CN103617425B - For monitoring the generation method of the movable variation track of aurora - Google Patents

Abstract

The invention discloses a kind of generation method for monitoring the movable variation track of aurora, the extracting method of a kind of all-sky aurora image motion field based on aurora fluid behaviour and multiple dimensioned movement characteristic is proposed, this method can be according to the resolution of current data and the character adaptive selection Regularization Strategy of aurora activity, the sports ground being then based on extracting realizes characterizing aurora video, and the difference between tolerance aurora video sequence is in order to monitor the change of aurora activity further.Characterizing method based on sports ground effectively embodies the abundant Two-dimensional morphology feature of aurora and motor pattern, the variation track of generation and then can position the position that aurora are undergone mutation exactly, provides new means for the research of space physics.

Description

For monitoring the generation method of the movable variation track of aurora

Technical field

The invention belongs to pattern recognition and Intelligent treatment technical field, the sports ground relating to video extracts and dynamic texture table Levy, can be used for monitoring the movable change of aurora and estimating the activity cycle of aurora, be used for monitoring aurora particularly to one movable The generation method of variation track.

Background technology

Aurora are the geophysical phenomenas with polar region feature that people uniquely can with the naked eye be directly observed, and are polar regions The form of expression concentrated most on day ground physical process (particularly magnetosphere-ionosphere interact).At present, obtain aurora data to have Various ways, including all-sky imaging device, auroral radar system, magnetometer array, cosmic noise measuring device array and son Noon line Scanning photometer array etc..And the advantage of the all-sky aurora data that this project uses is to obtain the form of two dimension Information, it is achieved the Continuous Observation movable to aurora.

For the characteristic of fluid, in 2002, Corpetti et al. document " Corpetti, T., M é min,,Pé rez,P.Dense estimation of fluid flows.IEEE Transactions on Pattern Analysis And Machine Intelligence, 24 (3): 365-380,2002. " in, introduce the continuity equation in hydrodynamics As the basis of composition data constraint, extract the sports ground of aerography picture in conjunction with second order divergence-curl regularization method.Thing In reality, aurora are interacted by solar wind and magnetic field of the earth and produce, and period produces and discharge up to 10⁶The energy of MW, These energy flow to the upper atmosphere by electric current along electric field line.These high energy charged particles are transported under the effect of electric field line Dynamic, the period number of photons flux of release is the highest, and the brightness of image is the highest, so aurora have the character of fluid.Therefore, pole is estimated During photomovement, data constraint equation can be based on continuity equation.But, in terms of regularization method, the side of Corpetti et al. Method have employed single second order divergence-curl regularization factors, it is impossible to meets the feature that aurora motion is multiple dimensioned.

Because single order smooths regularization factorsLimit motion vector in neighborhood spatial variations (Represent Spatial gradient), use single order to smooth regularization factors and the motion vector in neighborhood is reached unanimity.It should be noted that single order Regularization factorsWith single order divergence-curl regularization factorsIn optimization problem Lagrange's equation identical (this conclusion see document " Corpetti, T., M é min,,Pérez,P.Dense estimation of fluid flows.IEEE Transactions on Pattern Analysis and Machine Intelligence, 24 (3): 365-380,2002. "), so using the motion that single order regularization method can more be smoothed Vector field, and there is less divergence and curl.Second order divergence-curl regularization methodPunishment Divergence and the change of curl, be not intended to divergence and curl itself, and the most directly limit the spatial variations of motion vector.So it is right In smooth motion, use single order to smooth regularization factors and can obtain real sports ground, and use second order divergence-curl canonical Changing the factor equally to obtain close to real sports ground, different regularization methods, the sports ground difference obtained is little.And it is right In rotary motion, use second order divergence-curl regularization factors can obtain real sports ground, use single order regularization factors Then can limit divergence and the curl of motion.

To sum up, existing method for estimating is primarily present following two problem for Auroral video image:

1) during aurora move, the volume of energetic particles stream, brightness and shape all can change.These grains Subflow is compressible, transparent, or instantaneous, and object of observation does not have rigid nature.So, traditional based on brightness phase The sports ground method of estimation of closing property is not suitable for aurora data.

2) use the sports ground that the motion estimation algorithm of single order regularization method can more be smoothed, and use second order The sports ground algorithm for estimating of regularization method is more suitable for rotary motion.Present stage, almost all of sports ground method of estimation is all adopted Solve the nonlinear estimation problem of sports ground with multi-scale method, Pyramid transform will be carried out, then different by video image Motion estimation algorithm is implemented on Pyramid transform layer.But these researchs for the motion of different scale have employed duplicate just Then changing method, this just gives tacit consent to aurora and has identical motion feature under different scale.The motion of aurora is extremely complex, different chis The forms of motion of degree has different features.So, estimate that aurora motion uses unified regularization method to be inappropriate.

