CN106067022A - Remote sensing image harbor ship detection false alarm eliminating method based on DPM algorithm - Google Patents
Remote sensing image harbor ship detection false alarm eliminating method based on DPM algorithm Download PDFInfo
- Publication number
- CN106067022A CN106067022A CN201610366241.7A CN201610366241A CN106067022A CN 106067022 A CN106067022 A CN 106067022A CN 201610366241 A CN201610366241 A CN 201610366241A CN 106067022 A CN106067022 A CN 106067022A
- Authority
- CN
- China
- Prior art keywords
- wave filter
- image
- layer
- root
- score
- 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 25
- 238000001514 detection method Methods 0.000 title claims abstract description 19
- 238000012549 training Methods 0.000 claims abstract description 23
- 238000005457 optimization Methods 0.000 claims abstract description 4
- 230000008030 elimination Effects 0.000 claims description 7
- 238000003379 elimination reaction Methods 0.000 claims description 7
- 238000005070 sampling Methods 0.000 claims description 5
- 239000011159 matrix material Substances 0.000 claims description 4
- 238000006073 displacement reaction Methods 0.000 claims description 3
- 238000012360 testing method Methods 0.000 claims description 3
- 239000002131 composite material Substances 0.000 claims description 2
- 230000007850 degeneration Effects 0.000 claims description 2
- 239000000203 mixture Substances 0.000 claims description 2
- 230000000877 morphologic effect Effects 0.000 abstract description 3
- 230000003287 optical effect Effects 0.000 abstract description 2
- 238000004364 calculation method Methods 0.000 description 6
- 238000005516 engineering process Methods 0.000 description 5
- 238000000605 extraction Methods 0.000 description 3
- 238000009432 framing Methods 0.000 description 2
- 238000003384 imaging method Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 230000007423 decrease Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 239000011800 void material Substances 0.000 description 1
- 239000003643 water by type Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/255—Detecting or recognising potential candidate objects based on visual cues, e.g. shapes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
A remote sensing image harbor ship detection false alarm rejection method based on a DPM algorithm is used for constructing an efficient harbor ship target false alarm rejection method by combining an existing morphological feature-based candidate region detection algorithm aiming at an intra-harbor ship target in an optical remote sensing image. Firstly, establishing a ship target training database in an off-line manner, training and initializing a ship model; constructing a feature image pyramid through an HOG algorithm; then calculating a root model response score on the l-layer resolution image of the characteristic pyramid; finding out a double-resolution characteristic image corresponding to the l-layer characteristic image, calculating the response score of the component model on the image, and expanding the high-resolution area; and finally, obtaining the comprehensive score of each layer of the feature image pyramid according to the scores of the root model and the part model, and obtaining the position of the ship target on the candidate area image through iterative loop optimization.
Description
Technical field
The present invention relates to reaching port in the processing method of remote sensing images, particularly optical satellite remote sensing images automatic business processing
Ship target detection false-alarm elimination method.
Background technology
In recent years, along with empty sky carrying platform and the fast development of sensor technology, utilized remote sensing image to harbour
Region carries out the demand of ship detecting to be increased day by day, and this technology can realize port area boat on a large scale with relatively low cost of labor
The dynamic surveillance of fortune safety.The Major Technology used at present includes: best entropy automatic threshold method (KSW), two-parameter permanent void
Alert (CFAR) algorithm, multipolarization detection method etc..
In having studied, the ship detecting research of optically-based remote sensing images is main with the ship detecting in the scene of ocean
Being main, these targets are positioned in marine background gray feature, and feature substantially and is prone to extract.But, the boats and ships under the background of harbour
Object detection method, the freshest rare method relates to.
