CN104408482B - A kind of High Resolution SAR Images object detection method - Google Patents
A kind of High Resolution SAR Images object detection method Download PDFInfo
- Publication number
- CN104408482B CN104408482B CN201410743668.5A CN201410743668A CN104408482B CN 104408482 B CN104408482 B CN 104408482B CN 201410743668 A CN201410743668 A CN 201410743668A CN 104408482 B CN104408482 B CN 104408482B
- Authority
- CN
- China
- Prior art keywords
- target
- subgraph
- roi
- classification
- image
- 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.)
- Active
Links
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
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (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)
- Astronomy & Astrophysics (AREA)
- Remote Sensing (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The embodiment of the present invention discloses a kind of High Resolution SAR Images object detection method, is applied to technical field of image processing, mainly solves that the existing detection false-alarm to solve existing SAR image object detection method is higher and the defect of position inaccurate.This method comprises: carrying out overlap partition to original SAR image, multiple subgraphs are obtained;The feature for extracting training sample is trained classifier, carries out catalog classification to multiple subgraphs with the classifier after training;Obtain the rarity feature of weighting for the pixel that catalogue classification is the subgraph comprising target, Threshold segmentation is carried out to the subgraph comprising target, Morphological scale-space is carried out to the bianry image after segmentation, it obtains the region of interest ROI that catalogue classification is the subgraph comprising target to be modified the ROI, obtains object detection results.The present invention is applied to the High Resolution SAR image object detection more than target type.
Description
Technical field
The invention belongs to technical field of image processing, are related to a kind of High Resolution SAR Images object detection method.It can be used for
Object detection and recognition in High Resolution SAR image, computer vision and field of intelligent control.
Background technique
The detection of synthetic aperture radar (SAR) image object is gray scale, texture, shape, edge and the direction etc. using target
Information determines its position in SAR image, by the image processing techniques of target and background separation.SAR image target detection be into
The premise of one step target identification is the hot issue of current intelligent control and computer vision field research.Due to SAR imaging
Particularity, so that SAR image is very different with optical imagery, such as SAR image includes more redundancies, and there are spots
Spot noise, SAR target azimuthal are very sensitive etc..So how from SAR image accurately to detect that target is always SAR
The difficult point of image interpretation.
As SAR image resolution ratio is continuously improved, explosive increase is presented in target information, and the difficult point of target detection is main
Have: (1) image scene becomes increasingly complex, and the data volume that need to be handled is increasing;(2) target is by low in original single channel single polarization
Point target on image in different resolution, becomes the Area Objects with abundant minutia and scattering signatures, the type of target signature and
Unstability increases;(3) redundancy that image includes increases, if handled image all pixels, arithmetic speed meeting
It is affected.
CFAR method based on constant false alarm is classical SAR image object detection method.This method be in image overall or
Under the conditions of local background is known, estimates the clutter distribution parameter of background, adaptively choose the threshold value of detection, then using solid
The sliding window traversal image for determining size, judges whether pixel belongs to target.When image detail information is unobvious, background is more uniform
When, this method can be realized in the case where the prior informations such as no target type, target property and background characteristics to target
Detection, and preferable testing result can be obtained.
Although CFAR method accurate in the case where simply utilizing picture amplitude information can detect target,
For High Resolution SAR image, detailed information is very rich, and CFAR is primarily present following defect in the case of this:
(1) target detection false-alarm is higher, is unsuitable for detailed information High Resolution SAR image abundant.CFAR method is simple
Target detection is carried out using the amplitude information of image, can be misjudged when the gray value of the gray value of background pixel and target is close
For target, testing result is caused higher false-alarm occur.Meanwhile testing result is easily affected by noise, after being unfavorable for image
Continuous processing, such as target identification.
(2) what is detected is suspected target region all in image, for true target position inaccurate.
Summary of the invention
The embodiment of the present invention provides a kind of High Resolution SAR Images object detection method, to solve existing SAR image mesh
Mark detection method detection false-alarm is higher and the defect of position inaccurate.
