CN101814139A - Raindrop identifying method - Google Patents

Abstract

The invention provides a raindrop identifying method, belongs to the field of digital image identification, and aims to automatically identify a raindrop target in the raindrop image and extract related raindrop information so as to improve the automation and precision of raining weather phenomenon observation. The method is applied to automatic observation and acquisition of a raindrop spectrum in the raining weather phenomenon observation. The method comprises the following steps in sequence: 1, image acquisition; 2, image pretreatment; 3, image differentiation; 4, extraction of the edge gradient image; 5, binarization of the edge gradient image; 6, image morphologic treatment; and 7, extraction of the raindrop information. In the method of the invention, the anisotropic diffusion smoothing filter is used as a method for the image pretreatment, the edge gradient information is used as effective characteristics of the raindrop target, an image morphologic method is also utilized to effectively identify the raindrop target and extract the related raindrop information, and thus the automaticity and the precision on the raining weather phenomenon observation can be improved.

Description

A kind of raindrop recognition methods

Technical field

The invention belongs to digital picture identification field, be specifically related to a kind of raindrop recognition methods, be used for discerning automatically the raindrop target of raindrop image and extract relevant raindrop information based on edge gradient.

Background technology

Rainfall observation is the important component part in the weather phenomenon observation, in development of the national economy service, important effect is arranged, as the design of flood control, drought resisting, mitigation, hydraulic engineering, industrial and agricultural production etc., simultaneously to climate change many aspects important influence such as monsoon research particularly.

The very important point will extract the pairing raindrop size distribution of rainfall exactly in rainfall observation, and and then utilizes it that Rainfall Characteristics such as rainfall amount, rainfall intensity, rain types and rainfall microstructure etc. are analyzed.Raindrop size distribution is meant in the unit space volume, and diameter is at the number of the raindrop of D～D+ Δ d, i.e. the distribution of raindrop size in the unit volume, and English is Rain Drop SizeDistribution (D SD).

As far back as the nineties in 19th century, external many scientific research personnel begin to pay close attention to the research of ground raindrop size distribution observation procedure, in the paper of delivering, have 5 kinds at least about measuring the method for raindrop size and distribution: dynamic method [Scheleusener P E.Drop Size Distribution and Energy ofFalling Raindrops from a Medium Pressure Irrigation Sprinkler.MichiganState University, East Lansing, 1967:45-48], mottling method [Hall M J.Use of thestain method in determining of the drop-size distribution of coarse liquidsprays.Trans ASAE, 1970,13 (1): 33-37], flour method [Kohl R A.Drop sizedistribution from medium-sized agricultural sprinklers.Trans ASAE, 1974,17 (4): 690-693], photographic process [Roels J M.Personal Communication.Laboratory ofPhysical Geography, Geographical Institute, University of Utrecht, Netherlands, 1981:127-136] and infusion method [McCooll D K.PersonalCommunication.USDA-ARS.Agricultural Engineering Department, Washington State University, Pullman, 1982:67-82].Recently existing instrument also can measure the terminal-velocity of raindrop when surveying raindrop size, shape.Dynamic method is only applicable to measure the homogeneous drop-size distribution, does not measure and be suitable for the bigger drop-size distribution of spectrum width.The theoretical foundation of mottling method is that the standard drop-size distribution by a broad spectrum width calculates actual raindrop, and its error range is equivalent to 0.037～37mg 6%～14%.The shortcoming of flour method is that typical curve can change along with the flour of different qualities, need correct at any time.Photographic process is to take the raindrop image with a single-lens reflex camera to carry out manual analysis, this method is a kind of method of direct measurement, be used to measure the size and the shape of water droplet individuality on the homogeneous plane, but the spectrum of measuring raindrop distributes, need obtain the individual data among the whole group of numerous quantity simultaneously, the inefficiency of manual analysis and technical difficulty is arranged, this also is the major technique bottleneck of its application of restriction.The subject matter of infusion method is that too fast big of falling speed dripped and can be broken through surface tension and immerse in the liquid, and this method is mainly used to calibrate the instrument that is used to measure the single-size raindrop size distribution, and can not be applied directly in the raindrop size distribution observation and analysis of natural precipitation.

