A method for evaluating the quality of a track based on waveform entropy

Abstract

The present application belongs to the field of radar signal processing, and particularly relates to a point track quality evaluation method based on waveform entropy method. According to a large number of samples of correct target point track blocks, statistical characteristics of probability density functions of various features (distance spread, azimuth spread, Doppler channel stationarity, signal-to-noise ratio and detection point number ratio) of the point track blocks are analyzed; confidence degrees of corresponding features of each point track block are calculated according to the probability density functions of the features; weights of the corresponding features are calculated by using a tomographic analysis method, and the features are weighted and summed to obtain a quality evaluation of the point track blocks. The point track quality evaluation result can be used to judge and suppress clutter point tracks, and improve data association accuracy.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of radar signal processing, and particularly relates to a method for evaluating the quality of a plot, which is suitable for the process of plot extraction and clutter plot suppression, and can be used for judging and associating targets and non-targets in the process of track association. BACKGROUND

[0002] After the radar completes signal processing on the received echo signal, a large number of detection points are obtained, in which there are a large number of clutter detection points of different attribute categories and other multi-target detection points. The centroid of a primary plot is obtained through the process of plot condensation, and is used for the process of back-end data processing. Due to the existence of clutter plots, a large number of clutter plots and possible target plots appear in the gate, which brings a computational burden to the process of data association and produces false tracks. Therefore, the quality of plots is evaluated in the process of plot processing, and part of the non-target plots is removed according to the evaluation results. The quality information is transmitted to the terminal data processing module to provide the basis for data association. SUMMARY

[0003] For the evaluation of the quality of plots after condensation, the present application proposes a method for evaluating the quality of plots based on waveform entropy. According to a large number of samples of correct target plot blocks, the statistical characteristics of the probability density functions of various features (range spread, azimuth spread, Doppler channel stationarity, signal-to-noise ratio and the ratio of the number of detection points) of the plot blocks are analyzed. The confidence of each feature of each plot block is calculated according to the probability density functions of the features. The weights of the respective features are calculated by using the method of tomographic analysis, and the features are weighted and summed to obtain the quality evaluation of the plot block.

[0004] The technical scheme of the present application is: a method for evaluating the quality of plots based on waveform entropy.

[0005] Step 1: Collect a large number of correct target plot samples (non-clutter samples), and fit the probability density functions of various features of the plot blocks. The probability density functions of four types of features, i.e., the range spread, the azimuth spread, the stationarity of all azimuth Doppler channels on the distance unit where the maximum amplitude detection point is located, and the ratio of the maximum signal-to-noise ratio to the number of detection points, of all plot blocks of the collected correct targets are respectively counted, and the data is saved.

[0006] Step 2: Calculate the confidence of each plot block of the four types of features.

[0007]

[0008] In the formula, θ i is the confidence of the i th feature; is the probability density value corresponding to the i th feature value of the plot block obtained according to the probability density function of the i th feature; The maximum probability density value of the i-th feature probability density function, wherein i = 1, 2, 3, 4.

[0009] Step 3, based on the waveform entropy method, the weight corresponding to each feature is calculated.

[0010] According to the probability density function of the four types of features (distance spread, azimuth spread, Doppler channel stationarity, and signal-to-noise ratio to detection point number ratio) of the target track obtained in step 1, the waveform entropy e i The sum E of the four types of feature waveform entropies is calculated, and the weight W corresponding to each feature is calculated according to the waveform entropy e i of each feature and the sum E of the four types of feature waveform entropies.

[0011] The discrete sequence value of the probability density function of a certain feature i is x i (n), (n = 1, 2, 3, …, N-1), then:

[0012]

[0013] The sequence x i (n) is:

[0014]

[0015] The sum E of the waveform entropies is calculated according to the waveform entropy of each feature:

[0016]

[0017] wherein x i (n) represents the discrete sequence value of the probability density function of a certain feature i in the four types of features, wherein i = 1, 2, 3, 4; the sequence length is N; e i represents the waveform entropy value corresponding to the i-th feature probability density function; E represents the sum of the waveform entropies corresponding to the four types of features.

[0018] The weight W corresponding to each feature is calculated according to e i and E:

[0019]

[0020] W = (w1, w2, w3, w4)'

[0021] wherein w i represents the weight corresponding to the i-th feature probability density function; w1 represents the weight of the track distance spread feature; w2 represents the weight of the track azimuth spread feature; w3 represents the weight of the track Doppler channel stationarity feature; w4 represents the weight of the track signal-to-noise ratio to detection point number ratio feature.