Ship target in harbour, has two kinds of existences, and one is internal waters at harbour, and two is to rest against harbour code
On head.The former background is almost identical with marine site, ocean background, it is easy to detection.But the latter's background is complicated, is directly connected to harbour.Harbour
The texture in region is the most similar to ship target to gray feature, cause can the extraction of distinguishing characteristic the most difficult, and can cause
A large amount of false-alarms.Therefore, the detection of boats and ships of reaching port is the technical difficult points of yard craft detection.The boats and ships that reach port in having studied
Detection method, the target candidate district mostly directly extraction obtained is as testing result, or assumes some simple geometric properties
Carrying out preliminary false-alarm rejecting, but reach port naval vessel and be connected between harbour the composition relation with complexity, these existing features are difficult to
Reject false-alarm region., cause ship target some ship detecting methods of reaching port improved recently of cannot being reached port accurately, though
So it is taken based on the thinking that boats and ships local feature confirms, the decision method of the morphological characteristic of fore as sharp-pointed in boats and ships.But in reality
Ship target wide variety in application, many Ship Types do not have sharp-pointed fore feature, as it is shown in figure 1, and same in harbour
There is the facility edge that angle is the most sharp-pointed in sample.Additionally, remote sensing image is affected by factors such as equipment platform imaging, weather
Relatively big, this type of method performance is difficult to ensure that, thus affects rejecting efficiency and the accuracy of ship detecting false-alarm of reaching port.Ultimately result in
A large amount of false-alarms that object candidate area detection-phase produces are difficult to reject.
Summary of the invention
Present invention aims to the difficult point of above-mentioned technology, it is provided that a kind of based on DPM (deformable part sub-model,
Deformable Parts Model) remote sensing images of algorithm reach port ship detecting false-alarm elimination method, by using boats and ships mesh
Mark model inspection, and combine existing based on morphological characteristic candidate region detection algorithm, builds one and reaches port efficiently boats and ships mesh
Target false-alarm elimination method.
The present invention is directed to situation described above, use and carry out picking of false-alarm targets in port based on DPM Algorithm for Training model
Remove.
The method that the present invention provides includes training pattern, sets up characteristics of image pyramid, root wave filter RESPONSE CALCULATION, parts
Wave filter RESPONSE CALCULATION and based on comprehensive score suspected target confirm five steps.The step specifically included is as follows:
The first step: training pattern
This step is that off-line completes, and the target in image is outlined according to handmarking's frame, and stores the dependency number of correspondence
According to.By repeating a large amount of above-mentioned steps, obtain the tranining database of ship target.By the mass data in training storehouse, obtain
Ship model.And initialize ship model, optimize ship model by alternative manner.Finally give in ship model, including one
Individual root model (root wave filter), a partial model (parts wave filter) containing n widget and a distorted pattern.
Second step: set up characteristic image pyramid
By HOG (histograms of oriented gradients, Histograms of Oriented Gradients) algorithm, calculate target
The candidate regions image that candidate region detection-phase produces histograms of oriented gradients under different resolution, and these gradients are special
Levy image to arrange according to the size order of resolution, form characteristic image pyramid.
3rd step: root wave filter RESPONSE CALCULATION
Based on the extraordinary image pyramid obtained in previous step, (root filters to use the root model in boats and ships training pattern
Device), its l layer image in different resolution calculates its response score.
4th step: parts wave filter RESPONSE CALCULATION
According to the resolution in previous step, from characteristic image pyramid, find out two resolution characteristic images of correspondence,
And the widget (parts wave filter) used in boats and ships training pattern in partial model, calculate the response of each widget thereon
Score.Spend further according to deformation, high score region is expanded.
5th step: confirm based on comprehensive score suspected target
By the response score of root wave filter with through expansion after each widget wave filter response score and be added, obtain
The score of l layer, it is judged that score eminence is suspected target position.By repeating the process of the 3rd step to the 5th step, iteration optimization obtains
The position of ship target on candidate regions image.
According to an aspect of the invention, it is provided one remote sensing images based on deformable part sub-model (DPM) algorithm
Reach port ship detecting false-alarm elimination method, and the ship target in harbour on detection candidate regions image automatically, including following step
Rapid:
A) off-line sets up ship target tranining database, according to the training sample set in this data base, trains and initializes
Ship model；
B) extract under different resolution by histograms of oriented gradients (HOG) algorithm, candidate regions image direction Gradient Features
Image, thus set up characteristic image pyramid；
C) on characteristic image pyramid l layer image in different resolution, root model response score is calculated；
D) find the twice resolution characteristics image corresponding with l layer characteristic image, this image calculates each parts mould
The response score of type, and according to degeneration cost, high subregion is expanded；
E) according to the score of root model Yu partial model, the comprehensive score of every layer of characteristic image pyramid is obtained, by repeatedly
The position of ship target on candidate regions image is obtained for loop optimization.