Summary of the invention

The present invention provides a kind of generation method for monitoring the movable variation track of aurora, proposes a kind of based on aurora stream The extracting method of the all-sky aurora image motion field of bulk properties and multiple dimensioned movement characteristic, the data constraint of this method is based on even Continuous equation, and can be according to the resolution of current data and the character adaptive selection Regularization Strategy of aurora activity.So After realize aurora video is characterized based on the sports ground that extracts, the difference between tolerance aurora video sequence is in order to supervise further Survey the change that aurora are movable, for the means that the research offer of space physics is new.

For achieving the above object, the technical scheme is that

Step 1: input one section of aurora video, and Auroral video image original to each frame carries out pretreatment, original whole day The size of empty aurora image is 512 × 512, and the dynamic range of pixel intensity is [0,18000].Original all-sky aurora image Preprocessing process is divided into five steps: (1) deducts dark current；(2) edge noise is removed；(3) gray scale stretching；(4) image rotation Turn；(5) image cutting-out.Image size through pretreatment is 440 × 440, and gray scale dynamic range is [0,4000], the figure obtained As Sequence composition all-sky aurora video sequence S_a=I (x, t), x ∈ Ω, t=1,2 ..., T}, wherein Ω represents all-sky pole The circular masked areas of light image, T represents the length of aurora video sequence, and x represents the locus of pixel, t express time；

Step 2: build Pyramid transform layer decision training collection, every piece image I that Pyramid transform layer decision training is concentrated Set up this Pyramid transform of six floor heights, obtain I₀,…,I₅, wherein I₀For the image of original scale, I₀=I, I₁For carry out a floor height this The image of Pyramid transform, I₂For carrying out the image of two these Pyramid transform of floor height, I₃For carrying out the image of three these Pyramid transform of floor height, I₄For carrying out the image of four these Pyramid transform of floor height, I₅For carrying out the image of five these Pyramid transform of floor height；

Step 3: utilize the mean flow rate of aurora image and dutycycle the two characteristic function to judge that gaussian pyramid decomposes Whether layer retains the essential information of aurora, and the flex point of characteristic function is the position that development trend starts to change, by finding The flex point of two characteristic functions obtains the optimum number of strata l of Pyramid transform_opt；

Step 4: smooth regularization factors and the feature of second order divergence-curl regularization factors according to single order, constructs one New index regularization Scatter-difference D_regJudge the character that each Gauss Decomposition layer moves, thus adaptive selection regularization side Case；

Step 5: randomly select 500 width picture construction regularization scheme decision-making value training from all-sky aurora data base Collection, to the image configuration different scale in training set and the translational motion of different directions and rotary motion, asks translational motion respectively Regularization Scatter-difference D with rotary motion_reg, D to regularization program decisions Threshold-training collection_regIt is trained, obtains canonical The decision-making value Z of change scheme；

Step 6: for one section of aurora video sequence S to be monitored_a, set up 0-l_optLayer Pyramid transform layer, all meets D_reg(i) The Pyramid transform layer of ＞ Z, selects second order divergence-curl regularization factors；Meet D_regI the Pyramid transform layer of ()≤Z, selects one Rank smooth regularization factors.On each decomposition layer, utilize data constraint based on continuity equation, and combining adaptive is just selecting Then change the factor, try to achieve motion vector field sequence V={v (x, t), x ∈ Ω, t=1,2 ..., T-1}, v (x, t) represent be positioned at space Coordinate x and the motion vector of time coordinate t；

Step 7: the change movable in order to monitor aurora, is 2t by window width_wTime sliding window slide to from video original position End position, the step-length every time slided is a frame, and the aurora motion vector field sequence in the t time sliding window is divided into front t_wFrame and after t_wFrame subsequence, extracts these two sections of sequences Statistical Charateristics f based on partial vector difference_pre(t) and f_post(t)；

Step 8: measure the difference of aurora video sequence before and after in current sliding window, derive f_pre(t) and f_post(t) Chi-square statistical distance, is assigned to the variation track value in t by value:

d_chg(t)=χ²(f_pre(t), f_post(t))

Chi-square statistical distance is defined as:

${\mathrm{\χ}}^2(p,q)={\mathrm{\Σ}}_i\frac{{(p_i-q_i)}^2}{p_i+q_i},$

Wherein, i represents the index of two vectorial p and q, d_chgT () reflects the movable situation of change in t of aurora.