By foregoing as seen, in photographic process, needing manual analysis raindrop image is the principal element of its practicality of restriction and validity, if can discern the raindrop target in the raindrop image automatically and extract relevant raindrop information then can address this problem effectively, and further improve its accuracy.

Can image pre-service denoising and target effective Feature Extraction be two important steps in the image recognition, also be the key that effectively discern the raindrop target.Because the raindrop image is subjected to dust, illumination, imaging to hold the influence of rain glass sheet self texture or camera when taking in the open, the raindrop image self contains certain noise, can cause adverse influence to raindrop identification.Therefore, need carry out the pre-service denoising to the raindrop image.Anisotropy diffusion smoothing filtering technique [Perona P, Malik J.Scale-space and edge detection using anisotropic diffusion[J] .IEEETransactions on Pattern Analysis and Machine Intelligence, 1990,12 (7): 629～639] compare with traditional image smoothing denoising method, both remove noise effectively, and can keep edge of image and important detailed information preferably again.The edge is the most basic feature of image, and it is present in target and background, target and target, zone and zone, between primitive and the primitive.In the raindrop image, significant border is arranged between raindrop target and the background, the marginal information that therefore can extract the raindrop target is used as describing its validity feature.Suo Beier operator (Sobel operator) [Sobel, I., Feldman, G.A 3x3Isotropic Gradient Operator for ImageProcessing.presented at a talk at the Stanford Artificial Project in 1968, unpublished but often cited, orig.in Pattern Classification and Scene Analysis, Duda, R.and Hart, P., John Wiley and Sons, ' 73:271-272] be a kind of effective edge extracting operator.

Summary of the invention

The invention provides a kind of raindrop recognition methods based on edge gradient, purpose is to overcome existing problem in the existing method, by raindrop target in the automatic identification raindrop image and the relevant raindrop information of extraction, to improve the automaticity and the precision of the observation of rainy weather phenomenon.A kind of raindrop recognition methods based on edge gradient of the present invention comprises the steps:

(1) image acquisition step is promptly obtained the raindrop image to be identified and the pairing background image thereof of size unanimity respectively from imaging device;

(2) image pre-treatment step is promptly carried out smoothing denoising to above-mentioned raindrop image to be identified that obtains and corresponding background image thereof respectively, obtain result images be respectively I (x, y) and B (x, y);

(3) image difference step, promptly to I (x, y) and B (x y) carries out difference operation, obtain difference image S (x, y)=| I (x, y)-B (x, y) |;

(4) edge gradient image extraction step, promptly extract difference image S (x, y) pairing edge gradient image E (x, y);

(5) edge gradient image binaryzation step, with threshold value Thd to edge gradient image E (x y) carries out binaryzation operation, obtain binaryzation edge gradient image T (x, y):

(6) morphological image treatment step, utilize the morphological image method to the edge gradient image T of binaryzation (x y) handles, and obtains final raindrop recognition result, promptly T (x, y) in gray-scale value be 255 connected domain, detailed process is as follows:

(6.1) (x y) carries out the morphological image ON operation to T;

(6.2) with T (x, y) in area be that the gray scale of 255 connected domain is changed to 0 less than threshold value MinArea and gray-scale value;

(6.3) (x y) carries out the morphological image closed operation to the T after handling through step (6.2);

(6.4) to the T after step (6.3) is handled (x, y) in gray-scale value be that the gray scale that 255 pixel proportion surpasses the image row or column of threshold value Ratio is changed to 0;

(6.5) (x y) carries out the morphological image closed operation to T again;

(6.6) just the T after step (6.5) is handled (x, y) in the length and/or wide corresponding predetermined value and the gray-scale value of surpassing of minimum boundary rectangle be 255 connected domain C _iGray scale be changed to 0;

(6.7) to the T after step (6.6) is handled (x, y) in gray-scale value be that 255 connected domain is carried out inner hole and filled;

(6.8) will through the T after step (6.7) is handled (x, y) in the long and short shaft length of minimum external ellipse be changed to 0 than the gray scale that greater than threshold value Ratio1 and gray-scale value is 255 connected domain;

(6.9) will through the T after step (6.8) is handled (x, y) in area be that the gray scale of 255 connected domain is changed to 0 less than threshold value MinArea and gray-scale value;

At this moment, T (x, y) in gray-scale value be that 255 connected domain is the raindrop that identify.