[0022] Step 4: Calculate the quality of each dot block.

[0023] By combining the confidence scores of the four types of features of a certain point obtained in step 2 with the feature weights W obtained in step 3, the overall quality of the point block is calculated.

[0024]

[0025] In the formula, θ i w represents the confidence level of the i-th feature of the dot matrix block. i The weight corresponding to the i-th feature is given by i = 1, 2, 3, 4; quality is the overall quality of the current point block.

[0026] This invention, starting from the perspective of target clutter features and combining waveform entropy method, achieves objective quality evaluation of each aggregated clutter point. First, due to the diverse types of clutter, including ground clutter, sea clutter, meteorological clutter, and echoes from other targets in the surrounding environment, and the significant differences between these clutter types depending on the environment, statistical analysis of clutter point patch features is difficult, and the distribution range of the lattice feature probability density function is large and lacks representativeness. Considering the relatively stable and concentrated feature distribution characteristics of target point patches, this invention compares and comprehensively judges various types of aggregated clutter points from the perspective of target characteristics. Second, considering that some clutter points and target points share significant similarities in certain features, this invention employs a multi-feature comprehensive judgment method, drawing on the idea of ​​multi-feature joint probability density, which can more accurately distinguish between target and clutter points. Finally, using the point quality evaluation results of this invention, clutter points can be judged and suppressed, reducing the workload of data processing to a certain extent and improving the accuracy of data association. Attached Figure Description

[0027] Fig. 1 A probability distribution map of the distance to the correct target point trace sample.

[0028] Fig. 2 Expand the probability distribution map of the correct target point trace sample orientation.

[0029] Fig. 3 The probability distribution of Doppler channel consistency for the correct target point trace sample.

[0030] Fig. 4 This is a probability distribution diagram of the correct target point trace samples (signal-to-noise ratio / detection point).

[0031] Fig. 5 Display of multiple detection points before spot suppression.

[0032] Fig. 6 This is a display of multiple detection points after spot suppression. Detailed Implementation

[0033] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0034] This invention provides a point quality assessment method based on waveform entropy, which can objectively assess the quality of a single point and has a certain filtering capability for non-target points.

[0035] The experimental process in this embodiment is based on measured data from near-shore detection using dual-band radar. For the detection video information obtained after signal processing, point clustering and point quality assessment are performed. Assume the total number of point samples obtained from point clustering is R, where the correct aircraft target samples are B, and other clutter samples are C, and R = BUC.

[0036] Step 1: Collect a large number of correct target point samples B, and fit the probability density function of the four features of this type of point block (distance broadening, azimuth broadening, Doppler channel stability, signal-to-noise ratio to the number of detection points).

[0037] pdf dis_around =P(B dis_around )

[0038] pdf azi_around =P(B azi_around )

[0039] pdf doupler =P(B doupler )

[0040] pdf snr / Ep =P(B snr / Ep )

[0041] Among them, pdf dis_around The probability density distribution representing the widening of the point trace distance; pdf azi_around The probability density distribution representing the azimuth broadening of the dot pattern; pdf doupler The probability density distribution of Doppler channel stationarity of the trace; pdf snr / Ep B represents the probability density distribution of the ratio of the signal-to-noise ratio of the dots to the number of detection dots; dis_around This represents the distance-expanded sample, which consists of the distance-expanded values ​​of each point in the target point sample B; B azi_around This represents the azimuth-expanded sample formed by the azimuth expansion values ​​of each point in the target point sample B; B doupler This represents the Doppler channel stationarity sample formed by the Doppler channel stationarity values ​​of each point in the target point sample B; B snr / Ep P(·) represents the set of discrete feature sample sets (B) whose signal-to-noise ratio (SNR) is equal to the number of detection points for each point in the target point sample B; P(·) represents the discrete feature sample set (B).dis_around B azi_around B doupler B snr / Ep The probability density distribution function of the corresponding feature is mapped to the corresponding feature.

[0042] Step 2: Calculate the confidence scores of the four features for each point in the total sample R of points.

[0043] Suppose that the distance of a trace k is widened by k. dis The azimuth expansion size is k azi The Doppler channel stationarity is k. douple The ratio of signal-to-noise ratio to the number of detection points is k. snr / Ep Find the probability density value of a feature from the probability density distribution corresponding to that feature.

