CN106650629A - Kernel sparse representation-based fast remote sensing target detection and recognition method - Google Patents
Kernel sparse representation-based fast remote sensing target detection and recognition method Download PDFInfo
- Publication number
- CN106650629A CN106650629A CN201611039128.4A CN201611039128A CN106650629A CN 106650629 A CN106650629 A CN 106650629A CN 201611039128 A CN201611039128 A CN 201611039128A CN 106650629 A CN106650629 A CN 106650629A
- Authority
- CN
- China
- Prior art keywords
- remote sensing
- sensing target
- sparse
- nuclear
- 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.)
- Pending
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
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/513—Sparse representations
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Astronomy & Astrophysics (AREA)
- Remote Sensing (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kernel sparse representation-based fast remote sensing target detection and recognition method. The method includes the following steps that: S1, four RGB characteristic channels are created; S2, the four-phase Fourier transformation of the four characteristic channels of a given image is calculated, a phase spectrum is extracted, the images of the four characteristic channels are reestablished through inverse Fourier transformation, and a saliency map can be generated; S3, binaryzation division is performed on the saliency map obtained in the step S2, and candidate regions of interest are extracted; S4, a search box is scanned through an effective sub-window search algorithm, so that image blocks to be detected are obtained, so that a remote sensing target image block training set is obtained; S5, SIFT features are extracted from the remote sensing target image block training set, and a sparse dictionary is generated; S6, a spatial pyramid is adopted to map the SIFT features; S7, kernel sparse representation is obtained; S8, the kernel sparse representation is solved; S9, the space pyramid vector representation of a target is performed; and S10, a linear support vector machine classification algorithm is used in combination to complete a recognition task.
Description
Technical field
The present invention relates to remote sensing images analysis field, more particularly to a kind of rapid remote sensing target based on nuclear sparse expression
Detection recognition method.
Background technology
Due to the complexity of remote sensing image, detect that multiclass interesting target thing is one in the remote sensing image of view picture large format
Individual more difficult subject study direction.In computer vision field, BOVW models have been extensively studied and have been applied to image point
In class and pattern recognition task.In recent years, BOVW models are also introduced in remote sensing target detection identification field, and are achieved
Good performance.But the method for being proposed at present is both without the remote sensing image before examinations recognizer, to large format
Pre-processed the prediction i.e. to targets of interest region to process, also fail to fully capture institute's extraction feature in the feature coding stage
Between similarity.This means that respectively a large amount of insignificant redundant computations lose the negative effect to detecting recognition performance with information.
More importantly it is that these methods or model are both for what a certain specific object was proposed, and this is just to a great extent
Limit their range of application.At the same time, existing document points out that the method for notable figure can be used to predict that potential candidate is emerging
The regional location of interesting object, this is because targets of interest thing generally has significantly different on vision mechanism from background.Together
When, input data, by kernel function, stealthyly is mapped to high dimensional feature empty by the method that existing document proposes nuclear sparse expression
Between, while ensureing to become more sparse in the sparse of input data of feature space, this sparse property is sparse for correcting
The defect of expression, the distinctive and lifting classification performance that improve between data have effect well.
However, above-mentioned method carries out pre- place both without before examinations recognizer to the remote sensing image of large format
Reason is processed the prediction in targets of interest region, also fails to fully to be captured in the feature coding stage similar between institute's extraction feature
Degree.This means that respectively a large amount of insignificant redundant computations lose the negative effect to detecting recognition performance with information.More increase
Want, the method or model of the above are both for what a certain specific object was proposed, and this is just limited to a great extent
Their range of application.
The content of the invention
Goal of the invention：It is an object of the invention to provide a kind of can solve the problem that the dilute based on core of defect present in prior art
Dredge the rapid remote sensing target detection recognition methods for representing.
Technical scheme：Rapid remote sensing target detection recognition methods based on nuclear sparse expression of the present invention includes following
Step：
S1：Create four RGB feature passages；
S2：Four phase Fourier transformations of four feature passages of given image are calculated, phase spectrum is extracted, by inverse Fourier
The image of four feature passages of transform reconstruction, so as to generate notable figure；
S3：The notable figure obtained to step S2 carries out binaryzation division, extracts candidate's interest region；
S4：Testing image block is obtained by effective sub- window searching algorithm scanning search frame, remote sensing target image block instruction is formed
Practice collection；
S5：SIFT feature extraction is carried out to remote sensing target image block training set, sparse dictionary is generated；
S6：SIFT feature is mapped using spatial pyramid；
S7：Nuclear sparse expression；
S8：Solve nuclear sparse expression；
S9：The spatial pyramid vector representation of target；
S10：Identification mission is completed with reference to linear support vector cassification algorithm.