The first aspect of the present invention provides a kind of High Resolution SAR image object detection method, comprising:
Overlap partition is carried out to the original SAR image of input, obtains multiple subgraphs;
Extract training sample feature classifier is trained, with the classifier after training to multiple subgraphs into
Column catalogue classification;
The rarity feature of weighting that obtains the pixel that catalogue classification is the subgraph comprising target, to described comprising target
Subgraph carries out Threshold segmentation, carries out Morphological scale-space to the bianry image after segmentation, and obtaining catalogue classification is to include target
The region of interest ROI of subgraph;
The ROI is modified, object detection results are obtained.
According in a first aspect, in the first possible implementation, the original SAR image of described pair of input is overlapped
Piecemeal includes:
Overlap partition is carried out to original SAR image according to formula 1;
Wherein, m, n are respectively the length and width of original SAR image, and the size of piecemeal is x × x, and p% is Duplication, is obtained
Subgraph is denoted as Si, i=1,2 ..., N, N are the number for obtaining multiple subgraphs.
According in a first aspect, in the second possible implementation, the feature for extracting training sample is to classifier
It is trained, catalog classification is carried out to multiple subgraphs with the classifier after training, comprising:
Catalogue classification number is determined, wherein being M in the available SAR image of prior information by the target type number for including, often
A kind of corresponding two kinds of catalogue classifications of target type, i.e., comprising this target and not comprising this target, so, total catalogue classification is shared
2M, each catalogue classification is denoted as Cq, q ∈ 2M, wherein C1,C2,...,CMFor the catalogue category label comprising target;
It randomly selects training sample for the 2M catalogue classification respectively and extracts feature vector;
The feature vector for extracting the subgraph carries out catalog classification to the subgraph.
According to second of possible implementation of first aspect, in the third possible implementation, the difference
Training sample is randomly selected for the 2M catalogue classification and extracts feature vector, comprising:
Weber partial descriptions feature WLD is extracted as feature vector to each training sample, is denoted asAnd to classifier into
Row training;
Correspondingly, the feature vector for extracting the subgraph, carries out catalog classification to the subgraph, comprising:
The WLD feature for extracting the subgraph carries out catalog classification to the subgraph using trained classifier.
According in a first aspect, in the fourth possible implementation, the subgraph to described comprising target carries out
Threshold segmentation, comprising:
Threshold segmentation, given threshold Th are carried out to the subgraph comprising targetu, image is divided as the following formula
It cuts, the image after segmentation is denoted as BW, such as formula 2:
In formula 2, UkFor the rare value of weighting of k-th of pixel.
According in a first aspect, in a fifth possible implementation, the rarity feature of weighting, comprising:
The spatial distribution weight for calculating k-th of pixel, is denoted as ωk, k=msub*nsub, ωkCalculate such as formula 3:
Wherein, in formula 3, msub,nsubRespectively subgraph is long and wide, IkFor the gray value of k-th of pixel of subgraph, I0
For the pixel value of subgraph center pixel, num be in subgraph with IkThe number of equal pixel;
The rare value for calculating k-th of pixel, is denoted as uk, k=msub*nsub, ukCalculate such as formula 4:
Wherein, in formula 4, msub,nsubRespectively subgraph is long and wide, unkFor time that k-th of pixel occurs in subgraph
Number;
The rare value of weighting for calculating k-th of pixel, is denoted as Uk, k ∈ m*n, UkCalculate such as formula 5;
Uk=ωk*uk (5)
According in a first aspect, in a sixth possible implementation, described be modified to the ROI includes:
ROI number of acquisition is denoted as L, and t-th of ROI is ROIt;
Judge ROIs, s ∈ L and ROIt, whether t ∈ L is connected region, wherein s ≠ t.If ROIsWith ROItFor connected region
Domain then merges them, obtains a new ROI;This step is repeated, until having traversed all ROI.
