An extended object detection method

Abstract

The embodiment of the application discloses a kind of extended target detection methods, belong to radar detection technical field, solve the problem that existing extended target detection method is in energy accumulation or introduces too much noise, or miss target energy, causes detection performance to decline.Problems include, based on echo matrix, the dynamic programming accumulation detection model of range extended target is constructed;Based on minimum energy loss criterion, the trade-off parameter is configured to value function, a double-probability joint constraint optimization model about target energy and noise energy is established to determine the trade-off parameter by solving the optimization model;Based on trade-off parameter, the value function and state transition matrix of joint state variable between multiple pulses are updated by recursive operation;After the value function obtained after updating is completed, the corresponding energy ridge line is extracted, and an approximate probability model of the change amplitude of the energy ridge line is constructed, and the optimal change threshold under the preset false alarm probability is obtained by solving the approximate probability model.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of radar detection, and particularly relates to an extended target detection method. BACKGROUND

[0002] Radar detection technology has a wide range of applications in the fields of security and civil use, and high-resolution radar plays an important role in improving target detection and recognition capabilities. Compared with low-resolution radar, the large transmission bandwidth and long time accumulation of high-resolution radar enable high-resolution radar to simultaneously obtain high range resolution and high Doppler frequency resolution, which can provide high measurement accuracy, strong echo coherence and extensive target structure information. However, when the range resolution of high-resolution radar is smaller than the target size, the target echo will be scattered on multiple range cells, that is, the phenomenon of range spreading occurs. For range-extended targets, point target detection algorithms may suffer performance loss due to the shadowing effect of adjacent scattering points or the spillover of adjacent range cell target energy.

[0003] In the prior art, most extended target detection methods based on long-time accumulation usually preset the distance extension length of the target to be known or fixed, however, in actual applications, due to factors such as target attitude change, radar viewing angle change, etc., the scattering characteristics of the target will change dynamically, resulting in mismatch between the preset extension length model and the actual target echo. This model mismatch will cause the detector to introduce too much noise cell or miss part of the target energy during energy accumulation, thereby causing detection performance to decline. And with the increase of the mismatch degree of the target scattering model, the detector based on the prior knowledge of the target may accumulate too much noise-only distance cell and suffer collapse loss, resulting in a decline in detection performance. SUMMARY

[0004] The embodiments of the present application provide an extended target detection method, which is used to solve the technical problem that the existing extended target detection method presets a fixed extension length, which is not matched with the actual dynamic change, resulting in too much noise being introduced or target energy being missed during energy accumulation, and causing detection performance to decline.

[0005] The embodiments of the present application adopt the following technical solutions:

[0006] The embodiment of the present application provides an extended target detection method. The method comprises the following steps: obtaining an echo matrix collected by a radar, and constructing a dynamic programming accumulation detection model of a range extended target based on the echo matrix; wherein, a state of the target in the echo matrix is defined as a joint state variable comprising a range cell index and a range extension length; based on a minimum energy loss criterion, a trade-off parameter is configured for a value function, a double probability joint constraint optimization model about target energy and noise energy is established, the trade-off parameter is determined by solving the optimization model, and noise energy accumulation in the range extension dimension is suppressed when accumulating target energy; based on the trade-off parameter, a value function and a state transition matrix of the joint state variable between multiple pulses are updated through recursive operation; the value function obtained after the updating is completed is extracted to obtain a corresponding energy ridge line, an approximate probability model of an energy ridge line change amplitude is constructed, an optimal change threshold under a preset false alarm probability is obtained by solving the approximate probability model, and a target extension length value is obtained.

[0007] In an implementation manner of the present application, an echo matrix collected by a radar is obtained, and a dynamic programming accumulation detection model of a range extended target is constructed based on the echo matrix, specifically comprising the following steps: M performing frequency domain splicing on a frame step frequency signal collected by the radar in a coherent processing interval to obtain the echo matrix;

[0008] the echo in the first n range cell in the echo matrix is represented as:

[0009] ;

[0010] wherein, a subscript represents the th pulse, a superscript represents the th range cell, is a target state corresponding to the th range cell of the th pulse; is an observation value of the target in the cell; based on the range extension characteristic, a target detection problem is modeled as a binary hypothesis testing problem, so as to construct the dynamic programming accumulation detection model of the range extended target; wherein, the hypotheses include that all range cells are noise, there is a target and distance migration correction and extension length estimation are accurate, there is a target but there is distance migration correction error and extension length estimation error.

