CN103049764A - Low-altitude aircraft target identification method - Google Patents
Low-altitude aircraft target identification method Download PDFInfo
- Publication number
- CN103049764A CN103049764A CN2012105411951A CN201210541195A CN103049764A CN 103049764 A CN103049764 A CN 103049764A CN 2012105411951 A CN2012105411951 A CN 2012105411951A CN 201210541195 A CN201210541195 A CN 201210541195A CN 103049764 A CN103049764 A CN 103049764A
- Authority
- CN
- China
- Prior art keywords
- aircraft
- sample
- identification method
- target identification
- 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
Landscapes
- Image Analysis (AREA)
Abstract
The invention relates to a low-altitude aircraft target identification method which comprises the following steps of: selecting an aircraft sample; extracting the characteristics of the aircraft sample; training a support vector machine; and performing aircraft target identification with the support vector machine based on detection and tracking. Obviously, the low-altitude aircraft target identification method can be used for identifying both a single target and multiple targets and can provide references for target identification in other conditions using intelligent video monitoring systems.
Description
Technical field
The present invention relates to the video image identification technical field, particularly relate to a kind of low flyer target identification method.
Background technology
In modern national defense and war, the real-time follow-up of low flyer identification to low-level defence, the performance grasping battlefield supremacy and correctly strike target and improve significantly the following system that commands troops, is significant.
The low latitude monitoring is the dead angle of conventional radar systems, along with developing rapidly of computer vision technique, the intelligent video monitoring system can be born the monitoring task fully, uses maturation at aspects such as traffic administration, public safeties, has carried out gradually at the low latitude monitoring and measuring application.
The target of low target identification is exactly to identify different machines, such as helicopter, unmanned aerial vehicle, dalta wing etc.The method of at present image object identification has multiple, such as template matching method, statistic decision method, syntactic pattern method of identification, Fuzzy Pattern Identification, neural network, support vector machine, and the bionical recognition methods that proposes in recent years, chaotic neural network method etc.Certainly for concrete recognition object, the recognition methods of required utilization is not necessarily identical, mainly sees its practicality, accuracy and rapidity, be not newer recognition methods, just can solve well some identification problems, the method for identification that yes is more simple better.
Template matching method is exactly that each classification to be identified is drawn the typical standard template as criterion of identification, and the shortcoming of this kind method is exactly high to the recognition system memory requirement, and calculated amount is large during identification, in addition to noise-sensitive.Fuzzy Pattern Recognition is exactly that system determines degree of membership according to the characteristics of pattern, then with degree of membership fuzzy set is divided into some subsets, corresponding to each classification, realizes classification according to principle of subsidiarity at last.Fuzzy Pattern Recognition allows things to be identified that the interference of certain degree is arranged, and often is difficult to but will accurately reasonably set up subordinate function.It is very complicated that neural network can be processed some environmental informations, and background knowledge is unclear, and the indefinite problem of inference rule, but neural metwork training unstable result need great amount of samples during training.Support vector machine has been avoided the situation of neural metwork training unstable result, and it is less that required training sample amount is compared other algorithms, and simultaneously, the support vector machine method training result is some texts, and what requirement storage is not almost had.
Summary of the invention
Technical matters to be solved by this invention provides a kind of low flyer target identification method, can identify single goal and multiple goal, and the target identification that can be other intelligent video monitoring system application scenarios provides reference.
The technical solution adopted for the present invention to solve the technical problems is: a kind of low flyer target identification method is provided, may further comprise the steps:
(1) chooses the aircraft sample, extract the feature of sample, Training Support Vector Machines;
(2) on the basis of detection and tracking, adopt support vector machine to carry out the identification of aircraft target.
Described step (1) also comprises following substep:
(11) choose the aircraft sample, each aircraft sample standard deviation is chosen the picture of multiple different attitudes, through intercepting, convergent-divergent, picture is carried out normalized;
(12) adopt the Gradient Features descriptor to extract feature to all samples that carry out after the normalized;
(13) adopt a class that unnecessary method is trained.
Also comprise the step of choosing the birds samples pictures and carrying out normalized in the described step (11).
Described step (12) specifically comprises picture is equally divided into some, the Gradient Features that picture is carried out at every turn monolithic extracts, obtain the Grad of each pixel, every is equally divided into several cell elements, the histogram of gradients of 9 directions of statistics in each cell element, finish successively the processing of all cell elements, finally complete picture processing.
Described step (13) specifically comprises first a kind of aircraft as positive sample, other is negative sample all, trains, and obtains a training result text, change afterwards a kind of aircraft and carry out same processing as positive sample, until all aircraft are disposed.