The method that the present invention moves with existing monitoring aurora compares, and has the advantage that

(1) present invention take into account the fluid properties of aurora, feature changeable for aurora form, self luminous, introduces base In continuity equation as data constraint, overcome that the constant hypothesis of brightness that existing method uses is incongruent with aurora phenomenon asks Topic, improves the accuracy that aurora sports ground is estimated.

(2) present invention considers the feature that aurora motion is extremely complex, designs the selection strategy of multiple dimensioned regularization scheme. According to current resolution and the movement characteristic of object of observation, self adaptation determines the number of plies of Pyramid transform and determines every layer of employing Regularization factors so that the motion of different under different scale aurora is recovered the most accurately.

(3) motion of aurora has uncertain timeliness and spatiality, but in neighborhood, the motion of pixel has substantially Dependency.The present invention, based on the aurora sports ground extracted, uses space-time statistics based on partial vector difference to characterize aurora sequence Row, partial vector difference considers the Local Phase in neighborhood to sports ground.

(4) simulation result shows, the present invention has efficiently extracted the transient motion field of aurora, and it is poor based on partial vector to use Space-time statistics characterize aurora sequence, the difference between tolerance aurora video sequence, located aurora exactly and undergo mutation Position.Because sports ground reflects the variation tendency of each pixel on image, compare the method for existing Keogram figure more Effectively, the change that aurora are movable is shown all sidedly.

Accompanying drawing explanation

Fig. 1 is the general flow chart that the present invention obtains for monitoring the movable variation track of aurora；

Fig. 2 is the sub-process figure that in the present invention, all-sky aurora video sequence sports ground extracts；

Fig. 3 is the sub-process figure generating aurora variation track in the present invention；

Fig. 4 is the algorithm schematic diagram that in the present invention, aurora video sequence based on partial vector difference characterizes；

Fig. 5 is that the present invention is for monitoring the comparison of the movable variation track of 2003/12/24 aurora and Keogram figure；

Fig. 6 is that the present invention is for monitoring the comparison of the movable variation track of 2003/12/25 aurora and Keogram figure；

Fig. 7 is the catastrophe one that the Keogram figure using the variation track of the present invention to monitor fails to show；

Fig. 8 is the catastrophe two that the Keogram figure using the variation track of the present invention to monitor fails to show.

Specific implementation method

The technical thought realizing the present invention is: first original All-sky image is carried out pretreatment, to pretreated All-sky aurora sequence uses data constraint equation based on continuity equation, the non-uniform regularization scheme of combining adaptive, extracts The transient motion field of all-sky Auroral video image, then the partial vector of calculation of motion vectors is poor, and adds up office on time-space domain The two-dimensional histogram of portion's vector difference is used for characterizing aurora sequence, finally calculates the χ of adjacent two sections of aurora videos²Distance, obtains pole The variation track of photo-reactive, is embodied as step as follows:

Below by specific embodiment, the present invention will be further described:

As it is shown in figure 1, the generation method for monitoring the movable variation track of aurora of the present invention, step is as follows:

Step 1: input one section of all-sky aurora video, each two field picture is carried out pretreatment.Take from Arctic Yellow River Station The size of original all-sky aurora image is 512 × 512, and the dynamic range of pixel grey scale is [0,18000].Original image pre- Processing procedure is divided into five steps: (1) deducts dark current；(2) edge noise is removed；(3) gray scale stretching；(4) image rotation；(5) Image cutting-out.The circular masks utilizing radius to be 220 removes the noises such as the light around Yellow River Station, mountain range, is then passed through cutting out Removing the extraneous region outside mask, the image size through pretreatment is 440 × 440, and gray scale dynamic range is [0,4000].Treat Preprocessed image construction sequence S obtained of aurora video sequence of monitoring_a=I (x, t), x ∈ Ω, t=1,2 ..., T}, its Middle Ω represents the circular masked areas of all-sky aurora image, and T represents the length of aurora video sequence.

Step 2: with reference to Fig. 2, builds Pyramid transform layer decision training collection, first calculates the all-sky aurora figure in training set As in the mean flow rate of six Pyramid transform layers and duty cycle parameters, thereby determining that the Pyramid transform number of plies of optimum.To training set In all-sky aurora image configuration different scale and the translation in direction and rotary motion, calculate regularization Scatter-difference respectively, really Determine regularization method decision-making value Z, adaptively determine regularization scheme according to Z, in conjunction with data constraint based on continuity equation Obtain the transient motion field of all-sky aurora video sequence.

