CN107316018A - A kind of multiclass typical target recognition methods based on combiner model - Google Patents
A kind of multiclass typical target recognition methods based on combiner model Download PDFInfo
- Publication number
- CN107316018A CN107316018A CN201710486694.8A CN201710486694A CN107316018A CN 107316018 A CN107316018 A CN 107316018A CN 201710486694 A CN201710486694 A CN 201710486694A CN 107316018 A CN107316018 A CN 107316018A
- Authority
- CN
- China
- Prior art keywords
- target
- subassembly
- feature
- model
- carried out
- 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
- 238000012706 support-vector machine Methods 0.000 claims abstract description 4
- 230000008569 process Effects 0.000 claims description 9
- 230000003252 repetitive effect Effects 0.000 claims description 3
- 238000001514 detection method Methods 0.000 description 3
- 239000013598 vector Substances 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 2
- 230000008901 benefit Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000006073 displacement reaction Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/13—Satellite images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/50—Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
- G06V10/507—Summing image-intensity values; Histogram projection analysis
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Astronomy & Astrophysics (AREA)
- Remote Sensing (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of multiclass typical target recognition methods based on combiner model, the combiner model includes overall profile part and subassembly, mainly comprised the following steps：First, the overall profile part and subassembly of target in image are described using histogram of gradients feature；2nd, a hidden variable is set for subassembly position, the training of target image sample is carried out using hidden SVMs, sorter model is drawn；3rd, after model is trained, under original resolution, convolution algorithm is carried out to the overall profile feature of target using root wave filter；Under down-sampled resolution ratio, convolution algorithm is carried out to the feature of each subassembly using subfilter；4th, the calculating offset by feature convolution and position, maximum score region is target critical position position.The present invention can be achieved with the secondary identification to target significant points when overall goals are identified, and armament systems can select the key position struck target, obtain maximum damage effectiveness according to important subassembly position.
Description
Technical field
The invention belongs to computer image recognition technology field, and in particular to a kind of many special dictionarys based on combiner model
Type target identification method.
Background technology
TV guidance due to its intuitive and convenient, meets the protrusions such as human vision custom as a kind of new aiming means
Advantage, is widely applied in the Modern weapon systems such as all kinds of guided bombs, guided missile.
Some large-scale targets (airfield runway, bridge, large ship etc.) and complex target (tank group, missile emplacements etc.),
Deng guidance end target, shared region is larger in guidance image, if last strike point selection is in these target critical portions
Position, then can provide preferable damage effectiveness.
At the initial stage of target following, due to target, occupied area is very small in visual field in itself, therefore either which kind of mesh
Mark, be all the entirety using target as its tracking characteristics, but arrived tracking middle and later periods, as target is constantly amplified, its
The characteristic information of itself gradually enrich when, the final point of attack and key position is possible to can have deviation, in order to up to
Hit to higher damage effectiveness, it is necessary to which key position is marked, therefore be accomplished by being modified hit position.
The content of the invention
The purpose of the present invention can not recognize that target is closed precisely in order to solving current guided munition, especially TV guidance ammunition
The defect at key position, it is proposed that each subassembly that target is constituted in whole mesh target area is identified and is marked
Method, when carrying out overall goals identification with regard to that can obtain the important sub-portion of target where.
The invention provides a kind of multiclass typical target recognition methods based on combiner model, the combiner mould
Type includes overall profile part and subassembly, mainly comprises the following steps：
First, the overall profile part and subassembly of target in image are described using histogram of gradients feature；
2nd, a hidden variable is set for subassembly position, the instruction of target image sample is carried out using hidden SVMs
Practice, draw sorter model；
3rd, after model is trained, under original resolution, the overall profile feature of target is entered using root wave filter
Row convolution algorithm；Under down-sampled resolution ratio, convolution algorithm is carried out to the feature of each subassembly using subfilter；
4th, the calculating offset by feature convolution and position, maximum score region is that target critical position institute is in place
Put.
Further, dimension-reduction treatment is carried out to histogram of gradients characteristic use PCA in step one.
Further, in step 2, training includes initialization procedure, repetitive exercise process and last handling process.
Further, the convolution algorithm of root wave filter is carried out under original resolution；The convolution algorithm of subfilter is in drop
Carried out under sampling resolution.
Recognition methods of the invention based on combiner model, just can be by each subassembly of target when recognizing target
All identify, convolution algorithm is carried out respectively to overall profile feature and subassembly feature using sorter model, so as to calculate
First target critical position position, realizes the secondary identification to target significant points.Armament systems can be according to important subassembly
Position, selects the key position struck target, obtains maximum damage effectiveness.