[0011] In an implementation manner of the present application, the dynamic programming accumulation detection model of the range extended target is as follows:

[0012]

[0013] wherein, is a zero hypothesis; is the first type of hypothesis; is the second type of hypothesis; is the pulse sequence number; M is the total number of pulses within the coherent processing interval; is the range cell index; i is the range cell index; j are both general range cell index variables; represents the clutter, represents the target echo; is the range migration correction error; is the extension length estimation error; is the range extension length; is the total number of cells containing target scatterers; represents the range extension function; represents the set of all range cell indices in the echo matrix; represents the set of all range cell indices in the noise data set; represents the true target position corresponding range cell index.

[0014] In an implementation form of the present application, based on the minimum energy loss criterion, a trade-off parameter is configured for the value function, a double probability joint constraint optimization model about target energy and noise energy is established, and the trade-off parameter is determined by solving the optimization model, specifically including:

[0015] The value function calculation equation is defined as:

[0016] ;

[0017] Based on the minimum energy loss criterion, a double probability joint constraint optimization model is constructed:

[0018] ;

[0019] The optimization problem is constructed:

[0020] ;

[0021] By solving the optimization problem, the optimal trade-off parameter is adaptively determined;

[0022] wherein, f (·) represents the value function; is the observation value of the target in the cell; represents the assumed instantaneous range extension length of the target at the m th pulse when the target head is located at the n th cell; represents the assumed instantaneous range extension length of the target at the m th pulse when the target head is located at the jecho signals of the distance unit; for the joint optimization target probability; for the weight coefficient; T for the weight coefficient; for the first probability constraint, representing a probability that the weight coefficient is greater than the noise unit energy, used for suppressing noise energy accumulation; for the second probability constraint, representing a probability that the target unit energy is greater than the weight coefficient, used for ensuring that the target energy is effectively accumulated; for T optimal value.

[0023] In an implementation form of the present application, the first probability constraint is:

[0024] ;

[0025] The second probability constraint is:

[0026] ;

[0027] Based on the first probability constraint and the second probability constraint, a double-probability joint constraint model is obtained as:

[0028] ;

[0029] wherein, is the noise energy; is the target echo energy; is a shape parameter; is a scale parameter; represents the noise power; is a gamma function.

[0030] In an implementation form of the present application, based on the weight parameter, the value function and the state transition matrix of the joint state variable between multiple pulses are updated through recursive operation, specifically including: based on the echo matrix of the first pulse received by the radar, initializing the value function and the state transition matrix corresponding to all joint state variables; for each subsequent pulse, for the joint state variable corresponding thereto, the following recursive operation is performed: determining all predecessor states that transition to the current state in each subsequent pulse, forming a state transition space; in the state transition space, determining an optimal predecessor state with a maximum sum of the value function of the predecessor state and the observation value function of the current state, and updating the maximum sum as the value function of the current state; recording the optimal predecessor state to the state transition matrix; after recursively processing all pulses, the update of the value function and the state transition matrix is completed.

[0031] In an implementation manner of the present application, the value function obtained after the updating is performed is used to extract a corresponding energy ridge line, an approximate probability model of a change amplitude of the energy ridge line is constructed, and an optimal change threshold under a preset false alarm probability is solved by solving the approximate probability model, so that a target extension length value is obtained, specifically including: based on the value function obtained after the updating is performed, an energy ridge line thereof is extracted; wherein the energy ridge line is used to reflect a change rule of the value function with respect to a distance extension length; based on a change amplitude of the energy ridge line in a noise region, an approximate probability model of the change amplitude of the energy ridge line is established, and an optimal change threshold is determined by solving an optimization problem based on a preset false alarm probability constraint; the change amplitude of the energy ridge line is compared with the optimal change threshold, a demarcation point of a target region and the noise region is determined, and an extension length corresponding to the demarcation point is taken as the target extension length value.

[0032] In an implementation manner of the present application, the approximate probability model of the change amplitude of the energy ridge line is:

[0033] ;

[0034] The optimization problem is constructed as:

[0035] ;

[0036] The optimal change threshold is determined by solving the optimization problem;

[0037] wherein, is a probability function; V is a change amplitude threshold; is a number of continuous noise units; is a preset false alarm probability; denotes a set of positive integers; is a constraint condition.

