CN102914769A - Joint fractal-based method for detecting small target under sea clutter background - Google Patents

Abstract

The invention provides a joint fractal-based method for detecting a small target under a sea clutter background. The joint fractal-based method is higher in detection probability. The detection problem of a non-additive model is transformed into a classification problem, i.e. whether a target exists or not is equivalent to belong to a class in which a pure sea clutter exists, and a characteristic joint detection algorithm is provided. A bilogarithmic graph is established by using a trend fluctuation method through sea clutter data, a slope, namely a Hurst index, is fitted by using a least square method within a scale-free interval, and is used as a characteristic scalar, a nodal increment of a keypoint in the bilogarithmic graph is used as another characteristic scalar, therefore, a double-scalar obtained by each group of sea clutter data corresponds to one point in the bilogarithmic graph, n groups of corresponding points (i=1,...n) of the pure sea clutter data are obtained by using the steps, a space optimal classification line omega is obtained by using a convex hull function, sea clutters of regions in which the target possibly exits are obtained by using the same steps, and finally, by using whether the points exist in the space optimal classification line omega or not as a criterion, when the points exists in the space optimal classification line omega, no target exists, and when the points are outside the space optimal classification line omega, the target exists.

Description

Based on uniting small target detecting method under the fractal extra large clutter background

Technical field:

The present invention relates to the radar data process field, especially, relate to a kind of naval target detections of radar and extra large clutter detection method.

Background technology:

The sea clutter namely comes from the backscattering echo on a slice sea of being shone by the radar emission signal, and radar is detecting air space above sea or during near the target of air space above sea, must overcome the interference of the echo in sea own.When the sea or when very little near the radar reflection sectional area (RCS) of the weak target on sea, its radar return usually is buried in extra large clutter and the noise, and extra large clutter is also because of factor affecting such as its polarization radar mode, frequency of operation, antenna look angle and sea condition, wind directions, present obvious non-stationary, non-Gauss is so that a kind of fixedly statistical model is difficult to describe extra large clutter.

By the Accurate Analysis to extra large clutter time domain, explore the difference of essence between pure extra large clutter (referring to the radar backscattering echo that the driftlessness signal exists and not affected by object element) and the target sea clutter, make up the algorithm that to distinguish pure extra large clutter and target sea clutter.

Traditional additive model feature detection mainly contains Non-linear, to allow extra large clutter sample training go out as far as possible accurately neural network or supporting vector machine model, utilize the predicated error realize target to detect, the method prerequisite is that the chaotic characteristic of extra large clutter does not obtain theoretical validation fully, and most of forecast models that training obtains do not have ergodicity.

Summary of the invention:

The invention provides a kind ofly based on small target detecting method under the fractal extra large clutter background of associating, this method detection probability is higher.

Concrete technical scheme of the present invention is as follows:

Whether the present invention changes into classification problem with the test problems of non additivity model, be about to target and exist and be equivalent to the class that whether belongs to pure extra large clutter place, has proposed a kind of characteristic binding detection algorithm.Utilize the trend Fluctuation Method to set up bilogarithmic graph extra large clutter data, in without scaling interval, utilize least square fitting to draw slope, this is the Hurst index, with it as a characteristic scalar, select the intercept of key point in the bilogarithmic graph as another characteristic scalar, each organizes the two scalar that extra large clutter data obtain like this, and a some ξ utilizes above step to obtain the corresponding point ξ that n organizes pure extra large clutter data on the corresponding diagram _i(i=1...n), utilize the convex closure function to obtain space optimal classification line Ω, utilize same steps as to obtain ξ the extra large clutter of target possibility region, at last with a ξ whether in the Ω the inside as criterion, when ξ is in Ω, be considered as driftlessness and exist, when ξ is outside Ω, be considered as target and existed.

The present invention has following beneficial effect:

Feature detection algorithm such as time domain fractal method based on the non additivity model then are to extract fractal geometry to carry out the naval target detection from the seasonal effect in time series amplitude, but classic method is by determining behind the fractal dimension that fixed threshold relatively judges having or not of target in the extra large clutter, cause false-alarm probability one regularly, detection probability is lower.

The present invention compares with Non-linear, need not prior modeling, compare with traditional fractal method, the present invention only selects the interior fractal dimension of one section Fractal scale scope and the intercept of minute key point, and combine the convex closure diagnostic method, through test, no matter radar is in the extra large clutter that HH or VV polarization mode obtain, thinking the probability of finding target by mistake when radar necessarily is false-alarm probability one timing, and the present invention compares with traditional fractal method all has higher detection probability.

