture identification of marine sediment sonar image | |
| Hu et al. | A study of automatic recognition and localization of pipeline for ground penetrating radar based on deep learning | |
| CN112164093A (en) | Automatic person tracking method based on edge features and related filtering |
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: 20190910 |
|
| RJ01 | Rejection of invention patent application after publication |