Brief description of the drawings
Fig. 1 is the flow chart of the invention based on partial model target detection identification process；
Fig. 2 is the flow chart of the invention based on partial model target training process.
Embodiment
The present invention is described in further detail below in conjunction with the accompanying drawings：
Referring to accompanying drawing 1-2, the present invention proposes a kind of multiclass typical target recognition methods based on combiner model.
In the present invention, combiner model refer to regard as target assembled by many subassemblies and an overall profile part and
Into object, have relative position relationship between each subassembly, and allow the presence of certain position deviation between subassembly.
By taking large-scale military boats and ships as an example, whole naval vessel is made up of several more obvious parts, such as fore, hull and ship
Tail etc., and different types of military boats and ships overall profile is essentially identical, therefore can be very good using variable combination partial model
Naval vessel is described.
Referring to accompanying drawing 1, recognition methods of the invention specifically includes herein below：
First, subassembly feature is extracted
The overall profile part and subassembly of target in image are described using histogram of gradients (HOG) feature.
Gradient orientation histogram is a kind of feature extraction side that histograms of oriented gradients is locally normalized based on image lattice
Method.Whole image is divided into the grid of fixed size first, and each grid is calculated, each in grid is drawn
The gradient intensity and directional information of pixel, then produce one dimensional histograms by gradient intensity and directional information.Finally by these nets
Lattice histogram is combined, and forms complete target signature description vectors.
In order to preferably adapt to the interference of the factors such as illumination variation and shade, office is used before complete characteristic vector is drawn
Portion's feature normalization is to obtain more preferable characteristic information.In use, to these histogram of gradients (HOG) characteristic use principal components
Analytic approach carries out dimension-reduction treatment, removes influence of noise.
2nd, sorter model is trained
A hidden variable is set for subassembly position, for the information of global characteristics, optimal local feature is extracted
Position makes Score maxi-mation, and the training of target image sample is carried out using hidden SVMs.During training in figure label target
The position at place, and composition and the position of crucial subassembly are marked simultaneously.The position model of other subassemblies is then supported by hidden
The training of vector machine is drawn, so can not only reduce mark amount but also can reduce position marking the identification mistake brought by mistake.
Training process is as shown in Fig. 2 including initialization procedure, repetitive exercise process and last handling process.
When it is implemented, can for example gather model ship image totally 80, wherein horizontally toward and being vertically oriented to situation
Each 20 of image, each 10 images for separately having four directions (positive northeast, the positive southeast, just southwestern, positive northwest), marks warship in figure
Position where ship, while image pattern of the collection without naval vessel is as negative sample 10, afterwards to the naval vessel figure of different directions
Decent is trained (see accompanying drawing 2), by taking the naval vessel grader of level of training direction as an example, you can draw sorter model.
Similarly, self-propelled gun model image totally 80 can be gathered, wherein the image horizontally toward with the situation that is vertically oriented to is each
20, each 10 images of each each 20 images, four directions in direction (positive northeast, the positive southeast, just southwestern, positive northwest), in figure
Position where middle mark self-propelled gun, while sample 10 of the collection without cannon, afterwards to the self-propelled gun of horizontal direction
Sample is trained, so as to draw sorter model.
3rd, target identification is carried out using sorter model
After model is trained, the feature of different parts is described respectively from subfilter using root wave filter.Root is filtered
Ripple device is exactly the overall profile feature of target, and subfilter is exactly the feature of each subassembly.
In recognition detection, the feature under two kinds of different resolutions is used to original image：Original resolution (i.e. coarse resolution
Rate) under, the overall profile feature of target is obtained, convolution algorithm is carried out using root wave filter；In down-sampled resolution ratio, (i.e. subdivision is distinguished
Rate) under obtain target subassembly feature, utilize subfilter carry out convolution algorithm.For the mesh of each detection to be identified
Mark, its final Model Identification score includes the convolution score of overall profile feature and the convolution score sum of subassembly feature,
Also need to subtract the inner product of partial model relative displacement and offset weight coefficient, finally along with model overall offset value.Pass through
The calculating that feature convolution is offset with position, the score region for searching out maximum is target critical position position.
Because the present invention is using subassembly knowledge method for distinguishing, its identification model contains the composition of all subassemblies, so
When recognizing target, whole mesh target area is not only can recognize that, and can recognize that each subassembly of composition target, therefore
As long as the subassembly of object module is carried out into artificial judgement in training, the subassembly of target significant points is marked,
In identification with regard to that can obtain the important sub-portion of target where, realize the secondary identification to target significant points.Armament systems can be according to
According to important subassembly position, the key position struck target is selected, maximum damage effectiveness is obtained.