[0038] The above at least one technical solution adopted by the embodiments of the present application can achieve the following beneficial effects: the embodiments of the present application jointly define a distance unit index of a target and an extension length as a joint state variable, and construct a dynamic programming accumulation detection framework, so as to realize a synchronous and adaptive optimal estimation of a target motion trajectory and an extension scale, solve the prior assumption limitation of presetting a target extension length in the prior art, and effectively cope with dynamic changes in scattering characteristics. Secondly, the embodiments of the present application construct a double-probability joint constraint optimization model, suppress the accumulation of additional noise energy caused by overestimating the size of the target, and reduce the false alarm probability, which is especially suitable for weak target detection under a low signal-to-noise ratio condition. Furthermore, the embodiments of the present application analyze an energy ridge line of a value function formed after dynamic programming accumulation, construct an approximate probability model of a change amplitude of the energy ridge line, solve an optimal change threshold for distinguishing a target region and a noise region under a preset constant false alarm probability constraint, and realize accurate estimation of a target distance extension length. BRIEF DESCRIPTION OF DRAWINGS

[0039] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments described in the present application, and other drawings can be obtained by those skilled in the art without creative labor. In the drawings:

[0040] Figure 1 An extended target detection overall flowchart is provided for the embodiments of the present application.

[0041] Figure 2 An extended target detection method flowchart is provided for the embodiments of the present application.

[0042] Figure 3 A dual-probability optimization problem solving diagram is provided for the embodiments of the present application.

[0043] Figure 4 A derivative diagram of a dual-probability constraint problem is provided for the embodiments of the present application.

[0044] Figure 5 A value function solving diagram is provided for the embodiments of the present application.

[0045] Figure 6 A false alarm probability change curve diagram under different noise cell numbers is provided for the embodiments of the present application.

[0046] Figure 7 An optimal change threshold solving result diagram is provided for the embodiments of the present application.

[0047] Figure 8 A value function energy ridge line diagram is provided for the embodiments of the present application.

[0048] Figure 9 An extended state estimation result diagram is provided for the embodiments of the present application.

[0049] Figure 10 A detection performance comparison diagram is provided for the embodiments of the present application.

[0050] Figure 11 A trajectory estimation accuracy comparison diagram is provided for the embodiments of the present application.

[0051] Figure 12 An extended length estimation error diagram is provided for the embodiments of the present application. DETAILED DESCRIPTION

[0052] The embodiments of the present application provide an extended target detection method.

[0053] In order for those skilled in the technical field to better understand the technical solutions in the present application, the technical solutions in the embodiments of the present application will be clearly and completely described in the following with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative labor should fall within the scope of protection of the present application.

[0054] Figure 1 An overall flow chart of an extended target detection method provided by an embodiment of the present application is shown in FIG. 1. First, radar echoes are acquired, and a dynamic programming model of joint state variables is introduced to synchronously process target motion and size change. A double probability optimization based on a minimum energy loss criterion is used to adaptively configure key parameters to improve detection performance. Finally, target size is accurately estimated from a value function ridge line, and a target trajectory is traced back from a state history, realizing integrated processing of extended targets from detection, tracking to feature extraction. Figure 1 Figure 1 Trade-off Parameter is a trade-off coefficient; TheRidge of Merit Function is a value function ridge line, which is a key characteristic curve representing target position and length extracted from accumulated data; Segmentation Point is a segmentation point; Range spread is a range spread.

[0055] Figure 2 An overall flow chart of an extended target detection method provided by an embodiment of the present application is shown in FIG. 1. First, radar echoes are acquired, and a dynamic programming model of joint state variables is introduced to synchronously process target motion and size change. A double probability optimization based on a minimum energy loss criterion is used to adaptively configure key parameters to improve detection performance. Finally, target size is accurately estimated from a value function ridge line, and a target trajectory is traced back from a state history, realizing integrated processing of extended targets from detection, tracking to feature extraction. Figure 2

[0056] Step 101, an echo matrix acquired by a radar is acquired, and a dynamic programming accumulation detection model of an extended target is constructed based on the echo matrix.

[0057] In an implementation manner of the present application, it is assumed that a radar transmits a stepped frequency signal in one CPI (Coherent Processing Interval), and receives, at each frame of the stepped frequency signal, one HRRP (High-Resolution Range Profile) through frequency domain splicing, each HRRP having a number of sampling points, and thus an echo matrix is finally obtained. .