Described step also comprises the step that the training result text is tested after (13), if training result is relatively poor, adds new sample, and again training is until obtain good effect.
Described step (2) comprises following substep:
(21) according to the result of detection and tracking, the tracking target region is amplified a little, and this zone is identified;
(22) selected areas is carried out the zone identification of specific size at every turn, each extracted region Gradient Features descriptor feature is processed; Described specific size is the size after the picture normalized;
(23) adopt first a training result text that target is identified, affiliated aircraft then end of identification if find, if not, just adopt second training result text to identify, until all training result texts all identify finish after, all do not find target, then think birds.
In the described step (22) selected areas is carried out convergent-divergent and process, so that the target of different sizes can both be identified.
Beneficial effect
Owing to adopted above-mentioned technical scheme, the present invention compared with prior art has following advantage and good effect: the present invention can identify single goal and multiple goal, and has consuming time fewly, and real-time is high, the advantage that accuracy of identification is high.Embodiment
Below in conjunction with specific embodiment, further set forth the present invention.Should be understood that these embodiment only to be used for explanation the present invention and be not used in and limit the scope of the invention.Should be understood that in addition those skilled in the art can make various changes or modifications the present invention after the content of having read the present invention's instruction, these equivalent form of values fall within the application's appended claims limited range equally.
Embodiments of the present invention relate to a kind of low flyer target identification method, may further comprise the steps:
Step 1 is chosen sample, extracts feature, Training Support Vector Machines, and its detailed implementation procedure is:
(1) gets the aircraft sample, each machine is chosen three thousand sheets samples pictures of getting different attitudes and size, through intercepting, convergent-divergent, picture is all normalized to the size of 96*64, in order to distinguish birds, the samples pictures unification of also getting three thousand sheets birds simultaneously normalizes to the size of 96*64.For example, from image or video, select three machines (helicopter, unmanned aerial vehicle, dalta wing) and the birds picture that may occur as training sample, through intercepting, normalize at last the picture of 96*64 size.
(2) adopt Gradient Features descriptor (HOG) to extract feature to all training samples, the piece that the picture of 96*64 yardstick is carried out 16*16 at every turn carries out Gradient Features and extracts, obtain the Grad of each pixel, the histogram of gradients of 9 directions of statistics in each 8*8 cell element, the size of stepping 8*8 after handling, the same processing, after finishing the processing of a line direction, just move down 16 row, proceed to process, finally complete picture processing, obtain one 2772 column vector of tieing up.
(3) in order to accelerate the target recognition speed, adopt a class that unnecessary method is trained, allow first a machine as positive sample, other all is negative sample, train, obtain a training result text, allow again another machine carry out same processing as positive sample, until all machines are disposed.For example: choosing helicopter for the first time is positive sample, and other is negative sample all, obtains first training result text; For the second time choosing unmanned aerial vehicle is positive sample, and other is negative sample all, obtains second training result text; Choosing for the third time dalta wing is positive sample, and other is negative sample all, obtains the 3rd training result text.
(4) the training result text is tested, if training result is relatively poor, added new sample, again training, so repeatedly several times, until obtain good effect.
Step 2 on the basis of detection and tracking, adopts support vector machine to carry out the identification of aircraft target, and its detailed implementation procedure is:
(5) according to the result of detection and tracking, the tracking target region is amplified a little, only this zone is identified, and do not needed full figure is identified, saved the plenty of time, improved the real-time of identification;
(6) selected areas is carried out the zone identification of 96*64 at every turn, each 96*64 extracted region HOG feature is processed.Simultaneously selected areas is advanced convergent-divergent and process, the target that reaches different sizes can both be identified.
(7) adopt first a training result text that target is identified, if find it is affiliated machine, with regard to end of identification, if not, just then adopt second training result text, if until all training result texts all identify finish after, all do not find target, then think birds.
Be not difficult to find, when identification, utilize on support vector machine method and the basis of training result sample in detection and tracking, target identification is carried out in the appointed area, at every turn with 96*64 size scanned picture.Simultaneously convergent-divergent can be carried out in the appointed area, dwindle processing first after, dwindle at most four times, then carry out target identification amplifying to process at each yardstick, can both identify with the targets that reach different sizes, thereby realize single goal and multiobject identification.In identifying, do not need full figure is identified, saved the plenty of time, improved the real-time of identification.
Claims (8)
1. a low flyer target identification method is characterized in that, may further comprise the steps:
(1) chooses the aircraft sample, extract the feature of sample, Training Support Vector Machines;
(2) on the basis of detection and tracking, adopt support vector machine to carry out the identification of aircraft target.