Further, in step S7 shown in nuclear sparse expression such as formula (1)：
Wherein v=[v1,v2,v3,…,vN]∈RN×KFor sparse coding coefficient, ξ is reconstructed residual,K (■, ■) be mercer kernel functions, x=[x1,x2,x3,…,xN]∈RN×DFor
The SIFT feature of training image collection,For the sparse dictionary after mapping,For the feature after mapping
Collection, U=[u1,u2,u3,…,uK]∈RK×DFor visual dictionary.
Further, in step S8, using based on the orthogonal matching pursuit algorithm of principal component analysis nuclear sparse expression is solved：
Setting 0<ξ ＜ ＜ 1,0<ω ＜ ＜ K, 0<H≤K, K are degree of rarefication, and h is that principal component analyses the dimension after dimensionality reduction, calculates sparse dictionary
Nuclear matrix Q ∈ RK×KCorresponding regularization characteristic vector, for each regularization characteristic vector, calculates its core vector Y ∈ R1 ×K, then dimensionality reduction is carried out to Q and Y by matrix B, as a result respectively A=QBT=(a1,a2,a3,…,aK)∈RK×hAnd y=YBT∈
R1×h, then using the new matrix A for producing as dictionary, vectorial y generates sparse coefficient v ∈ as input feature vector with OMP algorithms
R1×KRepresent as image vector.
Further, in step S3, the method for extracting candidate's interest region is：Binaryzation notable figure, will be wherein positive
Pixel number exceedes the region of the threshold value of setting as candidate's interest region.
Further, the threshold value is set according to formula (2)：
Threshold=Mean+n × Variance, (2)
Wherein Mean is the average of image intensity value, and Variance is the variance of image intensity value, and n is artificial setting value.
Further, in step S5, SIFT feature is calculated using K-means++ clustering algorithms, generates sparse dictionary.
Further, in step S9, the spatial pyramid vector representation of target is dilute to core using maximum assemblage method
Thin coding carries out final convergence, shown in maximum assemblage method therein such as formula (3)：
Max:R=max (| v1|,|v1|,|v2|,|v3|,…,|vM|) (3) wherein viIt is V=that step S8 is obtained
(v1,v2,v3,…,vM)∈RM×(21K)In element, 1≤i≤M, M for SIFT feature number, K is degree of rarefication, r ∈ R1×(21K)
For the final coding vector of the image block.
Beneficial effect：Compared with prior art, the present invention has following beneficial effect：
(1) present invention has been extracted with predictive candidate's interest region, is accelerated the speed of target detection identification, is carried
High target identification precision；
(2) view-based access control model dictionary of the present invention reduces redundant computation, accelerates with nuclear sparse expression image feature representation model
Iamge description speed；
(3) present invention accelerates nuclear sparse expression.
Description of the drawings
Fig. 1 is the method schematic diagram of the specific embodiment of the invention；
The schematic diagram that Fig. 2 maps for the spatial pyramid of the specific embodiment of the invention.
Specific embodiment
This specific embodiment discloses a kind of rapid remote sensing target detection recognition methods based on nuclear sparse expression, such as schemes
Shown in 1, comprise the following steps：
S1：Four RGB features passage RG (t), BY (t), I (t) and M (t) are created, as shown in formula (1)-(4)；
RG (t)=R (t)-G (t) (1)
BY (t)=R (t)-G (t) (2)
M (t)=| I (t)-I (t- τ) | (4)
Wherein, the RGB triple channels of r (t), g (t), b (t) difference representative image, R (t), G (t) and B (t) such as formulas (5)-
(7) shown in, I (t) represents the image of t；
S2：Four phase Fourier transformations of four feature passages of given image are calculated, phase spectrum is extracted, by inverse Fourier
The image of four feature passages of transform reconstruction, so as to generate notable figure.
S3：The notable figure obtained to step S2 carries out binaryzation division, extracts candidate's interest region；Extract candidate's region of interest
The method in domain is：Binaryzation notable figure, wherein positive pixel number is exceeded the region of the threshold value of setting as candidate's interest
Region；Threshold value therein is set according to formula (8)：
Threshold=Mean+n × Variance, (8)
Wherein Mean is the average of image intensity value, and Variance is the variance of image intensity value, and n is artificial setting value.