The present invention can distinguish pure extra large clutter and target sea clutter under radar HH polarization mode fully, and under radar VV polarization mode, when false-alarm probability one timing, detection probability is significantly better than classic method.

Table 1

Description of drawings

This figure of Fig. 1 be full coherent X-band IPIX radar during in November, 1993, gather in the bay, the Atlantic in Nova Scotia, Canada Dartmouth city the differentiation design sketch of extra large clutter data under the HH polarization mode.

The critical region that Ω zone utilizes method is tried to achieve among the present invention two dimensional character vector to form at characteristic plane for driftlessness sea clutter among Fig. 1, and the two dimensional character vector that target sea clutter arranged at the point on the characteristic plane all outside critical region Ω.

This figure of Fig. 2 be full coherent X-band IPIX radar during in November, 1993, gather in the bay, the Atlantic in Nova Scotia, Canada Dartmouth city the differentiation design sketch of radar sea clutter data under the VV polarization mode.

The Ω zone is the critical region that driftlessness sea clutter two dimensional character vector forms among Fig. 2, and the two dimensional character vector that target sea clutter arranged in the some overwhelming majority on the characteristic plane outside critical region Ω, determine among the figure that alarmed falsely probability is that 1/10 o'clock detection probability reaches 96.1%.

From figure, obviously find out the premium properties of time domain fractal characteristic associated detecting method among the present invention, even (VV polarization) still can well distinguish target sea clutter and pure extra large clutter under Low SNR.

Embodiment

Step 1 as input, is designated as x (i) i=1 with the echo data of full phase parameter radar, and 2 ..., N(N generally gets 2 for the length of input echo data ¹⁵), x (i) is deducted mean value, ask local and, set up a new sequence.

$Y\left(i\right)=\underset{k=1}{\overset{i}{\mathrm{\&Sigma;}}}[x\left(k\right)-<x>],(i=\mathrm{1,2},\&CenterDot;\&CenterDot;\&CenterDot;,N)---\left(1\right)$

Wherein＜x〉is input echo data average.

Step 2 is divided into the N that length is m with new sequence Y (i) _mThe integral multiple of span m is not necessarily separated in the individual disjoint isometric sub-range of=int (N/m) because of sequence length N, do not lose for guaranteeing former sequence information, and this sequence is back separated once from the terminal beginning of former sequence again, and can obtain like this length is 2N _mThe sub-range.

Step 3 with each sub-range s (s=1,2 ..., 2N _m) data utilize existing least square method to carry out match, obtain local trend function y _s(i), be designated as DFA2, eliminate trend in each sub-range, calculate its mean variance.

$Y_v(s,m)=\frac{1}{m}\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}{\left\{Y\right[(s-1)m+i]-y_s\left(i\right)\}}^2---\left(2\right)$

Here m is the separation span in the step 2, s=1, and 2 ... N _m

$Y_v(s,m)=\frac{1}{m}\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}{\left\{Y\right[N-(s-N_m)m+i]-y_s\left(i\right)\}}^2---\left(3\right)$

Here m is the separation span in the step 2, s=N _m+ 1,2 ... 2N _m

Step 4 is determined the 2 rank wave functions of complete sequence Y (i).

$F\left(m\right)={\left[\frac{1}{2N_m}\underset{s=1}{\overset{2N_m}{\mathrm{\&Sigma;}}}{Y}_v(s,m)\right]}^{1/2}\mathrm{\&alpha;}{m}^H---\left(4\right)$

Wherein H is the Hurst index in the self similar processes, embodied separate length m and between power function relationship, s=1,2 ... 2N _m

Step 5 F (m) changes along with separating length m, F (m) and m are got log-log coordinate after, the gained oblique distance is the Hurst exponential quantity of Y (i)

$H=\frac{{\log }_{2}F\left(m\right)}{{\log }_{2}m}---\left(5\right)$

Wherein H is the Hurst index in the self similar processes, has characterized the similarity relation between the data.

Step 6 is selected m (2 ²)～m (2 ⁴) yardstick is as without scaling interval, utilizes existing least square fitting to obtain slope as one of them parameter scalar at this without scaling interval, selects m (2 in the bilogarithmic graph ⁸) intercept located is as another parameter scalar.Form a two dimensional character vector, a point on each two dimensional character vector character pair plane.Horizontal ordinate on the characteristic plane is intercept, and ordinate is fractal dimension.