Raindrop information be can further extract to the above-mentioned raindrop that identify, raindrop target numbers, raindrop target mean diameter and mean diameter distribution profile comprised, wherein raindrop mean diameter MeanDia _iBe defined as:

${\mathrm{MeanDia}}_i=\sqrt{{\mathrm{MajorL}}_i*{\mathrm{MinL}}_i}$

MajorL _iBe the minimum external long axis of ellipse length of raindrop, MinL _iBe minor axis length;

The mean diameter distribution profile is defined as:

If the codomain of raindrop target mean diameter is RN=[MinDia, MaxDia], RN is divided into Num1 sub-codomain { RN _j, j=1,2 ... Num1-2, Num1-1, Num1}, raindrop target mean diameter distribution profile promptly refers to raindrop mean diameter MeanDia _iBe distributed in the number on each subvalue territory among the RN.

In the above-mentioned steps (2), described smoothing denoising is realized by adopting the filtering of anisotropy diffusion smoothing.

In the above-mentioned steps (4), described edge gradient image E (x, y) extract by Suo Beier operator (Sobel):

$E(x,y)=\sqrt{({(S(x,y)\⊗{\mathrm{ES}}_V)}^2+{(S(x,y)\⊗{\mathrm{ES}}_H)}^2)}$

Wherein

The expression convolution operation, ES _HWith ES _VIt is respectively horizontal and vertical template of Sobel operator.

In the above-mentioned steps (6), the morphological structure element SE that is adopted in the described morphological image operation is:

$\mathrm{SE}=\left[\begin{array}{ccc}0& 1& 0\\ 1& 1& 1\\ 0& 1& 0\end{array}\right]$ Or $\mathrm{SE}=\left[\begin{array}{ccccc}0& 0& 1& 0& 0\\ 0& 1& 1& 1& 0\\ 1& 1& 1& 1& 1\\ 0& 1& 1& 1& 0\\ 0& 0& 1& 0& 0\end{array}\right]$ Or $\mathrm{SE}=\left[\begin{array}{ccccccc}0& 0& 0& 1& 0& 0& 0\\ 0& 1& 1& 1& 1& 1& 0\\ 0& 1& 1& 1& 1& 1& 0\\ 1& 1& 1& 1& 1& 1& 1\\ 0& 1& 1& 1& 1& 1& 0\\ 0& 1& 1& 1& 1& 1& 0\\ 0& 0& 0& 1& 0& 0& 0\end{array}\right];$

Description of drawings

Fig. 1 is a schematic flow sheet of the present invention;

Fig. 2 is a raindrop image to be identified;

Fig. 3 is the pairing background image of raindrop image to be identified;

Fig. 4 is the anisotropy diffusion smoothing filtering result images of raindrop image to be identified;

Fig. 5 is the anisotropy diffusion smoothing filtering result images of background image;

Fig. 6 is a difference image;

Fig. 7 is the pairing edge gradient image of difference image;

Fig. 8 is the edge gradient image of binaryzation;

Fig. 9 is the result of morphological image ON operation;

Figure 10 is the result after the little connected domain of removal;

In the above-mentioned steps (6.6), establish described connected domain C_iThe height of minimum boundary rectangle be H_i, width is W_i，MT ₁、NT ₁、MT ₂And NT₂Be preset value, then:

If H_i≥MT ₁And W_i≥NT ₁Then with C_iGray scale be set to 0;

If H_i≥MT ₂Or W_i≥NT ₂Then with C_iGray scale be set to 0.

