CN109801253A - A kind of adaptive cloud sector detection method of pair of high-resolution optical remote sensing image - Google Patents
A kind of adaptive cloud sector detection method of pair of high-resolution optical remote sensing image Download PDFInfo
- Publication number
- CN109801253A CN109801253A CN201711116482.7A CN201711116482A CN109801253A CN 109801253 A CN109801253 A CN 109801253A CN 201711116482 A CN201711116482 A CN 201711116482A CN 109801253 A CN109801253 A CN 109801253A
- Authority
- CN
- China
- Prior art keywords
- remote sensing
- cloud
- resolution optical
- optical remote
- 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
Abstract
The invention discloses the adaptive cloud sector detection methods of a kind of pair of high-resolution optical remote sensing image, wherein, high-resolution optical remote sensing image is high score No.1 remote sensing images, comprising the following steps: S1: to high-resolution optical remote sensing image F, calculates its cloud and mist thickness chart F1；S2: according to the relationship between the cutoff frequency f for the high-pass filter applied during the radial energy of F1 spectrum and homomorphic filtering, the value of f is calculated；S3: the cutoff frequency of high-pass filter during using f as homomorphic filtering carries out homomorphic filtering to F1, obtains filtered image F2；S4: it calculates the whitness index of each pixel in F2 and therefrom filters out the pixel that whitness index is greater than a preset threshold, obtain image F3；S5: circular configuration element is selected, closed operation is first carried out to F3 and carries out opening operation again, to carry out morphology optimization to it, obtains cloud sector recognition result F4.The present invention can carry out the efficient cloud sector of mass to high-resolution optical remote sensing image and detect, and overall recognition accuracy is up to 93.81%.
Description
Technical field
The present invention relates to technical field of remote sensing image processing, in particular to a kind of pair of high-resolution optical remote sensing figure
The adaptive cloud sector detection method of picture.
Background technique
Optical remote satellite sensor pass through frequently with visible near-infrared wave band (0.38-0.90 μm) cloud layer can not be penetrated,
It is difficult to obtain the terrestrial object information in cloud cover region, international cloud climatology plan (International in practical application
Satellite Cloud Climatology Project, ISCCP) according to the ISCCP-FD (ISCCP-Flux of its foundation
Data) radiation data sets are to the cloud overlay capacity estimation of remotely-sensed data the results show that the optical remote sensing data cloud that the whole world obtains every year
Covering is about 66% [1].The intensive situation of cloud cover can seriously affect interpretation, interpretation precision and the application of remotely-sensed data.
Therefore, important step of the cloud sector identification technology as the previous basis of remote sensing image data applied analysis, can improve image number
According to utilization rate, while also saving data collection cost.
Currently, it includes high score one that Chinese high-resolution earth observation systems, which have emitted and obtained the satellite of optical remote sensing data,
Number (GF-1), high score two (GF-2) and high score four (GF-4).High score series of optical satellite realizes round-the-clock round-the-clock
Earth observation, extreme enrichment high-resolution domestic spatial data set content.GF-1 multispectral data imaging band includes four
Wave band: blue wave band (0.45-0.52 μm), green wave band (0.52-0.59 μm), red wave band (0.63-0.69 μm) and near-infrared wave
Section (0.77-0.89 μm).Some middle spaceborne images of low resolution used at present in the world, such as AVHRR, MODIS, Landsat
The data such as series and Sentinel, are conducive to identify the thermal infrared and water vapor absorption wave band [2- of cloud and mist comprising one or more
4].Under the limitation of limited wavelength band, cloud sector judgement is carried out to high-resolution multi-spectral data, easily causing has with other in image
The earth's surface object (such as road, house, bare area) of high reflectance is obscured.Therefore, how to increasing high-resolution
Satellite remote sensing date carries out effective and efficient cloud sector automatic identification, and being still one very has challenge and urgently to be resolved ask
Topic.
In terms of the identification of cloud sector, common cloud sector recognition methods generally includes the identification of the cloud sector based on multidate image and base
It is identified in the cloud sector of single width image.Cloud sector recognition methods based on multidate image needs the cloudless data using different phases
As reference, accuracy of identification is higher；Based on the cloud sector recognition methods of single width image without reference to data, directly to image itself into
Row processing produces cloud sector mask data, therefore has stronger applicability and actual effect.Currently, homomorphic filtering method is still most
The common cloud sector recognition methods [5-6] for being directed to single width remote sensing image, homomorphic filtering are inhibited in frequency domain by high-pass filter
The low frequency region of cloud sector distribution, so that cloud sector radiation information is weakened, non-cloud sector radiation information is enhanced.Image passes through homomorphism
After filtering, the higher cloud layer reflectivity of brightness is reduced, and original relatively darker cloudless region reflectivity is got higher, by with it is original
Input is compared, and can obtain the mask result in cloud sector.But meanwhile homomorphic filtering go cloud effect greatly rely on high pass filter
The cutoff frequency of wave device, the research of forefathers' majority all use empirical value to cutoff frequency, it is clear that experience cutoff frequency can not adapt to
The Remote Sensing Data Processing of high-volume cloud sector level complexity, and to the cloud sector of different-thickness, such as spissatus and thin cloud, using empirical value
It is unable to reach best identified effect.