High Resolution SAR Images object detection method provided in an embodiment of the present invention, by carrying out overlap partition to image,
The unified subgraph of size is obtained, then all subgraph is carried out cataloguing, that is, according to subgraph whether includes mesh
It is marked with and the type comprising target carries out catalog classification to it, the mesh comprising all subgraphs is obtained after image directory
Record.Finally from catalogue selection comprising target subgraph carry out target detection, obtain interested region (ROI), to ROI into
Row amendment obtains final testing result, removes redundancy, reduces influence of the complex background to target detection, examine target
The specific aim of survey enhances, and obtained result is more accurate.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described.
Fig. 1 is the flow chart of High Resolution SAR image object detection provided in an embodiment of the present invention;
Fig. 2 is the original SAR image that emulation experiment of the invention uses;
Fig. 3 is the result for carrying out target detection to image using CFAR algorithm;
Fig. 4 is image block schematic diagram used in the present invention;
Fig. 5 is the μ that the present invention usesl,μr,μd,μuThe specific location schematic diagram of included pixel;
Fig. 6 is the present invention to the sorted result figure of sub- image directory, C1Class is that meadow includes target, C2Class be meadow not
Containing target, C4Class is that field is free of target, C6Class is that forest is free of target, C8Class is that lake is free of target;
Fig. 7 is the present invention to class C1Subgraph comprising target carries out the result of target detection.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.
It is inaccurate it is an object of the invention to overcome the detection false-alarm of existing SAR image object detection method higher and position
True deficiency proposes a kind of object detection method based on the classification of High Resolution SAR image directory.Image directory classification is original
On the basis of image block, according to the difference of information included in the subgraph that piecemeal obtains, sub-image assigns to different mesh
It records under classification.The advantage of catalog classification is embodied in: (1) removing redundancy.It, only need to be to including mesh after image carries out catalog classification
Target subgraph carries out target detection, reduces data volume to be treated, ensure that the real-time of detection algorithm;(2) it reduces
Influence of the complex background to target detection, enhances the specific aim of target detection, obtained result is more accurate.
The technical scheme is that carrying out overlap partition to image first, the unified subgraph of size is obtained, it is then right
Whether all subgraph carries out cataloguing, that is, according to subgraph include that target and type comprising target carry out it
Catalog classification, obtain after image directory one include all subgraphs catalogue.Finally selection includes target from catalogue
Subgraph carries out target detection, obtains interested region (ROI), is modified to obtain final testing result to ROI.Below
The present embodiment is described in detail.Fig. 1 is the process of High Resolution SAR image object detection method provided in an embodiment of the present invention
Figure.With reference to Fig. 1, this method is mainly comprised the steps that
Step 10, overlap partition is carried out to the original SAR image of input, obtains multiple subgraphs.
Reference formula 1,
In formula 1, m, n are respectively the length and width of original SAR image, and the size of piecemeal is x × x, and p% is Duplication, is obtained
To subgraph be denoted as Si, i=1,2 ..., N, N are the number for obtaining multiple subgraphs.
Step 20, the feature for extracting training sample is trained classifier, with the classifier after training to multiple subgraphs
As carrying out catalog classification, the classification of catalog classification includes comprising target and not including target.
Firstly, determining catalogue classification number: being M in the available SAR image of prior information by the target type number for including, often
A kind of corresponding two kinds of catalogue classifications of target type, i.e., comprising this target and not comprising this target, so, total catalogue classification is shared
2M, each catalogue classification is denoted as Cq, q ∈ 2M, wherein C1,C2,...,CMFor the catalogue category label comprising target.
Secondly, selecting training sample and extracting feature vector: randomly selecting training sample for 2M catalogue classification respectively.