Accompanying drawing explanation
Ship target fore examples of features in Fig. 1 remote sensing image
Fig. 2 remote sensing images based on DPM algorithm reach port ship detecting false-alarm elimination method flow chart
Detailed description of the invention:
It is described below how be embodied as the method that the present invention provides, Fig. 2 is the FB(flow block) of the method that the present invention provides.
Processing procedure is as follows:
The first step: training pattern
(1.1st) step obtains suspected target region and the foundation in storehouse: the kind of definition training objective is c.P is indicia framing
Positive sample database, each P comprises a pair (I, B), and wherein I is the image needing labelling, and B is targeted species in image I
Indicia framing for the target of c.N is negative sample data base.
(1.2nd) step model initialization: according to length-width ratio, P is divided into m group, is labeled as P1,…,Pm.Use SVM to calculate simultaneously
Method training m group root wave filter F1,…,Fm.These root wave filter are merged into a root wave filter, retains positive sample P and negative sample
N there is flag parameters more.The method using coordinate to decline, by the label in these root wave filter of checker, thus counts
Calculate and obtain optimal root filter location.The root wave filter obtained is carried out difference and is amplified to its two resolution.Use with
The method training that training root wave filter is identical obtains the parts wave filter containing n widget, by this n in parts wave filter
Widget is categorized as lh (left-hand, left hand) and rh (right-hand, the right hand) two classes, is individually positioned in root model
Mandrel both sides.Finally, set 6 predetermined rectangle positions, by iterative cycles, calculate each parts wave filter in precalculated position
Weighted value, determines n widget position respectively in parts wave filter according to weight limit value.
Second step: set up characteristics of image pyramid
By HOG algorithm, calculate the histograms of oriented gradients of the candidate regions image that object candidate area detection-phase produces,
Setting the pyramidal number of plies of feature as N, by image by resolution order arrangement from small to large, pyramid top is former point of input
The histograms of oriented gradients characteristic pattern of resolution.Meanwhile, λ is set as sampling specification.
3rd step: root wave filter RESPONSE CALCULATION
By carrying out the first step training for positive and negative sample set, it is thus achieved that include a root model (root wave filter), one
The individual partial model (parts wave filter) containing n widget and the object module of corresponding distorted pattern.Training
To the object module comprising n widget be expressed as (n+2) dimension group (F0,P1,…,Pn,b).Wherein F0For root wave filter, Pi
For i-th parts wave filter, b is the real part of deviation value.Root wave filter responds to be divided into:
R0,l0(x0,y0)=F0'·φ(H,p)
P=(x0,y0,l0)
Wherein, p refers to characteristic image pyramid l0On layer (layer corresponding to root wave filter), position is (x0,y0) point；H is
The characteristic image pyramid set up in previous step；(H p) is the characteristic vector on p position in H to φ；F0' for root wave filter F0Row
Sequence.
4th step: parts wave filter RESPONSE CALCULATION
Parts wave filter score be intended to search out one in the range of anchor point, comprehensive matching and the optimum position of deformation.
(4.1st) step calculates the response of each widget: parts wave filter response score is as follows:
Ri,l(x, y)=Fi'·φ(H,p)
P=(x, y, l)
L=l0-λ
Wherein, due to the twice that the training image of parts wave filter is root wave filter training image resolution, thus corresponding
Feature pyramidal layer l of parts wave filter, for root wave filter place layer l0Deduct sampling specification λ.P is characterized on pyramid l layer,
Position is (x, some y)；(H p) is the characteristic vector on p position in H to φ；Fi' for parts wave filter FiLine order.These are little
Unit response is added, and just obtains the parts wave filter response score of l layer.
(4.2nd) step expands high response region: according to the region that score in previous step is higher, by outside for these high subregions
Expand:
(dx, dy)=(x, y)-(2 (x0,y0)+v)
φd(dx, dy)=(dx, dy, dx2,dy2)
Wherein (x y) is the ideal position of parts wave filter；(dx, dy) is the displacement relative to anchor point, (x0,y0) it is
The position of root wave filter in 3rd step, due to the resolution of parts wave filter place feature pyramidal layer, for root wave filter place
The twice of layer, so needing to be multiplied by 2 multiplying factors, v is the relative position of parts wave filter anchor point and root wave filter；diFor offseting to
Amount；φdCost weight for skew.