[1]: Zhang Y, Rossow W B, Lacis A A, et al.Calculation of radiative
fluxes from the surface to top of atmosphere based on ISCCP and other global
data sets:Refinements of the radiative transfer model and the input data[J]
.Journal of Geophysical Research:Atmospheres,2004,109(D19).
[2]: Ackerman S A, Strabala K I, Menzel W P, et al.Discriminating clear
sky from clouds with MODIS[J].Journal of Geophysical Research:Atmospheres,
1998,103(D24):32141-32157.
[3]: Derrien M, Farki B, Harang L, et al.Automatic cloud detection
applied to NOAA-11/AVHRR imagery[J].Remote Sensing of Environment,1993,46(3):
246-267.
[4]: Zhu Z, Woodcock C E.Object-based cloud and cloud shadow detection
in Landsat imagery[J].Remote Sensing of Environment,2012,118:83-94.
[5]: Shen H, Li H, Qian Y, et al.An effective thin cloud removal
procedure for visible remote sensing images[J].ISPRS Journal of
Photogrammetry and Remote Sensing,2014,96:224-235.
[6]: Lv H, Wang Y, Shen Y.An empirical and radiative transfer model
based algorithm to remove thin clouds in visible bands[J].Remote Sensing of
Environment,2016,179:183-195.
Summary of the invention
The present invention provides the adaptive cloud sector detection method of a kind of pair of high-resolution optical remote sensing image, is solved now with effective
There is the problem of Homomorphic Filtering Algorithm can not adaptively adjust filter cutoff frequency according to input image in technology.
In order to achieve the above objectives, the present invention provides the detections of the adaptive cloud sector of a kind of pair of high-resolution optical remote sensing image
Method, wherein high-resolution optical remote sensing image is high score No.1 remote sensing images comprising following steps:
S1: to high-resolution optical remote sensing image F, its cloud and mist thickness chart F1 is calculated；
S2: according between the cutoff frequency f for the high-pass filter applied during the radial energy of F1 spectrum and homomorphic filtering
Relationship, calculate the value of f；
S3: the cutoff frequency of high-pass filter during using f as homomorphic filtering carries out homomorphic filtering to F1, is filtered
Image F2 after wave；
S4: calculating the whitness index of each pixel in F2 and therefrom filter out the pixel that whitness index is greater than a preset threshold,
Obtain image F3；
S5: selecting circular configuration element, closed operation first carried out to F3 and carries out opening operation again, excellent to carry out morphology to it
Change, obtains cloud sector recognition result F4.
In one embodiment of this invention, in step S1, following calculate first is carried out:
Wherein, I (x, y)=2Ib(x,y)-0.95Ig(x, y), Ib(x,y)、Ig(x, y) is respectively that high-resolution optical is distant
Feel the blue wave band and green band of image F, I (x, y) is input wave band, is located at point (x ', y centered on W (x ', y ')′) w ×
The local window of w,
Recycle median filter to HTM (x later0,y0) carry out smoothly, and original image size is sampled, obtain cloud and mist
Thickness chart F1.
In one embodiment of this invention, the median filter uses 3 × 3 filter windows.
In one embodiment of this invention, w=5.
In one embodiment of this invention, the high-pass filtering applied during radial energy spectrum R (r) and homomorphic filtering of F1
Relationship between the cutoff frequency f of device is as follows:
Wherein, R (u, v), I (u, v) are respectively to cloud and mist thickness
Degree figure F1 carries out the real part and imaginary part of the complex spectrum obtained after Fourier transformation.
In one embodiment of this invention, it is the circular configuration element of 7 pixels that diameter is selected in step S5.
In one embodiment of this invention, step S4 includes the following steps:
S41: spectral concentration integral is carried out to the visible light wave range in F2, calculates cloud brightness fBr,
Wherein, λiFor wavelength, λmax、λminMaximum wavelength and minimum wavelength respectively in visible light region, unit are
Nm, BvFor the wavelength band of visible light, f (λi) expression wavelength be λiThe pixel value of wave band；
S42: whitness index f is calculatedWh,
Wherein, e (λi)=| f (λi)-fBr|；
S43: by fWhThe corresponding object of the pixel of > β is judged as non-white object and filters out it from F2, after obtaining filtering
Image F3.