2.1) number of plies of Pyramid transform in regularization scheme is determined:

2.1.1) data randomly selecting continuous 12 hours from all-sky aurora data base (comprise 4320 width all-skies Image) constitute Pyramid transform layer decision training collection.Each width all-sky aurora image I that Pyramid transform layer decision training is concentrated Carry out Gauss turriform 6 layers decomposition, be designated as I respectively₀,I₁,I₂,I₃,I₄,I₅, wherein I₀For the image of original scale (440 × 440), I₀=I.I₁Image size be 220 × 220, by that analogy, I₅Image size be 14 × 14.Original image I₀As Gauss gold The zero layer of word tower, the l layer of gaussian pyramid constructs as follows: existing by I_l-1With a window function having low-pass characteristic Carry out convolution, then convolution results made the interlacing down-sampling every row, it may be assumed that

$I_l=\underset{m=-2}{\overset{2}{\mathrm{\&Sigma;}}}\underset{n=-2}{\overset{2}{\mathrm{\&Sigma;}}}w(m,n)I_{l-1}(2i+m,2j+n),0<l<\mathrm{6,1}\&le;i,j<C_l$

Wherein, (i, j) be in image i-th row jth row coordinate, C_lFor line number and the columns of turriform l tomographic image, w (m, n) be 5 × 5 window function, the present invention use w (m, n) is defined as:

$w=\left[\begin{array}{ccccc}0.0030& 0.0133& 0.0219& 0.0133& 0.0030\\ 0.0133& 0.0596& 0.0983& 0.0596& 0.0133\\ 0.0219& 0.0983& 0.1621& 0.0983& 0.0219\\ 0.0133& 0.0596& 0.0983& 0.0596& 0.0133\\ 0.0030& 0.0133& 0.0219& 0.0133& 0.0030\end{array}\right]$

2.1.2) each decomposition layer to image Gauss turriform, seeks mean flow rate f in circular masked areas Ω_a(l):

$f_a\left(l\right)=\underset{(i,j)\&Element;{\mathrm{\&Omega;}}_l}{\mathrm{\&Sigma;}}{I}_l(i,j)/\mathrm{\&pi;}{R}_l^2$

Wherein f_aL () is the mean flow rate function about the change of number of plies l, Ω_lIt it is the circular masks district of l floor monument diagram picture Territory, (i, j) ∈ Ω_lRepresent the pixel only adding up circular masked areas, R_lIt it is the radius of the circular masked areas of l layer turriform.

2.1.3) utilize partitioning algorithm based on Darwin's particle group optimizing that each decomposition layer image of Gauss turriform is carried out Image is split, Q_lRepresent the auroral region segmentation result of l tomographic image.Calculate the dutycycle of each decomposition layer image of Gauss turriform Parameter f_b(l):

$f_b\left(l\right)=\underset{(i,j)\&Element;{\mathrm{\&Omega;}}_l}{\mathrm{\&Sigma;}}Y(i,j)/\mathrm{\&pi;}{R}_l^2.Y(i,j)=\left\{\begin{array}{cc}1,& p(i,j)\&Element;Q_l\\ 0,& p(i,j)\&NotElement;Q_l\end{array}\right.$

Wherein f_bL () is the duty cycle functions about the change of number of plies l, represent that the various aurora occurred in All-sky image exist Ratio shared in whole sky background, p (i, j) be coordinate (i, j) corresponding pixel, Y (i, j) represent pixel p (i, j) whether Belonging to the logical value of auroral region, (i, j)=1 means that (i, j) belongs to aurora generation area to pixel p, and (i, j)=0 represents Y to Y (i j) belongs to sky background region to pixel p.

2.1.4) in order to determine function f_a(l) and f_bL () development trend starts the number of plies changed, calculate f respectively_a(l) And f_bL the second dervative of (), tries to achieve mean flow rate and the flex point of dutycycle development trend generation great change:

l_a=argmax (f_a″(l))-1

l_b=argmax (f_b″(l))-1

2.1.5) select in the decomposition layer that two function flex points occur less for optimal Decomposition layer l_opt:

l_opT=min [l_a,l_b]

Above formula explanation mean picture brightness and duty cycle functions are respectively at l_a+ 1 and l_b+ 1 layer there is great change, imageLost too much information, have been out characterizing the ability of the aerial aurora in sky.Experimental result is showed more than The image that the Pyramid transform layer decision training of 88.96% is concentrated selects to carry out four layers of Pyramid transform, will the image of original resolution It is decomposed into I₀,…,I₃。

2.2) from all-sky aurora data base, randomly select 500 width All-sky images constitute regularization scheme decision-making value Training set, smooths the sports ground method of estimation of regularization factors respectively with data constraint based on continuity equation associating single order (fluid-1) and data constraint based on continuity equation associating second order divergence-curl regularization factors sports ground method of estimation (fluid-2) solve known motion, according to regularization Scatter-difference, determine the regularization used on this Pyramid transform layer of each floor height Method.

