CN105957049A - Remote sensing image changing detection method based on sparse expression classification - Google Patents
Remote sensing image changing detection method based on sparse expression classification Download PDFInfo
- Publication number
- CN105957049A CN105957049A CN201610077917.0A CN201610077917A CN105957049A CN 105957049 A CN105957049 A CN 105957049A CN 201610077917 A CN201610077917 A CN 201610077917A CN 105957049 A CN105957049 A CN 105957049A
- Authority
- CN
- China
- Prior art keywords
- value
- error
- change
- remote sensing
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10032—Satellite or aerial image; Remote sensing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20212—Image combination
- G06T2207/20224—Image subtraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30181—Earth observation
Landscapes
- Engineering & Computer Science (AREA)
- Quality & Reliability (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The present invention relates to a remote sensing image changing detection method based on sparse expression classification. The method comprises the following steps: 1) inputting the remote sensing images after changing and prior to changing through correction processing; 2) obtaining the difference image of the images before the change and after the change; 3) selecting few of sample points of the changing area in the difference image; 4) calculating the estimation values of each pixel element obtained by sparse expression by employing the sample points; 5) calculating the difference value of the real value and the estimation value of each sample point; and 6) if the difference value is smaller than a given threshold value, allowing the sample points to belong to the changing area, or else, allowing the sample points to belong to the unchanging area. The remote sensing image changing detection method based on the sparse expression classification provides a rapid detection method for determining whether there is change in the remote sensing image or not, and only needs few of sample points to obtain accurate changing detection area; and moreover, the method provided by the invention only considers the characteristics of changing area and has high anti-interference capability for a complex background.
Description
Technical field
The present invention relates to field of remote sensing image processing, particularly relate to the change detecting method of a kind of Multitemporal Remote Sensing Images.
Background technology
Utilize remote sensing images that the change on earth's surface is identified and range detection, for hazard scope statistics, loss appraisal etc., there is important practical significance.Owing to different atural object presents different spectral signatures in remote sensing images, the region changed can be identified by the difference in Multitemporal Remote Sensing Images and detects.When being changed detection, in most cases, during as fire, big flood, landslide etc. are detected, it is most important that region of variation is quickly identified, for the background parts not changed and be not concerned with.Thus, herein propose one and carry out quick high accuracy knowledge method for distinguishing for region of variation.
Summary of the invention
The purpose of the present invention is to propose to a kind of judge in remote sensing images with or without the method for quick changed.This method, based on sparse representation theory, carries out semi-automatic detection to the region of variation in remote sensing images.In change-detection, it is conceived to the region changed is analyzed and finely identifies, it is only necessary to region of variation being chosen a small amount of sample, and without considering complicated, the diversified background changed residing for atural object, there is strong capacity of resisting disturbance.
For reaching above-mentioned purpose, technical scheme provides a kind of change detecting method based on rarefaction representation classification, said method comprising the steps of:
1) the remote sensing images X before inputting the change of corrected process and after change1And X2.The image of former and later two phases has to pass through geometric correction and radiant correction, it is ensured that time front and back, the geometric error of phase images is less than a pixel, and atural object radiation characteristic is more consistent;
2) the error image X of image before and after acquisition changet, Xt=X1-X2, and difference is done normalized to 0-1 codomain scope;
3) in error image, region of variation is chosen n sample point, obtain its image coordinate value [i, j], and obtain the numerical value X of this coordinate in error imaget[i, j] is as training sample;
4) to each pixel y, X is utilizedtIt is carried out sparse linear represent and obtain estimated valueWith the sparse situation of estimated result and initial value similarity degree and coefficient for constraint, solving the factor alpha of above rarefaction representation, its target equation is:
Wherein, λ is regular coefficient, for balancing similarity and the sparse degree of coefficient of estimated value and original value.With the SPAMS workbox in Matlab, rarefaction representation coefficient is solved;
5) residual values r (y) of each sample point actual value and estimated value is calculated:
6) residual error is carried out curve of error analysis and determine threshold values.To residual values withFor increment, obtain 10000 threshold values t, each threshold values is calculated verification and measurement ratio and false drop rate, and makes curve of error.Verification and measurement ratio (longitudinal axis) be actually detected go out change account for the ratio of total change, false drop rate (transverse axis) is the ratio that detected error change number accounts for unchanged district.Curve of error trend is similar to the logarithmic function curve risen, and when false drop rate is 0, verification and measurement ratio is 0, and when false drop rate is 1, verification and measurement ratio levels off to 1.Increasing detection with false drop rate to take the lead in being significantly increased, time to a certain extent, slope reduces.Taking corresponding to the verification and measurement ratio at curve of error flex point and false drop rate is the value that threshold values, i.e. curve of error are put down by abrupt change for t value, takes this value and can obtain minimum false drop rate in the case of ensureing compared with high detection rate as threshold values.