In one embodiment of this invention, β=0.1.
In one embodiment of this invention, in step S2, the high-pass filter applied during homomorphic filtering is Gaussian difference
Divide high-pass filter, the high-frequency gain of difference of Gaussian high-pass filter is 1, low-frequency gain 0.05.
Adaptive cloud sector detection method provided by the invention to high-resolution optical remote sensing image passes through to a large amount of high scores
No.1 (GF-1) remote sensing images carry out analysis of experiments, have obtained the radial energy spectrum of image and the height applied during homomorphic filtering
Quantitative relationship between the cutoff frequency of bandpass filter can adaptively adjust cutoff frequency according to different input pictures, real
The batch cloud sector identification of existing GF-1 remote sensing images.It is efficient that the present invention can carry out mass to high-resolution optical remote sensing image
Cloud sector detection, overall recognition accuracy is up to 93.81%.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention for those of ordinary skill in the art without creative efforts, can be with
It obtains other drawings based on these drawings.
Fig. 1 a- Fig. 1 d is the specific embodiment for selecting circular configuration element to carry out morphology optimization to cloud sector, and Fig. 1 a is 7
The circular configuration element of pixel, Fig. 1 b are that treated as a result, Fig. 1 d is Fig. 1 c by closed operation by F3 for cloud sector example F3, Fig. 1 c
Using opening operation processing as a result, i.e. image F4；
Fig. 2 a- Fig. 2 c is the example optimized using whitness index and morphology operations to homomorphic filtering result F2,
Fig. 2 a is original input picture F, and Fig. 2 b is that homomorphic filtering result F2, Fig. 2 c is the image F4 optimized after post treatment；
Fig. 3 a is the high-resolution optical remote sensing image (1024 × 1024) of 8 width difference cloud amount covering, and 8 width image cloud amount cover
Lid is successively reduced from upper left to bottom right；
Fig. 3 b is the cloud sector testing result using method of the invention to Fig. 3 a, is image F in Fig. 3 a, is image in Fig. 3 b
Co-located image corresponds in F4, Fig. 3 a, Fig. 3 b；
The preceding two width figure of first row is that the high split-phase motor of 8m resolution multi-spectral (PMS) in GF-1 satellite collects in Fig. 4
Remote sensing images F, rear two width figure is the collected remote sensing images F of 16m resolution multi-spectral wide cut camera (WFV), to first row
Using the cloud sector mask result F4 of secondary series is obtained after method processing of the invention, third is classified as from the distant of first row raw video
Feel and removes the result image behind cloud sector in image.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, those of ordinary skill in the art are obtained every other under that premise of not paying creative labor
Embodiment shall fall within the protection scope of the present invention.
Before specific embodiment, the principle for the homomorphic filtering being applied in the present invention is first introduced:
In cloud and mist region, remotely-sensed data imaging model includes two parts: solar radiation is through cloud layer reflective portion and too
Sun radiation penetrates the part of cloud layer again after clutter reflections.The signal that sensor receives, f are indicated with f (x, y)i(x, y) is cloud
Layer reflectivity, is able to reflect the distribution of cloud, fr(x, y) indicates clutter reflections rate, then the imaging model of cloud can simplify are as follows:
F (x, y)=fi(x,y)·fr(x,y)
In frequency domain, the f of cloud layer distribution is indicatedi(x, y) concentrates on low frequency, reflects the f of image detail contentr(x, y) point
Cloth is in high-frequency region.Homomorphic filtering passes through the separation of low-frequency component and radio-frequency component, and the low frequency component for inhibiting cloud to be distributed and increasing
Strong high-frequency information realizes the cloud removing in airspace.
By taking logarithm to obtain on above formula both sides:
In (f (x, y))=In (fi(x,y))+In(fr(x,y))
Fourier transformation is carried out again carries out space-frequency conversion:
F { In (f (x, y)) }=F { In (fi(x,y))}+F{In(fr(x, y)) }, be abbreviated as F (u, v)=L (u, v)+H (u,
v)
High-pass filter is used to frequency spectrum, inhibits the low-frequency component of cloud layer reflectivity on frequency domain:
S (u, v)=L (u, v) Filter (u, v)+H (u, v) Filter (u, v)
Wherein Filter (u, v) is the transmission function of high-pass filter, and filter function uses difference of Gaussian high-pass filter,
In formula: γ H is high-frequency gain, and γ L is low-frequency gain, be usually arranged H>=1 γ and γ L<<1, adopt in the present embodiment
With γ H=1, γ L=0.05；D(u,V) distance that point (u, v) arrives Fourier transformation central point is represented；D0It is cut for high-pass filter
Only frequency.