[0058] It is assumed that an echo in the i th row and j th column of the echo matrix is denoted as x ij. ​​​​​​, the echo matrix can be expressed as:

[0059] (1)

[0060] (2)

[0061] where the subscript denotes the th pulse, the superscript denotes the th range cell, is the target state corresponding to the th range cell of the th pulse; is the observation value of the target in the cell.

[0062] Let denote the real target position corresponding to the range cell index. It is assumed that is the range extension length, is the total number of cells containing target scatterers, and generally, , and are unknown quantities. It is assumed that denotes the set of all range cell indices in the echo matrix . The noise data set is denoted as , and denotes the set of all range cell indices in the noise data set. The number of pulses of the noise data set is an integer multiple of the number of pulses of the echo matrix, and the noise data set has the same noise background as the echo matrix .

[0063] Subsequently, a single echo is modeled as:

[0064] (3)

[0065] where denotes the clutter, denotes the target echo.

[0066] During the energy accumulation process, the range migration correction error will cause the target energy accumulation to be incomplete, thereby causing the detection probability to decrease; and the extension length estimation error will cause additional noise cells to be accumulated, thereby increasing the false alarm rate. Let denote the range extension function.

[0067] (4)

[0068] Finally, the binary detection model can be expressed as:

[0069] (5)

[0070] where the state of the target in the echo matrix is defined as a joint state variable containing the range cell index and the range extent length. Wherein, is the zero hypothesis; is the first hypothesis; is the second hypothesis; is the pulse sequence number; M is the total number of pulses within the coherent processing interval; is the range cell sequence index; i and j are both general range cell index variables; denotes clutter, denotes target echo; is the range migration correction error; is the extent length estimation error; is the range extent length; is the total number of cells containing target scatterers; denotes the range extent function; denotes the set of all range cell indexes in the echo matrix; denotes the set of all range cell indexes in the noise data set; denotes the true target position corresponding range cell index.

[0071] In an implementation manner of the present application, based on the echo matrix of the first pulse received by the radar, the value function corresponding to all joint state variables and the state transition matrix are initialized.

[0072] Specifically, the noise energy probability distribution obeys the exponential distribution , represents the noise power, as follows:

[0073] (6)

[0074] The target energy probability distribution obeys , is the shape parameter, is the scale parameter, and the relationship is . Wherein, SNR is the signal-to-noise ratio.

[0075] (7)

[0076] Based on the extended target signal model, an extended target dynamic programming accumulation detection framework is established. First, the value function and state transition matrix are initialized as follows:

[0077] (8)

[0078] (9)

[0079] in, Represents a value function. This is a calculation function based on observations. It is the distance expansion length used as a hidden variable; This is the distance unit index for the first pulse.

[0080] Step 102: Based on the minimum energy loss criterion, configure the trade-off parameters for the value function and establish a dual-probability joint constraint optimization model for target energy and noise energy. By solving the optimization model, the trade-off parameters are determined, thereby suppressing the accumulation of noise energy in the distance extension dimension while accumulating target energy.

[0081] In one implementation of this application, the value function calculation equation proposed in this embodiment is:

[0082] (10)

[0083] This represents the tradeoff coefficient, whose physical meaning is to introduce an attenuation greater than the noise in the distance extension dimension of the value function, as shown below:

[0084] (11)

[0085] (12)

[0086] in, Extend the distance to the target by a certain length; As a regulating factor; Noise power; For the first m The pulse, the first n Noise energy per distance cell.

[0087] The method for determining the tradeoff coefficients is given below. Let... Represents the target unit energy. Represents noise energy. For a distance spread length of... The distance extension target, its The echo can be expressed by the following relationship:

[0088] (13)

[0089] The maximum value of the ideal value function can be expressed as:

[0090] (14)

[0091] The adjacent state can be expressed as:

[0092] (15)

[0093] The trade-off parameter in the embodiment of the present application needs to suppress the accumulation of noise energy in the range expansion dimension while accumulating target energy, and is constructed according to the following two constraint conditions:

[0094] (16)

[0095] (17)

[0096] According to the probability model, two constraint probabilities can be obtained:

[0097] (18)

[0098] (19)

[0099] Finally, a double probability constraint can be constructed, as shown below:

[0100] (20)

[0101] (21)

[0102] wherein, f (·) represents the value function; is the observation value of the target in the unit; represents the assumed instantaneous range expansion length of the target when the target head is located in the m th unit at the n th pulse; represents the echo signal of the m th range unit at the j th pulse; is the joint optimization target probability; is the weight coefficient; T is the trade-off coefficient; is the first probability constraint, representing the probability that the trade-off coefficient is greater than the noise unit energy, used to suppress the accumulation of noise energy; is the second probability constraint, representing the probability that the target unit energy is greater than the trade-off coefficient, used to ensure that the target energy is effectively accumulated; is the optimal value of T ; and is the noise energy; For the target echo energy; These are shape parameters; It is a scale parameter; Represents noise power; This is a gamma function.