7) if difference r (y) is less than given threshold values, then this sample point belongs to variation zone, otherwise, belongs to unchanged district.
The invention have the benefit that
1) present invention only needs region of variation is chosen the i.e. available change-detection region accurately of a small amount of sample, it is not necessary to obtaining the background ground object sample outside region of variation, the training to method is relatively simple.
2) spectral characteristic of region of variation is only analyzed and describes by the present invention, and finds classification boundary, it is not necessary to considers the characteristic of background area, has strong capacity of resisting disturbance to complex background.
3) present invention uses rarefaction representation sorting technique to be described region of variation, plays rarefaction representation technology and utilizes a few sample point to can be carried out the advantage that perfect information describes, is accurately identified as a classification in the region changed.
Accompanying drawing explanation
Fig. 1 is the schematic flow sheet of the method for detecting change of remote sensing image based on rarefaction representation classification described in the embodiment of the present invention.
Fig. 2 is error curve diagram.
Detailed description of the invention
Below according to accompanying drawing, the present invention is described in further detail.
As it is shown in figure 1, a kind of method for detecting change of remote sensing image described in the embodiment of the present invention, said method comprising the steps of:
1) the remote sensing images X before inputting the change of corrected process and after change1And X2.The image of former and later two phases has to pass through geometric correction and radiant correction, it is ensured that time front and back, the geometric error of phase images is less than a pixel, and atural object radiation characteristic is more consistent;
2) the error image X of image before and after acquisition change12=X1-X2, and to difference X12Do normalized to 0-1 codomain scope;
3) in error image, region of variation is chosen no less than 50 sample points, obtain its image coordinate value [i, j], and obtain the numerical value X of this coordinate in error imaget[i, j] is as training sample;
4) to each pixel y, utilize the mode of rarefaction representation to calculate and utilize region of variation training sample that it is carried out estimated value y=X that rarefaction representation obtainstα, the target equation solved is:
Wherein, λ is regular coefficient, for balancing the similarity of estimated value and original value, and the sparse degree of coefficient, take λ=1 herein.|| ||2For squared and, | | | |1For the sum that takes absolute value.
With the SPAMS workbox in Matlab, rarefaction representation coefficient is solved.
5) residual values of each sample point actual value and estimated value is calculated:
6) residual error is carried out curve of error analysis and determine threshold values.To residual values withFor increment, obtain 10000 threshold values t, each threshold values is calculated verification and measurement ratio and false drop rate, and makes curve of error.Verification and measurement ratio (longitudinal axis) be actually detected go out change account for the ratio of total change, false drop rate (transverse axis) is the ratio that detected error change number accounts for unchanged district.Curve of error trend is similar to the logarithmic function curve risen, and when false drop rate is 0, verification and measurement ratio is 0, and when false drop rate is 1, verification and measurement ratio levels off to 1.Increasing detection with false drop rate to take the lead in being significantly increased, time to a certain extent, slope reduces (the curve of error case that enforcement this method obtains is as shown in Figure 2).Taking corresponding to the verification and measurement ratio at curve of error flex point and false drop rate is the value that threshold values, i.e. curve of error are put down by abrupt change for t value, takes this value and can obtain minimum false drop rate in the case of ensureing compared with high detection rate as threshold values.
7) if difference r (y) is less than given threshold values, then this sample point belongs to variation zone, otherwise, belongs to unchanged district.
Method for detecting change of remote sensing image of the present invention has the following characteristics that
1, use the rarefaction representation sorting technique of single class, only region of variation need to be taken a small amount of sample and can be changed detection, step 3) in, it is only necessary to region of variation is carried out sample point and chooses.
2, the similarity of each pixel and variation zone is obtained based on rarefaction representation technology.In described step 4) in, each pixel is obtained its estimated value represented with variation zone sample, this value can be used for judging whether this pixel belongs to variation zone.If this pixel belongs to variation zone, then this pixel value can carry out high-precision rarefaction representation by the sample of variation zone;If being not belonging to variation zone, then estimated value and actual value have bigger residual error.