5th step: confirm based on comprehensive score suspected target
(5.1st) step calculates each layer and comprehensively responds score: the root wave filter obtained according to the 3rd step and the 4th step and parts
The score of wave filter, can calculate the comprehensive score of characteristic image every layer testing result:
The comprehensive decile of every layer, for l0The root wave filter response score of layer, adds the parts through expanding and after down-sampling
Wave filter response score, wherein b is that the skew for making parts wave filter align spends.
(5.2nd) step determines suspected target based on comprehensive score: according to every layer of composite score of pyramid obtained in the previous step,
In order to obtain the comprehensive score of the candidate regions image that object candidate area detection-phase produces, by every layer of characteristic image pyramid
Detect comprehensive decile to be iterated optimizing:
By choosing the part of comprehensive highest scoring, the matrix of mark that will obtain.This matrix is mapped back in image,
Ship target position in the image of candidate region may finally be determined.
The present invention has the advantage that compared with existing detection method
The Gradient Features used in this method, is compared to gray scale, Texture eigenvalue, and noise resisting ability is strong, to image
Imaging effect requires low, it is adaptable to the remote sensing image under the conditions of vast.Obtain additionally, extract from low-resolution image
Direction gradient feature, can preferably describe the appearance profile of object on candidate regions image.Meanwhile, from high-resolution candidate's figure image
The direction gradient feature that middle extraction obtains, it is possible to the minutia of each object in careful depiction picture.Therefore, direction gradient
Feature and the pyramidal combination of characteristic image, be demonstrated by the appearance profile of ship target and boats and ships mesh in candidate regions image accurately
Mark the structural relation of self, improve the discrimination of ship target and harbour, effectively raise the accuracy that false-alarm is rejected.
By the training to mass data, in the ship target model obtained, root model can preferable matching candidate district figure
The suspected target that in Xiang, profile is similar to Ship Target, partial model can carry out fine ratio to suspected target internal structure
Relatively, deformation simultaneously spends deformation and the change in displacement that can control partial model each with comparison, so that ship model has
Well adapting to property and morphotropism, be not limited to single fore feature, it is possible to adapt to polytype ship target, thus carry
High detection and the efficiency identified, achieve the process that false-alarm is rejected efficiently.
Claims (4)
1. remote sensing images based on deformable part sub-model (DPM) algorithm reach port a ship detecting false-alarm elimination method, are used for
Automatically the ship target in harbour on detection candidate regions image, comprises the following steps:
A) off-line sets up ship target tranining database, according to the training sample set in this data base, trains and initializes boats and ships
Model；
B) extract under different resolution by histograms of oriented gradients (HOG) algorithm, candidate regions image direction Gradient Features image,
Thus set up characteristic image pyramid；
C) on characteristic image pyramid l layer image in different resolution, root model response score is calculated；
D) find the twice resolution characteristics image corresponding with l layer characteristic image, this image calculates each partial model
Response score, and according to degeneration cost, high subregion is expanded；
E) according to the score of root model Yu partial model, obtain the comprehensive score of every layer of characteristic image pyramid, followed by iteration
Ring optimization obtains the position of ship target on candidate regions image.
Remote sensing images the most according to claim 1 reach port ship detecting false-alarm reject detection method, it is characterised in that step
C) including:
By repeat step A) training, obtain by a root model (root wave filter), the parts containing n widget
Model (parts wave filter) and the object module of corresponding distorted pattern composition
The object module comprising n widget is expressed as (n+2) dimension group (F0,P1,…,Pn, b), wherein F0For root wave filter,
PiFor i-th parts wave filter, b is the real part of deviation value,
Root wave filter response score is defined as:
P=(x0,y0,l0) (2)
Wherein, p refers to characteristic image pyramid l0On layer (layer corresponding to root wave filter), position is (x0,y0) point；H is built
Vertical characteristic image pyramid；(H p) is the characteristic vector on p position in H to φ；F0' for root wave filter F0Line order.