2.2.1) every piece image that regularization program decisions Threshold-training is concentrated, enforcement 4 directions up and down, 10 The translational motion of individual different scale, is designated as M_S(i), 1≤i≤40；And construct both direction clockwise and anticlockwise, 20 angles The rotary motion of degree, is designated as M_R(i), 1≤i≤40.Thus 80 the different known motions obtaining piece image, regularization side 500 width images in case decision-making value training set just obtain 40000 known motions, including 20000 translational motions and 20000 Individual rotary motion.

2.2.2) being utilized respectively fluid-1 and fluid-2 and estimate this two groups of known motions, translational motion utilizes fluid-1 The sports ground obtained with fluid-2 method is designated as respectivelyWithAnd rotary motion utilizes fluid-1 and fluid-2 method The sports ground obtained is designated as respectivelyWith

2.2.3) the regularization Scatter-difference of translational motion is calculatedAnd rotary motion Regularization Scatter-difference $|{\mathrm{ave}}_{\mathrm{\&Omega;}}\left(\mathrm{div}\left(V_R^1\right)\right)-{\mathrm{ave}}_{\mathrm{\&Omega;}}\left(\mathrm{div}\left(V_R^2\right)\right)|.$

2.2.4) for 20000 translational motions, the distribution histogram P of the regularization Scatter-difference of translational motion is obtained_S, pin To 20000 rotary motions, obtain the distribution histogram P of the regularization Scatter-difference of rotary motion_R.Ask that both moves about The intersection point of the histogram functions of regularization Scatter-difference, intersection point location is regularization method decision-making value Z:

Z=arg (P_R(x)=P_S(x))

2.3) All-sky image sequence is set up 0-l_optPyramid transform layer, and calculate 0-l_optThe Gauss Pyramid transform of layer Regularization Scatter-difference D_reg(l).All meet D_regL the Pyramid transform layer of ()≤Z, selects single order to smooth regularization factors；Meet D_regL the Pyramid transform layer of () ＞ Z, selects second order divergence-curl regularization factors.On each Pyramid transform layer, perform base In the sports ground method of estimation of the non-uniform regularization factors of data constraint combining adaptive of continuity equation, obtain whole day to be monitored Motion vector field sequence V={v that empty aurora video sequence is corresponding (x, t), x ∈ Ω, t=1,2 ..., T-1}, global object side Cheng Wei:

$\min ({\&Integral;}_{\mathrm{\&Omega;}}{f}_1\left({\mathrm{\&xi;}}_1\right)+\mathrm{\&alpha;}{f}_2\left({\mathrm{\&xi;}}_2^i\right))$

Wherein

ξ₁=E (x+d (x), t+ Δ t) exp (divd (x))-E (x, t)

${\mathrm{\&xi;}}_2^1=|\&dtri;d(x,t)|$

${\mathrm{\&xi;}}_2^2={|\&dtri;\mathrm{div}d|}^2+{|\&dtri;\mathrm{curl}d|}^2$

Wherein ξ₁It is data constraint equation based on continuity equation,It is that single order smooths regularization factors,It is that second order dissipates Degree-curl regularization side the factor, (x, t) represents the brightness of image to E, and x indicates locus, t express time, and d estimates for needs Sports ground, and f₁And f₂Being respectively the penalization equation of secondary, α represents weighting parameter, and it is for coordination data constraint and smooths Relation between the factor.

Step 3: with reference to Fig. 3, time sliding window is slided to end position from video original position, by the pole in the t time sliding window Light video sequence is divided into the subsequence that before and after two segment length is equal, calculates the based on partial vector difference of the two subsequence respectively Statistical Charateristics, calculates the difference between the two continuous sequence.Gradually move sliding window, until video end, exporting change Track.

3.1) inputting the motion vector field sequence that video to be monitored is corresponding, making window width is 2t_wTime sliding window from sequence Beginning position starts mobile, and the step-length every time slided is a frame.

3.2) by t(t_w<t≤T-t_w,t_w< < T) vector field sequence corresponding in secondary sliding window is divided into first half cross-talk sequence (t- t_w..., t-1) and second half section subsequence (t ..., t+t_w-1), V it is designated as respectively_pre(t) and V_postT (), length is all t_wFrame.Point Do not calculate V_pre(t) and V_postT () is at time-space domain { Ω × (t-t_w..., t-1) } and Ω × (t ..., t+t_w-1) } based on office The Statistical Charateristics of portion's vector difference, is then normalized operation, and this algorithm flow is with reference to Fig. 4.