3, by the way of error analysis, the threshold values of change-detection is obtained: in step 6) in, residual error is carried out threshold values detection, each threshold values to be selected is calculated curve of error, and obtains optimum threshold by the trend analysis of curve of error, to realize high-precision change-detection.
As can be seen from the above embodiments, embodiment of the present invention technical characterstic based on rarefaction representation, by being analyzed the characteristic of single classification, it is only necessary to region of variation is chosen a small amount of sample and can be changed the accurately detecting in region, need input few, and good Detection results can be obtained.
The present invention is illustrated by above embodiment, but, the present invention is not limited to particular example as described herein and embodiment.Any those of skill in the art are easy to be further improved and perfect without departing from the case of distributing bright spirit and scope, therefore distributing bright the content by the claims in the present invention and scope is limited, its intention contains the alternative in all spirit and scope of the invention being included in and being limited by appendix claim and equivalent.
Claims (1)
1. a method for detecting change of remote sensing image based on rarefaction representation classification, it is characterised in that: the method
Comprise the following steps:
1) the remote sensing images X before inputting the change of corrected process and after change1And X2;Former and later two time
The image of phase has to pass through geometric correction and radiant correction, it is ensured that time front and back, the geometric error of phase images is less than
One pixel, and atural object radiation characteristic is more consistent;
2) the error image X of image before and after acquisition changet, Xt=X1-X2, and difference is done normalized
To 0-1 codomain scope;
3) in error image, region of variation is chosen n sample point, obtains its image coordinate value [i, j],
And obtain the numerical value X of this coordinate in error imaget[i, j] is as training sample;
4) to each pixel y, X is utilizedtIt is carried out sparse linear represent and obtain estimated valueTo estimate
Meter result is constraint with the sparse situation of initial value similarity degree and coefficient, enters the factor alpha of above rarefaction representation
Row solves, and its target equation is:
Wherein, λ is regular coefficient, for balancing similarity and the sparse degree of coefficient of estimated value and original value;
With the SPAMS workbox in Matlab, rarefaction representation coefficient is solved;
5) residual values r (y) of each sample point actual value and estimated value is calculated:
6) residual error is carried out curve of error analysis and determine threshold values;To residual values withFor increasing
Amount, obtains 10000 threshold values t, each threshold values is calculated verification and measurement ratio and false drop rate, and makes curve of error;
Verification and measurement ratio (longitudinal axis) be actually detected go out change account for the ratio of total change, false drop rate (transverse axis) is detection
The mistake change number gone out accounts for the ratio in unchanged district;Curve of error trend is similar to the logarithmic function curve risen,
When false drop rate is 0, verification and measurement ratio is 0, and when false drop rate is 1, verification and measurement ratio levels off to 1;Verification and measurement ratio is increased with false drop rate
First being significantly increased, time to a certain extent, slope reduces;Take the verification and measurement ratio at curve of error flex point and false drop rate institute
Corresponding is the value that threshold values, i.e. curve of error are put down by abrupt change for t value, takes this value and can ensure relatively as threshold values
Minimum false drop rate is obtained in the case of high detection rate;
7) if difference r (y) is less than given threshold values, then this sample point belongs to variation zone, otherwise, belongs to unchanged
District.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610077917.0A CN105957049B (en) | 2016-02-03 | 2016-02-03 | A kind of method for detecting change of remote sensing image based on rarefaction representation classification |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610077917.0A CN105957049B (en) | 2016-02-03 | 2016-02-03 | A kind of method for detecting change of remote sensing image based on rarefaction representation classification |
Publications (2)
Publication Number | Publication Date |
---|---|
CN105957049A true CN105957049A (en) | 2016-09-21 |
CN105957049B CN105957049B (en) | 2018-10-23 |
Family
ID=56917287
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610077917.0A Active CN105957049B (en) | 2016-02-03 | 2016-02-03 | A kind of method for detecting change of remote sensing image based on rarefaction representation classification |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105957049B (en) |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102915436A (en) * | 2012-10-25 | 2013-02-06 | 北京邮电大学 | Sparse representation face recognition method based on intra-class variation dictionary and training image |