[0044]

[0045] Calculate the maximum probability density value corresponding to the probability density distributions of the four feature attributes (range broadening, azimuth broadening, Doppler channel stationarity, and signal-to-noise ratio to number of detection points).

[0046] Where i represents the i-th type of feature attribute (range broadening, azimuth broadening, Doppler channel stationarity, signal-to-noise ratio to number of detection points), i = 1, 2, 3, 4; k i Let represent the numerical value of the i-th type of feature attribute of a certain trace k. The horizontal axis represents the probability density distribution of the i-th type of feature attribute; The probability density value corresponding to the value of the i-th type of feature attribute of a certain trace k.

[0047] Calculate the confidence level of the four features of a point k based on the probability density values ​​of the four features and the maximum probability density value of the probability density function of the corresponding features.

[0048]

[0049] In the formula, θ i Let be the confidence level of the i-th feature of the trace k; The probability density value corresponding to the value of the i-th type feature attribute of a certain trace k; Let be the maximum probability density value of the probability density function of the i-th feature, where i = 1, 2, 3, 4.

[0050] Step 3: Calculate the weights corresponding to the four types of features based on the waveform entropy method.

[0051] Based on the probability density functions (pdf) of the four types of features of the target point obtained in step 1 dis_around ,pdfazi_around ,pdf doupler ,pdf snr / Ep ), calculate the waveform entropy e corresponding to each feature probability density function respectively. i Calculate the sum E of the waveform entropy for the four types of features, and then calculate the waveform entropy e corresponding to each feature. i The weight W corresponding to each feature is calculated by taking the sum of the entropies of the four types of feature waveforms, E, and E.

[0052] The discrete sequence value of the probability density function of a certain feature i is pdf i Given a sequence (n), (n = 1, 2, 3, ..., N), we have:

[0053]

[0054] Then the probability density function pdf of a certain feature i in the sequence i The waveform entropy of (n) is:

[0055]

[0056] Calculate the total waveform entropy E based on the waveform entropy corresponding to each feature:

[0057]

[0058] Among them, pdf i (n) represents the discrete sequence values ​​of the probability density function of a feature i among the four features, where i = 1, 2, 3, 4; the sequence length is N; e i E represents the waveform entropy value corresponding to the probability density function of the i-th feature; E represents the sum of the waveform entropies corresponding to the four types of features.

[0059] According to e i E calculates the weight W for each feature:

[0060]

[0061] W = (w1, w2, w3, w4)'

[0062] Among them, w i w1 represents the weight corresponding to the probability density function of the i-th feature; w2 represents the weight of the point distance broadening feature; w3 represents the weight of the point Doppler channel stationarity feature; and w4 represents the weight of the point signal-to-noise ratio to the number of detection points feature.

[0063] Step 4: Calculate the quality of each dot block.

[0064] By combining the confidence scores of the four types of features of a certain point k obtained in step 2 with the feature weights W obtained in step 3, the overall quality of the point block is calculated.

[0065]

[0066] In the formula, θ i w represents the confidence level of the i-th feature of the dot matrix block. i The weights corresponding to the i-th feature among the four types of features are i = 1, 2, 3, 4; quality is the overall quality of the current point k.

[0067] This invention uses the analytic hierarchy process (AHP) to comprehensively calculate the confidence levels of various physical characteristics of dot blocks, thereby enabling quality assessment of the dots. Figs. 1-4 The probability density distributions of four types of physical features of the target sample's trace blocks are displayed. It can be seen that the feature distribution of the target trace blocks is limited to a certain range. By calculating a certain type of feature of the trace block and comparing it with the probability distribution function of that feature, the target confidence level corresponding to that feature can be obtained. Fig. 5 and Fig. 6 The display screen shows the results, evaluates the quality of each aggregated point, sets a quality threshold, and suppresses non-target points. It can be seen that after point suppression, clutter points are significantly reduced and target point loss is minimal.

[0068] The beneficial effects of this invention are:

[0069] In step 1, the distinction between target and clutter clutter is based on a physical analysis of various characteristic probabilistic statistical models. However, due to the large variety of clutter types and the significant differences in the distribution of their clutter clutter characteristic probabilistic models, a unified model cannot be established. Since the characteristics of target clutter are related to the signal processing process, if the signal processing process is determined, a unified characteristic model for each type of target clutter detected by the radar can be established. This established target clutter characteristic model can then be used as an evaluation criterion during the comparison and discrimination process between clutter and targets.