3.2.1) spatial neighborhood Ν is calculated_xInterior partial vector is poor:

$v_{\mathrm{LVD}}^d(x,t)=v(x,t)-v(x+d,t)$

Wherein pixel x and the distance of other pixels in field centered by d, it illustrates a kind of neighborhood relationships, and x, x+d ∈Ν_x, Ν_xIt is chosen in t frame the Square Neighborhood of pixel centered by x.

3.2.2) on time-space domain, the partial vector under current neighborhood relation is added up poor.By space-time neighborhood { Ω × (t- t_w..., t-1) } in allIt is projected on the polar coordinate plane split in advance, adds up each regionObtain the space-time statistic histogram of partial vector difference under current neighborhood relationIn like manner obtain

3.2.3) cascade obtaining space-time statistic histogram under difference neighborhood relationships And it is normalized operation, obtains the first half section sports ground sequence vector V in the t time sliding window_preDuring the partial vector difference of (t) Empty statistical nature, is designated as f_pre(t).In like manner obtain second half section sports ground sequence vector V_postThe partial vector difference space-time statistics of (t) Feature, is designated as f_post(t)。

3.3) measure the difference of aurora video sequence before and after in current sliding window, calculate f_pre(t) and f_postThe chi-of (t) Square statistical distance, is assigned to the variation track value in t by value:

d_chg(t)=χ²(f_pre(t), f_post(t))

Chi-square statistical distance is defined as:

${\mathrm{\&chi;}}^2(p,q)={\mathrm{\&Sigma;}}_i\frac{{(p_i-q_i)}^2}{p_i+q_i},$

Wherein, i represents the index of characteristic vector, d_chgT () reflects the movable situation of change in t of aurora.

3.4) time sliding window is moved to subsequent time, it is judged that whether sliding window moves to video end, if it is, output becomes Change track, otherwise, return to step 3.2.

Advantages of the present invention can be described further by following experiment:

Experimental data used in the present invention is from the Arctic Yellow River Station observation (2003 11 of surviving the winter of 2003-2004 years -2004 years 02 month moon).All-sky aurora image is the data of G-band (557.7nm), adjacent two two field pictures be spaced apart 10 seconds. This cover system during polar night can 24 hours every days Continuous Observation, whole day can obtain the most about 8640 width all-sky aurora figures Picture, including 4320 width days side (03:00-15:00UT(universal time)) observe data and 4320 width sides at night observation data.

The present invention utilizes said method to monitor the aurora activity change of observation of surviving the winter for 2003-2004 years, in order to verify this The effectiveness of invention, contrasts the variation track of output with corresponding Keogram figure.

Experiment one: the variation track of aurora activity and the comparison of Keogram figure

Keogram figure is the change movable in order to observe aurora, so as to the activity trend of fast browsing aurora, and The generation of location aurora event, a kind of traditional method used in space physics field.It is the all-sky aurora to each section Video, extracts auroral intensity data along the magnetic meridian of earth magnetism North and South direction in each frame aurora image, with the time For sequence, it is arranged in aurora activity diagram.Keogram figure can substantially reflect the change of aurora brightness of image, but it only extracts The information of a line in piece image, the aurora activity of reflection is limited.Variation track is to measure aurora activity therewith The situation of change that the aurora of front generation are movable, the change of the value the biggest expression aurora on curve is the most violent, is worth the least explanation aurora Motor pattern occur change the least.

Fig. 5 be the present invention for monitor the movable variation track of 2003/12/2411:00-15:00 (UT) aurora with The comparison of Keogram figure.The variation track of Fig. 5 mainly has 5 local peakings, and the Keogram of comparison diagram 5 figure is it can be seen that the One local peaking is positioned at 12:23(UT), at 12:23(UT) in the past, there is (the all-sky pole, region to the north of Yellow River Station in aurora The top half of light image), and the brightness of aurora and movement velocity the most relatively relax.In first local peaking and Between two local peakings (12:55UT), the aurora of multi sphere structure start to the south mobile, and finally this movement tendency gradually disappears Moving back, the brightness of aurora starts dimmed.From the beginning of second local peaking, the activity of aurora there occurs significantly change, becomes clear Arc structure occurs again, and moves to direction to pole.3rd innings portion's peak value occurs at 13:07(UT), the last period Aurora activity end, aurora suddenly disappear.From 13:23(UT) from the beginning of, auroral brightness brightens suddenly, and aurora have pole to fortune Dynamic trend.This trend continues up to last local peaking (13:53(UT)).After 13:53(UT), aurora are little by little Disappear.Obviously, the brightness of these period aurora, shape and generation position the most all have a very large change, and variation track is all Can accurate measurements to change generation.

