CN106651847A - Remote sensing image fog information detection method - Google Patents

Abstract

The invention discloses a remote sensing image fog information detection method. The method comprises the following steps of: inputting a remote sensing image, and obtaining rough fog estimation information; obtaining a rough transmittance of a diagram of an image scene and an atmosphere daylight value of the scene by utilizing dark channel prior, and obtaining a refined transmittance diagram; extracting atmospheric component in an atmosphere multiple scattering image model according to the model, superposing the component with the rough fog estimation information so as to obtain a fog space strength distribution diagram; and carrying out data fitting by utilizing a log function according to observation data obtained at ground observation points and numerical values of corresponding points in the fog space strength distribution diagram, so as to obtain a mapping relationship between a detection physical quantity required by actual scene and fog space strength distribution information and then realize the quantified measurement of fog atmosphere information. The invention provides a convenient and effective detection method for the monitoring of urban air environment pollution according to the atmosphere fog strength and space distribution information, obtained on the image, of the scenes on the basis of image processing.

Description

A kind of cloud and mist information detection method of remote sensing images

Technical field

The present invention relates to remote sensing and Computer Image Processing are calculated, more particularly to one kind is obtained using image procossing Take the detection method of remote sensing images cloud and mist information.

Background technology

Haze and PM2.5 pollutions in today's society is the keypoint treatment problem in Chinese majority cities region.But Each city employs a variety of Mitigation options, including automobile odd-and-even license plate rule, reduces carbon emission etc., but the air quality of entirety Do not change therewith.And it is exactly that monitoring during air contaminant treatment and law enforcement dynamics are inadequate to trace it to its cause.

Generally for the detection of the atmospheric pollution such as haze, often it is monitored using the Continuous Observation of ground monitoring point. But Atmospheric components characteristic etc. has abundant dynamic characteristic, light depends the data of several monitoring points alone, it is difficult to grasp on a large scale The details and spatial characteristics of air pollution.Simultaneously Atmospheric Flow is also resulted in, and the data of Continuous Observation receive peripheral ring The possibility of border pollution, so that monitoring is difficult to determine the accurate location of air pollution source.Therefore how air is effectively grasped The position of polluter and intensity, and the haze origin cause of formation to different cities is analyzed, and is the important need of current social environmental monitoring Ask.

The content of the invention

The technical problem to be solved in the present invention is to provide a kind of technological means, defends for the high-resolution of certain weather condition Star chart picture, obtains its corresponding high-resolution air cloud and mist intensity and space distribution information.

To solve above-mentioned technical problem, the technical solution used in the present invention is：A kind of cloud and mist information detection of remote sensing images Technology, implementation step is as follows：

(1) remote sensing images are input into, coarse cloud and mist are obtained using low pass filter and Steerable filter device and is estimated figure.

(2) the coarse transmitance figure of image scene and the air day light value of scene are obtained using dark primary priori, based on this Coarse transmitance combines the coarse cloud and mist estimated in step (1) and estimates figure, and using Steerable filter device fine transmitance figure is obtained.

(3) based on the fine transmitance value and the daylight obtained in the input figure and step (2) after Filtering Processing in step (1) Value, according to air Multiple Scattering iconic modelCan calculate and recover image Simultaneously, it is also possible to lift the air component in modelIt is coarse by what is obtained in the same step of component (1) Cloud and mist estimated information is superimposed, the cloud and mist spatial intensity distribution information for exactly extracting from remote sensing images.

(4) cloud and mist obtained in the observation data for obtaining according to ground observation point in remote sensing images scene and step (3) is empty Between in intensity distribution corresponding point numerical value, carry out data fitting using logarithmic function, detect needed for so as to obtain actual scene Mapping relations between physical quantity and cloud and mist spatial intensity distribution information, realize the measurement of cloud and mist atmospheric information.

Further, the detailed step in the step (1) includes：

A) remote sensing images are input into, Fourier transformation is carried out to it, its spectrogram is obtained, while building following low pass filtered Ripple device is filtered process to image：

H (u, v) is filter function in formula, and u, v are frequency coordinates, σ₀It is off frequency.The non-homogeneous cloud of filtering output Mist background is as follows：

Wherein B_cloudIt is the low-frequency information of input picture, I is input picture,WithIt is respectively Fourier transformation and Fu In two operators of leaf inverse transformation.