The present invention has been incorporated into the digital picture automatic identification technology in the observation of rainy weather phenomenon, can automatically identify the raindrop target in the raindrop image and extract relevant raindrop information. The present invention can be applicable to the observation of raindrop size distribution and obtains, thereby overcomes the problem that manual analysis raindrop image efficient is low and difficulty is big in the process of existing photographic process observation raindrop size distribution. Binding time information, the present invention can further be applied to rainfall statistics, in real time raininess monitoring and rainfall phase and microstructure analysis etc., in development of the national economy service important effect is arranged.

Figure 11 is the result of morphological image closed operation;

Figure 12 is the result that hole is filled;

Figure 13 is the result who removes the false target that shape do not meet the demands;

Shown in Figure 14 is the final recognition result of raindrop target.

Embodiment

The present invention with the filtering of anisotropic diffusion smoothing as the pretreated method of image, can discern the raindrop target effectively and extract relevant raindrop information by the combining image morphology methods as the validity feature of raindrop target with edge gradient information, treatment scheme as shown in Figure 1:

(1) image acquisition step is obtained the raindrop image to be identified and the pairing background image thereof of size unanimity respectively from imaging device.After background image is meant and has removed all raindrop in the recording geometry by certain condition, the image that imaging device is captured.Shown in Figure 2 is raindrop image to be identified, and shown in Figure 3 is the pairing background image of Fig. 2;

(2) image pre-treatment step is subjected to dust, illumination, imaging to hold the influence of rain glass sheet self texture or camera when taking owing to the raindrop image in the open, and the raindrop image self contains certain noise, can cause adverse influence to raindrop identification.Therefore, need carry out the pre-service denoising to the raindrop image.The present invention will utilize the technology of anisotropy diffusion smoothing filtering to treat identification raindrop image respectively and background image carries out smoothing denoising.Compare with traditional image smoothing denoising method, the advantage of anisotropy diffusion smoothing filtering is both to remove effectively noise, can keep edge of image and important detailed information preferably again.Fig. 4 and Fig. 5 shown in respectively are the anisotropy diffusion smoothing filtering results of raindrop image to be identified and background image;

(3) image difference step is carried out difference operation to the anisotropy diffusion smoothing filtering result images of raindrop image to be identified and background image thereof, obtains difference image.In the difference image raindrop target obtained outstanding, thereby and background is weakened even eliminate and help the raindrop identification of targets:

The anisotropy diffusion smoothing filtering result images of raindrop image to be identified and background image thereof be respectively I (x, y) and B (x, y), to both carry out the difference image S that difference operation obtains (x y) is defined as:

S(x，y)＝|I(x，y)-B(x，y)|

Shown in Figure 6 is difference image;

(4) edge gradient image extraction step, utilize the edge gradient image of edge extracting operator extraction difference image correspondence, present embodiment preferably utilize Suo Beier operator extraction difference image S (x, y) pairing edge gradient image E (x, y):

$E(x,y)=\sqrt{({(S(x,y)\⊗{\mathrm{ES}}_V)}^2+{(S(x,y)\⊗{\mathrm{ES}}_H)}^2)}$

Wherein

The expression convolution operation, ES _HWith ES _VIt is respectively horizontal and vertical template of Sobel operator.In the present embodiment, can choose ES _HWith ES _VIn the definition any one group, preferably be chosen for:

${\mathrm{ES}}_H=\left(\begin{array}{ccc}-1& -2& -1\\ 0& 0& 0\\ -1& -2& -1\end{array}\right)$ ${\mathrm{ES}}_V=\left(\begin{array}{ccc}-1& 0& -1\\ -2& 0& -2\\ -1& 0& -1\end{array}\right)$

The present invention will mainly utilize the validity feature of marginal information as the raindrop target, finish the identification to raindrop.Shown in Figure 7 is the pairing edge gradient image of difference image;

(5) edge gradient image binaryzation step, with threshold value Thd to edge gradient image E (x y) carries out binaryzation operation, obtain binaryzation edge gradient image T (x, y):

(x, y) middle gray-scale value is that 255 connected domain is potential raindrop target to T.Thd gets the integer between [15,20] in concrete the enforcement.Shown in Figure 8 is the edge gradient image of binaryzation;