CN103065320A (en) * | 2013-01-09 | 2013-04-24 | 西安电子科技大学 | Synthetic aperture radar (SAR) image change detection method based on constant false alarm threshold value |
CN103116880A (en) * | 2013-01-16 | 2013-05-22 | 杭州电子科技大学 | Image super resolution rebuilding method based on sparse representation and various residual |
CN103455825A (en) * | 2013-09-08 | 2013-12-18 | 西安电子科技大学 | Method for detecting SAR image changes based on neighborhood clustering kernels |
CN103810704A (en) * | 2014-01-23 | 2014-05-21 | 西安电子科技大学 | SAR (synthetic aperture radar) image change detection method based on support vector machine and discriminative random field |
CN104700411A (en) * | 2015-03-15 | 2015-06-10 | 西安电子科技大学 | Sparse reconstruction-based dual-time phase remote-sensing image change detecting method |
-
2016
- 2016-02-03 CN CN201610077917.0A patent/CN105957049B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102915436A (en) * | 2012-10-25 | 2013-02-06 | 北京邮电大学 | Sparse representation face recognition method based on intra-class variation dictionary and training image |
CN103065320A (en) * | 2013-01-09 | 2013-04-24 | 西安电子科技大学 | Synthetic aperture radar (SAR) image change detection method based on constant false alarm threshold value |
CN103116880A (en) * | 2013-01-16 | 2013-05-22 | 杭州电子科技大学 | Image super resolution rebuilding method based on sparse representation and various residual |
CN103455825A (en) * | 2013-09-08 | 2013-12-18 | 西安电子科技大学 | Method for detecting SAR image changes based on neighborhood clustering kernels |
CN103810704A (en) * | 2014-01-23 | 2014-05-21 | 西安电子科技大学 | SAR (synthetic aperture radar) image change detection method based on support vector machine and discriminative random field |
CN104700411A (en) * | 2015-03-15 | 2015-06-10 | 西安电子科技大学 | Sparse reconstruction-based dual-time phase remote-sensing image change detecting method |
Non-Patent Citations (2)
Title |
---|
WEI LI ET AL: "ADAPTIVE SPARSE REPRESENTATION FOR HYPERSPECTRAL IMAGE CLASSIFICATION", 《IGARSS》 * |
刘瑞 等: "基于模糊局部信息C均值的SAR图像变化检测", 《激光杂志》 * |
Also Published As
Publication number | Publication date |
---|---|
CN105957049B (en) | 2018-10-23 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN101980293B (en) | Method for detecting MTF of hyperspectral remote sensing system based on edge image | |
CN106127930B (en) | A kind of method and system detecting banknote security line integrality | |
CN105427448B (en) | A kind of method and device obtaining bank note safety line boundary | |
CN106097361A (en) | A kind of defective area detection method and device | |
US20010055418A1 (en) | Image-correspondence position detection device, distance measuring device and apparatus using the same | |
CN105466927A (en) | Identification method, correction method and alarm method for turbidimetry abnormal reaction curve | |
JP2011248885A5 (en) | ||
CN103093458B (en) | The detection method of key frame and device | |
CN110058222B (en) | Double-layer particle filter tracking-before-detection method based on sensor selection | |
CN104966349A (en) | ATM platform-based method of detecting new bank note and old bank note | |
CN104899892A (en) | Method for quickly extracting star points from star images | |
CN102360503B (en) | SAR (Specific Absorption Rate) image change detection method based on space approach degree and pixel similarity | |
CN103559715B (en) | A kind of method for detecting abnormality of high spectrum image and device | |
CN110852243A (en) | Improved YOLOv 3-based road intersection detection method and device | |
CN105374047A (en) | Improved bilateral filtering and clustered SAR based image change detection method | |
CN103065320A (en) | Synthetic aperture radar (SAR) image change detection method based on constant false alarm threshold value | |
CN109727226A (en) | A kind of position table automatic generation method based on machine learning | |
CN106646634B (en) | Method and device for correcting abnormity of micro-resistivity scanning imaging logging data | |
CN103500453A (en) | SAR(synthetic aperture radar) image significance region detection method based on Gamma distribution and neighborhood information | |
CN106960433B (en) | It is a kind of that sonar image quality assessment method is referred to based on image entropy and the complete of edge | |
CN101751674A (en) | Change detection method of remote sensing image based on Graph-cut and general gauss model (GGM) | |
CN103903002A (en) | Environment brightness detection method and system | |
CN105184829A (en) | Closely spatial object detection and high-precision centroid location method | |
CN104715160B (en) | Soft sensor modeling data exception point detecting method based on KMDB | |
CN105204047B (en) | A kind of detecting and repairing method of the single rough error of observed quantity in satellite navigation system |
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 |