Therefore, when certain region of the inswept view picture remote sensing image of sliding window, if in the corresponding binaryzation region of the image just
Pixel number is more than Threshold, then extract the image block in now scanning window as candidate's interest region, the otherwise area
Domain is considered as background area.
S4：Testing image block is obtained by effective sub- window searching algorithm (ESS algorithms) scanning search frame, remote sensing target is formed
Image block training set；
S5：SIFT feature extraction is carried out to remote sensing target image block training set, is calculated using K-means++ clustering algorithms
SIFT feature, generates sparse dictionary.；
S6：SIFT feature is mapped using spatial pyramid：For the SIFT feature of image is given birth to by clustering algorithm
Into length be the dictionary of K, each image is divided into into 3 tomographic image sub-block such as Fig. 2 according to 1 × 1,2 × 2,4 × 4, and for each layer
Weights are assigned, weight is followed successively by 1/4,1/4,1/2.Dictionary table is carried out to the feature of each sub-block and is connected into 21K length up to after converging
Dictionary represent.
S7：Nuclear sparse expression：By the way that simultaneously input feature vector and dictionary base are mapped to into higher-dimension or infinite dimensional feature space
Middle F carries out similarity mode so as to obtain more distinctive sparse coefficient.Shown in nuclear sparse expression such as formula (9)：
Wherein v=[v1,v2,v3,…,vN]∈RN×KFor sparse coding coefficient, ξ is reconstructed residual,K (■, ■) be mercer kernel functions, x=[x1,x2,x3,…,xN]∈RN×DFor
The SIFT feature of training image collection,For the sparse dictionary after mapping,For the feature after mapping
Collection, U=[u1,u2,u3,…,uK]∈RK×DFor visual dictionary.
S8：Solve nuclear sparse expression：Core is solved using based on the orthogonal matching pursuit algorithm (KPOMP algorithms) of principal component analysis
Rarefaction representation：Setting 0<ξ《1,0<ω《K, 0<H≤K, K are degree of rarefication, and h is that principal component analyses the dimension after dimensionality reduction, calculates sparse word
The nuclear matrix Q ∈ R of allusion quotationK×KCorresponding regularization characteristic vector, for each regularization characteristic vector, calculates its core vector Y ∈
R1×K, then dimensionality reduction is carried out to Q and Y by matrix B, as a result respectively A=QBT=(a1,a2,a3,…,aK)∈RK×hAnd y=YBT
∈R1×h, then using the new matrix A for producing as dictionary, vectorial y generates sparse coefficient v as input feature vector with OMP algorithms
∈R1×KRepresent as image vector.
S9：The spatial pyramid vector representation of target：The spatial pyramid vector representation of target is using maximum convergence side
Method carries out final convergence to core sparse coding, shown in maximum assemblage method therein such as formula (10)：
Max:R=max (| v1|,|v1|,|v2|,|v3|,…,|vM|) (10)
Wherein viIt is V=(v that step S8 is obtained1,v2,v3,…,vM)∈RM×(21K)In element, 1≤i≤M, M is SIFT
The number of feature, K is degree of rarefication, r ∈ R1×(21K)For the final coding vector of the image block.
S10：Identification mission is completed with reference to linear support vector cassification algorithm.
Claims (7)
1. a kind of rapid remote sensing target detection recognition methods based on nuclear sparse expression, it is characterised in that：Comprise the following steps：
S1：Create four RGB feature passages；
S2：Four phase Fourier transformations of four feature passages of given image are calculated, phase spectrum is extracted, by inverse Fourier transform
The image of four feature passages is rebuild, so as to generate notable figure；
S3：The notable figure obtained to step S2 carries out binaryzation division, extracts candidate's interest region；
S4：Testing image block is obtained by effective sub- window searching algorithm scanning search frame, remote sensing target image block training set is formed；
S5：SIFT feature extraction is carried out to remote sensing target image block training set, sparse dictionary is generated；
S6：SIFT feature is mapped using spatial pyramid；
S7：Nuclear sparse expression；
S8：Solve nuclear sparse expression；
S9：The spatial pyramid vector representation of target；
S10：Identification mission is completed with reference to linear support vector cassification algorithm.
