CN105574855A - Method for detecting infrared small targets under cloud background based on temperate filtering and false alarm rejection - Google Patents

Abstract

The invention diWscloses a method for detecting infrared small targets under a cloud background based on temperate filtering and false alarm rejection. The method comprises the following steps: firstly, obvious noises are removed from an image through maximum median filtering to complete image preprocessing; secondly, Robinson template filtering is used to suppress the background and highlight targets; thirdly, cloud partitioning of the original image is carried out, binarization processing of a result obtained after the Robinson template filtering is carried out through a low threshold in a cloud partition, and binarization processing of the result obtained after the Robinson template filtering is carried out through a high threshold in a non-cloud partition; finally, a plurality of ''false target points'' generated by a same target are further removed from the binarization results to complete ''coarse detection'', and time domain processing is continued for adjacent frame images that have gone through space domain processing to complete ''fine detection'', so that the detection of the infrared small targets is achieved. The method provided by the invention has the advantage that the constant false alarm rejection algorithm is added in interframe track correlation, so that the detection false alarm rate is greatly reduced.

Description

Based on the infrared small target detection method of template convolution and false alarm rejection under cloud background

Technical field

The invention belongs to infrared image detection and processing technology field, particularly under a kind of cloud background based on the infrared small target detection method of template convolution and false alarm rejection.

Background technology

Infrared small target detection is a core technology in the systems such as Infra-Red Search & Track System, Large visual angle object detection system, satellite remote sensing, disaster alarm, fire control and disaster rescue.Be subject to the factor impacts such as air, radiation from sea surface, operating distance and noise of detector due to infrared sensor, make remote target size on infrared image less, even present point-like; In addition, the signal to noise ratio (S/N ratio) of image is lower, adds background more complicated under normal circumstances, target to be easy to by noise and background clutter flood, make the detection of infrared small target become more difficult.

The detection probability and the reduction false alarm rate that how to improve target are mainly devoted to the research of infrared small target detection method.Traditional detection method is divided into two large classes: Time Domain Processing and spatial processing.Time Domain Processing, based on the multiple image sequence do not gathered in the same time, fully takes into account the correlativity between picture frame, can suppress between consecutive frame to false-alarm, keeps higher accuracy of detection.Conventional time-domain processing method has entropy difference method (Wang Guangjun, field inscription on ancient bronze objects, Liu Jian. based on the infrared image small target deteection [J] of local entropy. infrared and laser engineering, 2000,04:26-29.), sequence image detection method (Sun Jigang. sequence image infrared small target detection and track algorithm are studied [D]. Postgraduate School, Chinese Academy of Sciences (Changchun optical precision optical machinery and physics Institute), 2014.) etc.But must carry out registration operation between the method consecutive frame, intractability is comparatively large, and algorithm is comparatively complicated, and is not suitable for the detection of single-frame images.Spatial processing then utilizes object pixel to there is significant difference with neighborhood territory pixel in gray scale, and there is not the feature of correlativity with background, directly in its spatial domain, carry out Pixel-level process to single-frame images, owing to usually utilizing template operation, therefore algorithm is easy to be transplanted in hardware.Conventional method has morphology filter method (to spend the profit autumn, Zhang Ying, Lin Xiaochun. based on the infrared small target detection method [J] of shape filtering. laser and infrared, 2005,06:451-453.), high-pass filtering method (Dong Hongyan, Li Jicheng, Shen Zhenkang etc. based on the small target deteection [J] of high-pass filtering and Order Filtering. systems engineering and electronic technology, 2004,26 (5) .) etc.But the false alarm rate of the method is higher, and robustness is poor, and cannot be applied in the process of video sequence.

Summary of the invention

The invention provides the infrared small target detection method based on template convolution and false alarm rejection under a kind of cloud background, the infrared small target under cloud background accurately can be detected.

The present invention is the technical scheme solving prior art problem: based on the infrared small target detection method of template convolution and false alarm rejection under a kind of cloud background, for single-frame images, adopt the spatial domain operation of template convolution, first carry out max-medium filter to image to remove significant noise and complete Image semantic classification, secondly with Robinson's template convolution Background suppression, outstanding target, then cloud sector division is carried out to former figure, Low threshold is adopted to carry out binary conversion treatment in cloud sector part to the filtered result of Robinson, but not cloud sector part then adopts high threshold process, finally the result after binaryzation is rejected further multiple " the pseudo-impact points " of the generation of same target, thus complete " rough detection ", the consecutive frame image carrying out spatial processing is continued to take time domain operation process, namely track association is carried out in interframe, and the difference in gamma characteristic and kinetic characteristic carries out CFAR suppression operation for real goal and false-alarm point, complete " essence detects ", thus realize the detection of infrared small target.

Compared with prior art, its remarkable advantage is in the present invention: spatial processing and Time Domain Processing combine by (1), have concurrently simultaneously spatial processing algorithm complex low, be easy to Hardware and realize and Time Domain Processing accuracy of detection is high, false alarm rate is low feature.(2) preprocessing part, the basis of medium filtering proposes a kind of max-medium filter method, thus considers the gamma characteristic in image all directions.(3) target rough detection part: introduce Robinson's filter method Background suppression, outstanding target; Divide cloud sector and non-cloud sector, employing dual threshold carries out binary conversion treatment to different piece, ensures that Small object in cloud sector is not by blindness filtering.(4) target essence detecting portion, utilizes the difference in gamma characteristic and kinetic characteristic between real goal and false-alarm point, adds CFAR Restrainable algorithms, greatly reduce detection false alarm rate in interframe track association.

Below in conjunction with accompanying drawing, the present invention is described in further detail.

Accompanying drawing explanation

Fig. 1 is the process flow diagram based on the infrared small target detection method of template convolution and false alarm rejection under cloud background of the present invention.

Fig. 2 is the infrared image comprising Small object under the complicated cloud background of single frames.

Fig. 3 is the design sketch after max-medium filter process.

Fig. 4 is the design sketch after Robinson's filtering process.

Fig. 5 is the design sketch after cloud sector differentiates.

Fig. 6 is the design sketch after binaryzation operation.

Fig. 7 is the design sketch after adjacent point target combination.

Fig. 8 is the design sketch after interframe track association.

Fig. 9 is the design sketch after false alarm rejection.

Figure 10 is false alarm rate statistical graph.

Figure 11 is detectivity statistical graph.

Embodiment

Composition graphs 1, based on the infrared small target detection method of template convolution and false alarm rejection under cloud background of the present invention, for single-frame images, adopt the spatial domain operation of template convolution, first carry out max-medium filter to image to remove significant noise and complete Image semantic classification, secondly with Robinson's template convolution Background suppression, outstanding target, then cloud sector division is carried out to former figure, Low threshold is adopted to carry out binary conversion treatment in cloud sector part to the filtered result of Robinson, but not cloud sector part then adopts high threshold process, " neighborhood non-maxima suppression " principle is finally utilized to reject multiple " the pseudo-impact points " of same target generation further to the result after binaryzation, thus complete " rough detection ", the consecutive frame image carrying out spatial processing is continued to take time domain operation process, namely track association is carried out in interframe, and the difference in gamma characteristic and kinetic characteristic carries out CFAR suppression operation for real goal and false-alarm point, complete " essence detects ", thus realize the detection of infrared small target.Its concrete implementation step is as follows:

1. Image semantic classification.Max-medium filter is carried out to input picture as shown in Figure 2, remove the remarkable noise in image, i.e. max-medium filter water intaking flat, vertical, left 45 degree, 45 degree of four filtering directions, the right side, the maximal value of getting pixel grey scale intermediate value in all directions gives center pixel.Max-medium filter has fully taken into account the pixel grey scale distribution in multiple directions, and retains the energy of target in image better, and the basis not destroying the original gray feature of target realizes efficient denoising, and result as shown in Figure 3.

For the max-medium filter template of (2N+1) * (2N+1), its computing formula is:

f _max-med(i,j)＝max(z ₁,z ₂,z ₃,z ₄)(1)

In formula (1),

z ₁＝med[f(i,j-N),...,f(i,j),...,f(i,j+N)]

z ₂＝med[f(i-N,j),...,f(i,j),...,f(i+N,j)]

(2)

z ₃＝med[f(i+N,j-N),...,f(i,j),...,f(i-N,j+N)]

z ₄＝med[f(i-N,j-N),...,f(i,j),...,f(i+N,j+N)]

Wherein, pixel coordinate centered by (i, j), med is for getting median operation, and max is for getting maxima operation, and N characterizes template size size, and in the inventive method, for N=2, namely template size is 5*5.

2. rough detection, its process is:

(1) Robinson's template convolution is carried out to the image after max-medium filter, suppress cloud background, and outstanding target.This template utilizes this characteristic comparatively large of the gray difference between real goal and cloud background, the maximum gradation value of central pixel point gray-scale value and surrounding pixel is compared and arithmetical operation.If the gray scale of center pixel is comparatively strong, then can be retained, otherwise will be suppressed.Meanwhile, template-setup has isolation strip, can ensure that the gamma characteristic of real goal is not destroyed.Its result as shown in Figure 4.

The method of Robinson's template convolution is: for Robinson's Filtering Template of (2N+1) * (2N+1), its computing formula is:

$f_{Robison}(i,j)=\left\{\begin{array}{c}f(i,j)-max(z_1,z_2,z_3,z_4),f(i,j)>max(z_1,z_2,z_3,z_4)\\ f(i,j)-min(z_1,z_2,z_3,z_4),f(i,j)<max(z_1,z_2,z_3,z_4)\\ 0,else\end{array}\right.---\left(3\right)$

In formula (3),

z ₁＝max[f(i-N:i+N,j-N)]

z ₂＝max[f(i-N:i+N,j+N)]

(4)

z ₃＝max[f(i-N,i-N:j+N)]

z ₄＝max[f(i+N,i-N:j+N)]

Wherein, pixel coordinate centered by (i, j), max is for getting maxima operation, and N characterizes in template size size the inventive method, and for N=4, namely template size is 9*9.

(2) cloud sector differentiation is carried out to original image.Because cloud sector usually presents wire or stratiform in infrared image, utilize this feature, can take the method for template matches filtering in cloud, cloud exterior domain distinguished.This template is provided with blank isolation strip on the one hand and is not damaged to protect the gamma characteristic of real goal; on the other hand by the gray scale sum of all for central row pixels and the pixel grey scale sum of the top a line or bottom a line are carried out doing difference operation and normalizing to [0; 255] interval, and utilize binaryzation to operate cloud sector segmentation.Particularly, this operation is divided into following 3 steps:

I () carries out matched filtering operation to original image.For the matched filtering template of (2M+1) * (2N+1) size, its computing formula is:

$f_c(i,j)=\left\{\begin{array}{c}{z}_2-z_3,z_1>z_3\\ z_2-z_1,else\end{array}\right.---\left(5\right)$

In formula (5),

z ₁＝sum(f(i-N,j-M:j+M))

z ₂＝sum(f(i,j-M:j+M))(6)

z ₃＝sum(f(i+N,j-M:j+M))

Wherein, pixel coordinate centered by (i, j), sum is sum operation, and M, N characterize template size size, and in the present invention, M gets 5, N and gets 1, and namely template size is 11*3.

(ii) gray scale normalization process is carried out to the result after matched filtering.Normalization formula is:

$f_c^{\&prime;}(i,j)=\left\{\begin{array}{cc}255,& f_c(i,j)>255\\ 0,& f_c(i,j)<0\\ f_c(i,j),& else\end{array}\right.---\left(7\right)$

(iii) binarization segmentation is carried out to the result after normalization.The formula of binaryzation operation is:

$f_{cloud}(i,j)=\left\{\begin{array}{cc}255,& f_c^{\&prime;}(i,j)>f_{threshold}\\ 0,& else\end{array}\right.---\left(8\right)$

F in formula (8) _thresholdfor binary-state threshold.Herein, gray-scale value be 255 pixel represent cloud sector, and the pixel that gray-scale value is 0 represents non-cloud sector.

Through aforesaid operations, the cloud sector that can obtain as shown in Figure 5 differentiates result.

(3) in the filtered image of Robinson, different threshold value is adopted to carry out binarization segmentation to cloud sector and non-cloud sector part.Because the contrast difference between cloud sector target and cloud background is less, but not cloud sector part is then comparatively large, therefore takes Low threshold binarization segmentation in cloud sector, takes high threshold binarization segmentation in non-cloud sector.Concrete formula is:

Cloud sector part: $f_{seg}(i,j)=\left\{\begin{array}{cc}255,& f_{Robinson}(i,j)>f_L\\ 0,& else\end{array}\right.---\left(9\right)$

Non-cloud sector part: $f_{seg}(i,j)=\left\{\begin{array}{cc}255,& f_{Robinson}(i,j)>f_H\\ 0,& else\end{array}\right.---\left(10\right)$

F in formula (9) (10) _land f _hrepresent the binary-state threshold of two different sizes respectively.Wherein the former is Low threshold, and the latter is high threshold.As shown in Figure 6, comparatively complete extraction effect can be obtained after Double Thresholding Segmentation.

(4) continuous point target combination is carried out to the result after binarization segmentation, guarantees that target only stays next pixel size " isolated point ".Concrete operations are, are the pixel of 255 to gray-scale value after binaryzation, record its coordinate position, and be mapped in the filtered image of Robinson, pixel centered by it, offers the template (L refers to template length) of a L*L, as offered the template of a 5*5.If the gray-scale value of center pixel is the maximal value of all pixels in template, then retained in binary image, otherwise just rejected (namely utilizing " neighborhood non-maxima suppression " principle to reject multiple " the pseudo-impact points " of same target generation further to the result after binaryzation).As shown in Figure 7, each continuity point target is successfully merged into an isolated point to its result.

3. the step that essence detects is:

(1) to the image that continuous multiple frames operates through above-mentioned spatial domain, interframe track association is carried out.First set up flight path manager with list structure, from the second two field picture, successively the point coordinate in present frame and the point coordinate in previous frame are compared, if drop on certain some institute opened round Bo Mennei in previous frame, be then considered as being successfully associated.The ripple door of tradition track association is chosen and is usually needed the method utilizing prediction of speed dynamically to generate, but the number of pixels that infrared small target occupies in the picture is little, the amount of movement of interframe is also very little, and therefore ripple door radius can be taken as a definite value (a general desirable 1-2 pixel).As shown in Figure 8, the suspected target obtained after each rough detection all defines stable flight path (flight path is red annotate portions) in interframe.

(2) CFAR is carried out to the result after track association and suppress operation.The inventive method utilizes the difference that there is kinetic characteristic and gamma characteristic two aspect between real goal and false-alarm point, distinguishes both.Suppose that current time is k, object is v in the speed in k moment _k, in the speed sum in front k-1 moment be order (wherein, alpha+beta=1).Because noise motion meets standardized normal distribution, so its Δ v levels off to 0 within a period of time, and the Δ v of real goal will level off to a positive number.Setting speed threshold value V _t, as Δ v > V _ttime, be real goal by this target-recognition, otherwise, be then false-alarm noise (kinetic characteristic);

Gamma characteristic aspect, because real goal is approach ours gradually in visual field, in the process that this draws near, its gray scale is inevitable is also ascending conversion, and the gray scale of noise remains unchanged substantially.Investigate gamma characteristic for " shade of gray " this physical quantity: suppose that current time is k, object is I in the gray scale in k moment _k, be I in the gray scale in k-1 moment _k-1, then the shade of gray J of current time object _k=I _k-I _k-1, in the gradient sum in k-1 moment be before this order (wherein, alpha+beta=1).For noise, within a period of time, its Δ J levels off to 0, and the Δ J convergence of real goal is a positive number.Setting Grads threshold J _t, as Δ J > J _ttime, be real goal by this target-recognition, otherwise, be then considered as false-alarm noise.Based on above two criterions, the CFAR just completed in track association process suppresses operation.As shown in Figure 9, the flight path produced by false-alarm is not labeled as real goal by yellow box to concrete result.

For the advantage of the present invention in infrared small target detection reliability is described, use the inventive method to carry out simulation process to the mobile Small object video under the cloud background of 90 ° of Large visual angle infrared thermovision systems collection, and carry out statistical computation for false alarm rate and detectivity two indexs.Wherein, false alarm rate is defined as the detection number of times of interior false target per hour; Verification and measurement ratio is defined as the number percent of correct detection destination number relative to realistic objective quantity.

As shown in Figure 10, horizontal ordinate is hourage, and ordinate is wrong report number of times, sets up false alarm rate statistical graph.As calculated, under this scene, false alarm rate can control at 1.7 times/hour by the inventive method.

As shown in figure 11, horizontal ordinate is hourage, and ordinate is detectivity, sets up detectivity statistical graph.As calculated, under this scene, detectivity can control in 96% by the inventive method.

Claims (7)

1. under a cloud background based on the infrared small target detection method of template convolution and false alarm rejection, it is characterized in that for single-frame images, adopt the spatial domain operation of template convolution, first carry out max-medium filter to image to remove significant noise and complete Image semantic classification, secondly with Robinson's template convolution Background suppression, outstanding target, then cloud sector division is carried out to former figure, Low threshold is adopted to carry out binary conversion treatment in cloud sector part to the filtered result of Robinson, but not cloud sector part then adopts high threshold process, finally the result after binaryzation is rejected further multiple " the pseudo-impact points " of the generation of same target, thus complete " rough detection ", the consecutive frame image carrying out spatial processing is continued to take time domain operation process, namely track association is carried out in interframe, and the difference in gamma characteristic and kinetic characteristic carries out CFAR suppression operation for real goal and false-alarm point, complete " essence detects ", thus realize the detection of infrared small target.

2. according to the infrared small target detection method based on template convolution and false alarm rejection under the cloud background described in claim 1, it is characterized in that Image semantic classification process is: max-medium filter is carried out to input picture, remove the remarkable noise in image, i.e. max-medium filter water intaking flat, vertical, left 45 degree, 45 degree of four filtering directions, the right side, the maximal value of getting pixel grey scale intermediate value in all directions gives center pixel.

3., according to the infrared small target detection method based on template convolution and false alarm rejection under the cloud background described in claim 2, it is characterized in that the max-medium filter template for (2N+1) * (2N+1), its computing formula is:

f _max-med(i,j)＝max(z ₁,z ₂,z ₃,z ₄)(1)

In formula (1),

z ₁＝med[f(i,j-N),...,f(i,j),...,f(i,j+N)]

z ₂＝med[f(i-N,j),...,f(i,j),...,f(i+N,j)](2)

z ₃＝med[f(i+N,j-N),...,f(i,j),...,f(i-N,j+N)]

z ₄＝med[f(i-N,j-N),...,f(i,j),...,f(i+N,j+N)]

Wherein, pixel coordinate centered by (i, j), med is for getting median operation, and max is for getting maxima operation, and N characterizes template size size.

4., according to the infrared small target detection method based on template convolution and false alarm rejection under the cloud background described in claim 1, it is characterized in that the process of rough detection is:

(1) Robinson's template convolution is carried out to the image after max-medium filter, suppress cloud background, and outstanding target, this template utilizes this characteristic comparatively large of the gray difference between real goal and cloud background, the maximum gradation value of central pixel point gray-scale value and surrounding pixel is compared and arithmetical operation, if the gray scale of center pixel is comparatively strong, then can be retained, otherwise will be suppressed; Meanwhile, template-setup has isolation strip, ensures that the gamma characteristic of real goal is not destroyed;

(2) cloud sector differentiation is carried out to original image, take the method for template matches filtering in cloud, cloud exterior domain distinguished, this template is provided with blank isolation strip on the one hand and is not damaged to protect the gamma characteristic of real goal, on the other hand by the gray scale sum of all for central row pixels and the pixel grey scale sum of the top a line or bottom a line are carried out doing difference operation and normalizing to [0,255] interval, and utilize binaryzation to operate cloud sector segmentation;

(3) in the filtered image of Robinson, adopt different threshold value to carry out binarization segmentation to cloud sector and non-cloud sector part, take Low threshold binarization segmentation in cloud sector, take high threshold binarization segmentation in non-cloud sector, be specially:

Cloud sector part: $f_{seg}(i,j)=\left\{\begin{array}{cc}255,& f_{Robinson}(i,j)>f_L\\ \mathrm{0,}& else\end{array}\right.---\left(9\right)$

Non-cloud sector part: $f_{seg}(i,j)=\left\{\begin{array}{cc}255,& f_{Robinson}(i,j)>f_H\\ 0,& else\end{array}\right.---\left(10\right)$

F in formula (9) (10) _land f _hrepresent the binary-state threshold of two different sizes respectively, wherein the former is Low threshold, and the latter is high threshold;

(4) continuous point target combination is carried out to the result after binarization segmentation, guarantee that target only stays next pixel size " isolated point ", namely be the pixel of 255 to gray-scale value after binaryzation, record its coordinate position, and be mapped in the filtered image of Robinson, pixel centered by it, offer the template of a L*L, if the gray-scale value of center pixel is the maximal value of all pixels in template, then retained in binary image, otherwise just rejected.

5. according to the infrared small target detection method based on template convolution and false alarm rejection under the cloud background described in claim 4, it is characterized in that the method for Robinson's template convolution is: for Robinson's Filtering Template of (2N+1) * (2N+1), its computing formula is:

$f_{Robison}(i,j)=\left\{\begin{array}{c}f(i,j)-max(z_1,z_2,z_3,z_4),f(i,j)>max(z_1,z_2,z_3,z_4)\\ f(i,j)-min(z_1,z_2,z_3,z_4),f(i,j)<max(z_1,z_2,z_3,z_4)\\ 0,else\end{array}\right.---\left(3\right)$

In formula (3),

z ₁＝max[f(i-N:i+N,j-N)]

z ₂＝max[f(i-N:i+N,j+N)](4)

z ₃＝max[f(i-N,i-N:j+N)]

z ₄＝max[f(i+N,i-N:j+N)]

Wherein, pixel coordinate centered by (i, j), max is for getting maxima operation, and N characterizes template size size.

6., according to the infrared small target detection method based on template convolution and false alarm rejection under the cloud background described in claim 4, it is characterized in that the step that cloud sector differentiates is:

I () carries out matched filtering operation to original image, for the matched filtering template of (2M+1) * (2N+1) size, its computing formula is:

$f_c(i,j)=\left\{\begin{array}{c}{z}_2-z_3,z_1>z_3\\ z_2-z_1,else\end{array}\right.---\left(5\right)$

In formula (5),

z ₁＝sum(f(i-N,j-M:j+M))

z ₂＝sum(f(i,j-M:j+M))(6)

z ₃＝sum(f(i+N,j-M:j+M))

Wherein, pixel coordinate centered by (i, j), sum is sum operation, and M, N characterize template size size;

(ii) carry out gray scale normalization process to the result after matched filtering, normalization formula is:

$f_c^{\&prime;}(i,j)=\left\{\begin{array}{cc}255,& f_c(i,j)>255\\ 0,& f_c(i,j)<0\\ f_c(i,j),& else\end{array}\right.---\left(7\right)$

(iii) carry out binarization segmentation to the result after normalization, the formula of binaryzation operation is:

$f_{cloud}(i,j)=\left\{\begin{array}{cc}255,& f_c^{\&prime;}(i,j)>f_{threshold}\\ 0,& else\end{array}\right.---\left(8\right)$

F in formula (8) _thresholdfor binary-state threshold, herein, gray-scale value be 255 pixel represent cloud sector, and the pixel that gray-scale value is 0 represents non-cloud sector.

7., according to the infrared small target detection method based on template convolution and false alarm rejection under the cloud background described in claim 4, it is characterized in that the step that essence detects is:

(1) to the image that continuous multiple frames operates through spatial domain, carry out interframe track association, first flight path manager is set up with list structure, from the second two field picture, successively the point coordinate in present frame and the point coordinate in previous frame are compared, if drop on certain some institute opened round Bo Mennei in previous frame, be then considered as being successfully associated; Ripple door radius is taken as a definite value;

(2) carry out CFAR to the result after track association and suppress operation, utilize the difference that there is kinetic characteristic and gamma characteristic two aspect between real goal and false-alarm point, distinguish both, namely suppose that current time is k, object is v in the speed in k moment _k, in the speed sum in front k-1 moment be order (wherein, alpha+beta=1); Because noise motion meets standardized normal distribution, so its Δ v levels off to 0 within a period of time, and the Δ v of real goal will level off to a positive number; Setting speed threshold value V _t, as Δ v > V _ttime, be real goal by this target-recognition, otherwise, be then false-alarm noise;

For gamma characteristic aspect, investigate gamma characteristic for " shade of gray " this physical quantity: suppose that current time is k, object is I in the gray scale in k moment _k, be I in the gray scale in k-1 moment _k-1, then the shade of gray J of current time object _k=I _k-I _k-1, in the gradient sum in k-1 moment be before this order (wherein, alpha+beta=1), for noise, within a period of time, its Δ J levels off to 0, and the Δ J convergence of real goal is a positive number; Setting Grads threshold J _t, as Δ J > J _ttime, be real goal by this target-recognition, otherwise, be then considered as false-alarm noise; Based on above two characteristics, just complete the false alarm rejection operation in track association process.