2. low flyer target identification method according to claim 1 is characterized in that, described step (1) also comprises following substep:
(11) choose the aircraft sample, each aircraft sample standard deviation is chosen the picture of multiple different attitudes, through intercepting, convergent-divergent, picture is carried out normalized;
(12) adopt the Gradient Features descriptor to extract feature to all samples that carry out after the normalized;
(13) adopt a class that unnecessary method is trained.
3. low flyer target identification method according to claim 1 is characterized in that, also comprises the step of choosing the birds samples pictures and carrying out normalized in the described step (11).
4. low flyer target identification method according to claim 1, it is characterized in that, described step (12) specifically comprises picture is equally divided into some, the Gradient Features that picture is carried out at every turn monolithic extracts, obtain the Grad of each pixel, every is equally divided into several cell elements, the histogram of gradients of 9 directions of statistics in each cell element, finish successively the processing of all cell elements, finally complete picture processing.
5. low flyer target identification method according to claim 1, it is characterized in that, described step (13) specifically comprises first a kind of aircraft as positive sample, other all is negative sample, train, obtain a training result text, change afterwards a kind of aircraft and carry out same processing as positive sample, until all aircraft are disposed.
6. low flyer target identification method according to claim 1 is characterized in that, described step also comprises the step that the training result text is tested after (13), if training result is relatively poor, add new sample, again training is until obtain good effect.
7. low flyer target identification method according to claim 2 is characterized in that, described step (2) comprises following substep:
(21) according to the result of detection and tracking, the tracking target region is amplified a little, and this zone is identified;
(22) selected areas is carried out the zone identification of specific size at every turn, each extracted region Gradient Features descriptor feature is processed; Described specific size is the size after the picture normalized;
(23) adopt first a training result text that target is identified, affiliated aircraft then end of identification if find, if not, just adopt second training result text to identify, until all training result texts all identify finish after, all do not find target, then think birds.
8. low flyer target identification method according to claim 7 is characterized in that, in the described step (22) selected areas is carried out convergent-divergent and processes, so that the target of different sizes can both be identified.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2012105411951A CN103049764A (en) | 2012-12-13 | 2012-12-13 | Low-altitude aircraft target identification method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2012105411951A CN103049764A (en) | 2012-12-13 | 2012-12-13 | Low-altitude aircraft target identification method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN103049764A true CN103049764A (en) | 2013-04-17 |
Family
ID=48062396
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN2012105411951A Pending CN103049764A (en) | 2012-12-13 | 2012-12-13 | Low-altitude aircraft target identification method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN103049764A (en) |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104159031A (en) * | 2014-08-19 | 2014-11-19 | 湖北易瓦特科技有限公司 | Method and equipment of locating and tracking target object |
CN104331691A (en) * | 2014-11-28 | 2015-02-04 | 深圳市捷顺科技实业股份有限公司 | Vehicle logo classifier training method, vehicle logo recognition method and device |
CN105759834A (en) * | 2016-03-09 | 2016-07-13 | 中国科学院上海微系统与信息技术研究所 | System and method of actively capturing low altitude small unmanned aerial vehicle |
WO2016201359A1 (en) * | 2015-06-12 | 2016-12-15 | Foina Aislan Gomide | A low altitude aircraft identification system |
CN107016690A (en) * | 2017-03-06 | 2017-08-04 | 浙江大学 | The unmanned plane intrusion detection of view-based access control model and identifying system and method |
CN108038452A (en) * | 2017-12-15 | 2018-05-15 | 厦门瑞为信息技术有限公司 | A kind of quick detection recognition method of household electrical appliances gesture based on topography's enhancing |
CN108549076A (en) * | 2018-03-12 | 2018-09-18 | 清华大学 | A kind of multiple types unmanned plane scene recognition method for gathering figure based on speed section |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101383008A (en) * | 2008-10-23 | 2009-03-11 | 上海交通大学 | Image classification method based on visual attention model |
CN102004922A (en) * | 2010-12-01 | 2011-04-06 | 南京大学 | High-resolution remote sensing image plane extraction method based on skeleton characteristic |
CN102622607A (en) * | 2012-02-24 | 2012-08-01 | 河海大学 | Remote sensing image classification method based on multi-feature fusion |