2. the rapid remote sensing target detection recognition methods based on nuclear sparse expression according to claim 1, it is characterised in that：
Shown in nuclear sparse expression in step S7 such as formula (1)：
WhereinFor sparse coding coefficient, ξ is reconstructed residual,
K (■, ■) be mercer kernel functions, x=[x1,x2,x3,…,xN]∈RN×DFor the SIFT feature of training image collection,For the sparse dictionary after mapping,For the feature set after mapping, U=[u1,u2,u3,…,uK]∈RK×DFor visual dictionary.
3. the rapid remote sensing target detection recognition methods based on nuclear sparse expression according to claim 2, it is characterised in that：
In step S8, using based on the orthogonal matching pursuit algorithm of principal component analysis nuclear sparse expression is solved：Setting 0<ξ ＜ ＜ 1,0<
ω ＜ ＜ K, 0<H≤K, K are degree of rarefication, and h is that principal component analyses the dimension after dimensionality reduction, calculates the nuclear matrix Q ∈ R of sparse dictionaryK×KIt is right
The regularization characteristic vector answered, for each regularization characteristic vector, calculates its core vector Y ∈ R1×K, then by matrix B to Q
Dimensionality reduction is carried out with Y, as a result respectively A=QBT=(a1,a2,a3,…,aK)∈RK×hAnd y=YBT∈R1×h, then newly producing
Matrix A as dictionary, vectorial y generates sparse coefficient v ∈ R as input feature vector with OMP algorithms1×KAs image vector table
Show.
4. the rapid remote sensing target detection recognition methods based on nuclear sparse expression according to claim 1, it is characterised in that：
In step S3, the method for extracting candidate's interest region is：Binaryzation notable figure, wherein positive pixel number is exceeded and sets
The region of fixed threshold value is used as candidate's interest region.
5. the rapid remote sensing target detection recognition methods based on nuclear sparse expression according to claim 4, it is characterised in that：
The threshold value is set according to formula (2)：
Threshold=Mean+n × Variance, (2)
Wherein Mean is the average of image intensity value, and Variance is the variance of image intensity value, and n is artificial setting value.
6. the rapid remote sensing target detection recognition methods based on nuclear sparse expression according to claim 1, it is characterised in that：
In step S5, SIFT feature is calculated using K-means++ clustering algorithms, generate sparse dictionary.
7. the rapid remote sensing target detection recognition methods based on nuclear sparse expression according to claim 1, it is characterised in that：
In step S9, the spatial pyramid vector representation of target is carried out final using maximum assemblage method to core sparse coding
Converge, shown in maximum assemblage method therein such as formula (3)：
Max:R=max (| v1|,|v1|,|v2|,|v3|,…,|vM|) (3)
Wherein viIt is V=(v that step S8 is obtained1,v2,v3,…,vM)∈RM×(21K)In element, 1≤i≤M, M is SIFT feature
Number, K is degree of rarefication, r ∈ R1×(21K)For the final coding vector of the image block.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611039128.4A CN106650629A (en) | 2016-11-21 | 2016-11-21 | Kernel sparse representation-based fast remote sensing target detection and recognition method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611039128.4A CN106650629A (en) | 2016-11-21 | 2016-11-21 | Kernel sparse representation-based fast remote sensing target detection and recognition method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106650629A true CN106650629A (en) | 2017-05-10 |
Family
ID=58812372
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611039128.4A Pending CN106650629A (en) | 2016-11-21 | 2016-11-21 | Kernel sparse representation-based fast remote sensing target detection and recognition method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106650629A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108052957A (en) * | 2017-11-07 | 2018-05-18 | 聊城大学 | A kind of spacecraft target method for quickly identifying |
| CN108681999A (en) * | 2018-05-22 | 2018-10-19 | 浙江理工大学 | SAR image target shape generation method based on depth convolutional neural networks model |
| CN112614056A (en) * | 2020-12-31 | 2021-04-06 | 中国科学院生物物理研究所 | Image super-resolution processing method |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103020647A (en) * | 2013-01-08 | 2013-04-03 | 西安电子科技大学 | Image classification method based on hierarchical SIFT (scale-invariant feature transform) features and sparse coding |
| CN103177458A (en) * | 2013-04-17 | 2013-06-26 | 北京师范大学 | Frequency-domain-analysis-based method for detecting region-of-interest of visible light remote sensing image |