Step 7 is collected the coherent radar echo in N (N gets 1000 at least) group driftlessness marine site, and every group of radar return is that length is 2 ¹⁵Time series, utilize the N group two dimensional character vector that above-mentioned steps asks, be shown as N point at characteristic plane, will comprise on the characteristic plane N the minimal convex polygon of putting as critical region Ω.

Step 8 is in realistic objective is judged, utilize step 1-6 to draw two dimensional character vector ξ the coherent radar echo in a certain marine site to be measured, judge the relation of ξ and above-mentioned critical region Ω, be considered as this marine site driftlessness and exist when ξ is positioned at critical region Ω inside, being considered as this marine site when ξ is positioned at critical region Ω outside has target to exist.

When selecting m (2 ⁸)～m (2 ¹²) yardstick is as without scaling interval, utilize existing least square fitting to obtain slope (H urst index) as unique parameter scalar at this without scaling interval, set up again thresholding to utilize this parameter scalar to differentiate the existence of target sea clutter, can find out that by form the method works as false-alarm probability one regularly, detection probability is extremely low.Time domain fractal characteristic associated detecting method among the present invention has a high detection probability identical false-alarm probability is next by comparison.

Claims (1)

1. based on uniting small target detecting method under the fractal extra large clutter background, the method may further comprise the steps:

Step 1 as input, is designated as x (i) i=1 with the echo data of full phase parameter radar, and 2 ..., N, N deducts mean value for the length of input echo data with x (i), ask local and, set up a new sequence;

$Y\left(i\right)=\underset{k=1}{\overset{i}{\mathrm{\&Sigma;}}}[x\left(k\right)-<x>],(i=\mathrm{1,2},\&CenterDot;\&CenterDot;\&CenterDot;,N)---\left(1\right)$

Wherein＜x〉is input echo data average.

Step 2 is divided into the N that length is m with new sequence Y (i) _mBack separate this sequence once from the terminal beginning of former sequence in the individual disjoint isometric sub-range of=int (N/m) again, and obtaining length is 2N _mThe sub-range;

Step 3 with each sub-range s (s=1,2 ..., 2N _m) data utilize existing least square method to carry out match, obtain local trend function y _s(i), be designated as DFA2, eliminate trend in each sub-range, calculate its mean variance;

$Y_v(s,m)=\frac{1}{m}\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}{\left\{Y\right[(s-1)m+i]-y_s\left(i\right)\}}^2---\left(2\right)$

Here m is the separation span in the step 2, s=1, and 2 ... N _m

$Y_v(s,m)=\frac{1}{m}\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}{\left\{Y\right[N-(s-N_m)m+i]-y_s\left(i\right)\}}^2---\left(3\right)$

Here m is the separation span in the step 2, s=N _m+ 1,2 ... 2N _m

Step 4 is determined the 2 rank wave functions of complete sequence Y (i);

$F\left(m\right)={\left[\frac{1}{2N_m}\underset{s=1}{\overset{2N_m}{\mathrm{\&Sigma;}}}{Y}_v(s,m)\right]}^{1/2}\mathrm{\&alpha;}{m}^H---\left(4\right)$

Wherein H is the Hurst index in the self similar processes, embodied separate length m and between power function relationship, s=1,2 ... 2N _m

Step 5 F (m) changes along with separating length m, F (m) and m are got log-log coordinate after, the gained oblique distance is the Hurst exponential quantity of Y (i)

$H=\frac{{\log }_{2}F\left(m\right)}{{\log }_{2}m}---\left(5\right)$

Wherein H is the H urst index in the self similar processes;

Step 6 is selected m (2 ²)～m (2 ⁴) yardstick is as without scaling interval, utilizes existing least square fitting to obtain slope as one of them parameter scalar at this without scaling interval, selects m (2 in the bilogarithmic graph ⁸) intercept located is as another parameter scalar; Form a two dimensional character vector, a point on each two dimensional character vector character pair plane; Horizontal ordinate on the characteristic plane is intercept, and ordinate is fractal dimension;

Step 7 is collected the coherent radar echo in N group driftlessness marine site, and every group of radar return is that length is 2 ¹⁵Time series, utilize the N group two dimensional character vector that above-mentioned steps asks, be shown as N point at characteristic plane, will comprise on the characteristic plane N the minimal convex polygon of putting as critical region Ω.

Step 8 is in realistic objective is judged, utilize step 1-6 to draw two dimensional character vector ξ the coherent radar echo in a certain marine site to be measured, judge the relation of ξ and above-mentioned critical region Ω, when ξ is positioned at critical region Ω inside, be considered as driftlessness existence in this marine site, when ξ is positioned at critical region Ω outside, be considered as having in this marine site target to exist.

Images