Filtered result is subjected to inverse fourier transform:
S (x, y)=In ' (fi(x,y))+In′(fr(x,y))
Exponential transform is carried out again:
F ' (x, y)=fi′(x,y)+fr′(x,y)。
Transformation results f ' (x, y) is stretched to the amplitude of original input data:
Wherein maxiAnd miniThe respectively maxima and minima of input data f (x, y), maxoAnd minoRespectively f '
The maxima and minima of (x, y), and thus obtain cloud mask result:
Cutoff frequency for the high-pass filter applied during homomorphic filtering, is described as follows:
Cutoff frequency value difference can bring different filter result, and the key of homomorphic filtering is will be according to different defeated
Enter to determine the cutoff frequency of suitable high-pass filter.If cutoff frequency value is lower, high-pass filter can not completely inhibit low frequency
Luminance information；If cutoff frequency value is higher, the detailed information of earth's surface object can be high-pass filtered device and excessively cut down.
At the cutoff frequency point of high-pass filter, filter understands the low frequency signal before significant attenuation cutoff frequency, makes defeated
Signal intensity attenuation is the half (- 3db) of input signal out, while the signal after cutoff frequency being allowed to pass through.Cutoff frequency
Value significantly impact homomorphic filtering as a result, and its selection spectrum energy corresponding with illumination field and mirror field is related, need
It largely to practice to determine.The present invention is from the illumination field of image and the spectrum energy of mirror field, by analyzing luminance component
Energy accounting, to select suitable cutoff frequency.
Image is after two-dimensional fast fourier transform, and by origin translation to spectral centroid, frequency origin value is indicated at this time
The average gray level of image, frequency origin value is bigger, and expression image averaging gray scale is higher, otherwise image averaging gray scale is lower.Fu Li
Transition respectively indicates frequency from low frequency gradually to high frequency variation to leaf spectral centroid around.The frequency spectrum of input image f (x, y) at this time
F (u, v) is indicated:
F (u, v)=| F (u, v) | ejθ
Wherein, amplitude spectrum is expressed asR (u, v) and I (u, v) is respectively Fourier transformation
The real part and imaginary part of complex spectrum, phase angleAssuming that input image then has u=m/ having a size of m × n
2 and v=n/2.Assuming that illumination condition is uniformly, then the Fourier spectrum of illumination field only has DC component, with uneven illumination
Evenness increases, and starts reflective variant occur, is expressed as on Fourier spectrum in addition to DC component, while increasing harmonic wave point
Amount.At this point, Fourier transformation radial direction frequency spectrum R (r) indicates the sum of the gross energy of all directions in image frequency domain harmonic annulus, reflection
Energy size in frequency domain in harmonic frequency annulus, as harmonic frequency gradually tends to nyquist frequency, R (r) amplitude is more next
Smaller and decaying tends to 0.Percentage R (r) % of R (r) He Qizhan gross energy is indicated are as follows:
In formula:θ=0o~360o.Cloud amount is whole compared with multitemporal image radial direction frequency spectrum
Body energy level is higher, and average gray value R [0] is also higher, at this point, embody the luminance component high concentration of cloud in low frequency part,
Therefore cutoff frequency is disposed proximate to relatively reasonable at the low frequency in the center of circle；When cloudless covering image radial direction spectrum energy level compared with
It is low, harmonic energy convergence level off to 0 speed it is slower, at this point, luminance component is weaker, embody the harmonic energy of reflecting component compared with
By force, then cutoff frequency should be arranged higher and sufficiently inhibit terrestrial object information, avoid owing filtering.
Inventor has found that positive is presented in average gray energy accounting R (r) % and cutoff frequency of image through practice
Close, i.e., it is lower to cover more (R (r) % is low) its cutoff frequency for cloud amount, when cloud amount is less (R (r) % high), cutoff frequency compared with
It is high.Inventor establishes diameter at R (r) % and cutoff frequency by analyzing the cutoff frequency of 79 width instance datas
To frequency spectrum andQuantitative model, so as to obtain the corresponding cutoff frequency value of every scape image.Regression model of the invention
Precision is more excellent, coefficient of determination R2Up to 0.9275, model is reliable, specifically sees the following detailed.
High-resolution optical remote sensing image in the present invention is high score No.1 (GF-1) remote sensing images, and GF-1 Seeds of First Post-flight has
Two 2m resolution panchromatics and the high split-phase motor of 8m resolution multi-spectral (PMS), 4 16m resolution multi-spectral wide cut cameras
(WFV)。
Detection method includes the following steps in adaptive cloud sector provided by the invention to high-resolution optical remote sensing image:
S1: to high-resolution optical remote sensing image F, its cloud and mist thickness chart F1 is calculated；
In the present invention, it is assumed that be constantly present the extremely low pixel of certain reflected values in certain scene, pixel value close to
0.It is influenced due to being reflected by cloud and mist, makes the pixel value of above-mentioned dark pixel and incomplete " black ", then it is assumed that such dark picture
The pixel value of member objectively reflects the thickness of cloud and mist in scene.