(6) morphological image treatment step, utilize the morphological image method to the edge gradient image T of binaryzation (x y) handles, and obtains final raindrop recognition result, promptly T (x, y) in gray-scale value be 255 connected domain:

(6.1) (x y) carries out the morphological image ON operation to T.The purpose of carrying out the morphological image ON operation is that (x, y) some interference components in and raindrop target are separated and weakened, so that further remove with T.The morphological structure element SE that is adopted in concrete the enforcement is defined as follows:

$\mathrm{SE}=\left[\begin{array}{ccc}0& 1& 0\\ 1& 1& 1\\ 0& 1& 0\end{array}\right]$ Or $\mathrm{SE}=\left[\begin{array}{ccccc}0& 0& 1& 0& 0\\ 0& 1& 1& 1& 0\\ 1& 1& 1& 1& 1\\ 0& 1& 1& 1& 0\\ 0& 0& 1& 0& 0\end{array}\right]$ Or $\mathrm{SE}=\left[\begin{array}{ccccccc}0& 0& 0& 1& 0& 0& 0\\ 0& 1& 1& 1& 1& 1& 0\\ 0& 1& 1& 1& 1& 1& 0\\ 1& 1& 1& 1& 1& 1& 1\\ 0& 1& 1& 1& 1& 1& 0\\ 0& 1& 1& 1& 1& 1& 0\\ 0& 0& 0& 1& 0& 0& 0\end{array}\right]$

That shown in Figure 9 is the result of morphological image ON operation;

(6.2) just the T after step (6.1) is handled (x, y) in area be that the gray scale of 255 connected domain is changed to 0 less than threshold value MinArea and gray-scale value.Here mainly be that (x, y) interference component in such as noise are residual in order to remove T.The span of threshold value MinArea is [0.14,0.19] in concrete the enforcement, and unit is a square millimeter.Shown in Figure 10 is removes result after the little connected domain;

(6.3) (x y) carries out the morphological image closed operation to the T after handling through step (6.2).Here mainly be more continuous, reduce disappearance for the edge that makes the raindrop target.Adopt the morphological structure element identical in concrete the enforcement with (6.1).That shown in Figure 11 is the result of morphological image closed operation;

(6.4) to the T after step (6.3) is handled (x, y) in gray-scale value be that the gray scale that 255 pixel proportion surpasses the image row or column of threshold value Ratio is changed to 0.Here mainly be in order to remove the interference at place, image border.The span of threshold value Ratio is [0.7,0.9] in concrete the enforcement;

(6.5) (x y) carries out the morphological image closed operation, adopts the morphological structure element identical with (6.3) in concrete the enforcement to the T after handling through step (6.4).；

(6.6) will through the T after step (6.5) is handled (x, y) in the minimum boundary rectangle length of side to surpass predetermined value and gray-scale value be that the gray scale of 255 connected domain is changed to 0.Here mainly be in order to remove the excessive connected domain of some area size's features.If gray-scale value is arbitrary connected domain C of 255 _iThe height of minimum boundary rectangle be H _i, width is W _i, MT ₁, NT ₁, MT ₂And NT ₂Be the predetermined value of setting in advance:

If (6.6.1) H _i〉=MT ₁And W _i〉=NT ₁Then with C _iGray scale be changed to 0.MT in concrete the enforcement ₁Span be [20,27], NT ₁Span be [20,27], unit be the millimeter;

If (6.6.2) H _i〉=MT ₂Or W _i〉=NT ₂Then with C _iGray scale be changed to 0.MT in concrete the enforcement ₂Span be [27,35], NT ₂Span be [27,35], unit be the millimeter;

(6.7) to the T after step (6.6) is handled (x, y) in gray-scale value be that 255 connected domain is carried out inner hole and filled.Here mainly be for the further hypostazation of raindrop target, so both avoided T (x, y) in raindrop target situation nested against one another also help the extraction of follow-up raindrop information.That shown in Figure 12 is the result that hole is filled;