It is obvious to a person skilled in the art that the invention is not restricted to the details of above-mentioned one exemplary embodiment, Er Qie
In the case of without departing substantially from spirit or essential attributes of the invention, the present invention can be realized in other specific forms.Therefore, no matter
From the point of view of which point, embodiment all should be regarded as exemplary, and be nonrestrictive, the scope of the present invention is by appended power
Profit is required rather than described above is limited, it is intended that all in the implication and scope of the equivalency of claim by falling
Change is included in the present invention.Any reference in claim should not be considered as to the claim involved by limitation.
Moreover, it will be appreciated that although the present specification is described in terms of embodiments, not each embodiment is only wrapped
Containing an independent technical scheme, this narrating mode of specification is only that for clarity, those skilled in the art should
Using specification as an entirety, the technical solutions in the various embodiments may also be suitably combined, forms those skilled in the art
It may be appreciated other embodiment.
The description to embodiment the invention is not restricted to more than, those skilled in the art according to disclosure of the present invention,
The improvement and modification that need not be carried out on the basis of the present invention by creative work, all should protection scope of the present invention it
It is interior.
Claims (4)
1. a kind of multiclass typical target recognition methods based on combiner model, the combiner model includes overall profile
Part and subassembly, it is characterised in that
Mainly comprise the following steps：
First, the overall profile part and subassembly of target in image are described using histogram of gradients feature；
2nd, a hidden variable is set for subassembly position, the training of target image sample is carried out using hidden SVMs, is obtained
Go out sorter model；
3rd, after model is trained, under original resolution, the overall profile feature of target is rolled up using root wave filter
Product computing；Under down-sampled resolution ratio, convolution algorithm is carried out to the feature of each subassembly using subfilter；
4th, the calculating offset by feature convolution and position, maximum score region is target critical position position.
2. a kind of multiclass typical target recognition methods based on combiner model according to claim 1, its feature exists
In：
In step one, dimension-reduction treatment is carried out to histogram of gradients characteristic use PCA.
3. a kind of multiclass typical target recognition methods based on combiner model according to claim 1, its feature exists
In：
In step 2, the training includes initialization procedure, repetitive exercise process and last handling process.
4. a kind of multiclass typical target recognition methods based on combiner model according to claim 1 or 2, its feature
It is：
The convolution algorithm of root wave filter is carried out under original resolution；The convolution algorithm of subfilter enters under down-sampled resolution ratio
OK.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710486694.8A CN107316018A (en) | 2017-06-23 | 2017-06-23 | A kind of multiclass typical target recognition methods based on combiner model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710486694.8A CN107316018A (en) | 2017-06-23 | 2017-06-23 | A kind of multiclass typical target recognition methods based on combiner model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107316018A true CN107316018A (en) | 2017-11-03 |
Family
ID=60180135
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710486694.8A Pending CN107316018A (en) | 2017-06-23 | 2017-06-23 | A kind of multiclass typical target recognition methods based on combiner model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107316018A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115631388A (en) * | 2022-12-21 | 2023-01-20 | 第六镜科技(成都)有限公司 | Image classification method and device, electronic equipment and storage medium |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8363913B2 (en) * | 2008-09-05 | 2013-01-29 | Purdue Research Foundation | Dietary assessment system and method |
| CN104134071A (en) * | 2014-06-20 | 2014-11-05 | 北京工业大学 | Deformable part model object detection method based on color description |
| CN104200236A (en) * | 2014-08-22 | 2014-12-10 | 浙江生辉照明有限公司 | Quick target detection method based on DPM (deformable part model) |
| CN104850850A (en) * | 2015-04-05 | 2015-08-19 | 中国传媒大学 | Binocular stereoscopic vision image feature extraction method combining shape and color |
| CN105095905A (en) * | 2014-04-18 | 2015-11-25 | 株式会社理光 | Target recognition method and target recognition device |
| CN105335716A (en) * | 2015-10-29 | 2016-02-17 | 北京工业大学 | Improved UDN joint-feature extraction-based pedestrian detection method |
| CN105718889A (en) * | 2016-01-21 | 2016-06-29 | 江南大学 | Face recognition method based on GB(2D)2PCANet deep convolution model |