[0103] noise power The second dataset can be used to estimate the value, therefore the optimization problem is about... , and The function.

[0104] The detection framework in this application is a noncoherent accumulation method, with a detection performance threshold of 0dB; target parameters This can be obtained through estimation of the target parameters. Subsequently, based on simulation experiments, this embodiment of the application determines the following parameters under the constraints of the physical meaning of the tradeoff coefficients and the existence of zeros in the first derivative of the optimization problem: The method of determining the value of the coefficients ultimately optimized the tradeoff coefficients of the value function.

[0105] Figure 3 This is a schematic diagram of a biprobabilistic optimization problem provided in an embodiment of this application. Figure 4 A schematic diagram of the derivative of a biprobabilistic constraint problem provided in an embodiment of this application is shown below. Figure 3 As shown, Figure 3 The graph displays the function curves of the joint optimization objective probability as a function of the adjustment factor under different weighting coefficients, and marks the optimal solution for each curve. The horizontal axis represents the adjustment factor, and the vertical axis represents the overall probability. Figure 4 As shown, Figure 4 The condition for the first derivative to be zero is given, thus verifying the result. Figure 3 The existence and uniqueness of the optimal solution are shown in the figure. The horizontal axis represents the adjustment factor, and the vertical axis represents the derivative value of the adjustment factor. Multiple solid lines represent the trend of the derivative under different weight coefficient values, while the zero dashed line represents the baseline where the derivative is zero. For each weight coefficient value, the derivative curve starts from a positive value, monotonically decreases as the adjustment factor increases, crosses zero, and then enters the negative range. The intersection of each derivative curve with the zero line corresponds exactly to... Figure 3 The adjustment factor value corresponding to the point where the probability of the target objective is maximized when the curves of the same color are jointly optimized.

[0106] Step 103: Based on the trade-off parameters, update the value function and state transition matrix of the joint state variables across multiple pulses through recursive operations.

[0107] In an implementation of the present application, for each subsequent pulse, the following recursive operation is performed for its corresponding joint state variable. All predecessor states that transition to the current state in each subsequent pulse are determined, forming a state transition space. In the state transition space, the optimal predecessor state is determined that maximizes the sum of the value function of the predecessor state and the observation function of the current state, and the maximum sum is updated as the value function of the current state, and the optimal predecessor state is recorded to the state transition matrix. After all pulses are processed recursively, the update of the value function and the state transition matrix is completed.

[0108] Specifically, determine which states can transition to the current state in the first m -1 pulse (i.e., the last pulse). The present application defines a state transition space, which is determined by physical constraints, i.e., the maximum number of distance units that the target can cross between adjacent pulses due to movement, and the reasonable range of the extension length in a short time. Through these two constraints, all candidate states that can be the predecessor of the current state are filtered out to form a set of predecessor states. The accumulated energy of each predecessor state in the set is calculated, and then the sum of the energy of the k distance units covered by the current state according to the optimal trade-off parameter is added. All predecessor states are traversed, the sums are compared, and the maximum value is determined as the new value function of the current state. This maximum value represents the maximum cumulative energy that can be obtained from the starting pulse to the current pulse with the current state as the end point. When the predecessor state that maximizes the cumulative energy is determined, the selection is recorded, and the state transition matrix is a database for storing backtracking information. For the current state, the position of the optimal predecessor state and the extension length are recorded in the corresponding position in the matrix. All possible states of the current pulse, i.e., all possible combinations of positions and extension lengths, are traversed, and the above steps are repeated for each state. After updating all states of the current pulse, proceed to the next pulse and repeat the entire process again. This iterative cycle advances from one pulse to the next pulse until all M pulses in the coherence processing interval are processed. When the value function of all states of the last pulse and the state transition matrix are updated, the entire recursive operation process is completed.

[0109] Further, after initialization, the recursive accumulation of the value function and the update of the state transition matrix are started, as follows:

[0110] (22)

[0111] (23)

[0112] wherein the state and in the first The state transition space in one pulse repetition period is The corresponding state transition matrix is .