B) contrast expansion is carried out to cloud and mist distributed intelligence image according to a preliminary estimate, it is as follows that it adjusts accordingly formula：

WhereinBe adjust formula threshold value, max (B_cloud) and min (B_cloud) It is respectively the maximum and minima of non-homogeneous cloudy background image according to a preliminary estimate, and k₁, k₂With the parameter that λ is adjustment formula.

And zero-frequency compensation is carried out to it, and final preliminary estimation air cloud and mist distributed intelligence is obtained, its compensation formula is such as Under：

B′_cloud=B_cloud-offset (4)

Wherein offset is the compensation constant of zero-frequency compensation, then by the use of input picture as guiding figure, is filtered using being oriented to Ripple device is processed the air cloud and mist scattergram for obtaining, and is obtained one and is protected the smooth preliminary cloud and mist spatial distribution estimation figure in side.

C) air cloud and mist distributed intelligence according to a preliminary estimate is removed from input picture：

I'(x, y)=I (x, y)-B '_cloud(x,y) (5)

Using the image for obtaining, further result is extracted in the distributed intelligence of optimization air cloud and mist in ensuing process.

Further, the detailed step in the step (2) includes：

A) according to the dark of the image finally obtained in dark primary priori extraction step 1：

Wherein I_darkFor the image dark channel for obtaining, I^cEach Color Channel component of image is represented, Ω (x) is represented with x pixels Local neighborhood centered on position.

B) the front 0.1% brightest area pixel in part is extracted from above-mentioned image dark channel, each passage in the region is calculated Pixel average is used as day light value A.

C) according to the property of dark primary passage for its value of fogless region is close to 0.Therefore image initial transmitance is calculated It is as follows：

Wherein t' is calculated initial transmission, and ω is constant coefficient, A^cFor the channel components of day light value.

D) the preliminary cloud and mist spatial distribution calculated by the use of in step (1) estimates figure as guiding figure, using Steerable filter device To the initial transmission for obtaining, the fine high-resolution scene transmitance figure of optimization acquisition that becomes more meticulous is carried out.

Further, according to air multiple scattering model in the step (3), its form is as follows：

Wherein I deducts the image after preliminary cloud and mist distribution estimating for input picture, and A and t is the day light value obtained in step 2 With transmitance image.APSF_oAnd APSF_aIt is respectively aerosol optical depth and daylight point spread function, using the Gauss point of broad sense Cloth is being solved, and its form is as follows：

Wherein x, y are image coordinate location, and Γ (.) is gamma function,p It is with σ and atmospheric parameters, is respectively calculated as follows：

P=kT (10)

T is atmosphere optical thickness in formula, according to transmitance and atmosphere optical thickness relation t=e^-TResolve and go out, k is parameter Constant, q is the forward scattering factor, is a constant with weather condition correlation.Its optics for aerosol optical depth Thickness value is-log (t), and its optical thickness values is-log (1-t) for daylight point spread function.So we can count Calculate the air component in model：

The final preliminary air distributed intelligence with reference to acquisition in step 1, we can obtain final air cloud and mist distribution Hum pattern：

B″′_cloud=B '_cloud+B″_cloud (13)

Further, according to input remote sensing images in the step (4), according to its scene imaging scope, obtain correspondingly Face observation station remote sensing images imaging during real atmosphere measurement data (such as PM2.5 concentration, air quality index etc.) as sample Notebook data Y, then takes the respective point in the cloud and mist spatial intensity distribution figure of the acquisition in step (3) as sample data X, utilizes Below the data of above-mentioned acquisition are fitted by the exponential function of form：

Y=-Alog (B (1-x))+C (14)

Wherein y is atmosphere measurement data value, and x is the cloud and mist information detection diagram data value for obtaining, and A, B, C are to need to be fitted The parameter for arriving.After obtaining fitting parameter, the image to obtaining in step (3) carries out pointwise mapping, final to obtain the big of quantification Gas measurement data scattergram.

The invention has the beneficial effects as follows：The present invention proposes a kind of cloud and mist information detection method of remote sensing images, realizes The cloud and mist distributed intelligence of corresponding scene is obtained by the remote sensing images under the certain atmospheric condition of input.In the present invention, only need Remote sensing images are simply input, without the need for introducing other measurement parameters, by the algorithm of Digital Image Processing, remote sensing images is made While becoming clear, the cloud and mist distributed intelligence of air is got.

Description of the drawings

Fig. 1 is the schematic flow sheet of the embodiment of the present invention.

Fig. 2 is the original image of the embodiment of the present invention.

Fig. 3 is the remote sensing images of the preliminary cloud and mist distribution of removal that the embodiment of the present invention is obtained.