(6.8) will through the T after step (6.7) is handled (x, y) in the long and short shaft length of minimum external ellipse be changed to 0 than the gray scale that greater than predetermined value Ratio1 and gray-scale value is 255 connected domain.Here mainly be in order to remove the false raindrop target that shape does not meet the demands.The span of predetermined value Ratio1 is [4,8] in concrete the enforcement.Shown in Figure 13 is the result of being correlated with;

(6.9) will through the T after step (6.8) is handled (x, y) in area be that the gray scale of 255 connected domain is changed to 0 less than threshold value MinArea and gray-scale value.Here mainly be that (x, y) interference component in such as noise are residual in order to remove T.The span of threshold value MinArea is [0.14,0.19] in concrete the enforcement, and unit is a square millimeter, obtains final recognition result.T (x, y) in gray-scale value be that 255 connected domain is the raindrop target that is finally identified.Shown in Figure 14 is the colored final raindrop recognition result that shows;

(7) raindrop information extraction step, extract raindrop target numbers, raindrop target mean diameter and mean diameter distribution profile thereof:

(7.1) the number N um of raindrop target, promptly T (x, y) in gray-scale value be the number of 255 connected domain;

(7.2) establish raindrop target R _iMinimum external long axis of ellipse length is MajorL _i, minor axis length is MinL _i, the mean diameter MeanDia of this raindrop target then _iBe defined as:

${\mathrm{MeanDia}}_i=\sqrt{{\mathrm{MajorL}}_i\×{\mathrm{MinL}}_i}$

Add up the minimum external long axis of ellipse length M ajorL of each raindrop target _iWith minor axis length be MinL _i, can draw the mean diameter MeanDia of raindrop target _i

That (7.3) raindrop target mean diameter distribution profile DiaSpectrum describes is raindrop mean diameter MeanDia _iBe distributed in the number on each subvalue territory among the raindrop target mean diameter codomain RN.Wherein, the codomain of raindrop target mean diameter is RN=[MinDia in the raindrop image, MaxDia], MinDia is the minimum value of raindrop target mean diameter, MaxDia is the maximal value of raindrop target mean diameter, and RN is divided into Num1 sub-codomain { RN _j, j=1,2 ... Num1-2, Num1-1, Num1}, Num1 are natural number.Mean diameter MeanDia according to the raindrop target _iAnd raindrop target mean diameter distribution profile DiaSpectrum can be determined in each subvalue territory that above-mentioned steps draws.Num1 gets the integer between [20,30] in concrete the enforcement.

Claims (6)

1. the raindrop recognition methods based on edge gradient comprises the steps:

(1) image acquisition step is promptly obtained the raindrop image to be identified and the pairing background image thereof of size unanimity respectively from imaging device;

(2) image pre-treatment step is promptly carried out smoothing denoising to above-mentioned raindrop image to be identified that obtains and corresponding background image thereof respectively, obtain result images be respectively I (x, y) and B (x, y);

(3) image difference step, promptly to I (x, y) and B (x y) carries out difference operation, obtain difference image S (x, y)=| I (x, y)-B (x, y) |;

(4) edge gradient image extraction step, promptly extract difference image S (x, y) pairing edge gradient image E (x, y);

(5) edge gradient image binaryzation step, with threshold value Thd to edge gradient image E (x y) carries out binaryzation operation, obtain binaryzation edge gradient image T (x, y):

(6) morphological image treatment step, utilize the morphological image method to the edge gradient image T of binaryzation (x y) handles, and obtains final raindrop recognition result, promptly T (x, y) in gray-scale value be 255 connected domain, detailed process is as follows:

(6.1) (x y) carries out the morphological image ON operation to T;

(6.2) with T (x, y) in area be that the gray scale of 255 connected domain is changed to 0 less than threshold value MinArea and gray-scale value;

(6.3) (x y) carries out the morphological image closed operation to the T after handling through step (6.2);

(6.4) to the T after step (6.3) is handled (x, y) in gray-scale value be that the gray scale that 255 pixel proportion surpasses the image row or column of threshold value Ratio is changed to 0;

(6.5) (x y) carries out the morphological image closed operation to T again;

(6.6) just the T after step (6.5) is handled (x, y) in the length and/or wide corresponding predetermined value and the gray-scale value of surpassing of minimum boundary rectangle be 255 connected domain C _iGray scale be changed to 0;