[0113] After accumulation of one (Pulse Repetition Time, Pulse Repetition Time) accumulation, the accumulated value function is compared with the detection threshold, and all states greater than the threshold are found.

[0114] (24)

[0115] The complete track can be detected by backtracking using the state transition matrix. Among them, is the output target state set; represents the estimated state of the target; is the final value function.

[0116] Step 104, extracting the corresponding energy ridge line of the value function obtained after updating, and constructing an approximate probability model of the change amplitude of the energy ridge line, obtaining the target extension length value by solving the optimal change threshold of the approximate probability model under the preset false alarm probability.

[0117] In an implementation manner of the present application, based on the value function obtained after updating, the energy ridge line thereof is extracted; wherein the energy ridge line is used to reflect the change law of the value function with the distance extension length. Based on the change amplitude of the energy ridge line in the noise region, an approximate probability model of the change amplitude of the energy ridge line is established, and based on the preset false alarm probability constraint, the optimal change threshold is determined by solving the optimization problem. The change amplitude of the energy ridge line is compared with the optimal change threshold to determine the demarcation point of the target region and the noise region, and the extension length corresponding to the demarcation point is taken as the target extension length value.

[0118] Specifically, Figure 5 A value function solving schematic diagram provided by an embodiment of the present application is as shown in Figure 5 The three-dimensional surface graph depicts the complete landscape of the value function changing with the target distance unit index and the assumed extension length, and clearly forms an energy ridge line (red curve in the figure). The two-dimensional profile graph fixes the distance unit and shows the characteristics of the value function changing with the extension length, that is, in the target region, the value function converges the target energy and rapidly rises; if there is a noise unit in the distance extension region, the ridge line change will tend to be flat; and in the pure noise region, the value function only converges noise energy, and its value tends to be stable or gradually decreases. The real distance extension length is located at the region demarcation, not the peak position. Therefore, the energy change of the noise region can be approximately regarded as the sum of a noise unit, and the noise energy unit obeys an exponential distribution According to the additivity of the gamma distribution, The cumulative energy of the noise units obeys the gamma distribution Wherein, represents the number of noise units, and the value is a positive integer.

[0119] Embodiments of the present application estimate the target extension length by energy ridge line change. Set as the threshold of energy ridge line change. If the energy ridge line change amplitude in a certain region exceeds , it is determined as a target region, otherwise it is determined as a noise region. Analogous to the definition of false alarm probability in detection theory, the probability that the energy ridge line change amplitude in the noise region exceeds is defined as For a given value and any positive number , the solution is to find the optimal threshold , so that the probability that the energy ridge line change amplitude exceeds is less than the preset value

[0120] (25)

[0121] The formula can be simplified as:

[0122] (26)

[0123] (27)

[0124] The optimization problem can be represented as:

[0125] (28)

[0126] Subsequently, the optimal threshold is obtained by solving the problem through simulation verification. Wherein, is a probability function; V is a change amplitude threshold; is the number of continuous noise units; is a preset false alarm probability; represents a set of positive integers; is a constraint condition.

[0127] Figure 6 is a false alarm probability change curve provided by an embodiment of the present application under different noise unit numbers, as shown in Figure 6 Under different assumed continuous noise unit numbers, the probability that the energy ridge line change amplitude is misjudged as a target signal changes with the decision threshold. As the threshold increases, the false alarm probability monotonically decreases. Figure 7A result graph of solving an optimal change threshold provided by an embodiment of the present application is shown in Figure 7 , which presents the final result of optimization solving based on Figure 6 a probability model. The curve in the graph represents the envelope of the best decision performance with respect to the change of the threshold under the constraint of meeting the preset false alarm probability. The peak point marked on the curve is the optimal solution determined by maximizing the objective function. The threshold value corresponding to the point is most sensitive to the change of the target energy ridge line while strictly controlling the system false alarm. Figure 7 The effectiveness of the optimization problem is verified, and the key parameters for accurately estimating the target extension length are given.