Detection method the most according to claim 1, it is characterised in that step D) including:
Calculate the response of each widget: parts wave filter response score is as follows:
Ri,l(x, y)=Fi'·φ(H,p) (3)
P=(x, y, l) (4)
L=l0-λ (5)
Wherein, due to the twice that the training image of parts wave filter is root wave filter training image resolution, so corresponding component
Feature pyramidal layer l of wave filter, for root wave filter place layer l0Deducting sampling specification λ, p is characterized on pyramid l layer, position
For (x, some y)；(H p) is the characteristic vector on p position in H to φ；Fi' for parts wave filter FiLine order, by these widgets
Response is added, and just obtains the parts wave filter response score of l layer；
Expand high response region, high subregion outwards expanded:
(dx, dy)=(x, y)-(2 (x0,y0)+v) (7)
φd(dx, dy)=(dx, dy, dx2,dy2) (8)
Wherein (x y) is the ideal position of parts wave filter；(dx, dy) is the displacement relative to anchor point, (x0,y0) it is root filter
The position of ripple device, due to the resolution of parts wave filter place feature pyramidal layer, for the twice of root wave filter place layer, so
Needing to be multiplied by 2 multiplying factors, v is the relative position of parts wave filter anchor point and root wave filter；diFor offset vector；φdFor skew
Spend weight.
Detection method the most according to claim 1, it is characterised in that step E) including:
Calculate each layer and comprehensively respond score, including according to the root wave filter obtained and the score of parts wave filter, calculate characteristic pattern
Comprehensive score as every layer of testing result:
Wherein, the comprehensive decile of every layer, for l0The root wave filter response score of layer, adds the parts through expanding and after down-sampling
Wave filter response score, wherein b is that the skew for making parts wave filter align spends,
Suspected target is determined, including according to every layer of composite score of pyramid obtained in the previous step, by characteristic pattern based on comprehensive score
As the comprehensive decile of the detection of every layer of pyramid is iterated optimizing:
By choosing the part of comprehensive highest scoring, the matrix of mark that will obtain,
This matrix is mapped back in image, ship target position in the image of candidate region may finally be determined.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610366241.7A CN106067022A (en) | 2016-05-28 | 2016-05-28 | Remote sensing image harbor ship detection false alarm eliminating method based on DPM algorithm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610366241.7A CN106067022A (en) | 2016-05-28 | 2016-05-28 | Remote sensing image harbor ship detection false alarm eliminating method based on DPM algorithm |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106067022A true CN106067022A (en) | 2016-11-02 |
Family
ID=57420975
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610366241.7A Pending CN106067022A (en) | 2016-05-28 | 2016-05-28 | Remote sensing image harbor ship detection false alarm eliminating method based on DPM algorithm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106067022A (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106709914A (en) * | 2017-01-05 | 2017-05-24 | 北方工业大学 | SAR image ship detection false alarm eliminating method based on two-stage DEM sea-land reservoir |
| CN107220650A (en) * | 2017-05-31 | 2017-09-29 | 广州炒米信息科技有限公司 | Food image detection method and device |
| CN107368832A (en) * | 2017-07-26 | 2017-11-21 | 中国华戎科技集团有限公司 | Target detection and sorting technique based on image |
| CN107766792A (en) * | 2017-06-23 | 2018-03-06 | 北京理工大学 | A kind of remote sensing images ship seakeeping method |
| CN108171752A (en) * | 2017-12-28 | 2018-06-15 | 成都阿普奇科技股份有限公司 | A kind of sea ship video detection and tracking based on deep learning |
| CN108241869A (en) * | 2017-06-23 | 2018-07-03 | 上海远洲核信软件科技股份有限公司 | A kind of images steganalysis method based on quick deformable model and machine learning |
| CN108615028A (en) * | 2018-05-14 | 2018-10-02 | 北京主线科技有限公司 | The fine granularity detection recognition method of harbour heavy vehicle |
| CN109977965A (en) * | 2019-02-28 | 2019-07-05 | 北方工业大学 | Method and device for determining detection target in remote sensing airport image |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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) |
-
2016
- 2016-05-28 CN CN201610366241.7A patent/CN106067022A/en active Pending
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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) |
Non-Patent Citations (2)
