CN110390255A - High-speed rail environmental change monitoring method based on various dimensions feature extraction - Google Patents
High-speed rail environmental change monitoring method based on various dimensions feature extraction Download PDFInfo
- Publication number
- CN110390255A CN110390255A CN201910454218.7A CN201910454218A CN110390255A CN 110390255 A CN110390255 A CN 110390255A CN 201910454218 A CN201910454218 A CN 201910454218A CN 110390255 A CN110390255 A CN 110390255A
- Authority
- CN
- China
- Prior art keywords
- image
- speed rail
- remote sensing
- feature
- layer
- 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
- 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
- G06F18/211—Selection of the most significant subset of features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/24323—Tree-organised classifiers
-
- 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
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/182—Network patterns, e.g. roads or rivers
Abstract
The high-speed rail environmental change monitoring method based on various dimensions feature extraction that the invention discloses a kind of, comprising: S1, high-resolution remote sensing image pretreatment;S2, different phase Remote Sensing Images Matchings;S3, multidimensional variation characteristic extract;S4, feature figure layer are converted and are counted to pictureization;S5 chooses variation sample as data source using the feature figure layer to pictureization and carries out optimal characteristics screening;S6 is changed extracted region and optimization using change detection optimal characteristics interblock space building random forest grader;S7, optimization, the export of classification results, completes high-speed rail environmental change monitoring.The present invention can efficiently accomplish the earth's surface change detection along high-speed rail, compared with extracting method pixel-by-pixel, salt-pepper noise can effectively be inhibited, improve region of variation extraction accuracy, region of variation boundary and true atural object situation are more coincide simultaneously, the efficiency for effectively increasing high-speed rail ground surface environment variation monitoring has saved the artificial and economic cost of monitoring.
Description
Technical field
The present invention relates to high-speed rail environmental monitoring field, more particularly to a kind of high-speed rail environment based on various dimensions feature extraction
Variation monitoring method.
Background technique
With China's high-speed rail build fast development, high-speed rail mileage and coverage area are continuously increased, it is contemplated that the year two thousand twenty I
The total kilometrage of state's high-speed rail will be more than 30,000 kilometers.This high-speed transit means of transportation of high-speed railway while benefiting huge numbers of families,
Its operation security causes the highest attention of society.High-speed rail how effectively to be monitored along thread environment and rapidly extracting its along thread environment
Situation of change becomes the new research direction of numerous domestic scholar.
High-speed railway speed per hour is high, it is harsher to require operating environment.Unstable building body around railway, such as color steel
Watt, crop greenhouse, towering pylon, line bar, the ditch deep-cut, dense accumulation goods yard etc. in exceedingly odious weather (hurricane
Wind, typhoon, heavy rainfall, hail etc.) all the operation of high-speed rail may be caused to seriously threaten under 3, gently then cause bullet train urgent
Parking, it is heavy then cause train serious accident.Therefore the variation of the operating environment along high-speed rail is paid close attention to, is grasped at any time high
The distribution of risk source and position are most important to high-speed rail operation security along iron.
The mode of variation monitoring is mainly carried out by the way of manpower tour along existing high-speed rail, and this method is due to by people
The limitation operating efficiency of the factors such as the visual field and operating environment is low, and precision can not be guaranteed, increasingly with high-speed rail mileage
Increase, is unable to satisfy the demand of job task.Remote sensing technology has the characteristics that large scale, multidate, can be with using remotely-sensed data
The big regional space information of quick obtaining.With the research and development and application of China's high score satellite data platform, the space point of remotely-sensed data
Resolution is significantly improved, and passes through the available earth's surface meter level of high score data platform, sub-meter grade high spatial resolution multi-spectral
Remote sensing image data, therefore carry out ground surface environment monitoring along high-speed rail using multidate high score data and be possibly realized.
The correlative study of remote sensing variation monitoring mainly has image direct comparison method, classification and predicting method and Direct Classification method three
Kind.Direct comparison method is relatively simple common, main method be using two phase single band remote sensing images being registrated directly into
Row difference or ratio obtain earth's surface change information distributed image.Classification and predicting method is the remote sensing using the front and back phase after registration
Image is classified respectively, is then carried out spatial overlay analysis using sorted vector data and is obtained earth's surface change profile.Directly
Classification is connect, the information of former and later two phases is registrated and is superimposed, the image band group for comprehensively utilizing superposition is changed
Feature extraction, and using variation characteristic image be directly changed region, do not change territorial classification.
Direct method is simple but precision is poor, be easy by when equal factor influenced and generate more noise information, simultaneously
Multiple band class informations can not directly be comprehensively utilized.Classification and predicting method, process is relatively complicated, needs the atural object to two phases
Classify, the precision of change detection depends on the precision of classification, it is generally the case that not can avoid classification detail differences and produce
Raw noise region, result are chiefly used in classification variation statistical analysis.Direct Classification method, comprehensive utilization feature extraction and classification
Thought has the advantages such as noise is relatively fewer, extraction accuracy is high, and process is relatively simple compared with classification and predicting method.But either scheme
As direct comparison method, classification and predicting method or Direct Classification method, generallys use and carried out based on single pixel, it is distant for high-resolution
Sense image is easy to produce salt-pepper noise phenomenon.
Summary of the invention
The object of the present invention is to provide it is a kind of quickly, accurately and efficiently ground surface environment variation monitoring method along high-speed rail.
For this purpose, technical scheme is as follows:
A kind of high-speed rail environmental change monitoring method based on various dimensions feature extraction, comprising the following steps:
High-resolution remote sensing image pretreatment: S1 successively carries out spoke to the high-resolution remote sensing image of different phases respectively
Calibration, ortho-rectification, atmospheric correction and image co-registration are penetrated, the pretreatment of different phase remote sensing images is completed;
S2, different phase Remote Sensing Images Matchings: using computer vision algorithms make in step S1 pretreatment remote sensing image two
The high-resolution remote sensing image of a phase carries out pyramid image Feature corner extraction under different scale respectively, to the characteristic angle
Point carries out similarity measurement, obtains matching characteristic point, and enter next layer of more fine pyramid diagram using matching characteristic point
Layer matching, removes the matching characteristic point of mistake, using the matching characteristic o'clock finally retained as the tie point of two phase images,
Complete the Auto-matching of different phase images;
S3, multidimensional variation characteristic extract: to step S2 matching after two phases high-resolution remote sensing image respectively into
Row multidimensional variation characteristic calculates, and obtains corresponding feature figure layer;
S4, feature figure layer are converted and are counted to pictureization: carrying out super-pixel as data source using the characteristic pattern layer data that S3 is obtained
Segmentation obtains earth object unit, and the mean value and standard deviation of multidimensional variation characteristic are calculated using earth object as statistic unit;It is comprehensive
The information using multiple feature figure layers is closed, Image Segmentation parameter is set, completes conversion of the image from pixel unit to object unit;
S5 chooses variation sample as data source using the feature figure layer to pictureization and carries out optimal characteristics screening: to pictureization
Feature figure layer be data source, select the earth's surface of multiple types to change sample, and calculate each sample data to pictureization multidimensional
Euclidean distance in variation characteristic dimension carries out dissociable basis on this basis, obtains the combination of change detection optimal characteristics;
S6, using change detection optimal characteristics interblock space building random forest grader be changed extracted region with it is excellent
Change: using variation sample, multidimensional characteristic information carries out random forest disaggregated model instruction under optimal multidimensional variation characteristic space first
Disaggregated model is got, is classified using trained disaggregated model to whole region, obtains the change profile of whole region
Situation;
Optimization, the export of classification results: S7 optimizes the change profile situation that step S6 is obtained, after optimization
Result data includes that its category attribute is exported with the format of vector, completes high-speed rail environmental change monitoring.
Wherein, the step S1 includes:
1) radiation calibration:
After obtaining remote sensing image, radiance value calculation formula is converted by DN value are as follows:
Radition_value=Gain*Band_DN+offset............... (1)
Wherein: the sampling under each wave band that Radition_value is radiance value, Band_DN is sensor acquisition
Quantized value, Gain are the yield value of this wave band, and offset is the deviator of this wave band;
2) image distortion generated by hypsography ortho-rectification: is eliminated by ortho-rectification;
3) atmosphere moisture, carbon dioxide and tiny colloid and dust particles atmospheric correction: are eliminated by atmospheric correction
It is influenced on brought by clutter reflections, converts Reflectivity for Growing Season for radiance value;
4) remote sensing image fusion: by remote sensing image fusion by the multispectral image of low resolution with it is high-resolution panchromatic
Image carries out information fusion, to improve the spatial resolution of multi-spectrum remote sensing image.
Wherein, the step S2 specifically includes the following steps:
A the Robert gradient of each pixel) is successively calculated:
fx=f (x+1, y+1)-f (x, y) ... ... ... ... ... ... (2)
fx=f (x+1, y)-f (x, y+1) ... ... ... ... ... ... (3)
Robert operator neighborhood is as follows:
F (x, y) | F (x, y+1) |
F (x+1, y) | F (x+1, y+1) |
;
B) gray scale covariance matrix in sliding window is calculated by pixel:
Wherein dxAnd dyThe respectively differential in the direction x and the direction y;
C it) calculates and determines corner feature value W and U:
Wherein trN is the mark of covariance matrix, and given threshold Tw, Tu traverses image and obtains while meeting W > Tw and U >
The pixel point of Tu;
D the size of field window) is set, and using the maximum eigenvalue point in the window of field as feature angle point.Selection is special
Fixed sliding window, the similarity measurement of feature angle point, selects similitude highest between window carries out figure layer using mahalanobis distance
Point to as tie point, matched linking point is entered into next stage pyramid and carries out continuing to match.To final matching tie point
Error is estimated, the biggish tie point of error is removed.It is registrated according to above matching algorithm use research area remote sensing image,
Image data after being registrated.
Multidimensional variation characteristic in the step S3 include: brightness ratio (Bright_BI), minimal noise separation (MNF),
Normalized differential vegetation index ratio (NDVI_BI), normalization water body index ratio (NDWI_BI) and spectral modeling (SAM).
In step S4, the parameter for needing to be arranged include: participate in figure layer and weight, the cutting unit of Image Segmentation scale it is big
Compactness/smoothness of small, geometric shape/spectrum reference specific gravity factor parameter and cutting object.
Preferably, in step S4, brightness ratio, minimal noise separation, normalized differential vegetation index, normalization water body are referred to
Several and spectral modeling is respectively as follows: 1,1,1,1,1 as segmentation figure layer, the weight coefficient of figure layer.
Preferably, in step S4, the image data of 1m spatial resolution is assembled there are the scattered settlement place in part
Image is split unit using the division size of 80-150;
Preferably, in step S4, the reference specific gravity of geometric shape is 0.2~0.5.
Preferably, in step S4, the reference specific gravity of compactness is 0.4-0.6.
Preferably, in step S5, multidimensional variation characteristic is screened and is combined using area sample data, is counted respectively
Calculate the normalization Euclidean distance average value D of sample under each dimensional space:
Wherein i is used for the dimension in control combination space, xiFor i-th of variation characteristic, μiFor sample average under character pair,
By enumerative technique, the normalization Euclidean distance mean value of the variation characteristic combination under 1,2,3,4,5 five dimensional space is calculated separately
D selects the smallest feature combination of distance average to combine as optimal characteristics.
The invention has the following advantages:
The present invention changes computer vision Feature Matching, image superpixel partitioning algorithm, multidimensional Remote Sensing special
Sign is extracted, Machine learning classifiers algorithm is organically combined together, and by variation, does not change regional scope identification and attribute judgement
Process is separated, and the efficiency of high-speed rail ground surface environment variation monitoring is effectively increased, and has saved the artificial and economic cost of monitoring;
The present invention can efficiently accomplish the earth's surface change detection along high-speed rail, compared with extracting method pixel-by-pixel, can effectively press down
Salt-pepper noise processed improves region of variation extraction accuracy, while region of variation boundary and true atural object situation are more coincide.
Detailed description of the invention
Fig. 1 is that image connecting points extract and matching figure under the image pyramid of multiple scales in the present invention;
Contrast effect figure in Fig. 2 present invention before and after Image registration;
Fig. 3 is vegetation spectral curve;
Fig. 4 is earth's surface change detection effect picture along high-speed rail in method of the invention.
Specific embodiment
Monitoring method of the invention is described in detail below in conjunction with specific embodiment.
Embodiment one
By taking 1m spatial resolution high score remotely-sensed data as an example, the high-speed rail environmental change monitoring side based on various dimensions feature extraction
Steps are as follows for method:
S1, high-resolution remote sensing image pretreatment:
The pretreatment for carrying out different phase high-resolution remote sensing images, including four steps: 1) sensor parameters carry out absolutely
Radiation calibration, 2) the high-precision DEM progress ortho-rectification in use research area;3) it atmospheric correction: is passed using MODTRAN 4+ radiation
Defeated model eliminates influence of the atmosphere to image data wave spectrum;4) multispectral data and panchromatic wave-band data fusion.It is specific as follows:
1) radiation calibration:
After obtaining remote sensing image, quantization and resampling, the spoke that will acquire are carried out in order to carry out compression and the transmission of data
It penetrates brightness value and is converted into DN (Digital Number) value and stored and transmitted.Under normal circumstances, DN value does not have specific
Physical significance generally requires DN value being then converted to radiance value in the application of accurate remote sensing image data, this process
Referred to as radiation calibration, its calculation formula is:
Radition_value=Gain*Band_DN+offset............ (1)
Sampling under each wave band that wherein Radition_value is radiance value, Band_DN is sensor acquisition
Quantized value, Gain are the yield value of this wave band, and offset is the deviator of this wave band.Each wave is obtained for specific data
The gain and deviator of section.
2) ortho-rectification:
Since hypsography causes the satellite image collected under relatively fixed track that can generate distortion, topography is caused
Higher region is big compared with the lower regional percentage ruler of topography, or causes the distortion on plane space.It is needed to eliminate this distortion
Ortho-rectification is carried out to remote sensing image data.Under normal conditions, high-definition remote sensing data ortho-rectification needs surface control
The elements of exterior orientation of point (GCPs) or satellite sensor is established rational polynominal coefficient files (RPC), and utilizes digital elevation mould
Type is corrected.The high-definition remote sensing data of most of 1a rank include RPC data, therefore directly high using zone digit
Ortho-rectification can be completed in number of passes evidence.
3) atmospheric correction:
Atmospheric correction is to eliminate atmosphere moisture, carbon dioxide and tiny colloid, dust particles to clutter reflections
Brought influence, the radiance value that step 1) is obtained are converted into the process of Reflectivity for Growing Season or incidence.MODTRAN 4
The atmospheric correction of remote sensing image in Pixel-level 400nm -2500nm wavelength band may be implemented in+radiative transfer model.Wherein earth's surface
Average height according to zone digit terrain model determine;Its sensor height, the shared attribute that Pixel size is sensor;At
As the date, the time, centre coordinate, obtained according to header file;Aerosol parameters, the urban district of building dense are selected according to regional location
Urban aerosol model is selected, selects Rural if region large area is located at field;Atmospheric models are according to the longitude and latitude of image
With phase determine, wherein Atmospheric models include: Sub-Arctic Winter (SAW), Mid-Latitude Winter (MLW),
Sub-Arctic Summer (SAS), Mid-Latitude Sumer (MLS), Tropical (T), specific Atmospheric models determine
Parameter is as shown in table 1:
1 Atmospheric models parameter list of table
4) remote sensing image fusion:
Remote sensing image fusion typically refers to carry out the remote sensing image of low resolution EO-1 hyperion and high-resolution single band
Information fusion, the technology of high-resolution multi-spectrum remote sensing image is generated by resampling, remote sensing image has simultaneously so that treated
There are higher spatial resolution and spectral resolution.Color space transformation, NNDiffuse Pan Sharpening etc. can be used
Image interfusion method completes visual fusion and obtains the multispectral image data of 1m spatial resolution.Pass through the remote sensing to different phases
The Data duplication above process completes the pretreatment of different phase remotely-sensed datas.
S2, different phase Remote Sensing Images Matchings:
Image registration is also known as images match, is in the image for obtaining different sensors, different phases or different perspectives
Culture point of the same name is matched under identical coordinates of targets.Image automatic matching method is divided into image grayscale matching, based on feature
Match.Gray-scale Matching is computationally intensive under normal conditions, calculating speed is slow, influences vulnerable to ambient environmental change.Based on feature
With the features such as small with calculation amount, matching effect is good.Characteristic matching is divided into the matching based on Points And lines.The extraction of angle point is better than
The detection of line segment observes straight line by point this is because what aperture problem determined, only perpendicular to line segment when
It can just observe, and the observation of angle point is not limited then by angle.Angle point is grey scale change in all directions in image domains
Local maximum.There are many algorithms for Corner Detection, using more for Moravec, Harris and Forstner algorithm etc..Its
Middle Forstner algorithm arithmetic speed is very fast, precision is higher, is widely used in images match field, the algorithm is by pixel
The gray scale covariance matrix in Robert gradient and sliding window centered on the pixel is calculated, is chosen on this basis special
Sign point.As shown in Figure 1, the present invention based on Forstner algorithm, carries out image company under the image pyramid of multiple scales
Contact is extracted and is matched, the remote sensing image after being matched, the specific steps are as follows:
Building matching image and the image fusion pyramid for being matched image first, it is sharp respectively in phase data at two
The feature angle point in out to out in pyramid figure layer is extracted with Forstner algorithm, similarity measurement is carried out to characteristic angle point and is obtained
To matching characteristic point, and enter next layer of more fine pyramid figure layer using matching characteristic point and match, removes of mistake
With point.Using the matching characteristic o'clock finally retained as the tie point of two phase images, specific calculating process are as follows:
A the Robert gradient of each pixel) is successively calculated:
fx=f (x+1, y+1)-f (x, y) ... ... ... ... ... ... (1)
fx=f (x+1, y)-f (x, y+1) ... ... ... ... ... ... (2)
Robert operator neighborhood is as follows:
F (x, y) | F (x, y+1) |
F (x+1, y) | F (x+1, y+1) |
;
B) gray scale covariance matrix in sliding window is calculated by pixel:
Wherein dx and dy is respectively the differential in the direction x and the direction y;
C it) calculates and determines corner feature value W and U:
Wherein trN is the mark of covariance matrix, and given threshold Tw, Tu traverses image and obtains while meeting W > Tw and U >
The pixel point of Tu
D the size of field window) is set, and using the maximum eigenvalue point in the window of field as feature angle point.Selection is special
Fixed sliding window, the similarity measurement of feature angle point, selects similitude highest between window carries out figure layer using mahalanobis distance
Point to as tie point, matched linking point is entered into next stage pyramid and carries out continuing to match.To final matching tie point
Error is estimated, the biggish tie point of error is removed.It is registrated according to above matching algorithm use research area remote sensing image,
Image data after being registrated, the contrast effect before and after Image registration are as shown in Figure 2.By matching result it is found that region room
There is the region of obvious angle point to have obtained more accurately matching in room etc., its corner feature of the linear ground objects such as road is not ten clearly demarcated
Aobvious, matching effect is poor compared with building atural object, but generally matches and be greatly improved between its atural object.
S3, multidimensional variation characteristic extract:
Brightness ratio (Bright_BI) is carried out respectively using the multispectral image after fusion matching, minimal noise separates
(MNF), the multidimensional variation such as index index ratio (NDVI_BI), water body index ratio (NDWI_BI), spectral modeling (SAM) is normalized
Feature calculation obtains corresponding feature figure layer.Wherein:
Brightness ratio (Bright_BI) is retouched using the overall brightness value difference of wave band each in multispectral image data is different
State the situation of change of atural object.For specific remotely-sensed data, brightness Brightness's is defined as:
Wherein blue, green, red and nir are respectively the Reflectivity for Growing Season of blue and green light, feux rouges and near infrared band
Value.And brightness ratio (Bright_BI) is the ratio of the brightness of two phases, its calculation formula is:
Wherein Brighti、BrightjThe respectively brightness value of i, j phase time data.
Minimal noise separation (MNF) transformation is substantially the principal component transform being laminated twice, and first step transformation is for dividing
From and readjust correlation between the transformed noise wave band data of noise remove in data so that the variance of noise is minimum.
The purpose of second of transformation is that noise whitening data are for further processing, and the inherent dimension of data is judged by characteristic value.Finally
Data are divided into two large divisions, and first part is the relevant feature figure layer of maximum eigenvalue, and physical significance is in image
Main information, second part are that close characteristic value corresponds to wave band and noise accounts for the wave band figure layer of main body.To two phase remote sensing shadows
As figure layer group carries out minimal noise separation, main information of the figure layer maximum characteristic wave bands for image, i.e., not changed sky
Between main body terrestrial object information, and changing unit then shows the feature of noise due to the inconsistency of its spectral information, different
When alternate spatial variations information in noise figure layer show more apparent feature.Former and later two phases after accuracy registration are distant
Feel image and carry out space overlapping, figure layer after minimal noise separation is converted is carried out to superimposed figure layer group.Pass through change
Image after changing is it is found that main terrestrial object information is such as built, water body, road have obtained clearly on preceding several characteristic images
Embodiment, and the noise information in region is more obvious in each and every one rear several characteristic wave bands.By comparative observation former and later two when phase transformation
Change region, selects the MNF image that can represent variation characteristic as MNF variation characteristic.It is adjacent that 9 are carried out to MNF variation characteristic figure layer
Domain window median filter process, removal image abnormity value single-point noise are influenced on brought by information extraction.
By vegetation object spectrum CURVE STUDY it is found that vegetation has apparent reflection peak near infrared band, and red
Optical band absorbs paddy since chlorophyll has the absorption of red spectral band, as shown in Figure 3.Therefore it can use this spy
Property building normalized differential vegetation index NDVI carry out vegetation enhancing.Under different phases, coverage area, Yu Midu of vegetation etc. have bright
Aobvious difference, these differences have apparent difference on NDVI.Therefore the NDVI ratio of two phases can be used to describe
The difference degree of vegetation under two phases.For specific remotely-sensed data calculates NDVI and NDVI_BI, calculation formula are as follows:
Wherein nir, red are remote sensing image near-infrared and red spectral band Reflectivity for Growing Season value, NDVIi、NDVIjRespectively i, j
The NDVI figure layer of phase.
The calculating of water body normalized differential vegetation index ratio feature figure layer is completed with above-mentioned formula by pixel.
By forefathers to earth's surface water spectral CURVE STUDY it is found that the SPECTRAL REGION of water body, exists obvious in green light band
Reflection peak, near infrared band almost close to hypersorption.Therefore image can be carried out using normalization water body index (NDWI)
Water body enhancing.Pass through the difference of the range of water body under available two phases of the ratio NDWI_BI of the water body index of two phases
It is different.
Wherein nir, green are respectively the near-infrared and green light band data of remotely-sensed data.NDWIi、NDWIjRespectively i, j
The NDWI figure layer of phase.Water body normalization water body index ratio feature figure layer is completed with above-mentioned formula by pixel to calculate.
The corresponding curve of spectrum of each pixel can be considered as a multi-C vector, identical atural object in multispectral image
The similarity degree of the wave spectrum of two phase is bigger, shows on spectral vector angle, and the spectral modeling of similar atural object is small, differently
As the increase spectral modeling of the difference of the curve of spectrum increases between object.The calculation formula of spectral modeling (SAM) are as follows:
In the present embodiment, xiFor the spectral vector of i-th of pixel in first phase in No. two remotely-sensed datas of high score, yi
For the spectral vector of i-th of pixel in second phase in No. two remotely-sensed datas of high score.When by being calculated two by pixel
The spectral modeling feature figure layer of phase.
The calculating of two phases time multidimensional variation characteristic figure layer is completed by the above process.
S4, feature figure layer are converted and are counted to pictureization:
The characteristic pattern layer data obtained using S3 is carried out super-pixel segmentation as data source and obtains earth object unit, and with atural object
Object is the mean value and standard deviation that statistic unit calculates multidimensional variation characteristic;The information of multiple wave bands is comprehensively utilized, image is set
Partitioning parameters complete conversion of the image from pixel unit to object unit.The parameter for needing to be arranged is as follows:
The figure layer and weight for participating in Image Segmentation, in addition to fused visible light and near infrared band, panchromatic wave-band, image
The figure layer and multi-scale division can be participated in the figure layer that other image informations enhance that edge enhances, and according to segmentation object pair
Weight as each figure layer is arranged;
The scale size of cutting unit is opposite concept, without concrete unit, describes it and divides imaged object list
The size of position, value is bigger, and image unit is bigger;
The proportion coefficients parameter (0-1) of geometric shape/spectrum reference is image spectrum and geometric shape in image point
The sum of specific gravity parameter of reference specific gravity during cutting, geometric shape and spectral reference is 1, if geometrical body reference specific gravity
Parameter is set as x, then the reference specific gravity parameter of spectrum is 1-x;
Compactness/smoothness (0-1) of cutting object, this parameter describe the smoothness of cutting object cell edges,
The smoothness of its bigger cutting object of compactness is smaller, and the sum of smoothness and compactness are similarly 1.
The present invention specifically using the earth's surfaces variation characteristic figure layer such as Brihgt_BI, MNF, NDVI_BI, NDWI_BI and SAM as
Divide figure layer, the weight coefficient of figure layer is respectively as follows: 1,1,1,1,1.According to the type of ground objects of influence point for dividing scale
Cloth is selected, and under normal circumstances, there are adopt in the image of the scattered settlement place aggregation in part for the image data of 1m spatial resolution
It can satisfy application demand with the division size of 80-150;According to the empirical value that experiment obtains, the specific gravity of the reference of geometric shape
It is traditionally arranged to be 0.2~0.5, based on the spectral information of image, supplemented by shape information;Compactness selects 0.4-0.6, generally may be used
To meet application demand.
The above parameter is the empirical parameter of general data, and value can be adjusted according to specific application effect.It segments
Cheng Hou calculates separately the mean value and standard deviation of each figure layer object of Brihgt_BI, MNF, NDVI_BI, NDWI_BI and SAM.
S5, variation sample is chosen as data source using the feature figure layer to pictureization and carries out optimal characteristics screening:
Multidimensional variation characteristic describes the situation of change of earth's surface from multiple dimensions, such as brightness ratio from integral radiation brightness
Angle describes the situation of change of earth's surface, more sensitive to atural objects such as artificial structures, and minimal noise is separated from statistics for entire group
Earth's surface variation is described in angle, and NDVI_BI is then that earth's surface situation of change is described from the angle of vegetation variation,
NDWI_BI is that the angle of water body variation describes the situation of change of earth's surface.For different zones, from different dimension descriptions and
Extract region of variation can be improved the extraction accuracy of region of variation to a certain extent, but also bring a part of noise into simultaneously,
Therefore it needs that multidimensional variation characteristic is screened and combined using area sample data, to construct the optimal spy of change detection
Levy space.Its Computing Principle is the normalization Euclidean distance average value D for calculating separately sample under each dimensional space.It calculates public
Formula are as follows:
Wherein i is used for the dimension in control combination space, xiFor i-th of variation characteristic, μiFor sample average under character pair.
By enumerative technique, the normalization Euclidean distance mean value of the variation characteristic combination under 1,2,3,4,5 five dimensional space is calculated separately
D selects the smallest feature combination of distance average to combine as change detection optimal characteristics.Different type is selected under different zones
The sample data calculated optimal characteristics Spatial Coupling of institute there may be difference, therefore optimal characteristics group be combined into variable and very
Amount.
S6, using change detection optimal characteristics interblock space building random forest grader be changed extracted region with it is excellent
Change:
Random forest sorting algorithm (RF) is that a kind of close in large data sets effectively runs algorithm, preferably can handle and transport
It with multidimensional feature space, is operated without dimensionality reduction, there is preferable nicety of grading.In the present invention, random forest sorting algorithm
Utilization be divided into disaggregated model training and simulation classify two processes.First using variation sample under multidimensional variation characteristic space
Carry out the training of random forest disaggregated model.For corresponding region by sample data screening obtain Bright_BI, NDVI_BI and
The features group such as SAM is combined into the optimal characteristics space of one's respective area.Utilize mean value, the standard under each object unit under feature space
Independent variable of the difference as model, the variation of object/do not change classification are trained as dependent variable.Utilize trained classification mould
Type carries out the change profile situation that classification obtains whole region to whole region.
S7, the optimization of classification results, export:
The change detection result that step 6 obtains is optimized, main step is edge-smoothing, the conjunction of adjacent patch
And small patch processing.Since Object Segmentation is on the basis of Raster Images, so easily there is sawtooth boundary in earth object boundary.Benefit
It is carried out smoothly with operations such as object expansion, retractions;The adjacent identical patch of attribute is merged, and cancels its boundary line;
Small patch less than 4 pixels is merged, classification incorporates into as adjacent classification.It include it by the result data after optimization
Category attribute is exported with the format of vector.
Use 2015, two phase high spatial resolution remote sensing datas walk for experimental data according to 1-7 along high-speed rail in 2017
Suddenly it carries out processing and obtains experimental result, as a result as shown in figure 3, passing through interpretation of result it is found that the method is for the coloured silk along high-speed rail
The earth's surfaces variations such as steel room, concrete construction, surface vegetation variation have preferable extraction effect, have for variable noise preferable
Inhibiting effect.
The present embodiment is changed with ground surface environment along high-speed rail as research object, distant with two phases time 1m spatial resolution high definition
Sense image is data source, and Image Matching experiment generates 280 groups of match points, and images match totality root-mean-square error (RMS) is 0.31
A pixel.Precision test is carried out by randomly selecting 200 sample points, the overall accuracy of remote sensing variation monitoring is 91%.
By the analysis to experimental result: the shade of different angle caused by towering building under different phases,
The Yu Midu of state, vegetation under different phases of water body at different temperatures is the main mistake of remote sensing variation monitoring along high-speed rail
Poor source.
This method color steel tile room, vinyl house, building and the slum-dweller makeshift house of interest for variation monitoring along high-speed rail
Equal feature changes have preferable recognition effect, insensitive to the variation such as strong and weak difference of coupling relationship, light, along high-speed rail
Ground surface environment monitoring has preferable applicability.
Claims (10)
1. a kind of high-speed rail environmental change monitoring method based on various dimensions feature extraction, comprising the following steps:
S1, high-resolution remote sensing image pretreatment:
Radiation calibration, ortho-rectification, atmospheric correction and image are successively carried out respectively to the high-resolution remote sensing image of different phases
Fusion, completes the pretreatment of different phase remote sensing images;
S2, different phase Remote Sensing Images Matchings:
Distinguished using high-resolution remote sensing image of the computer vision algorithms make to two phases in step S1 pretreatment remote sensing image
Pyramid image Feature corner extraction under different scale is carried out, similarity measurement is carried out to the characteristic angle point, it is special to obtain matching
Point is levied, and enters next layer of more fine pyramid figure layer using matching characteristic point and matches, removes the matching characteristic point of mistake,
Using the matching characteristic o'clock finally retained as the tie point of two phase images, the Auto-matching of different phase images is completed;
S3, multidimensional variation characteristic extract:
Multidimensional variation characteristic calculating is carried out to the high-resolution remote sensing image of two phases after step S2 matching respectively, obtains phase
The feature figure layer answered;
S4, feature figure layer are converted and are counted to pictureization: carrying out super-pixel segmentation as data source using the characteristic pattern layer data that S3 is obtained
Earth object unit is obtained, and calculates the mean value and standard deviation of multidimensional variation characteristic using earth object as statistic unit;Comprehensive benefit
With the information of multiple feature figure layers, Image Segmentation parameter is set, completes conversion of the image from pixel unit to object unit;
S5 chooses variation sample as data source using the feature figure layer to pictureization and carries out optimal characteristics screening:
Using the feature figure layer to pictureization as data source, the earth's surface of multiple types is selected to change sample, and calculate each sample data
To the Euclidean distance in pictureization multidimensional variation characteristic dimension, dissociable basis is carried out on this basis, obtains change detection most
Excellent feature combination;
S6 is changed extracted region and optimization using change detection optimal characteristics interblock space building random forest grader:
Using variation sample, multidimensional characteristic information carries out random forest disaggregated model under optimal multidimensional variation characteristic space first
Training obtains disaggregated model, is classified using trained disaggregated model to whole region, obtains the variation point of whole region
Cloth situation;
S7, optimization, the export of classification results:
The change profile situation that step S6 is obtained is optimized, includes its category attribute with vector by the result data after optimization
Format exported, complete high-speed rail environmental change monitoring.
2. high-speed rail environmental change monitoring method according to claim 1, which is characterized in that the step S1 includes:
1) radiation calibration:
After obtaining remote sensing image, radiance value calculation formula is converted by DN value are as follows:
Radition_value=Gain*Band_DN+offset ... ... ... (1)
Wherein: the sample quantization under each wave band that Radition_value is radiance value, Band_DN is sensor acquisition
Value, Gain are the yield value of this wave band, and offset is the deviator of this wave band;
2) image distortion generated by hypsography ortho-rectification: is eliminated by ortho-rectification;
3) atmosphere moisture, carbon dioxide and tiny colloid and dust particles atmospheric correction: are eliminated over the ground by atmospheric correction
It is influenced brought by object reflection, converts Reflectivity for Growing Season for radiance value;
4) remote sensing image fusion:
The multispectral image of low resolution information is carried out with high-resolution panchromatic image by remote sensing image fusion to merge, with
Improve the spatial resolution of multi-spectrum remote sensing image.
3. high-speed rail environmental change monitoring method according to claim 1, which is characterized in that the step S2 specifically include with
Lower step:
A the Robert gradient of each pixel) is successively calculated:
fx=f (x+1, y+1)-f (x, y) ... ... ... ... ... ... (1)
fx=f (x+1, y)-f (x, y+1) ... ... ... ... ... ... (2)
Robert operator neighborhood is as follows:
;
B) gray scale covariance matrix in sliding window is calculated by pixel:
Wherein dxAnd dyThe respectively differential in the direction x and the direction y;
C it) calculates and determines corner feature value W and U:
Wherein trN is the mark of covariance matrix, and given threshold Tw, Tu traverses image and obtains while meeting W > Tw and U > Tu's
Pixel point;
D the size of field window) is set, and using the maximum eigenvalue point in the window of field as feature angle point.Selection is specific
Sliding window, the similarity measurement of feature angle point, selects the highest point of similitude between window carries out figure layer using mahalanobis distance
To as tie point, matched linking point is entered into next stage pyramid and carries out continuing to match.To final matching connection point tolerance
It is estimated, removes the biggish tie point of error.It is registrated, is obtained according to above matching algorithm use research area remote sensing image
Image data after registration.
4. high-speed rail environmental change monitoring method according to any one of claim 1-3, which is characterized in that in step S3
Multidimensional variation characteristic includes: brightness ratio (Bright_BI), minimal noise separation (MNF), normalized differential vegetation index ratio
(NDVI_BI), water body index ratio (NDWI_BI) and spectral modeling (SAM) are normalized.
5. high-speed rail environmental change monitoring method according to claim 4, which is characterized in that in step S4, need to be arranged
Parameter includes: to participate in the figure layer and weight, the scale size of cutting unit, geometric shape/spectrum reference specific gravity of Image Segmentation
Compactness/smoothness of factor parameter and cutting object.
6. high-speed rail environmental change monitoring method according to claim 5, which is characterized in that in step S4, by brightness ratio,
Minimal noise separation, normalized differential vegetation index ratio, normalization water body index ratio and spectral modeling are used as segmentation figure layer, figure layer
Weight coefficient is respectively as follows: 1,1,1,1,1.
7. high-speed rail environmental change monitoring method according to claim 6, which is characterized in that in step S4, for the space 1m
The image data of resolution ratio divides unit using the division size of 80-150 there are the image of the scattered settlement place aggregation in part
It cuts.
8. high-speed rail environmental change monitoring method according to claim 7, which is characterized in that in step S4, geometric shape
Reference specific gravity is 0.2~0.5.
9. high-speed rail environmental change monitoring method according to claim 8, which is characterized in that in step S4, the ginseng of compactness
Examining specific gravity is 0.4-0.6.
10. high-speed rail environmental change monitoring method according to claim 9, which is characterized in that utilize region sample in step S5
Notebook data is screened and is combined to multidimensional variation characteristic, and the normalization Euclidean distance of sample under each dimensional space is calculated separately
Average value D:
Wherein i is used for the dimension in control combination space, xiFor i-th of variation characteristic, μiFor sample average under character pair, pass through
Enumerative technique calculates separately the normalization Euclidean distance mean value D of the variation characteristic combination under 1,2,3,4,5 five dimensional space, choosing
The smallest feature combination of distance average is selected to combine as optimal characteristics.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910454218.7A CN110390255A (en) | 2019-05-29 | 2019-05-29 | High-speed rail environmental change monitoring method based on various dimensions feature extraction |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910454218.7A CN110390255A (en) | 2019-05-29 | 2019-05-29 | High-speed rail environmental change monitoring method based on various dimensions feature extraction |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110390255A true CN110390255A (en) | 2019-10-29 |
Family
ID=68284971
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910454218.7A Pending CN110390255A (en) | 2019-05-29 | 2019-05-29 | High-speed rail environmental change monitoring method based on various dimensions feature extraction |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110390255A (en) |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110910457A (en) * | 2019-11-22 | 2020-03-24 | 大连理工大学 | Multispectral three-dimensional camera external parameter calculation method based on angular point characteristics |
CN110956207A (en) * | 2019-11-22 | 2020-04-03 | 苏州中科天启遥感科技有限公司 | Optical remote sensing image full-element change detection method |
CN111091113A (en) * | 2019-12-30 | 2020-05-01 | 贵阳欧比特宇航科技有限公司 | Hyperspectral image data fusion method |
CN111091054A (en) * | 2019-11-13 | 2020-05-01 | 广东国地规划科技股份有限公司 | Method, system and storage medium for monitoring land type change |
CN111257854A (en) * | 2020-01-19 | 2020-06-09 | 中南林业科技大学 | Universal terrain correction optimization method based on remote sensing image segmentation unit |
CN111582642A (en) * | 2020-04-03 | 2020-08-25 | 中国水产科学研究院东海水产研究所 | Fish optimum environment judgment method, electronic device and storage medium |
CN112147078A (en) * | 2020-09-22 | 2020-12-29 | 华中农业大学 | Multi-source remote sensing monitoring method for crop phenotype information |
CN113822220A (en) * | 2021-10-09 | 2021-12-21 | 海南长光卫星信息技术有限公司 | Building detection method and system |
CN114283070A (en) * | 2022-03-07 | 2022-04-05 | 中国铁路设计集团有限公司 | Method for manufacturing terrain section by fusing unmanned aerial vehicle image and laser point cloud |
CN115481368A (en) * | 2022-09-29 | 2022-12-16 | 河北省科学院地理科学研究所 | Vegetation coverage estimation method based on full remote sensing machine learning |
Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1651860A (en) * | 2004-06-08 | 2005-08-10 | 王汶 | Symmetric system sampling technique for estimating area change by different scale remote sensing data |
CN104851087A (en) * | 2015-04-17 | 2015-08-19 | 华中农业大学 | Multi-scale forest dynamic change monitoring method |
CN107392925A (en) * | 2017-08-01 | 2017-11-24 | 西安电子科技大学 | Remote sensing image terrain classification method based on super-pixel coding and convolutional neural networks |
CN108195767A (en) * | 2017-12-25 | 2018-06-22 | 中国水产科学研究院东海水产研究所 | Estuarine wetland denizen monitoring method |
CN109523516A (en) * | 2018-10-19 | 2019-03-26 | 中国科学院遥感与数字地球研究所 | A kind of object level land cover pattern change detecting method based on double constraints condition |
CN109657610A (en) * | 2018-12-18 | 2019-04-19 | 北京航天泰坦科技股份有限公司 | A kind of land use change survey detection method of high-resolution multi-source Remote Sensing Images |
CN109685081A (en) * | 2018-12-27 | 2019-04-26 | 中国土地勘测规划院 | A kind of joint change detecting method of Remotely sensed acquisition black fallow |
CN109684929A (en) * | 2018-11-23 | 2019-04-26 | 中国电建集团成都勘测设计研究院有限公司 | Terrestrial plant ECOLOGICAL ENVIRONMENTAL MONITORING method based on multi-sources RS data fusion |
CN109784251A (en) * | 2019-01-04 | 2019-05-21 | 中国铁路总公司 | Small water remote sensing recognition method along high-speed rail |
-
2019
- 2019-05-29 CN CN201910454218.7A patent/CN110390255A/en active Pending
Patent Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1651860A (en) * | 2004-06-08 | 2005-08-10 | 王汶 | Symmetric system sampling technique for estimating area change by different scale remote sensing data |
CN104851087A (en) * | 2015-04-17 | 2015-08-19 | 华中农业大学 | Multi-scale forest dynamic change monitoring method |
CN107392925A (en) * | 2017-08-01 | 2017-11-24 | 西安电子科技大学 | Remote sensing image terrain classification method based on super-pixel coding and convolutional neural networks |
CN108195767A (en) * | 2017-12-25 | 2018-06-22 | 中国水产科学研究院东海水产研究所 | Estuarine wetland denizen monitoring method |
CN109523516A (en) * | 2018-10-19 | 2019-03-26 | 中国科学院遥感与数字地球研究所 | A kind of object level land cover pattern change detecting method based on double constraints condition |