| CN104392463A (en) * | 2014-12-16 | 2015-03-04 | 西安电子科技大学 | Image salient region detection method based on joint sparse multi-scale fusion |
-
2016
- 2016-11-21 CN CN201611039128.4A patent/CN106650629A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103020647A (en) * | 2013-01-08 | 2013-04-03 | 西安电子科技大学 | Image classification method based on hierarchical SIFT (scale-invariant feature transform) features and sparse coding |
| CN103177458A (en) * | 2013-04-17 | 2013-06-26 | 北京师范大学 | Frequency-domain-analysis-based method for detecting region-of-interest of visible light remote sensing image |
| CN104392463A (en) * | 2014-12-16 | 2015-03-04 | 西安电子科技大学 | Image salient region detection method based on joint sparse multi-scale fusion |
Non-Patent Citations (3)
| Title |
|---|
| 邹劲松，黄凯锋: "遥感图像分类中的核稀疏字典学习", 《计算机工程与设计》 * |
| 陆迎曙，贾林虎: "基于核稀疏表示和AdaBoost算法的自然场景识别", 《电子设计工程》 * |
| 马新星 等: "一种基于ROI的红外舰船目标定位方法", 《红外技术》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108052957A (en) * | 2017-11-07 | 2018-05-18 | 聊城大学 | A kind of spacecraft target method for quickly identifying |
| CN108052957B (en) * | 2017-11-07 | 2021-09-14 | 聊城大学 | Spacecraft target rapid identification method |
| CN108681999A (en) * | 2018-05-22 | 2018-10-19 | 浙江理工大学 | SAR image target shape generation method based on depth convolutional neural networks model |
| CN108681999B (en) * | 2018-05-22 | 2022-05-31 | 浙江理工大学 | SAR image target shape generation method based on deep convolutional neural network model |
| CN112614056A (en) * | 2020-12-31 | 2021-04-06 | 中国科学院生物物理研究所 | Image super-resolution processing method |
| CN112614056B (en) * | 2020-12-31 | 2023-09-05 | 北京纳析光电科技有限公司 | Image super-resolution processing method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107506761B (en) | Brain image segmentation method and system based on significance learning convolutional neural network | |
| CN107316031A (en) | The image characteristic extracting method recognized again for pedestrian | |
| CN110569738B (en) | Natural scene text detection method, equipment and medium based on densely connected network | |
| CN105574534A (en) | Significant object detection method based on sparse subspace clustering and low-order expression | |
| CN108154133B (en) | Face portrait-photo recognition method based on asymmetric joint learning | |
| CN114266794B (en) | Pathological section image cancer region segmentation system based on full convolution neural network | |
| CN104077742B (en) | Human face sketch synthetic method and system based on Gabor characteristic | |
| Chen et al. | Dr-tanet: Dynamic receptive temporal attention network for street scene change detection | |
| WO2021164280A1 (en) | Three-dimensional edge detection method and apparatus, storage medium and computer device | |
| CN113112416B (en) | Semantic-guided face image restoration method | |
| CN112396036B (en) | Method for re-identifying blocked pedestrians by combining space transformation network and multi-scale feature extraction | |
| CN110852327A (en) | Image processing method, image processing device, electronic equipment and storage medium | |
| CN104966054A (en) | Weak and small object detection method in visible image of unmanned plane | |
| KR101018299B1 (en) | Apparatus and method for detecting a plurality of objects in an image | |
| CN106650629A (en) | Kernel sparse representation-based fast remote sensing target detection and recognition method | |
| CN104637060B (en) | A kind of image partition method based on neighborhood principal component analysis-Laplce | |
| CN105354547A (en) | Pedestrian detection method in combination of texture and color features | |
| CN114373079A (en) | Rapid and accurate ground penetrating radar target detection method | |
| CN108876776B (en) | Classification model generation method, fundus image classification method and device | |
| CN114764870A (en) | Object positioning model processing method, object positioning device and computer equipment | |
| CN107679467B (en) | Pedestrian re-identification algorithm implementation method based on HSV and SDALF | |
| CN109902692A (en) | A kind of image classification method based on regional area depth characteristic coding | |
| CN111079516B (en) | Pedestrian gait segmentation method based on deep neural network | |
| CN111931689B (en) | Method for extracting video satellite data identification features on line | |
| CN114463732A (en) | Scene text detection method and device based on knowledge distillation |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170510 |
|
| RJ01 | Rejection of invention patent application after publication |