| Title |
|---|
| PEDRO F. FELZENSZWALB 等: "Object Detection with Discriminatively Trained Part-Based Models", 《IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE》 * |
| 王彦情 等: "光学遥感图像舰船目标检测与识别综述", 《自动化学报》 * |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106709914A (en) * | 2017-01-05 | 2017-05-24 | 北方工业大学 | SAR image ship detection false alarm eliminating method based on two-stage DEM sea-land reservoir |
| CN106709914B (en) * | 2017-01-05 | 2020-01-17 | 北方工业大学 | SAR image ship detection false alarm eliminating method based on two-stage DEM sea-land reservoir |
| CN107220650A (en) * | 2017-05-31 | 2017-09-29 | 广州炒米信息科技有限公司 | Food image detection method and device |
| CN107220650B (en) * | 2017-05-31 | 2020-07-17 | 广州炒米信息科技有限公司 | Food image detection method and device |
| CN107766792A (en) * | 2017-06-23 | 2018-03-06 | 北京理工大学 | A kind of remote sensing images ship seakeeping method |
| CN108241869A (en) * | 2017-06-23 | 2018-07-03 | 上海远洲核信软件科技股份有限公司 | A kind of images steganalysis method based on quick deformable model and machine learning |
| CN107766792B (en) * | 2017-06-23 | 2021-02-26 | 北京理工大学 | Remote sensing image ship target identification method |
| CN107368832A (en) * | 2017-07-26 | 2017-11-21 | 中国华戎科技集团有限公司 | Target detection and sorting technique based on image |
| CN108171752A (en) * | 2017-12-28 | 2018-06-15 | 成都阿普奇科技股份有限公司 | A kind of sea ship video detection and tracking based on deep learning |
| CN108615028A (en) * | 2018-05-14 | 2018-10-02 | 北京主线科技有限公司 | The fine granularity detection recognition method of harbour heavy vehicle |
| CN109977965A (en) * | 2019-02-28 | 2019-07-05 | 北方工业大学 | Method and device for determining detection target in remote sensing airport image |
| CN109977965B (en) * | 2019-02-28 | 2021-04-13 | 北方工业大学 | Method and device for determining detection target in remote sensing airport image |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106067022A (en) | Remote sensing image harbor ship detection false alarm eliminating method based on DPM algorithm | |
| CN108510467B (en) | SAR image target identification method based on depth deformable convolution neural network | |
| CN109711295B (en) | Optical remote sensing image offshore ship detection method | |
| CN109800735A (en) | Accurate detection and segmentation method for ship target | |
| CN114742820B (en) | Bolt loosening detection method, system and storage medium based on deep learning | |
| CN104331886B (en) | Port region ship and warship detection method based on high resolution SAR image | |
| CN106485651B (en) | The image matching method of fast robust Scale invariant | |
| CN107092890A (en) | Naval vessel detection and tracking based on infrared video | |
| CN110516605A (en) | Any direction Ship Target Detection method based on cascade neural network | |
| CN106056625B (en) | A kind of Airborne IR moving target detecting method based on geographical same place registration | |
| CN107909002B (en) | Sea-land segmentation method of infrared remote sensing image based on coastline matching | |
| CN108254727A (en) | A kind of radar plot condensing method based on Contour extraction | |
| CN109859178A (en) | A kind of infrared remote sensing image real-time target detection method based on FPGA | |
| Xie et al. | Fast ship detection from optical satellite images based on ship distribution probability analysis | |
| Zeng et al. | Research on mosaic image data enhancement for overlapping ship targets | |
| CN114821358A (en) | Optical remote sensing image marine ship target extraction and identification method | |
| CN109325958A (en) | A kind of offshore ship detection method for refining and improve generalised Hough transform based on profile | |
| CN116245872A (en) | Aircraft structure crack detection method based on convolutional neural network | |
| CN107657272A (en) | A kind of remote sensing images method for detecting targets at sea | |
| CN104809444B (en) | The half ridge anisotropic filter Method of IR Small Target Fusion Recognition based on tensor extension | |
| CN115471782B (en) | Unmanned ship-oriented infrared ship target detection method and device | |
| Zhang et al. | Ship detection and recognition in optical remote sensing images based on scale enhancement rotating cascade R-CNN networks | |
| Shuai et al. | A ship target automatic recognition method for sub-meter remote sensing images | |
| CN116051808A (en) | YOLOv 5-based lightweight part identification and positioning method | |
| Yang et al. | Method for building recognition from FLIR images |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20161102 |
|
| RJ01 | Rejection of invention patent application after publication |