Fig. 4 is that the embodiment of the present invention extracts the image transmission rate after the optimization for obtaining.

The air cloud and mist spatial intensity distribution result that Fig. 5 is obtained for the embodiment of the present invention.

Fig. 6 is that embodiment of the present invention surface-based observing station assumes the location drawing.

Fig. 7 is the matched curve in the embodiment of the present invention with ground data point.

Specific embodiment

Below in conjunction with the accompanying drawings the present invention is described in further detail with specific embodiment.

As shown in figure 1, the implementation steps of the remote sensing images defogging method under the conditions of the non-homogeneous cloud and mist of the present embodiment are as follows：

(1) the secondary remote sensing images (see Fig. 2) of input one, it is σ to select cut-off frequency parameter₀=1.7, build low pass filterLow-pass filtering operation is carried out to input picture.The parameter k of setting adjustment formula₁=5, k₂=5 (being directed to 8 bitmaps) and λ=2, are adjusted to the output image of low pass filter：

Zero-frequency compensation is carried out to the image of Tuning function output, final preliminary estimation air cloud and mist distribution letter is obtained Breath, its compensation formula is as follows：

B′_cloud=B_cloud-offset

Wherein offset is the compensation constant of zero-frequency compensation, and its value is each passage average of input picture, then using input Image is processed the air cloud and mist scattergram for obtaining as guiding figure using Steerable filter device, is obtained a guarantor side and is smoothed Preliminary cloud and mist spatial distribution estimate figure.Obtain simultaneously and eliminate the remote sensing images I'(x that thick buyun mist distribution is removed, y), its knot Fruit is schemed as shown in Figure 3.

(2) to the remote sensing images I'(x for obtaining in step 1, y), according to dark primary priori helping secretly for image is extracted Road：

Wherein I_darkFor the image dark channel for obtaining, I^cThe RGB channel components of image are represented, Ω (x) is represented with x pixels Local neighborhood centered on position.The front 0.1% brightest area pixel in part is extracted from above-mentioned image dark channel, the area is calculated The pixel average of each passage in domain is A as day light value A, in the present embodiment^R=255, A^G=254, A^B=250.

According to the property of dark primary passage for its value of fogless region is close to 0.Therefore image initial transmitance is calculated such as Under：

Wherein t' is calculated initial transmission, and ω is constant coefficient, A^cFor the channel components of day light value.

The preliminary cloud and mist spatial distribution calculated by the use of in step (1) estimates figure as guiding figure, using Steerable filter device pair The initial transmission for obtaining, carries out the fine high-resolution scene transmitance figure of optimization acquisition that becomes more meticulous, and sees Fig. 4.

(3) and then using air multiple scattering model, its form is as follows：

Wherein I ' deducts the image after preliminary cloud and mist distribution estimating for input picture, and A and t is the daylight obtained in step 2 Value and transmitance image.APSF_oAnd APSF_aIt is respectively aerosol optical depth and daylight point spread function, using the Gauss of broad sense It is distributed to be solved, its form is as follows：

Wherein x, y are image coordinate location, and Γ (.) is gamma function,p It is with σ and atmospheric parameters, is respectively calculated as follows：

P=kT (17)

T is atmosphere optical thickness in formula, according to transmitance and atmosphere optical thickness relation t=e^-TResolve and go out, k takes 0.5, Q is the forward scattering factor, and its value is as shown in the table, and 0.7 is taken as in the present embodiment.

0.0-0.2	0.2-0.7	0.7-0.8	0.8-0.85	0.85-0.9	0.9-1.0
						Air	Aerosols	Haze	Mist	Fog	Rain

Its optical thickness values is-log (1-t) for daylight point spread function.So we can calculate model In air component：

The final preliminary air distributed intelligence with reference to acquisition in step 1, we can obtain final air cloud and mist distribution Hum pattern, as shown in Figure 5：

B″′_cloud=B '_cloud+B″_cloud (19)

(4) as shown in Figure 6, it is assumed that there is A in the remote sensing graph region of input, surface air parameter observation station at B, C tri-, Air quality index (AQI) numerical value that it is detected during satellite imagery is respectively 170,142,155.It is obtained in step 3 To air cloud and mist distributed intelligence figure in the data of respective point position be respectively 0.4826,0.3518,0.4132.Using following shape Above-mentioned data are fitted by the logarithmic function of formula：

Y=-Alog (B (1-x))+C (20)

The parameter that fitting is obtained is respectively A=124.0137, B=15.1922, C=425.8834.The result of its fitting is such as Shown in Fig. 7.Based on this fit come logarithmic function we can calculate the air of optional position in remote sensing images scene domain Air quality index at performance figure, such as D is 142 (result of calculation is rounded), and the result at E is 152.