Fig. 6 is that the present invention is for monitoring 2003/12/2512:00-15:00(UT) the movable variation track of aurora with The comparison of Keogram figure.The continuity of the Keogram figure of Fig. 6 is relatively low, illustrates that the change of aurora activity is the most unexpected.From Fig. 6's Variation track is it can be seen that this section of aurora activity there occurs five great changes, and five local peakings of variation track occur Position located the moment that aurora are undergone mutation accurately.Moreover, the size of the value of variation track directly reflects aurora The severe degree of activity change.Such as, at 12:44(UT), the value of variation track has reached maximum, and the ripple of variation track Shape is also the most sharp-pointed, illustrates that movable there occurs in this moment of aurora is obviously suddenlyd change.Keogram figure shows 12:44 (UT), before, sky occurs almost without aurora in the air, 12:44(UT) after, the brightness of aurora increases severely, and auroral arc is towards pole Move to direction.By contrast, occur at 13:51(UT) local peaking relatively low, the waveform comparison of variation track relaxes, Illustrate that the change that aurora occur at this moment is the most violent.Comparison Keogram figure it is also seen that, although 13:51(UT) time The brightness of aurora and location all there occurs change, but mitigation is compared in change, and Keogram figure is also the most coherent.

Experiment two: the Keogram figure that variation track monitors fail display catastrophe

The left figure of Fig. 7 is that the generation using the variation track of the present invention to monitor is at 2004/01/1604:50-05:30 (UT) aurora activity change, right figure is the Keogram figure of corresponding period.Be can be seen that aurora are at this by the variation track on the left side One seasonal change is very frequent, and the Keogram figure on the right fails to demonstrate the aurora phenomenon that these are active.According to variation track Instruction, by observing aurora sequence it was found that occur at aurora form, brightness, position and the motion mould of this period Formula change is very fast.There is the position of peak-peak at 05:07:53(UT in variation track), the generation that bottom panel show of Fig. 7 exists All-sky aurora image near this moment.By shown aurora sequence it can be seen that 05:07:53(UT) before, aurora go out Showed obvious turbulent structure, 05:07:53(UT) after this special aurora structure be wholly absent.The continuous vortex knot of aurora Structure is a kind of important aurora phenomenon studying day ground physical process, but cannot monitor its by browsing Keogram figure Occur.Because Keogram figure is simply extracted the information of a line on aurora image, for the performance colourful two dimension of aurora Morphological characteristic and motor pattern, Keogram schemes then complete failure.

The left figure of Fig. 8 is that the generation using the variation track of the present invention to monitor is at 2004/01/1705:20-06:00 (UT) aurora activity change, the Keogram figure of Fig. 8 is shown that the situation of change of auroral brightness on same period meridian. Be can be seen that aurora are at 05:32(UT by variation track) near there occurs sudden change, this result can be by shown in figure below of Fig. 8 All-sky aurora sequence is confirmed.At 05:32:02(UT) before, the main hot spot region that sky occurs in the air in the upper right corner, warp Crossing the development of four minutes, this aurora speck fades away.From 05:32:02(UT) after, a bright auroral arc is from the left side Start occur, until disappearing, having the most again an auroral arc with pleated structure to start development from zenith position.Due to Keogram figure is the auroral intensity data extracted in each frame aurora image along the magnetic meridian of earth magnetism North and South direction, To the luminance patterns not appeared on the meridian of north and south, then cannot find that these aurora are movable by observing Keogram figure.Combine Upper described, if browsing aurora change by observing Keogram figure, it is most likely that many great aurora events can be missed, from And data cannot be provided to ensure for research aurora phenomenon physical mechanism behind.

Claims (3)

1., for monitoring a generation method for the movable variation track of aurora, step is as follows:

Step 1: input one section of aurora video, and Auroral video image original to each frame carries out pretreatment, original all-sky pole The size of light image is 512 × 512, and the dynamic range of pixel intensity is [0,18000], the pre-place of original all-sky aurora image Reason process is divided into five steps: (1) deducts dark current；(2) edge noise is removed；(3) gray scale stretching；(4) image rotation； (5) image cutting-out, the image size through pretreatment is 440 × 440, and gray scale dynamic range is [0,4000], the image obtained Sequence composition all-sky aurora video sequence S_a={ (x, t), x ∈ Ω, t=1,2, Λ, T}, wherein Ω represents all-sky aurora to I The circular masked areas of image, T represents the length of aurora video sequence, and x represents the locus of pixel, t express time；

Step 2: build Pyramid transform layer decision training collection, the every piece image I concentrating Pyramid transform layer decision training sets up This Pyramid transform of six floor heights, obtains I₀,Λ,I₅, wherein I₀For the image of original scale, I₀=I, I₁For carrying out this turriform of a floor height The image decomposed, I₂For carrying out the image of two these Pyramid transform of floor height, I₃For carrying out the image of three these Pyramid transform of floor height, I₄For Carry out the image of four these Pyramid transform of floor height, I₅For carrying out the image of five these Pyramid transform of floor height；