[0128] Figure 8 A value function energy ridge line graph provided by an embodiment of the present application is shown in Figure 8 , which shows the energy ridge line characteristics of the value function obtained after dynamic programming accumulation. The horizontal axis in the graph represents the change of the target distance extension length, and the vertical axis represents the value function output under the length assumption. The discrete ridge points are connected by a fitting curve, which clearly shows the typical change rule of the energy ridge line in the extension length dimension: in the initial stage, the value function increases rapidly and approximately linearly with the increase of the extension length, which indicates that the target energy is effectively accumulated; when the extension length exceeds the actual target size and enters the pure noise region, the fitting curve tends to be flat, and the value function growth stagnates or even decreases slightly. Figure 9 A result graph of extension state estimation provided by an embodiment of the present application is shown in Figure 9 , which presents the final result of accurate estimation of the target extension length based on the energy ridge line. The blue solid line in the graph reproduces the energy ridge line, and a segmentation point is automatically determined on the ridge line by solving the optimal change threshold. The point before the segmentation point is the effective target energy accumulation area, and the point after the segmentation point is the noise platform area. According to the segmentation point, the accurate estimation value of the target extension length can be obtained.

[0129] Figure 10 A detection performance comparison graph provided by an embodiment of the present application is shown in Figure 10 , which shows the detection probability comparison of the method in the embodiment of the present application and other existing target extension detection algorithms under different signal-to-noise ratios. As shown in Figure 10 , with the improvement of the signal-to-noise ratio, the detection probability of all methods shows an upward trend. The method in the embodiment of the present application shows a significant performance advantage in the entire signal-to-noise ratio interval, especially in the low signal-to-noise ratio region, and its detection probability is significantly higher than that of other comparison methods.

[0130] Figure 11 A trajectory estimation accuracy comparison graph provided by an embodiment of the present application is shown in Figure 11As shown, the estimation accuracy of different methods for the target trajectory is quantitatively evaluated, and the root mean square error is used as an evaluation index. In the figure, the method in the embodiment of the present application, the dynamic programming detection-before-tracking method based on local energy likelihood ratio, and the similar method based on local contrast likelihood ratio are compared. The results show that the trajectory of the method in the embodiment of the present application rapidly decreases with the increase of the signal-to-noise ratio and tends to a very low stable value, and the overall accuracy is obviously better than the other two methods.

[0131] Figure 12 An extended length estimation error graph is provided for the embodiment of the present application, as shown in Figure 12 The red dot broken line shows the trend of the length estimation error with the change of the signal-to-noise ratio. It can be clearly seen that in the low signal-to-noise ratio condition, due to the poor signal quality, the estimation error is relatively large; with the increase of the signal-to-noise ratio, the estimation error rapidly decreases and finally tends to a stable state close to zero. The smooth and monotonically decreasing characteristics of the curve prove that the length estimation method proposed based on the value function energy ridge line change rule and the approximate probability model in the embodiment of the present application has robustness and convergence, and can realize high-precision and stable estimation of the target physical size in a wide signal-to-noise ratio range.

[0132] Each of the embodiments in the present application is described in a progressive manner, and the same and similar parts between each embodiment can be referred to each other. Each embodiment mainly explains the difference from other embodiments. Especially, for the device, equipment and non-volatile computer storage medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the related parts can be referred to the part of the method embodiment.

[0133] The above only describes the embodiments of the present application and is not used to limit the present application. The embodiments of the present application can be variously changed and modified by those skilled in the art. The modifications or replacements do not make the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. An extended target detection method, characterized in that, The method includes: The radar echo matrix is ​​acquired, and a dynamic programming accumulation detection model for range-extended targets is constructed based on the echo matrix; wherein, the state of the target in the echo matrix is ​​defined as a joint state variable including the range cell index and the range extension length; Based on the minimum energy loss criterion, a trade-off parameter is configured for the value function, and a dual-probability joint constraint optimization model for target energy and noise energy is established. The trade-off parameter is determined by solving the optimization model, thereby suppressing the accumulation of noise energy in the distance extension dimension while accumulating target energy. Based on the aforementioned trade-off parameters, the value function and state transition matrix of the joint state variables across multiple pulses are updated through recursive operations; For the value function obtained after the update, the corresponding energy ridge line is extracted, and an approximate probability model of the change amplitude of the energy ridge line is constructed. By solving the optimal change threshold of the approximate probability model under the preset false alarm probability, the target extension length value is obtained. The process of acquiring the radar echo matrix and constructing a dynamic programming accumulation detection model for range-extended targets based on the echo matrix specifically includes: Data collected by radar within the coherent processing interval M The frame step frequency signals are spliced ​​in the frequency domain to obtain the echo matrix; The first in the echo matrix n The echo of a distance cell is represented as follows: ； Among them, subscript Indicates the first One pulse, superscript Indicates the first One distance unit, For the first The first pulse The target state corresponding to each distance unit; The observation value of the target in this cell; Based on the distance extension characteristics, the target detection problem is modeled as a binary hypothesis testing problem to construct a dynamic programming accumulation detection model for the distance-extended target; wherein, the hypothesis includes that all distance cells are noise, that there is a target and the distance migration correction and extension length estimation are accurate, and that there is a target but there are distance migration correction errors and extension length estimation errors; The dynamic programming accumulation detection model for the distance expansion target is as follows: in, Null hypothesis; This is a first-type hypothesis; This is a second type of hypothesis; This is the pulse sequence number; M This represents the total number of pulses within the coherent processing interval. This is the distance cell index; i and j These are all general distance cell index variables; Indicates clutter. Indicates the target echo; This is for distance migration correction error; To extend the length estimation error; For distance extension length; This represents the total number of units containing the target scatterer; Represents the distance spread function; This represents the set of all distance cell indices in the echo matrix; Represents the set of all distance cell indices in the noisy dataset; Indicates the actual target location The corresponding distance cell index.