In the present embodiment, in step S1, following calculate first is carried out:
Above formula is from left to right to find dark target from top to bottom by image by setting local window, to obtain image
HTM result.Wherein, I (x, y)=2Ib(x,y)-0.95Ig(x, y), Ib(x,y)、Ig(x, y) is respectively that high-resolution optical is distant
Feel the blue wave band and green band of image F, I (x, y) is input wave band, generallys use and is easily made by the blue wave band of Influence of cloud
To input wave band I (x, y), but detection is crossed since HTM result that blue wave band is calculated will cause cloud and mist thickness, herein using line
Property difference synthesize wave band I (x, y)=2Ib(x,y)-0.95Ig(x, y) can preferably be pressed down with the HTM result that the synthesis wave band calculates
The brightness of high reflection atural object in ground processed mitigates the effect excessively detected.It is located at w × the w's of point (x ', y ') centered on W (x ', y ')
Local window, w=5 in the present embodiment use 5 × 5 window.
Recycle median filter to HTM (x later0,y0) carry out smooth, median filter herein preferably uses 3 ×
3 filter windows, and original image size is sampled, obtain cloud and mist thickness chart F1.
S2: according between the cutoff frequency f for the high-pass filter applied during the radial energy of F1 spectrum and homomorphic filtering
Relationship, calculate the value of f；
In the present embodiment, the cutoff frequency for the high-pass filter applied during radial energy spectrum R (r) of F1 and homomorphic filtering
Relationship between rate f is as follows:
Wherein, R (u, v), I (u, v) are respectively to cloud and mist thickness
Degree figure F1 carries out the real part and imaginary part of the complex spectrum obtained after Fourier transformation.
S3: the cutoff frequency of high-pass filter during using f as homomorphic filtering carries out homomorphic filtering to F1, is filtered
Image F2 after wave；
By homomorphic filtering self-adaptive processing, the cloud sector of highlight reflection can be accurately identified, due to the height in other non-cloud sectors
Reflectivity atural object is equally existed by the risk of overidentified, therefore, is covered in next step using whitness index and Morphology Algorithm to cloud
Film PRELIMINARY RESULTS optimizes post-processing.
S4: calculating the whitness index of each pixel in F2 and therefrom filter out the pixel that whitness index is greater than a preset threshold,
Obtain image F3；
Whitness index is based on the most important characteristic of cloud, that is, highlighted white and visible light plateau is presented.It is highlighted
White indicates that the curve of spectrum in cloud sector is relatively high in the value of visible light wave range, by carrying out spectral concentration to visible light wave range
Integral, can obtain cloud brightness.
In the present embodiment, step S4 includes the following steps:
S41: spectral concentration integral is carried out to the visible light wave range in F2, calculates cloud brightness fBr,
Wherein, λiFor wavelength, λmax、λminMaximum wavelength and minimum wavelength respectively in visible light region, unit are
Nm, BvFor the wavelength band of visible light, f (λi) expression wavelength be λiThe pixel value of wave band；
" white spectra ", which is presented, in visible light curve indicates that the curve of spectrum is flat, then the first differential value of the curve of spectrum is lower,
Due to will receive different degrees of influence there are the precision of noise and quantization error, differential value.By each pixel and fBr's
The differential of error is as whitness index.
S42: whitness index f is calculatedWh,
Wherein, e (λi)=| f (λi)-fBr|；
e(λi) expression wavelength be λiThe pixel value and f of wave bandBrError.Obviously, cloud sector is whiter, e (λi) smaller, whiteness refers to
Number is lower, otherwise whitness index is higher.Homomorphic filtering result is optimized using whitness index, it can be higher but non-by pixel value
The object of white is rejected, and error recognition rate is reduced.
S43: by fWhThe corresponding object of the pixel of > β is judged as non-white object and filters out it from F2, after obtaining filtering
Image F3.
The value of β is chosen to be 0.1 in the present embodiment, in other embodiments, the value of β can be adjusted so that it is suitable
Answer real image.
S5: selecting circular configuration element, closed operation first carried out to F3 and carries out opening operation again, excellent to carry out morphology to it
Change, obtains cloud sector recognition result F4.
It is the circular configuration element of 7 pixels that diameter is selected in the present embodiment.
Morphologic filtering basic thought can be regarded as being filtered bianry image using structural element.Morphologic filtering effect
Fruit is related with structural element selection, since cloud sector typically exhibits irregular, present invention selection circular configuration element acquisition
Smooth cloud sector profile, and cloud sector mask result is post-processed using the strategy that first progress closed operation carries out opening operation again.