-
2012
- 2012-12-13 CN CN2012105411951A patent/CN103049764A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101383008A (en) * | 2008-10-23 | 2009-03-11 | 上海交通大学 | Image classification method based on visual attention model |
CN102004922A (en) * | 2010-12-01 | 2011-04-06 | 南京大学 | High-resolution remote sensing image plane extraction method based on skeleton characteristic |
CN102622607A (en) * | 2012-02-24 | 2012-08-01 | 河海大学 | Remote sensing image classification method based on multi-feature fusion |
Non-Patent Citations (1)
Title |
---|
黄洁,张海: "利用支持向量机的飞机目标检测", 《电光与控制》 * |
Cited By (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104159031A (en) * | 2014-08-19 | 2014-11-19 | 湖北易瓦特科技有限公司 | Method and equipment of locating and tracking target object |
CN104331691A (en) * | 2014-11-28 | 2015-02-04 | 深圳市捷顺科技实业股份有限公司 | Vehicle logo classifier training method, vehicle logo recognition method and device |
WO2016201359A1 (en) * | 2015-06-12 | 2016-12-15 | Foina Aislan Gomide | A low altitude aircraft identification system |
US10192451B2 (en) | 2015-06-12 | 2019-01-29 | Airspace Systems, Inc. | Low altitude aircraft identification system |
US10713959B2 (en) | 2015-06-12 | 2020-07-14 | Airspace Systems, Inc. | Low altitude aircraft identification system |
CN105759834A (en) * | 2016-03-09 | 2016-07-13 | 中国科学院上海微系统与信息技术研究所 | System and method of actively capturing low altitude small unmanned aerial vehicle |
CN105759834B (en) * | 2016-03-09 | 2018-07-24 | 中国科学院上海微系统与信息技术研究所 | A kind of system and method actively capturing low latitude small-sized unmanned aircraft |
CN107016690A (en) * | 2017-03-06 | 2017-08-04 | 浙江大学 | The unmanned plane intrusion detection of view-based access control model and identifying system and method |
CN107016690B (en) * | 2017-03-06 | 2019-12-17 | 浙江大学 | Unmanned aerial vehicle intrusion detection and identification system and method based on vision |
CN108038452A (en) * | 2017-12-15 | 2018-05-15 | 厦门瑞为信息技术有限公司 | A kind of quick detection recognition method of household electrical appliances gesture based on topography's enhancing |
CN108038452B (en) * | 2017-12-15 | 2020-11-03 | 厦门瑞为信息技术有限公司 | Household appliance gesture rapid detection and identification method based on local image enhancement |
CN108549076A (en) * | 2018-03-12 | 2018-09-18 | 清华大学 | A kind of multiple types unmanned plane scene recognition method for gathering figure based on speed section |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN103049764A (en) | Low-altitude aircraft target identification method | |
CN111460968B (en) | Unmanned aerial vehicle identification and tracking method and device based on video | |
Liu et al. | Insulator detection in aerial images based on faster regions with convolutional neural network | |
Sørensen et al. | Thistle detection using convolutional neural networks | |
CN103226835B (en) | Based on method for tracking target and the system of online initialization gradient enhancement regression tree | |
CN105989334B (en) | Road detection method based on monocular vision | |
CN111209935B (en) | Unsupervised target detection method and system based on self-adaptive domain transfer | |
CN103839065A (en) | Extraction method for dynamic crowd gathering characteristics | |
CN110175615A (en) | The adaptive visual position recognition methods in model training method, domain and device | |
CN105488468A (en) | Method and device for positioning target area | |
CN101221623A (en) | Object type on-line training and recognizing method and system thereof | |
CN101839722A (en) | Method for automatically recognizing target at medium and low altitudes and positioning carrier with high accuracy | |
CN112036384B (en) | Sperm head shape recognition method, device and equipment | |
Zhang et al. | CFANet: Efficient detection of UAV image based on cross-layer feature aggregation | |
CN103839078A (en) | Hyperspectral image classifying method based on active learning | |
US20220092869A1 (en) | System and method for generating a mask for object instances in an image | |
CN113378675A (en) | Face recognition method for simultaneous detection and feature extraction | |
CN106897730B (en) | SAR target model identification method based on fusion category information and local preserving projection | |
Ding et al. | A novel YOLOv3-tiny network for unmanned airship obstacle detection | |
Lin et al. | Small object detection in aerial view based on improved YoloV3 neural network | |
Bikmullina et al. | Stand for development of tasks of detection and recognition of objects on image | |
CN107273914B (en) | Efficient fusion identification method based on information source self-adaptive dynamic selection | |
CN105184257A (en) | Target detection method and device | |
Eriş et al. | Implementation of target tracking methods on images taken from unmanned aerial vehicles | |
KR20200075073A (en) | Apparatus and Method for Detecting/Analyzing a Blade of a Wind Turbine |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C02 | Deemed withdrawal of patent application after publication (patent law 2001) | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20130417 |