Compared to traditional method, this method does not need satellite sounding other data, only by Digital Image Processing skill Art realizes that the cloud and mist layer of image is extracted, and so as to realize remote sensing images is become clearly simultaneously, obtains high-resolution cloud and mist space Distribution and intensity map.Simultaneously just further it can be changed into quantitatively from qualitative observation if equipped with ground observation point data Observation.

Claims (5)

1. the cloud and mist information detection method of a kind of remote sensing images, it is characterised in that the method is comprised the following steps：

(1) remote sensing images are input into, coarse cloud and mist are obtained using low pass filter and Steerable filter device and is estimated figure.

(2) the coarse transmitance figure of image scene and the air day light value of scene are obtained using dark primary priori, based on coarse Cross rate figure and estimate figure with reference to the coarse cloud and mist obtained in step 1, using Steerable filter device fine transmitance figure is obtained.

(3) based on the fine transmitance figure and air day light value obtained in the input figure and step 2 after processing in step 1, according to Air Multiple Scattering iconic modelWhile calculating recovery image, extract The air component gone out in modelThe component is estimated into figure superposition with the coarse cloud and mist obtained in step 1, just It is the cloud and mist spatial intensity distribution figure extracted from remote sensing images.

(4) the cloud and mist spatial-intensity obtained in the observation data for obtaining according to ground observation point in remote sensing images scene and step 3 The numerical value of corresponding point in scattergram, using logarithmic function data fitting is carried out, and physical quantity is detected needed for so as to obtain actual scene And the mapping relations between cloud and mist spatial intensity distribution information, realize the measurement of cloud and mist atmospheric information.

2. the cloud and mist information detection method of remote sensing images as claimed in claim 1, it is characterised in that the step 1 is specially：

Being input into a pair has mist remote sensing images, and build following low pass filter and Tuning function carries out pretreatment to image：

$H(u,v)=\exp \[-\frac{u^2+v^2}{2{\mathrm{\σ}}_0^2}\]$

$B_{cloud}(x,y)=\left\{\begin{array}{c}{B}_{cloud}(x,y)+{\[\frac{B_{cloud}(x,y)-B_0}{\max \left(B_{cloud}\right)-B_0}\]}^{\mathrm{\λ}}\×k_1,B_{cloud}(x,y)>B_0\\ B_{cloud}(x,y)-{\[\frac{B_0-B_{cloud}(x,y)}{B_0-\min \left(B_{cloud}\right)}\]}^{\mathrm{\λ}}\×k_2,B_{cloud}(x,y)\≤B_0\end{array}\right.$

So as to the coarse scattergram of the fog for obtaining image, then combine Steerable filter using input picture and operate to coarse point of fog Butut carries out protecting side refinement to optimize cloud layer marginal information.

3. the cloud and mist information detection method of remote sensing images as claimed in claim 1, it is characterised in that the step 2 is specially：

The dark of image is extracted according to dark primary priori：Using dark, obtain Take the air day light value A and image transmitance t of image：Then obtain using in step 1 The coarse cloud and mist estimation figure for taking obtains the fine transmitance figure of scene using Steerable filter operation as guiding figure.

4. the cloud and mist information detection method of remote sensing images as claimed in claim 1, it is characterised in that the step 3 is specially：

Based on air multiple scattering modelTo through the distant of step 1 Filtering Processing Sense image carries out mist elimination recovery.The air component in model is extracted during recoveryAnd should With extracting in step 1, the coarse cloud and mist estimation figure that obtains is superimposed to obtain final remote sensing images cloud and mist spatial intensity distribution to component Figure.

5. the cloud and mist information detection method of remote sensing images as claimed in claim 1, it is characterised in that the step 4 is specially：

The real atmosphere measurement data of correspondence time is detected according to corresponding ground observation station in remote sensing images areas imaging, with reference to Respective point in the cloud and mist spatial intensity distribution figure of the acquisition in step 3, is fitted using the exponential function of following form：

Y=-Alog (B (1-x))+C

Wherein y is atmosphere measurement data value, and x is the cloud and mist spatial intensity distribution diagram data value for obtaining, and A, B, C are to need to be fitted The parameter for arriving.After obtaining fitting parameter, the image to obtaining in step 3 carries out pointwise mapping, the final air for obtaining quantification Measurement data scattergram.