The gap in cloud sector can be filled while not significantly change its area by first carrying out closed operation, then is carried out opening operation and can be eliminated cloud sector boundary
Discrete noise makes contour smoothing.After Morphological scale-space, relatively broken small patch cloud point is removed, and some wrong identifications highlight
Discrete point is also filtered out, and the profile around cloud sector show it is clearly more demarcated, obtain preferably post-process effect.
Fig. 1 a- Fig. 1 d is the specific embodiment for selecting circular configuration element to carry out morphology optimization to cloud sector, and Fig. 1 a is 7
The circular configuration element of pixel, Fig. 1 b are cloud sector example F3, wherein white is cloud pixel, black is backdrop pels, and Fig. 1 c is F3
By closed operation treated as a result, Fig. 1 d be Fig. 1 c using opening operation handle as a result, i.e. image F4；
Fig. 2 a- Fig. 2 c is the example optimized using whitness index and morphology operations to homomorphic filtering result F2,
Fig. 2 a is original input picture F, and Fig. 2 b is that homomorphic filtering result F2, Fig. 2 c is the image F4 optimized after post treatment；
Fig. 3 a is the high-resolution optical remote sensing image (1024 × 1024) of 8 width difference cloud amount covering, and 8 width image cloud amount cover
Lid is successively reduced from upper left to bottom right；
Fig. 3 b is the cloud sector testing result using method of the invention to Fig. 3 a, is image F in Fig. 3 a, is image in Fig. 3 b
Co-located image corresponds in F4, Fig. 3 a, Fig. 3 b；
The preceding two width figure of first row is that the high split-phase motor of 8m resolution multi-spectral (PMS) in GF-1 satellite collects in Fig. 4
Remote sensing images F, rear two width figure is the collected remote sensing images F of 16m resolution multi-spectral wide cut camera (WFV), to first row
Using the cloud sector mask result F4 of secondary series is obtained after method processing of the invention, third is classified as from the distant of first row raw video
Feel and removes the result image behind cloud sector in image.
Human interpretation is carried out by the cloud sector recognition result to 79 scape images, precision test is the results show that precision of the present invention
It is higher, overall accuracy 93.81%.Wherein, the precision highest in the exposure mask of cloud sector, up to 97.69%, cloud sector exposure mask buffer area precision
It is lower, it is 90.15%.Generate from the point of view of precision and error analysis, at the boundary profile of cloud sector compared with multiple error, close to cloud central point and
Buffer area point of distance precision is higher.From cloud sector center gradually to cloud sector border transition when, the accuracy of identification is gradually reduced；From
When cloud sector boundary is to buffer area edge transition, accuracy gradually rises again.Since cloud sector boundary is mostly irregular shape, work as cloud sector
When boundary pixel point and unobvious atural object comparison, the algorithm based on homomorphic filtering cannot be accurately identified effectively.Secondly, of the invention
The morphologic filtering processing of the make-before-break of middle use can also make cloud sector boundary generate variation.Totally apparently, the present invention can
The cloud sector position of relatively accurate identification GF-1 remotely-sensed data obtains more accurately cloud sector mask result.
Adaptive cloud sector detection method provided by the invention to high-resolution optical remote sensing image passes through to a large amount of high scores
No.1 (GF-1) remote sensing images carry out analysis of experiments, have obtained the radial energy spectrum of image and the height applied during homomorphic filtering
Quantitative relationship between the cutoff frequency of bandpass filter can adaptively adjust cutoff frequency according to different input pictures, real
The batch cloud sector identification of existing GF-1 remote sensing images.It is efficient that the present invention can carry out mass to high-resolution optical remote sensing image
Cloud sector detection, overall recognition accuracy is up to 93.81%.
Those of ordinary skill in the art will appreciate that: attached drawing is the schematic diagram of one embodiment, module in attached drawing or
Process is not necessarily implemented necessary to the present invention.
Those of ordinary skill in the art will appreciate that: the module in device in embodiment can describe to divide according to embodiment
It is distributed in the device of embodiment, corresponding change can also be carried out and be located in one or more devices different from the present embodiment.On
The module for stating embodiment can be merged into a module, can also be further split into multiple submodule.
Finally, it should be noted that the above embodiments are merely illustrative of the technical solutions of the present invention, rather than its limitations；Although
Present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that: it still may be used
To modify to technical solution documented by previous embodiment or equivalent replacement of some of the technical features；And
These are modified or replaceed, the spirit and model of technical solution of the embodiment of the present invention that it does not separate the essence of the corresponding technical solution
It encloses.