Weber partial descriptions feature (WLD) is extracted to each training sample and is used as feature vector, is denoted asAnd classifier is instructed
Practice.For the pixel in W × W window, WLD feature calculation such as formula 2:
In formula 2, wcFor the center pixel in window, wjFor j-th of pixel in window, μl,μr,μd,μuIn respectively
Mean value, the mean value of right all pixels, the mean value of lower section all pixels and the top all pixels of imago element left all pixels
Mean value.By using WLD feature, azimuth information is not only contained, but also embodies the difference of pixel Yu its neighborhood territory pixel
Property, it being capable of accurate characterization target.
Finally, extracting the WLD feature of subgraph, catalog classification is carried out to it using trained classifier.Catalog classification
After, given up for being not belonging to the other subgraph of any sort.
Step 30, the rarity feature of weighting for obtaining the pixel that catalogue classification is the subgraph comprising target, to including mesh
Target subgraph carries out Threshold segmentation, carries out Morphological scale-space to the bianry image after segmentation, and obtaining catalogue classification is to include mesh
The area-of-interest (Region Of Interest) of target subgraph.
In catalogue classification, the subgraph comprising target is marked as C1,C2,...,CMSubgraph.
For piece image, the number of object pixel is less relative to the number of background pixel, and object pixel is usually
Flock together rather than the distribution that disperses in the picture, while the gray value of target is greater than the gray value of background.According to above-mentioned
Analysis, calculates the rarity feature of weighting of pixel, is denoted as U.After obtaining the rare value of weighting of each pixel, threshold value is carried out to image
Segmentation, given threshold Thu, image is split as the following formula, the image after segmentation is denoted as BW, such as formula 3:
In formula 3, UkFor the rare value of weighting of k-th of pixel.Morphology is carried out to the bianry image BW that segmentation obtains to open
The processing such as closed operation, corrosion expansion, are accurately characterized the two-value template of interesting image regions.Pixel value is 1 in this template
Region correspond to the region of interest ROI in original image.
Step 40, ROI is modified, obtains object detection results.
ROI number of acquisition is denoted as L, and t-th of ROI is ROIt。
Judge ROIs, s ∈ L and ROIt, whether t ∈ L is connected region, wherein s ≠ t.If ROIsWith ROItFor connected region
Domain then merges them, obtains a new ROI.This step is repeated, until having traversed all ROI.
On the basis of above scheme, in step 30, the calculating such as formula 4 of rarity feature is weighted:
The spatial distribution weight for calculating k-th of pixel, is denoted as ωk, k=msub*nsub, ωkCalculate such as formula 4:
In formula 4, msub,nsubRespectively subgraph is long and wide, IkFor the gray value of k-th of pixel of subgraph, I0For subgraph
The pixel value of inconocenter pixel, num be subgraph in IkThe number of equal pixel.
The rare value for calculating k-th of pixel, is denoted as uk, k=msub*nsub, ukCalculate such as formula 5:
In formula 5, msub,nsubRespectively subgraph is long and wide, unkThe number occurred for k-th of pixel in subgraph.
The rare value of weighting for calculating k-th of pixel, is denoted as Uk, k ∈ m*n, UkCalculate such as formula 6:
Uk=ωk*uk (6)
The invention has the following advantages over the prior art:
1, detection method reduces the redundancy of image, shortens detection time, improves the execution efficiency of algorithm.
The present invention goes to from the processing to global image and handles subgraph, image directory classification is introduced, by size
Unified subgraph is categorized under different catalogues, is filtered out using priori knowledge to the subgraph for not including target, significantly
Reduce the data volume that need to be handled, ensure that the efficiency that algorithm executes.
2, object detection results are accurate, and false alarm rate reduces.
The present invention carries out overlap partition to image, High Resolution SAR image is divided into the unified subgraph of size, to each
Background is relatively uniform for subgraph, carries out target detection to image on this basis, and obtained testing result is more accurate.
3, object detection method is versatile.
Present invention introduces image directory classification method corresponding catalogue classification can be divided into sub-image, can artificially refer to
Fixed target category to be treated takes out subgraph therein and carries out target detection.High-resolution suitable for various complex scenes
SAR image.
Name the effect that specific embodiment illustrates the present invention.
Fig. 1 is the flow chart of High Resolution SAR image object detection provided in an embodiment of the present invention;
Fig. 2 is the original SAR image that emulation experiment of the invention uses;
Fig. 3 is the result for carrying out target detection to image using CFAR algorithm;
Fig. 4 is image block schematic diagram used in the present invention;
Fig. 5 is the μ that the present invention usesl,μr,μd,μuThe specific location schematic diagram of included pixel;
Fig. 6 is the present invention to the sorted result figure of sub- image directory, C1Class is that meadow includes target, C2Class be meadow not
Containing target, C4Class is that field is free of target, C6Class is that forest is free of target, C8Class is that lake is free of target;
Fig. 7 is the present invention to class C1Subgraph comprising target carries out the result of target detection.
Referring to Fig.1, a specific embodiment of the invention is as follows:
Step 1, overlap partition being carried out to the SAR image of input, the size of block is x × x, and it is p% that Duplication, which is manually set,
Obtained subgraph is Si, i=1,2 ..., N, N are subgraph sum.In the present embodiment, x=200, p=25.
Step 2, catalog classification is carried out to subgraph:
The weber partial descriptions feature (WLD) that (2a) calculates training sample divides SVM as the feature vector of characterization image
Class device is trained.For the pixel in W × W window, the calculation formula of WLD is as follows:
In the present embodiment, High Resolution SAR image scene classification M=4, respectively meadow, field, forest and lake.Catalogue
Classification is 2*M=8, and respectively meadow includes target (C1), meadow be free of target (C2), field include target (C3), field is free of
Target (C4), forest include target (C5), forest be free of target (C6), lake include target (C7) and lake be free of target (C8)。
(2b) extracts subgraph SiWLD feature input SVM, obtain SiGeneric label.After classification, for
The other subgraph of any sort is not belonging to be given up.After classification, C11 width subgraph of class, C2Class includes 12 width subgraphs, C3,
C5, C7Class includes 0 width subgraph, C4Class includes 6 width subgraphs, C6Class includes 9 width subgraphs, C8Class includes 5 width subgraphs.
Step 3, class C is selected according to priori knowledge1Subgraph under catalogue.The rare value of weighting of sub-image pixels is calculated,
It is denoted as:
Uk=ωk*uk
Wherein, unkFor the number that k-th of pixel in subgraph occurs, IkFor the gray value of k-th of pixel of subgraph, I0For
The pixel value of subgraph center pixel, num be subgraph in IkThe number of equal pixel, k ∈ 2002.Given threshold is
Thu=180, image is split as the following formula, the image after segmentation is denoted as BW:
The processing such as morphology opening and closing operation, corrosion expansion are carried out to BW, are accurately characterized the ROI template of image.
Step 4, ROI is modified, obtains final object detection results:
(4a) assumes after step 3, and ROI obtained number is L, remembers that t-th of ROI is ROIt;
(4b) judges ROIs, s ∈ L and ROIt, whether t ∈ L is connected region, wherein s ≠ t.If ROIsWith ROItFor even
Logical region, then merge into a new connected region for them.This step is repeated, until having traversed all ROI.In this experiment
Final ROI is indicated with blue rectangle frame.
The above description is merely a specific embodiment, but scope of protection of the present invention is not limited thereto, any
Those familiar with the art in the technical scope disclosed by the present invention, can easily think of the change or the replacement, and should all contain
Lid is within protection scope of the present invention.Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.
Claims (6)
1. a kind of High Resolution SAR image object detection method characterized by comprising
Overlap partition is carried out to the original SAR image of input, obtains multiple subgraphs;
The feature for extracting training sample is trained classifier, carries out mesh to multiple subgraphs with the classifier after training
Record classification;
The rarity feature of weighting for obtaining the pixel that catalogue classification is the subgraph comprising target, to the subgraph comprising target
As carrying out Threshold segmentation, Morphological scale-space is carried out to the bianry image after segmentation, acquisition catalogue classification is the subgraph comprising target
The region of interest ROI of picture;
The ROI is modified, object detection results are obtained.
2. the method according to claim 1, wherein the original SAR image of described pair of input carries out overlap partition
Include:
Overlap partition is carried out to original SAR image according to formula 1;
Wherein, m, n are respectively the length and width of original SAR image, and the size of piecemeal is x × x, and p% is Duplication, obtained subgraph
As being denoted as Si, i=1,2 ..., N, N are the number for obtaining multiple subgraphs.
3. the method according to claim 1, wherein the feature for extracting training sample instructs classifier
Practice, catalog classification carried out to multiple subgraphs with the classifier after training, comprising:
Determine catalogue classification number, wherein be M in the available SAR image of prior information by the target type number for including, each
Target type corresponds to two kinds of catalogue classifications, i.e., comprising this target and not comprising this target, so, total catalogue classification shares 2M
A, each catalogue classification is denoted as Cq, q ∈ 2M, wherein C1,C2,...,CMFor the catalogue category label comprising target;
It randomly selects training sample for the 2M catalogue classification respectively and extracts feature vector;
The feature vector for extracting the subgraph carries out catalog classification to the subgraph.
4. according to the method described in claim 3, it is characterized in that, described randomly select for the 2M catalogue classification respectively
Training sample simultaneously extracts feature vector, comprising:
Weber partial descriptions feature WLD is extracted as feature vector to each training sample, is denoted asAnd classifier is instructed
Practice;
Correspondingly, the feature vector for extracting the subgraph, carries out catalog classification to the subgraph, comprising:
The WLD feature for extracting the subgraph carries out catalog classification to the subgraph using trained classifier.
5. the method according to claim 1, wherein the subgraph to described comprising target carries out threshold value point
It cuts, comprising:
Threshold segmentation, given threshold Th are carried out to the subgraph comprising targetu, image is split as the following formula, is divided
Image afterwards is denoted as BW, such as formula 2:
In formula 2, UkFor the rare value of weighting of k-th of pixel.
6. the method according to claim 1, wherein described be modified to the ROI includes:
ROI number of acquisition is denoted as L, and t-th of ROI is ROIt;
Judge ROIs, s ∈ L and ROIt, whether t ∈ L is connected region, wherein s ≠ t;If ROIsWith ROItFor connected region, then
They are merged, a new ROI is obtained;This step is repeated, until having traversed all ROI.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410743668.5A CN104408482B (en) | 2014-12-08 | 2014-12-08 | A kind of High Resolution SAR Images object detection method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410743668.5A CN104408482B (en) | 2014-12-08 | 2014-12-08 | A kind of High Resolution SAR Images object detection method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN104408482A CN104408482A (en) | 2015-03-11 |
CN104408482B true CN104408482B (en) | 2019-02-12 |
Family
ID=52646113
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201410743668.5A Active CN104408482B (en) | 2014-12-08 | 2014-12-08 | A kind of High Resolution SAR Images object detection method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN104408482B (en) |
Families Citing this family (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104766334B (en) * | 2015-04-21 | 2017-12-29 | 西安电子科技大学 | Small IR targets detection tracking and its device |
CN106295478A (en) * | 2015-06-04 | 2017-01-04 | 深圳市中兴微电子技术有限公司 | A kind of image characteristic extracting method and device |
CN105976376B (en) * | 2016-05-09 | 2020-02-18 | 电子科技大学 | High-resolution SAR image target detection method based on component model |
WO2018141429A1 (en) * | 2017-02-03 | 2018-08-09 | Siemens Aktiengesellschaft | A method and apparatus for detecting objects of interest in images |
CN106886771B (en) * | 2017-03-15 | 2020-08-18 | 同济大学 | Image main information extraction method and face recognition method based on modular PCA |
CN107358262B (en) * | 2017-07-13 | 2020-01-14 | 京东方科技集团股份有限公司 | High-resolution image classification method and classification device |
CN107561533B (en) * | 2017-07-14 | 2019-11-22 | 清华大学 | A kind of C-band satellite-borne synthetic aperture radar motive target imaging method |
CN108509898B (en) * | 2018-03-29 | 2021-06-08 | 中国电子科技集团公司第五十四研究所 | Near-real-time remote sensing image online target detection method based on image stream |
CN109191424B (en) * | 2018-07-23 | 2022-04-22 | 哈尔滨工业大学(深圳) | Breast mass detection and classification system and computer-readable storage medium |
CN109859209B (en) * | 2019-01-08 | 2023-10-17 | 平安科技(深圳)有限公司 | Remote sensing image segmentation method and device, storage medium and server |
CN110135267B (en) * | 2019-04-17 | 2020-09-25 | 电子科技大学 | Large-scene SAR image fine target detection method |
CN110298865B (en) * | 2019-05-22 | 2023-07-07 | 深空探测科技(北京)有限责任公司 | Space-based starry sky background weak small point target tracking method based on threshold separation clustering device |
EP3800581A1 (en) * | 2019-10-03 | 2021-04-07 | Axis AB | A method and apparatus for generating an object classification for an object |
CN112991349B (en) * | 2019-12-17 | 2023-12-26 | 阿里巴巴集团控股有限公司 | Image processing method, device, equipment and storage medium |
CN111191730B (en) * | 2020-01-02 | 2023-05-12 | 中国航空工业集团公司西安航空计算技术研究所 | Method and system for detecting oversized image target oriented to embedded deep learning |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101650439A (en) * | 2009-08-28 | 2010-02-17 | 西安电子科技大学 | Method for detecting change of remote sensing image based on difference edge and joint probability consistency |
CN102426699A (en) * | 2011-11-04 | 2012-04-25 | 西安电子科技大学 | Level set synthetic aperture radar (SAR) image segmentation method based on edge and regional information |
CN103020978A (en) * | 2012-12-14 | 2013-04-03 | 西安电子科技大学 | SAR (synthetic aperture radar) image change detection method combining multi-threshold segmentation with fuzzy clustering |
US8422738B1 (en) * | 2008-08-25 | 2013-04-16 | The United States Of America As Represented By The Secretary Of The Navy | Adaptive automated synthetic aperture radar vessel detection method with false alarm mitigation |
CN103630885A (en) * | 2013-11-07 | 2014-03-12 | 北京环境特性研究所 | Target identification method and system for SAR (synthetic aperture radar) |
CN104020449A (en) * | 2014-05-15 | 2014-09-03 | 中国科学院电子学研究所 | Synthetic aperture radar interference phase diagram filtering method and equipment |
CN104036239A (en) * | 2014-05-29 | 2014-09-10 | 西安电子科技大学 | Fast high-resolution SAR (synthetic aperture radar) image ship detection method based on feature fusion and clustering |
-
2014
- 2014-12-08 CN CN201410743668.5A patent/CN104408482B/en active Active
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8422738B1 (en) * | 2008-08-25 | 2013-04-16 | The United States Of America As Represented By The Secretary Of The Navy | Adaptive automated synthetic aperture radar vessel detection method with false alarm mitigation |
CN101650439A (en) * | 2009-08-28 | 2010-02-17 | 西安电子科技大学 | Method for detecting change of remote sensing image based on difference edge and joint probability consistency |
CN102426699A (en) * | 2011-11-04 | 2012-04-25 | 西安电子科技大学 | Level set synthetic aperture radar (SAR) image segmentation method based on edge and regional information |
CN103020978A (en) * | 2012-12-14 | 2013-04-03 | 西安电子科技大学 | SAR (synthetic aperture radar) image change detection method combining multi-threshold segmentation with fuzzy clustering |