Step 3: utilize the mean flow rate of aurora image and dutycycle the two characteristic function to judge that gaussian pyramid decomposition layer is The essential information of no reservation aurora, the flex point of characteristic function is the position that development trend starts to change, by finding two The flex point of characteristic function obtains the optimum number of strata l of Pyramid transform_opt；

Step 4: smooth regularization factors and the feature of second order divergence-curl regularization factors according to single order, construct one new Index regularization Scatter-difference D_regJudge the character that each Gauss Decomposition layer moves, thus adaptive selection regularization scheme；

Step 5: randomly select 500 width picture construction regularization scheme decision-making value training sets from all-sky aurora data base, To the image configuration different scale in training set and the translational motion of different directions and rotary motion, ask translational motion and rotation respectively The regularization Scatter-difference D that transhipment is dynamic_reg, D to regularization program decisions Threshold-training collection_regIt is trained, obtains regularization side The decision-making value Z of case；

Step 6: for one section of aurora video sequence S to be monitored_a, set up 0-l_optLayer Pyramid transform layer, all meets D_reg(i) > Z Pyramid transform layer, selects second order divergence-curl regularization factors；Meet D_regI the Pyramid transform layer of ()≤Z, selects single order to smooth Regularization factors, on each decomposition layer, utilizes data constraint based on continuity equation, and the regularization that selects of combining adaptive because of Son, (x, t), x ∈ Ω, t=1,2, (x, t) expression is positioned at space coordinates x for Λ, T-1}, v to try to achieve motion vector field sequence V={v Motion vector with time coordinate t；

Step 7: the change movable in order to monitor aurora, is 2t by window width_wTime sliding window slide to stop bits from video original position Putting, the step-length every time slided is a frame, and the aurora motion vector field sequence in the t time sliding window is divided into front t_wFrame and rear t_wFrame Sequence, extracts these two sections of sequences Statistical Charateristics f based on partial vector difference_pre(t) and f_post(t)；

Step 8: measure the difference of aurora video sequence before and after in current sliding window, derive f_pre(t) and f_postThe chi-of (t) Square statistical distance, is assigned to the variation track value in t by value:

d_chg(t)=χ²(f_pre(t), f_post(t))

Chi-square statistical distance is defined as:

${\mathrm{\&chi;}}^2(p,q)={\mathrm{\&Sigma;}}_i\frac{{(p_i-q_i)}^2}{p_i+q_i},$

Wherein, i represents the index of two vectorial p and q, d_chgT () reflects the movable situation of change in t of aurora.

A kind of generation method for monitoring the movable variation track of aurora the most according to claim 1, it is characterised in that Wherein smooth regularization factors and the feature of second order divergence-curl regularization factors according to single order described in step 4, construct one New index: regularization Scatter-difference D_reg, carry out as follows:

D_reg=| ave_Ω(div(V¹))-ave_Ω(div(V²))|

For same motion, V¹It is to use to smooth the sports ground that the sports ground method of estimation of regularization factors obtains, V based on single order² It is the sports ground using sports ground method of estimation based on second order divergence-curl regularization factors to obtain, ave_Ω(div) meter is represented Calculate the average divergence in circular masked areas Ω.

A kind of generation method for monitoring the movable variation track of aurora the most according to claim 1, it is characterised in that The wherein D to regularization program decisions Threshold-training collection described in step 5_regIt is trained, obtains the decision-making threshold of regularization scheme Value Z, is carried out as follows:

(3a) from all-sky aurora data base, randomly select 500 width All-sky images constitute the training of regularization scheme decision-making value Collection, each width image configuration different scale that regularization program decisions Threshold-training is concentrated and the translational motion M in direction_S(i), And the rotary motion M in different scale and direction_R(i)；

(3b) smoothing the sports ground method of estimation of regularization factors with data constraint based on continuity equation associating single order is called Fluid-1, and claim the sports ground estimation side of data constraint based on continuity equation associating second order divergence-curl regularization factors Method is fluid-2；Be utilized respectively fluid-1 and fluid-2 and estimate this two groups of known motions, translational motion utilize fluid-1 and The sports ground that fluid-2 method obtains is designated asWithAnd rotary motion utilizes fluid-1 and fluid-2 method to obtain Sports ground is designated asWith

(3c) the regularization Scatter-difference of translational motion is calculatedRegularization with rotary motion Scatter-difference

(3d) seeking the intersection point of the histogram functions about regularization Scatter-difference that both moves, intersection point location is just Then change method decision-making value Z.