2. The extended target detection method according to claim 1, characterized in that, The method involves configuring trade-off parameters for the value function based on the minimum energy loss criterion, establishing a biprobabilistic joint constraint optimization model for the target energy and noise energy, and determining the trade-off parameters by solving the optimization model. Specifically, this includes: The equation for calculating the value function is defined as follows: ； Based on the minimum energy loss criterion, the dual-probability joint constraint optimization model is constructed as follows: ； Optimization problem: ； By solving the optimization problem, the optimal trade-off parameters are adaptively determined. in, f (·) denotes a value function; The observation value of the target in this cell; Indicates the first m The pulse, the target head is located at the... n When the target is in units, the assumed instantaneous distance extension length; Indicates the first m The pulse, the first j Echo signal of each distance unit; To jointly optimize the target probability; These are the weighting coefficients; T For the weighting factor; The first probability constraint represents the probability that the tradeoff coefficient is greater than the noise unit energy, which is used to suppress noise energy accumulation; The second probability constraint represents the probability that the target unit energy is greater than the tradeoff coefficient, which is used to ensure that the target energy is effectively accumulated. for T The optimal value.

3. The extended target detection method according to claim 2, characterized in that, The first probability constraint is: ； The second probability constraint is: ； Based on the first probability constraint and the second probability constraint, the dual-probability joint constraint optimization model is obtained as follows: ； in, For noise energy; For the target echo energy; These are shape parameters; It is a scale parameter; Represents noise power; This is a gamma function.

4. The extended target detection method according to claim 1, characterized in that, The step of updating the value function and state transition matrix of the joint state variables across multiple impulses through recursive operations based on the aforementioned trade-off parameters specifically includes: Based on the echo matrix of the first pulse received by the radar, initialize the value functions and state transition matrices corresponding to all joint state variables; For each subsequent pulse, perform the following recursive operation on its corresponding joint state variable: Identify all the preceding states that transition to the current state in each subsequent pulse, forming the state transition space; In the state transition space, the optimal predecessor state that maximizes the sum of the value function of the predecessor state and the observation function of the current state is determined, and the maximum sum is updated as the value function of the current state; Record the optimal predecessor state into the state transition matrix; After recursively processing all pulses, the logistic function and state transition matrix are updated.

5. The extended target detection method according to claim 1, characterized in that, The updated value function is used to extract the corresponding energy ridgeline, and an approximate probability model of the energy ridgeline's variation amplitude is constructed. By solving the optimal variation threshold of the approximate probability model under a preset false alarm probability, the target extension length value is obtained, specifically including: Based on the value function obtained after the update, its energy ridge is extracted; wherein, the energy ridge is used to reflect the change law of the value function with the distance expansion length; Based on the variation amplitude of the energy ridge line within the noise region, an approximate probability model of the variation amplitude of the energy ridge line is established, and based on the preset false alarm probability constraint, the optimal variation threshold is determined by solving the optimization problem. The variation amplitude of the energy ridge line is compared with the optimal variation threshold to determine the boundary point between the target region and the noise region. The extension length corresponding to the boundary point is taken as the target extension length value.

6. The extended target detection method according to claim 3, characterized in that, The approximate probabilistic model for the amplitude of the energy ridge variation is as follows: ； The optimization problem is: ； The optimal change threshold is determined by solving the optimization problem; in, It is a probability function; V The threshold for the range of change; This represents the number of continuous noise units. To preset the false alarm probability, Represents the set of positive integers; These are constraints.

Images