[0070] In step 3, the waveform entropy method is used to analyze and calculate the weights corresponding to the four types of features. This is done through comparison. Figs. 1-4 The Doppler channel stationarity probability density and signal-to-noise ratio / number of detection points probability density distribution of the target point block have a high degree of concentration, which can determine the range of concentration of most targets under this type of feature. By utilizing the concentration characteristics of the probability density function, the importance between various features can be objectively determined, and the weight corresponding to each feature can be obtained. The comprehensive quality of the current point block is obtained through confidence and weight. This comprehensive quality contains the main information of the target. The weight is adaptively calculated through the probability density function, making the judgment of point quality more objective in this invention.

Claims

1. A method for evaluating the quality of dots based on waveform entropy, characterized in that: Includes the following steps, Step 1: Collect target point trace samples and statistically analyze the probability density functions of four types of features of the collected target point trace blocks: distance broadening, azimuth broadening, Doppler channel stability above the distance cell where the maximum amplitude detection point is located, and the ratio of maximum signal-to-noise ratio to the number of detection points. Step 2: Calculate the confidence scores of the four features for each point block; ； In the formula, For the first Confidence level of each feature; According to the first The probability density function of the nth feature yields the point patch. The probability density values ​​corresponding to each eigenvalue. Indicates a trace of a certain point The The numerical value of the class feature attribute; For the first The maximum probability density value of each feature probability density function, where ; Step 3: Calculate the weight corresponding to each feature based on the waveform entropy method. The specific process is as follows: A certain feature The discrete sequence value of the probability density function For a sequence, then: ， Then the sequence The waveform entropy is: ； Calculate the sum of waveform entropy based on the waveform entropy corresponding to each feature. : ， in, Represents a feature among the four types of features The discrete sequence values ​​of the probability density function, where The sequence length is ; Indicates the first The waveform entropy value corresponding to each feature probability density function; This represents the sum of waveform entropy corresponding to the four types of features; according to , Calculate the weights corresponding to each feature : ， ， in, Indicates the first The weights corresponding to the probability density functions of each feature; The weights representing the point distance broadening feature; The weight representing the azimuth broadening feature of the dot pattern; The weights representing the stationarity characteristics of the Doppler channel at the dot mark; The weight of the feature representing the ratio of the signal-to-noise ratio of the detection points to the number of detection points; Step 4: Calculate the overall quality of each dot block: ， In the formula, For the first point of the trace block Confidence level of each feature; No. The weights corresponding to each feature, where ; This represents the overall quality of the current dot matrix block.

2. The point quality assessment method based on waveform entropy method according to claim 1, characterized in that: The calculation process for the maximum probability density value in step 2 is as follows: Find the probability density value corresponding to the feature value from the probability density distributions corresponding to the features of range broadening, azimuth broadening, Doppler channel stationarity, and the ratio of signal-to-noise ratio to the number of detection points. ， Calculate the maximum probability density value corresponding to the probability density distributions of the four types of feature attributes. ,in Indicates the first The horizontal axis value of the probability density distribution of class feature attributes; a certain trace The The probability density value corresponding to the numerical value of the class feature attribute.

3. The point quality assessment method based on waveform entropy method according to claim 2, characterized in that: The confidence level in step 2 is: ,in a certain trace The The probability density value corresponding to the numerical value of the class feature attribute; For the first The maximum probability density value of each feature probability density function, where .

4. The point quality assessment method based on waveform entropy method according to claim 1, characterized in that: In step 1, the probability density function of the four types of features of the dot patch is: ， ， ， ， in, This represents the probability density distribution of the point trace distance broadening; The probability density distribution representing the azimuth broadening of the dot trace; The probability density distribution of Doppler channel stationarity of the dot trace; The probability density distribution representing the ratio of the signal-to-noise ratio of the dots to the number of detection dots; Represents target point trace samples The distance-expanded sample is composed of the distance-expanded values ​​of each point in the sample. Represents target point trace samples The azimuth broadening sample is composed of the azimuth broadening values ​​of each point in the sample; Represents target point trace samples The Doppler channel stationarity sample is composed of the Doppler channel stationarity values ​​of each point in the sample. Represents target point trace samples The set of signals-to-noise ratios (SNRs) of each point in the detection array, which is a sample of the ratios of SNRs to the number of detection points. Represents a discrete feature sample set ( , , , The probability density distribution function of the corresponding feature is mapped to the corresponding feature.

Images