(6.7) to the T after step (6.6) is handled (x, y) in gray-scale value be that 255 connected domain is carried out inner hole and filled;

(6.8) will through the T after step (6.7) is handled (x, y) in the long and short shaft length of minimum external ellipse be changed to 0 than the gray scale that greater than threshold value Ratio1 and gray-scale value is 255 connected domain;

(6.9) will through the T after step (6.8) is handled (x, y) in area be that the gray scale of 255 connected domain is changed to 0 less than threshold value MinArea and gray-scale value;

At this moment, T (x, y) in gray-scale value be that 255 connected domain is the raindrop that identify.

2. raindrop recognition methods according to claim 1 is characterized in that, can further extract raindrop information to the above-mentioned raindrop that identify, and comprises raindrop target numbers, raindrop target mean diameter and mean diameter distribution profile, wherein raindrop mean diameter MeanDia _iBe defined as:

${\mathrm{MeanDia}}_i=\sqrt{{\mathrm{MajorL}}_i*{\mathrm{MinL}}_i}$

MajorL _iBe the minimum external long axis of ellipse length of raindrop, MinL _iBe minor axis length;

The mean diameter distribution profile is defined as:

If the codomain of raindrop target mean diameter is RN=[MinDia, MaxDia], RN is divided into Num1 sub-codomain { RN _j, j=1,2 ... Num1-2, Num1-1, Num1}, raindrop target mean diameter distribution profile promptly refers to raindrop mean diameter MeanDia _iBe distributed in the number on each subvalue territory among the RN.

3. a kind of raindrop recognition methods based on edge gradient as claimed in claim 1 or 2 is characterized in that, in the above-mentioned steps (2), described smoothing denoising is realized by adopting the filtering of anisotropy diffusion smoothing.

4. as the described a kind of raindrop recognition methods of one of claim 1-3, it is characterized in that based on edge gradient, in the above-mentioned steps (4), described edge gradient image E (x, y) extract by Suo Beier operator (Sobel):

$E(x,y)=\sqrt{({(S(x,y)\⊗{\mathrm{ES}}_V)}^2+{(S(x,y)\⊗{\mathrm{ES}}_H)}^2)}$

Wherein

The expression convolution operation, ES _HWith ES _VIt is respectively horizontal and vertical template of Sobel operator.

5. as the described a kind of raindrop recognition methods of one of claim 1-4, it is characterized in that in the above-mentioned steps (6), the morphological structure element SE that is adopted in the described morphological image operation is based on edge gradient:

$\mathrm{SE}=\left[\begin{array}{ccc}0& 1& 0\\ 1& 1& 1\\ 0& 1& 0\end{array}\right],$ $\mathrm{SE}=\left[\begin{array}{ccccc}0& 0& 1& 0& 0\\ 0& 1& 1& 1& 0\\ 1& 1& 1& 1& 1\\ 0& 1& 1& 1& 0\\ 0& 0& 1& 0& 0\end{array}\right]$ Or $\mathrm{SE}=\left[\begin{array}{ccccccc}0& 0& 0& 1& 0& 0& 0\\ 0& 1& 1& 1& 1& 1& 0\\ 0& 1& 1& 1& 1& 1& 0\\ 1& 1& 1& 1& 1& 1& 1\\ 0& 1& 1& 1& 1& 1& 0\\ 0& 1& 1& 1& 1& 1& 0\\ 0& 0& 0& 1& 0& 0& 0\end{array}\right].$

6. as the described a kind of raindrop recognition methods of one of claim 1-5, it is characterized in that, in the above-mentioned steps (6.6), establish described connected domain G based on edge gradient _iThe height of minimum boundary rectangle be H _i, width is W _i, MT ₁, NT ₁, MT ₂And NT ₂Be preset value, then:

If H _i〉=MT ₁And W _i〉=NT ₁Then with C _iGray scale be changed to 0;

If H _i〉=MT ₂Or W _i〉=NT ₂Then with C _iGray scale be changed to 0.

Images