Claims (9)
1. the adaptive cloud sector detection method of a kind of pair of high-resolution optical remote sensing image, wherein high-resolution optical remote sensing figure
As being high score No.1 remote sensing images, which comprises the following steps:
S1: to high-resolution optical remote sensing image F, its cloud and mist thickness chart F1 is calculated；
S2: according to the pass between the cutoff frequency f for the high-pass filter applied during the radial energy of F1 spectrum and homomorphic filtering
System, calculates the value of f；
S3: the cutoff frequency of high-pass filter during using f as homomorphic filtering carries out homomorphic filtering to F1, after obtaining filtering
Image F2；
S4: it calculates the whitness index of each pixel in F2 and therefrom filters out the pixel that whitness index is greater than a preset threshold, obtain
Image F3；
S5: circular configuration element is selected, closed operation is first carried out to F3 and carries out opening operation again, to carry out morphology optimization to it, is obtained
To cloud sector recognition result F4.
2. the adaptive cloud sector detection method according to claim 1 to high-resolution optical remote sensing image, feature exist
In in step S1, first progress is following to be calculated:
Wherein, I (x, y)=2Ib(x,y)-0.95Ig(x, y), Ib(x,y)、Ig(x, y) is respectively high-resolution optical remote sensing image
The blue wave band and green band of F, I (x, y) are input wave band, and the office of the w × w of point (x ', y ') is located at centered on W (x ', y ')
Portion's window,
Recycle median filter to HTM (x later0,y0) carry out smoothly, and original image size is sampled, obtain cloud and mist thickness
Scheme F1.
3. the adaptive cloud sector detection method according to claim 2 to high-resolution optical remote sensing image, feature exist
In the median filter uses 3 × 3 filter windows.
4. the adaptive cloud sector detection method according to claim 2 to high-resolution optical remote sensing image, feature exist
In w=5.
5. the adaptive cloud sector detection method according to claim 1 to high-resolution optical remote sensing image, feature exist
In the relationship between the cutoff frequency f for the high-pass filter applied during radial energy spectrum R (r) of F1 and homomorphic filtering is such as
Under:
Wherein,R (u, v), I (u, v) are respectively to cloud and mist thickness
Figure F1 carries out the real part and imaginary part of the complex spectrum obtained after Fourier transformation.
6. the adaptive cloud sector detection method according to claim 1 to high-resolution optical remote sensing image, feature exist
In selecting diameter in step S5 is the circular configuration element of 7 pixels.
7. the adaptive cloud sector detection method according to claim 1 to high-resolution optical remote sensing image, feature exist
In step S4 includes the following steps:
S41: spectral concentration integral is carried out to the visible light wave range in F2, calculates cloud brightness fBr,
Wherein, λiFor wavelength, λmax、λminMaximum wavelength and minimum wavelength respectively in visible light region, unit are nm, Bv
For the wavelength band of visible light, f (λi) expression wavelength be λiThe pixel value of wave band；
S42: whitness index f is calculatedWh,
Wherein, e (λi)=| f (λi)-fBr|；
S43: by fWhThe corresponding object of the pixel of > β is judged as non-white object and filters out it from F2, obtains filtered figure
As F3.
8. the adaptive cloud sector detection method according to claim 7 to high-resolution optical remote sensing image, feature exist
In β=0.1.
9. the adaptive cloud sector detection side according to claim 1 to 8 to high-resolution optical remote sensing image
Method, which is characterized in that in step S2, the high-pass filter applied during homomorphic filtering is difference of Gaussian high-pass filter, high
The high-frequency gain of this difference high-pass filtering device is 1, low-frequency gain 0.05.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711116482.7A CN109801253B (en) | 2017-11-13 | 2017-11-13 | Self-adaptive cloud area detection method for high-resolution optical remote sensing image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711116482.7A CN109801253B (en) | 2017-11-13 | 2017-11-13 | Self-adaptive cloud area detection method for high-resolution optical remote sensing image |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109801253A true CN109801253A (en) | 2019-05-24 |
| CN109801253B CN109801253B (en) | 2020-11-17 |
Family
ID=66554627
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711116482.7A Active CN109801253B (en) | 2017-11-13 | 2017-11-13 | Self-adaptive cloud area detection method for high-resolution optical remote sensing image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109801253B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114298945A (en) * | 2022-01-05 | 2022-04-08 | 首都师范大学 | Optical remote sensing image thick cloud removing method based on virtual image construction |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101359399A (en) * | 2008-09-19 | 2009-02-04 | 常州工学院 | Cloud-removing method for optical image |