CN109684929A (en) * | 2018-11-23 | 2019-04-26 | 中国电建集团成都勘测设计研究院有限公司 | Terrestrial plant ECOLOGICAL ENVIRONMENTAL MONITORING method based on multi-sources RS data fusion |
CN109657610A (en) * | 2018-12-18 | 2019-04-19 | 北京航天泰坦科技股份有限公司 | A kind of land use change survey detection method of high-resolution multi-source Remote Sensing Images |
CN109685081A (en) * | 2018-12-27 | 2019-04-26 | 中国土地勘测规划院 | A kind of joint change detecting method of Remotely sensed acquisition black fallow |
CN109784251A (en) * | 2019-01-04 | 2019-05-21 | 中国铁路总公司 | Small water remote sensing recognition method along high-speed rail |
Non-Patent Citations (4)
Title |
---|
张明哲等: "基于超像素分割和多方法融合的SAR图像变化检测方法", 《遥感技术与应用》 * |
温奇等: "高分辨率遥感影像的平原建成区提取", 《光学精密工程》 * |
肖明虹等: "超像素分割和多方法融合的遥感影像变化检测方法", 《测绘通报》 * |
蒋丽丽等: "京沪高铁周边环境安全隐患智能监测体系研究", 《铁路计算机应用》 * |
Cited By (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111091054B (en) * | 2019-11-13 | 2020-11-10 | 广东国地规划科技股份有限公司 | Method, system and device for monitoring land type change and storage medium |
CN111091054A (en) * | 2019-11-13 | 2020-05-01 | 广东国地规划科技股份有限公司 | Method, system and storage medium for monitoring land type change |
CN110910457B (en) * | 2019-11-22 | 2021-04-16 | 大连理工大学 | Multispectral three-dimensional camera external parameter calculation method based on angular point characteristics |
CN110956207A (en) * | 2019-11-22 | 2020-04-03 | 苏州中科天启遥感科技有限公司 | Optical remote sensing image full-element change detection method |
CN110910457A (en) * | 2019-11-22 | 2020-03-24 | 大连理工大学 | Multispectral three-dimensional camera external parameter calculation method based on angular point characteristics |
CN110956207B (en) * | 2019-11-22 | 2023-09-19 | 苏州中科天启遥感科技有限公司 | Method for detecting full-element change of optical remote sensing image |
CN111091113A (en) * | 2019-12-30 | 2020-05-01 | 贵阳欧比特宇航科技有限公司 | Hyperspectral image data fusion method |
CN111257854A (en) * | 2020-01-19 | 2020-06-09 | 中南林业科技大学 | Universal terrain correction optimization method based on remote sensing image segmentation unit |
CN111582642A (en) * | 2020-04-03 | 2020-08-25 | 中国水产科学研究院东海水产研究所 | Fish optimum environment judgment method, electronic device and storage medium |
CN112147078A (en) * | 2020-09-22 | 2020-12-29 | 华中农业大学 | Multi-source remote sensing monitoring method for crop phenotype information |
CN113822220A (en) * | 2021-10-09 | 2021-12-21 | 海南长光卫星信息技术有限公司 | Building detection method and system |
CN114283070A (en) * | 2022-03-07 | 2022-04-05 | 中国铁路设计集团有限公司 | Method for manufacturing terrain section by fusing unmanned aerial vehicle image and laser point cloud |
CN114283070B (en) * | 2022-03-07 | 2022-05-03 | 中国铁路设计集团有限公司 | Method for manufacturing terrain section by fusing unmanned aerial vehicle image and laser point cloud |
CN115481368A (en) * | 2022-09-29 | 2022-12-16 | 河北省科学院地理科学研究所 | Vegetation coverage estimation method based on full remote sensing machine learning |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110390255A (en) | High-speed rail environmental change monitoring method based on various dimensions feature extraction | |
CN108573276B (en) | Change detection method based on high-resolution remote sensing image | |
CN104915636B (en) | Remote sensing image road recognition methods based on multistage frame significant characteristics | |
Pu et al. | A comparative analysis of high spatial resolution IKONOS and WorldView-2 imagery for mapping urban tree species | |
CN107392130B (en) | Multispectral image classification method based on threshold value self-adaption and convolutional neural network | |
Wang et al. | Comparison of IKONOS and QuickBird images for mapping mangrove species on the Caribbean coast of Panama | |
Lu et al. | Land cover classification in a complex urban-rural landscape with QuickBird imagery | |
CN105335966B (en) | Multiscale morphology image division method based on local homogeney index | |
CN106339674B (en) | The Hyperspectral Image Classification method that model is cut with figure is kept based on edge | |
CN103679675B (en) | Remote sensing image fusion method oriented to water quality quantitative remote sensing application | |
CN105930772A (en) | City impervious surface extraction method based on fusion of SAR image and optical remote sensing image | |
CN107610164B (en) | High-resolution four-number image registration method based on multi-feature mixing | |
CN103632363A (en) | Object-level high-resolution remote sensing image change detection method based on multi-scale fusion | |
CN103839267B (en) | Building extracting method based on morphological building indexes | |
CN102842044B (en) | Method for detecting variation of remote-sensing image of high-resolution visible light | |
CN113963222B (en) | High-resolution remote sensing image change detection method based on multi-strategy combination | |
CN104361590A (en) | High-resolution remote sensing image registration method with control points distributed in adaptive manner | |
CN106971397B (en) | Based on the city high-resolution remote sensing image dividing method for improving JSEG algorithms | |
CN110309781A (en) | Damage remote sensing recognition method in house based on the fusion of multi-scale spectrum texture self-adaption | |
CN103077515A (en) | Multi-spectral image building change detection method | |
Zhang et al. | Remote sensing of impervious surfaces in tropical and subtropical areas | |
CN109598202A (en) | A kind of object-based satellite image multi objective built-up areas extraction method | |
CN102013095A (en) | Distribution transform-based multi-sensor image fusion method | |
CN112258523A (en) | Method for finely extracting enteromorpha coverage information of medium-low resolution remote sensing image | |
CN104573662B (en) | A kind of cloud sentences method and system |
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 | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191029 |