| US20160328838A1 (en) * | 2015-05-01 | 2016-11-10 | Applied Research LLC. | Automatic target recognition system with online machine learning capability |
| CN106296667A (en) * | 2016-08-01 | 2017-01-04 | 乐视控股（北京）有限公司 | Hand detection method and system |
| CN106384100A (en) * | 2016-09-28 | 2017-02-08 | 武汉大学 | Component-based fine vehicle model recognition method |
-
2017
- 2017-06-23 CN CN201710486694.8A patent/CN107316018A/en active Pending
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8363913B2 (en) * | 2008-09-05 | 2013-01-29 | Purdue Research Foundation | Dietary assessment system and method |
| CN105095905A (en) * | 2014-04-18 | 2015-11-25 | 株式会社理光 | Target recognition method and target recognition device |
| CN104134071A (en) * | 2014-06-20 | 2014-11-05 | 北京工业大学 | Deformable part model object detection method based on color description |
| CN104200236A (en) * | 2014-08-22 | 2014-12-10 | 浙江生辉照明有限公司 | Quick target detection method based on DPM (deformable part model) |
| CN104850850A (en) * | 2015-04-05 | 2015-08-19 | 中国传媒大学 | Binocular stereoscopic vision image feature extraction method combining shape and color |
| US20160328838A1 (en) * | 2015-05-01 | 2016-11-10 | Applied Research LLC. | Automatic target recognition system with online machine learning capability |
| CN105335716A (en) * | 2015-10-29 | 2016-02-17 | 北京工业大学 | Improved UDN joint-feature extraction-based pedestrian detection method |
| CN105718889A (en) * | 2016-01-21 | 2016-06-29 | 江南大学 | Face recognition method based on GB(2D)2PCANet deep convolution model |
| CN106296667A (en) * | 2016-08-01 | 2017-01-04 | 乐视控股（北京）有限公司 | Hand detection method and system |
| CN106384100A (en) * | 2016-09-28 | 2017-02-08 | 武汉大学 | Component-based fine vehicle model recognition method |
Non-Patent Citations (1)
| Title |
|---|
| 邱超: "提高可变形部件模型检测效果的研究", 《中国优秀硕士学位论文全文数据库（信息科技辑）》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115631388A (en) * | 2022-12-21 | 2023-01-20 | 第六镜科技(成都)有限公司 | Image classification method and device, electronic equipment and storage medium |
| CN115631388B (en) * | 2022-12-21 | 2023-03-17 | 第六镜科技(成都)有限公司 | Image classification method and device, electronic equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109654945A (en) | With trajectory expressive ability and injure multifarious confrontation fire analogue technique | |
| CN109344878B (en) | Eagle brain-like feature integration small target recognition method based on ResNet | |
| Zortea et al. | Oil-palm tree detection in aerial images combining deep learning classifiers | |
| CN108399420A (en) | A kind of visible light naval vessel false-alarm elimination method based on depth convolutional network | |
| CN103591852B (en) | Target scoring system and automatic correcting method for same | |
| JP5367244B2 (en) | Target detection apparatus and target detection method | |
| Zheng et al. | Object detection for low-resolution infrared image in land battlefield based on deep learning | |
| CN109583424A (en) | A kind of overlapping Ship Target candidate frame screening technique based on decaying confidence level | |
| CN116757999A (en) | A method of shooting counting and explosive point identification based on infrared camera | |
| CN107316018A (en) | A kind of multiclass typical target recognition methods based on combiner model | |
| KR20210143634A (en) | Divot detecting system for maintaining golf course and method using the same | |
| CN115482374A (en) | Target reporting method and system for single cannonball and computer storage medium | |
| CN114578849A (en) | Unmanned aerial vehicle target detection system, target detection method and computer readable storage medium | |
| Zin et al. | Image processing approach to automatic scoring system for archery targets | |
| KR102496737B1 (en) | Monte Carlo simulation-based optimal aiming point calculation method, system and computer readable recording medium | |
| KR20250072992A (en) | Systems and methods for automated target identification, classification, and scoring | |
| Hasan et al. | A Comparison between YOLOV7 and YOLOV5-based detection of Combat Targets and Camouflaged Militia | |
| Yang et al. | Method for building recognition from FLIR images | |
| CN115388705A (en) | Internet of things-based automatic control system and method for light weapon live ammunition shooting training | |
| Li | A new approach for firearm identification with hierarchical neural networks based on cartridge case images | |
| Dmytro | COMPUTER VISION MODELS FOR MILITARY VEHICLES DETECTION PROBLEM | |
| Chen et al. | Application and Implementation of CNN in Artillery Countermeasure Training System | |
| Stenhager et al. | Hit detection in sports pistol shooting | |
| Ma et al. | Assault rifle detection and identification based on convolutional neural network YOLOv3 | |
| KR102711217B1 (en) | Method and apparatus for recommending engagement strategy based on identification of non-linear maneuvers of aircraft |
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: 20171103 |
|
| RJ01 | Rejection of invention patent application after publication |