| CN102509262A (en) * | 2011-10-17 | 2012-06-20 | 中煤地航测遥感局有限公司 | Method for removing thin cloud of remote sensing image |
| CN104484859A (en) * | 2014-10-20 | 2015-04-01 | 电子科技大学 | Multispectral optical remote sensing image data thin-cloud removing method |
| CN105574826A (en) * | 2015-12-16 | 2016-05-11 | 中国科学院深圳先进技术研究院 | Thin cloud removing method of remote sensing image |
| US9846975B2 (en) * | 2016-02-18 | 2017-12-19 | Skycatch, Inc. | Generating filtered, three-dimensional digital ground models utilizing multi-stage filters |
-
2017
- 2017-11-13 CN CN201711116482.7A patent/CN109801253B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101359399A (en) * | 2008-09-19 | 2009-02-04 | 常州工学院 | Cloud-removing method for optical image |
| CN102509262A (en) * | 2011-10-17 | 2012-06-20 | 中煤地航测遥感局有限公司 | Method for removing thin cloud of remote sensing image |
| CN104484859A (en) * | 2014-10-20 | 2015-04-01 | 电子科技大学 | Multispectral optical remote sensing image data thin-cloud removing method |
| CN105574826A (en) * | 2015-12-16 | 2016-05-11 | 中国科学院深圳先进技术研究院 | Thin cloud removing method of remote sensing image |
| US9846975B2 (en) * | 2016-02-18 | 2017-12-19 | Skycatch, Inc. | Generating filtered, three-dimensional digital ground models utilizing multi-stage filters |
Non-Patent Citations (6)
| Title |
|---|
| PANG YONG ET AL.: "China typical forest aboveground biomass estimation by fusion of multi-platform data", 《IEEE》 * |
| 叶秋果 等: "基于同态滤波的遥感影像去云雾处理", 《海洋测绘》 * |
| 周小军 等: "基于改进同态滤波的遥感图像去云算法", 《信号与信息处理》 * |
| 沈文水 等: "基于同态滤波的遥感薄云去除算法", 《强激光与粒子束》 * |
| 王盼攀: "高空间分辨率遥感图像频域滤波增强算法研究及应用", 《中国优秀硕士学位论文全文数据库》 * |
| 蒙诗栎 等: "高分一号光学遥感数据自适应云区识别", 《红外与毫米波学报》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114298945A (en) * | 2022-01-05 | 2022-04-08 | 首都师范大学 | Optical remote sensing image thick cloud removing method based on virtual image construction |
| CN114298945B (en) * | 2022-01-05 | 2022-07-05 | 首都师范大学 | Optical remote sensing image thick cloud removing method based on virtual image construction |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109801253B (en) | 2020-11-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108596849B (en) | Single image defogging method based on sky region segmentation | |
| Vanmali et al. | Visible and NIR image fusion using weight-map-guided Laplacian–Gaussian pyramid for improving scene visibility | |
| CN104683767B (en) | Penetrating Fog image generating method and device | |
| Park et al. | Single image dehazing with image entropy and information fidelity | |
| CN105139367A (en) | Visible-light polarization image fusion method based on non-subsampled shearlets | |
| Gershikov et al. | Horizon line detection in marine images: which method to choose? | |
| US20110043603A1 (en) | System And Method For Dehazing | |
| CN102647941A (en) | Method and system for carrying out photoplethysmography | |
| CN108710910A (en) | A kind of target identification method and system based on convolutional neural networks | |
| CN109376600A (en) | Multi-spectrum remote sensing image comprehensive characteristics cloud detection method of optic and device | |
| Lu et al. | Underwater image descattering and quality assessment | |
| CN107403134B (en) | Local gradient trilateral-based image domain multi-scale infrared dim target detection method | |
| CN106846289A (en) | A kind of infrared light intensity and polarization image fusion method based on conspicuousness migration with details classification | |
| KR101928391B1 (en) | Method and apparatus for data fusion of multi spectral image and radar image | |
| CN108629757A (en) | Image interfusion method based on complex shear wave conversion Yu depth convolutional neural networks | |
| CN105426843A (en) | Single-lens palm vein and palmprint image acquisition apparatus and image enhancement and segmentation method | |
| Dümbgen et al. | Near-infrared fusion for photorealistic image dehazing | |
| CN112200787A (en) | Cloud detection method, storage medium and system for optical remote sensing image | |
| Lu et al. | Single underwater image descattering and color correction | |
| CN113989143B (en) | High-precision rapid focus detection method based on push-broom type underwater hyperspectral original image | |
| Lu et al. | Underwater scene enhancement using weighted guided median filter | |
| CN109801253A (en) | A kind of adaptive cloud sector detection method of pair of high-resolution optical remote sensing image | |
| CN115578304B (en) | Multi-band image fusion method and system combining saliency region detection | |
| Yang et al. | Infrared and visible image fusion based on QNSCT and Guided Filter | |
| CN108389191A (en) | The detection method in target shadow region in a kind of SAR image |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |