CN111796265A

CN111796265A - Low-speed target detection method based on improved area bilateral smoothing

Info

Publication number: CN111796265A
Application number: CN202010536205.7A
Authority: CN
Inventors: 孟凡; 曹阳; 贾倩茜
Original assignee: 724th Research Institute of CSIC
Current assignee: 724th Research Institute of CSIC
Priority date: 2020-06-12
Filing date: 2020-06-12
Publication date: 2020-10-20

Abstract

The invention discloses a low-speed target detection method based on improved area bilateral smoothing. The method is suitable for designing a detection area by taking a target extrapolation position as a center aiming at a key focus target under a noise environment, then preprocessing radar data and converting the radar data into image data, adopting an improved area bilateral smoothing algorithm, proposing a Gaussian function value corresponding to Doppler velocity as a weight term of a value domain, combining the weight term with a weight coefficient kernel of a spatial distance, enhancing the processing of the target area and improving the contrast, then extracting the contour of the target area from the background by using a self-adaptive threshold segmentation method, and carrying out target condensation by combining the Doppler velocity.

Description

Low-speed target detection method based on improved area bilateral smoothing

Technical Field

The invention belongs to the field of radar data processing.

Background

Due to the radar system and the influence of the target background environment, noise and clutter are often accompanied in the echo signal receiving process. The method is used for detecting the low-speed target under the clutter background, on one hand, radar clutter suppression performance needs to be improved, and on the other hand, echo energy needs to be accumulated.

In the traditional detection of moving targets, adaptive matched filtering methods are often adopted, such as a Generalized Likelihood Ratio detector (GLRT) and an adaptive matched filtering detector (AMF), and coherent accumulation can be realized through Doppler filtering, so that the signal-to-noise Ratio of target echoes is effectively improved, and the capability of detecting moving targets is enhanced. However, for low-speed targets, especially in clutter environment, the conventional signal processing method such as MTI or MTD is difficult to detect such targets because the MTI filter has zero-frequency notch depth and width, which cause the snr loss of the low-speed targets after filtering, and the constant false alarm processing after detecting the MTD generates false alarm due to clutter. The method is used for detecting the low-speed target in the clutter from the angle of image processing, firstly, region enhancement is carried out, the contrast between the target and the clutter is improved, the target is separated from the background to the greatest extent (the target is a foreground, and the clutter is the background), the target is completely extracted, and meanwhile, noise is removed as far as possible. When the target is in the clutter area, a traditional radar target detection method is adopted, a strict detection threshold is set, the clutter removal effect is good, but the target is also removed, and the detection fails. The invention introduces the image processing correlation theory into the radar target detection field aiming at the echo data after signal processing, adopts the improved bilateral region smoothing algorithm to detect the target in the detection region, simultaneously considers the airspace information and the speed similarity, and enhances the contrast ratio of the target region. Not only can extract the target, but also can achieve the effect of denoising.

Disclosure of Invention

The invention mainly solves the problem of detecting the low-speed moving target in the complex environment.

Aiming at a key focus target, a target extrapolation position is taken as a center, a detection area is designed according to the change of Doppler velocity, an improved bilateral smoothing technology is provided for carrying out target detection in the area, the Doppler velocity corresponding to each point is used for replacing the amplitude as a weight item of a Gaussian kernel function, weights are calculated from two angles of space proximity and velocity similarity respectively, a convolution template is obtained, the detection area is smoothed, and then a self-adaptive threshold value segmentation method is adopted for extracting a target contour to carry out target condensation.

Bilateral smoothing is based on gaussian filtering, which has the same smoothing degree in each direction, and does not change the edge trend of the original image, and simultaneously, the gaussian kernel functions are single-valued functions, which monotonically decrease in all directions (consistent with target characteristics), and the position coordinates of each position, i.e. the weight of each amplitude point, corresponds to the function value on the two-dimensional gaussian function, when the value of σ is small, as shown in fig. 1: when x is less than-1 and x is greater than 1, the value of f (x) is already close to 0, which means that the amplitude value of the more distant position has no meaning for calculating the amplitude of the current position, namely, the purpose of highlighting the target area is achieved.

In order to realize the purpose of extracting a target and removing noise, a double-Gaussian filter is adopted for bilateral smoothing, in an image flat area, the pixel value change is very small, the corresponding pixel range domain weight is close to 1, and at the moment, the spatial domain weight plays a main role and is equivalent to Gaussian blur; in the edge area of the image, the pixel value is greatly changed, the pixel range area weight is increased, so that the information of the edge is kept, and the Euclidean distance of the target amplitude is considered, and the difference between the amplitude in the detection area and the central amplitude is also considered.

However, in actual data, due to the influence of fluctuation of the amplitude, the amplitude similarity cannot be accurately used as a basis for target detection. Therefore, the invention uses the corresponding Doppler velocity to replace the amplitude as the weight item of the Gaussian kernel function, and calculates the weight from two angles of space proximity and velocity similarity respectively, thereby obtaining the convolution template and realizing the smoothness of the detection area.

When radial velocity exists between the target and the radar, Doppler frequency shift is generated on the receiver, so that the target velocity can be obtained by using the Doppler change rate, and the moving target is detected. By definition, the Doppler frequency f_dThe relationship to radial velocity is:

wherein v is_rIn the radial velocity, λ is the wavelength, f is the frequency of the target, and c is the propagation velocity of the electromagnetic wave, as shown in fig. 2.

Therefore, the target in the clutter region is enhanced from the space domain and the value domain, the target extrapolation position is the center of the Gaussian function, as shown in formula (3), namely, in the neighborhood, the point with the speed value closer to the speed value of the central point has larger weight, and the point with the larger speed value difference has smaller weight. The weight is determined by a value domain Gaussian function, and the weight coefficients of the two are multiplied to obtain the final convolution template.

(1) Spatial distance: the Euclidean distance between the current point and the central point is referred, and the spatial domain Gaussian function is as follows:

wherein (x)_i，y_i) Is the current point position, (x)_c，y_c) σ is the spatial domain standard deviation for the center point position.

(2) Speed distance: the absolute value of the difference between the current point speed and the central point speed is indicated, and the value domain Gaussian function is as follows:

wherein, doppler (x)_i，y_i) Current point speed value, doppler (x)_c，y_c) For center point velocity values, σ is the range standard deviation.

The width of the gaussian function is characterized by a parameter σ, the value of which determines the amplitude of the variation of the gaussian function, the larger σ the wider the band of the gaussian filter and the better the smoothness, preferably a 3 × 3 template σ value of 0.85.

According to the result of the enhancement of the target area, the target contour is extracted by adopting an adaptive threshold segmentation method, the method considers that an image consists of a foreground and a background or two groups of gray values, the gray values are distributed into peak states on a histogram, and the low valley between the two peaks is the threshold value of the image segmentation.

And finally, performing target condensation according to a threshold segmentation result, wherein the traditional radar target detection method is to calculate the target centroid by condensation by utilizing amplitude information. Because the amplitude information is easily interfered by static noise and the detection probability is often reduced, the method introduces Doppler velocity to carry out target condensation aiming at the detected target contour area.

Wherein D is_1nFor successive range cell values, A, of the orientation within the target profile_nAmplitude value after azimuthal agglomeration within the target contour, f_nIs the maximum Doppler velocity at a certain position in the target area in the target contour, and the Doppler velocity of the target after f' condensation.

The method of the invention is adopted to detect the focus target, can realize the enhancement of the target area under the complex environment, completely reserve the outline information of the target, remove the clutter and provide accurate measurement data for the target tracking.

Drawings

Fig. 1 is a diagram of doppler velocity.

Fig. 2 is a gaussian kernel function curve.

FIG. 3 is a flowchart of a low-speed target detection method based on improved bilateral smoothing of a region.

FIG. 4 is a diagram illustrating the effect of region object enhancement and threshold segmentation using the method of the present invention.

Fig. 5 is a histogram of the original detection area.

Fig. 6 is a histogram of the detected region after processing by the method of the present invention.

Detailed Description

The technical scheme of the invention is further explained by combining the drawings and the embodiment.

The invention provides a low-speed target detection method based on improved area bilateral smoothing, which comprises the following steps (see fig. 3):

(1) aiming at an important target, a detection area is designed according to the change of Doppler velocity by taking a target extrapolation position as a center, the width and the height are respectively equal to dsin (theta) + kh ', and equal to dcos (theta) + kw', and an alternative unit is formed according to the sub-units. Wherein d is the distance between the filtered value and the extrapolated value of the current period, h 'and w' are the height and width of the target initial detection region, k is a range coefficient, wherein,

Δ v is the difference between the radial velocity of the current period and the mean value of the radial velocities of the previous n periods, f is the maximum value of the Doppler velocity in the detection area of the current period,

is the mean value of the maximum values of the Doppler velocities of the first n periods, theta^′＝arctan(v_y/v_x)，v_xAnd v_yRadial velocities in the x and y directions, respectively.

(2) Preprocessing the data in the region and converting the data into image data;

(3) the target extrapolation position is used as the center of a Gaussian function, improved bilateral smoothing is adopted to enhance the target area, the test effect is shown in figure 4, the middle figure is the effect after enhancement treatment, noise and clutter are obviously smoothed, and the target area is completely stored;

(4) and adaptively selecting a threshold value. Fig. 5 is a histogram of an original target region, where only one peak value is present in the histogram, and the target and noise cannot be accurately distinguished, and fig. 6 is a histogram of a target region after an improved bilateral region smoothing process, where two peaks obviously appear in the histogram, a peak with an amplitude value between 30 and 40 represents smoothed background noise, a peak with an amplitude value around 50 is a target, and a trough between two peaks is a selection threshold.

(5) And extracting the target contour according to the result of threshold segmentation. If there are still multiple targets after the segmentation, as shown in the third diagram in fig. 4, it is described that the histogram distribution of the target coincides with the key target, and the target cannot be distinguished by the threshold segmentation, so that the target contour closest to the center of the region is extracted by using the euclidean distance, which is the key target of interest. Then, the original data is mapped according to the position of the target contour, and target agglomeration is performed by using equations (4) and (5).

Claims

1. The low-speed target detection method based on improved area bilateral smoothing is characterized by comprising the following steps: aiming at a key focus target, a target detection area is designed according to the change of Doppler velocity by taking a target extrapolation position as a center, an improved bilateral smoothing technology is provided for carrying out target detection in the area, the Doppler velocity corresponding to each point is used for replacing the amplitude as a weight item of a Gaussian kernel function, weights are calculated from two angles of space proximity and velocity similarity respectively, so that a convolution template is obtained, the detection area is smoothed, and finally a self-adaptive threshold value segmentation method is adopted for extracting a target contour to carry out target condensation.

2. The improved region bilateral smoothing-based low-speed target detection method according to claim 1, characterized in that: target agglomeration is performed using the doppler velocity for the detected target contour region, and is represented by the following equation:

wherein D is_1nFor successive range cell values, A, of the orientation within the target profile_nAs the amplitude value after azimuthal agglomeration within the target profile, f_nIs the maximum Doppler velocity at a certain position in the target profile, and f' is the Doppler velocity of the target after condensation.