CN103630885A (en) * | 2013-11-07 | 2014-03-12 | 北京环境特性研究所 | Target identification method and system for SAR (synthetic aperture radar) |
CN103630885B (en) * | 2013-11-07 | 2016-06-15 | 北京环境特性研究所 | The target identification method of synthetic aperture radar and system |
CN104020449A (en) * | 2014-05-15 | 2014-09-03 | 中国科学院电子学研究所 | Synthetic aperture radar interference phase diagram filtering method and equipment |
CN104036239A (en) * | 2014-05-29 | 2014-09-10 | 西安电子科技大学 | Fast high-resolution SAR (synthetic aperture radar) image ship detection method based on feature fusion and clustering |
Non-Patent Citations (4)
Title |
---|
SAR图像港口目标提取方法研究;陈琪;《中国博士学位论文全文数据库 信息科技辑》;20120715;第I136-116页 * |
SAR图像配准及变化检测技术研究;熊博莅;《中国博士学位论文全文数据库 信息科技辑》;20140415;第I136-61页 * |
Sea surface ship detection in SAR images;Liu Y J 等;《2004 IEEE International Geoscience and Remote Sensing Symposium》;20041231;第4723-4725页 * |
基于支持向量机的合成孔径雷达图像目标识别;吕金锐;《中国优秀硕士学位论文全文数据库 信息科技辑》;20110515;第I136-374页 * |
Also Published As
Publication number | Publication date |
---|---|
CN104408482A (en) | 2015-03-11 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN104408482B (en) | A kind of High Resolution SAR Images object detection method | |
Huang et al. | A new building extraction postprocessing framework for high-spatial-resolution remote-sensing imagery | |
Geetha et al. | Machine vision based fire detection techniques: A survey | |
Liu et al. | Multiscale morphological compressed change vector analysis for unsupervised multiple change detection | |
Li et al. | 3D parallel fully convolutional networks for real-time video wildfire smoke detection | |
Guo et al. | Weighted-RXD and linear filter-based RXD: Improving background statistics estimation for anomaly detection in hyperspectral imagery | |
Khazai et al. | An approach for subpixel anomaly detection in hyperspectral images | |
Dudhane et al. | C^ 2msnet: A novel approach for single image haze removal | |
Wan et al. | In-frame and inter-frame information based infrared moving small target detection under complex cloud backgrounds | |
Asokan et al. | Machine learning based image processing techniques for satellite image analysis-a survey | |
CN106780485A (en) | SAR image change detection based on super-pixel segmentation and feature learning | |
Qu et al. | A pedestrian detection method based on yolov3 model and image enhanced by retinex | |
Wang et al. | A sparse representation-based method for infrared dim target detection under sea–sky background | |
Qi et al. | Small infrared target detection utilizing local region similarity difference map | |
CN105512622A (en) | Visible remote-sensing image sea-land segmentation method based on image segmentation and supervised learning | |
Zhao et al. | An adaptation of CNN for small target detection in the infrared | |
Su et al. | A new local-main-gradient-orientation HOG and contour differences based algorithm for object classification | |
Shu et al. | Small moving vehicle detection via local enhancement fusion for satellite video | |
Nita et al. | CNN-based object detection and segmentation for maritime domain awareness | |
Xu et al. | COCO-Net: A dual-supervised network with unified ROI-loss for low-resolution ship detection from optical satellite image sequences | |
Wang et al. | Saliency detection using mutual consistency-guided spatial cues combination | |
Wang et al. | A background self-learning framework for unstructured target detectors | |
Zhang et al. | Region-of-interest extraction based on spectrum saliency analysis and coherence-enhancing diffusion model in remote sensing images | |
Mantini et al. | Camera Tampering Detection using Generative Reference Model and Deep Learned Features. | |
Moughal et al. | An automatic unsupervised method based on context-sensitive spectral angle mapper for change detection of remote sensing 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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |