A performance evaluation method and system of a seeker neural network tracking algorithm

Abstract

The application is a performance evaluation method and system of a seeker neural network tracking algorithm, belonging to the field of software evaluation, including simulating a moving target to test the seeker, obtaining multiple tracking performance index data of the seeker to the moving target based on the neural network tracking algorithm; based on the multiple tracking performance index data, converting each tracking performance index data into a single index score result through linear assignment and weighted fusion respectively; judging whether there is a single index score result lower than the lower threshold of the single index score, if there is, the comprehensive performance evaluation result is 0; if not, the single index score result higher than the upper threshold of the single index score is judged as full score, and the single index score result without arbitration conditions is weighted and fused to obtain the comprehensive performance evaluation result; judging whether the comprehensive performance score has arbitration conditions, if not, it is zero, otherwise, it is reserved, and the comprehensive performance evaluation result is obtained. The problem of lacking a special performance evaluation scheme for the seeker neural network tracking algorithm is solved.

Description

Technical Field

[0001] This invention belongs to the field of software evaluation technology, and in particular relates to a performance evaluation method and system for a seeker neural network tracking algorithm. Background Technology

[0002] In increasingly complex battlefield environments, the number of neural network recognition-tracking algorithms for seekers is growing, and the high complexity of their parameters and structures makes performance evaluation increasingly difficult. Traditional software evaluation methods do not consider the specific performance indicators of neural networks, nor do they address the closed-loop parameters of seeker tracking, making them unsuitable for the rapid development of seeker neural network tracking algorithms.

[0003] Research on the evaluation of seeker neural network tracking algorithms is limited, but the design of neural network tracking algorithms is developing rapidly. Existing literature includes some systems for evaluating the anti-interference performance of seekers, but these are limited to anti-interference performance evaluation and do not involve intelligent algorithms. Others provide evaluation schemes for neural network filtering algorithms, using grey relational analysis to eliminate the subjectivity of weight design, but these only address the anti-interference performance of seekers. In weighted fusion techniques, some literature provides ideas for the initial weight design in the evaluation of seeker neural network algorithms; this invention will also use a similar method to eliminate subjectivity.

[0004] There is still considerable room for research and understanding in the evaluation of seeker algorithms. However, with the continuous increase in algorithm complexity and the requirements of specialized seekers, performance evaluation is becoming increasingly important. For example, two object-oriented ship detectors based on fast regional convolutional neural networks have been proposed, trained from thousands of X-band airborne range compression radar data containing multiple ship signals. Adaptive filtering techniques are also proposed in seekers to mitigate the effects of interference, thus requiring consideration of spectral and spatially diverse wave signals to evaluate suppression performance. Synthetic aperture radar observation data has also been used to estimate soil moisture and surface roughness, thereby evaluating surface backscattering models. Some literature has derived the autocorrelation and cross-correlation sidelobes of seekers and integrated them into the evaluation theory of polyphase code orthogonality. In ground-penetrating radar, an anomaly detection algorithm based on a visual attention mechanism has been proposed multiple times and repeatedly verified and evaluated. However, the above-mentioned guidance system evaluation methods rarely consider the special characteristics of neural network tracking. Therefore, a specialized intelligent evaluation and optimization scheme must be designed for the neural network tracking algorithm of seekers.

[0005] Neural network recognition and tracking algorithms have developed rapidly in the field of motion control. Through training, neural networks can enable controllers to lock onto targets and adjust actuators for precise tracking. However, related evaluation techniques are rarely studied. For example, some studies provide large datasets of low-resolution and super-resolution images and use them to evaluate deep learning super-resolution models in terms of signal-to-noise ratio and upscaler. Some studies also specifically review current research on making deep neural networks secure and reliable, focusing on four aspects: verification, testing, adversarial attack and defense, and interpretability. Some studies train BP neural networks using the Matlab Neural Network Toolbox based on obtained features and evaluate the established continuous motion control models. Some studies propose many new PID (Proportional-Integral-Derivative) control wavelet neural network models to control servo motors and evaluate the performance of intelligent control in terms of tracking error and response time. However, the above evaluation techniques do not consider the special characteristics of the seeker. Summary of the Invention

[0006] Based on the above analysis, the embodiments of the present invention aim to provide a performance evaluation method and system for a seeker neural network tracking algorithm, to solve the problems in the prior art of lacking a dedicated intelligent performance evaluation scheme, lacking comprehensive performance evaluation, and lacking special performance evaluation for seekers in the performance evaluation of seeker neural network tracking algorithms.

[0007] This solution provides a performance evaluation method for a seeker neural network tracking algorithm, including the following steps:

[0008] Step S1: Simulate a moving target to test the seeker, and obtain multiple tracking performance index data of the seeker on the moving target based on the neural network tracking algorithm;

[0009] Step S2: Based on the multiple tracking performance index data, each tracking performance index data is converted into a single index score result through linear scoring and weighted fusion;

[0010] Step S3: Determine whether there are any single indicator scores below the lower threshold of the single indicator score. If so, the comprehensive performance evaluation result is 0. If not, the single indicator scores above the upper threshold of the single indicator score are judged as full marks, and then weighted and merged with the single indicator scores that do not meet the arbitration conditions to obtain the comprehensive performance evaluation result. Determine whether the comprehensive performance score meets the arbitration conditions. If not, the comprehensive performance evaluation result is set to zero; otherwise, it is retained to obtain the comprehensive performance evaluation result of the tracking performance indicator.

[0011] Furthermore, during testing, the x and y axis position information within the seeker's imaging plane and the angle information of the moving target are collected; the acquisition of multiple tracking performance index data of the seeker for the moving target based on the neural network tracking algorithm includes:

[0012] The three neurons in the tracking algorithm are input with the x and y axis position information and the angle information of the moving target according to the time sequence;

[0013] There are n output neurons, two of which contain tracking signals containing the position information of the target point on the x and y axes. The tracking signal values ​​are taken according to the time series to obtain multiple tracking performance index data of the seeker for the moving target.

[0014] The angle of the moving target is the angle between the line connecting the moving target and the center point of the seeker's main reflective surface and the central axis of the seeker.

[0015] Furthermore, the process of converting each tracking performance index data into a single index score result through linear scoring and weighted fusion includes:

[0016] Linear scoring transforms each performance tracking metric into a normalized dynamic score, specifically:

[0017] v = HI(u) = au + b

[0018]

[0019] Where u is the tracking performance index data value, v is the dynamic score value, a and b are real constants, and v max The maximum value u of the tracking performance index data determined after repeated experiments. max The corresponding rating, v * For normalized dynamic scoring;

[0020] After linear scoring transformation, the array of single-indicator score values ​​is converted into numerical values ​​using weighted fusion.

[0021] Furthermore, the step of converting the array of single-indicator score values ​​into numerical values ​​using weighted fusion includes:

[0022] F i =w(J v,i )J v,i +w(σ v,i )σ v,i

[0023] Among them, F i For a single indicator score, w(J) v,i ) and w(σ v,i ) represents the weights of the mean and variance of the normalized dynamic score v*, i = 1, 2, ..., Q, where Q is the sequence number of the tracking performance index data.

[0024] Further, step S3 includes:

[0025] Step S31: Preset the threshold range for single-index scoring and the lower threshold for the comprehensive performance score of the guide head neural network tracking index;

[0026] Step S32: Determine whether there are any single indicator scores below the lower threshold of the single indicator score. If there are, the overall performance evaluation result is 0; if not, the single indicator scores above the upper threshold of the single indicator score are judged as full marks.

[0027] Step S33: After judging the single indicator score results that are higher than the upper threshold of the single indicator score as full marks, the comprehensive performance score is obtained by weighted fusion calculation with the single indicator score results that do not meet the arbitration conditions.

[0028] Step S34: Determine whether the overall performance score is lower than the lower threshold of the overall performance score of the seeker neural network tracking index. If it is lower, the overall performance score of the seeker neural network tracking index is directly set to zero. If not, the overall performance score is retained.

[0029] Furthermore, the single-index scoring threshold range includes:

[0030] The single-index scoring for each tracking performance indicator is preset with a single-index scoring threshold range;

[0031] The single-index scoring threshold range includes an upper threshold and a lower threshold.

[0032] Further, step S33 includes:

[0033] After judging the single indicator score results that are higher than the upper threshold of the single indicator score as full marks, the grey relational analysis method is used to perform weighted fusion calculation on each single indicator score with the single indicator score results that do not meet the arbitration conditions to obtain the comprehensive performance score.

[0034] γ(u i ,u j Let q be the grey relational coefficient between the i-th and j-th tracking performance metrics, and let q be the corresponding grey relational depth coefficient function. ij for:

[0035]

[0036] Among them, γ(u i ,u j Let ) be the i-th tracking performance indicator u between the i-th indicator and indicator j. i With the j-th tracking performance metric u j The grey relational coefficient, where Q is the sequence number of the tracking performance index data;

[0037] The constraint of the weight value change range is:

[0038]

[0039] The constraint condition of the weight variance is:

[0040]

[0041] where D(q0) is the variance set of the full-index grey correlation depth coefficient set q0, and Q * is to satisfy Q * <Q represents any summation quantity less than the total index quantity, and D(q0) represents the fluctuation range of the index weight value as:

[0042]

[0043] Furthermore, the comprehensive performance evaluation result of the seeker neural network tracking algorithm is:

[0044]

[0045] where w e,i represents the overall weight of index i, and F represents the comprehensive performance evaluation result.

[0046] Furthermore, the tracking performance indexes include x-axis position information and y-axis position information;

[0047] The x-axis position information includes x-axis steady-state error, x-axis response time, and x-axis overshoot;

[0048] The y-axis position information includes y-axis steady-state error, y-axis response time, and y-axis overshoot.

[0049] This solution also provides a performance evaluation system for a seeker neural network tracking algorithm, including:

[0050] An acquisition module M1, which is used to simulate a moving target to test the seeker, and obtain multiple tracking performance index data of the seeker for the moving target based on the neural network tracking algorithm;;

[0051] A single-index scoring evaluation module M2, which is used to respectively convert each tracking performance index data into a single-index scoring result through linear scoring and weighted fusion based on the multiple tracking performance index data;

[0052] The comprehensive performance scoring and evaluation module M3 is used to determine whether there are any single indicator scores below the lower threshold of the single indicator score. If so, the comprehensive performance evaluation result is 0; if not, the single indicator scores above the upper threshold of the single indicator score are judged as full marks, and then weighted and merged with the single indicator scores that do not meet the arbitration conditions to obtain the comprehensive performance evaluation result; it is determined whether the comprehensive performance score meets the arbitration conditions. If not, the comprehensive performance evaluation result is set to zero; otherwise, it is retained to obtain the comprehensive performance evaluation result of the tracking performance indicator.

[0053] Compared with the prior art, the technical solution of the present invention can achieve at least one of the following beneficial effects:

[0054] (1) Comprehensive performance evaluation: This invention provides a method for evaluating the neural network tracking algorithm of a seeker, which can help researchers conduct a comprehensive performance evaluation of radar seekers, infrared seekers and optical seekers with target tracking capabilities;

[0055] (2) Objective weight setting: This invention provides a method for objective weight design. The grey relational method is used to determine the weight of each single indicator, avoiding the extremes and opaque operations of subjective weighting methods such as expert scoring.

[0056] (3) Arbitration Mechanism: This invention provides an arbitration mechanism that mimics the direct judgment of the judicial system on events. This mechanism directly judges the reasonable range of algorithm performance based on thresholds. For scores outside the upper and lower thresholds, it avoids cumbersome weighted fusion and repeated evaluation, reducing workload;

[0057] (4) Universality: This invention is universal and can be used to evaluate neural network recognition algorithms for seekers. By replacing the number of network nodes with other algorithm structure parameters, this invention can also be used to evaluate non-intelligent tracking algorithms. Furthermore, it is useful for control algorithms directly related to the closed-loop performance of seekers.

[0058] In this invention, the above-described technical solutions can be combined with each other to achieve more preferred combinations. Other features and advantages of this invention will be set forth in the following description, and some advantages may become apparent from the description or be learned by practicing the invention. The objects and other advantages of this invention can be realized and obtained from what is particularly pointed out in the description and drawings. Attached Figure Description

[0059] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts.

[0060] Figure 1 A flowchart illustrating a performance evaluation method for a seeker neural network tracking algorithm;

[0061] Figure 2 This is a schematic diagram of the BP neural network algorithm structure;

[0062] Figure 3a It is the x-axis tracking curve of the seeker neural network tracking algorithm;

[0063] Figure 3b It is the y-axis tracking curve of the seeker's neural network tracking algorithm;

[0064] Figure 4 It is a bar chart showing the corresponding weights of each indicator;

[0065] Figure 5 These are bar charts showing the individual scores and weighted overall performance scores for each indicator;

[0066] Figure 6 This is a schematic diagram of a performance evaluation system for a seeker neural network tracking algorithm. Detailed Implementation

[0067] Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which form part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not intended to limit the scope of the present invention.

[0068] This solution provides a comprehensive and effective closed-loop evaluation scheme to achieve local and global performance evaluation of the seeker neural network tracking algorithm in multiple scenarios and multiple segments.

[0069] The seeker is a key component used in guiding missiles, navigation systems, or aircraft. It includes radar seekers, infrared seekers, and optical seekers. A moving target is an object or target in motion whose position, velocity, or direction may change in a short period of time.

[0070] One of the tasks of a seeker is to track moving targets. The seeker's sensors are typically used to detect and track the position and motion information of moving targets. Once the seeker determines the target's position and motion, it can use this information to calculate and adjust the trajectory of the missile or aircraft to ensure it can accurately track and hit the moving target. The seeker guides the missile or aircraft to strike the moving target through its sensors and onboard neural network tracking algorithms.

[0071] Example 1:

[0072] This invention discloses a performance evaluation method for a seeker neural network tracking algorithm. Specifically, for the seeker neural network tracking algorithm, this invention proposes a comprehensive performance evaluation method with a single-index arbitration mechanism and a grey relational weighting mechanism. For a seeker specified by the user and with a clearly defined hardware environment, a moving target is simulated for testing, and multiple tracking performance index data within the view coordinates are obtained to evaluate the performance of the tracking algorithm.

[0073] Combined with the number of neurons, an indicator reflecting the complexity of neural network tracking algorithms, the indicator data is transformed into a single indicator score using a linear scoring method for each indicator.

[0074] The arbitration mechanism is designed to ensure that for each single indicator score that is not within the threshold range, a single indicator score below the lower threshold is directly determined to be zero, and the evaluation ends early; a single indicator score above the upper threshold is determined to be full marks, and then, together with the single indicator scores that do not meet the arbitration conditions, a grey relational method is used to assign weights to each indicator score, and a comprehensive performance score is calculated through weighted fusion.

[0075] If the overall performance score does not meet the arbitration criteria, i.e. it is lower than the threshold of the overall score of the seeker neural network algorithm, then the overall performance score is judged to be zero; otherwise, it is retained.

[0076] Ultimately, scores that do not meet the arbitration criteria generate an evaluation result containing single-index scores, a comprehensive performance score, and commentary explanations; scores that do meet the arbitration criteria have their comprehensive performance score reset to zero, thus creating conditions for improving the seeker neural network tracking algorithm.

[0077] like Figure 1 The diagram shows a flowchart of a performance evaluation method for a seeker neural network tracking algorithm. Multiple tracking performance indicators within the view coordinate system are obtained and combined with the number of neurons, then converted into single-indicator scores using linear scoring. An arbitration mechanism ensures that each indicator score outside a threshold range is directly determined as a comprehensive performance score or zeroed out. Furthermore, for scores that do not meet the arbitration criteria, a weighted fusion is used to calculate the comprehensive performance score, and the arbitration criteria are further evaluated. Finally, scores that do not meet the arbitration criteria generate an evaluation result containing single-indicator scores, a comprehensive performance score, and explanatory comments.

[0078] Including steps S1-S4, as follows:

[0079] Step S1: Simulate a moving target to test the seeker, and obtain multiple tracking performance index data of the seeker on the moving target based on the neural network tracking algorithm;

[0080] Step S2: Based on the multiple tracking performance index data, each tracking performance index data is converted into a single index score result through linear scoring and weighted fusion;

[0081] Step S3: Determine whether there are any single indicator scores below the lower threshold of the single indicator score. If so, the comprehensive performance evaluation result is 0. If not, the single indicator scores above the upper threshold of the single indicator score are judged as full marks, and then weighted and merged with the single indicator scores that do not meet the arbitration conditions to obtain the comprehensive performance evaluation result. Determine whether the comprehensive performance score meets the arbitration conditions. If not, the comprehensive performance evaluation result is set to zero; otherwise, it is retained to obtain the comprehensive performance evaluation result of the tracking performance indicator.

[0082] Step S4: Display of comprehensive evaluation results.

[0083] Step S1, specifically.

[0084] For a seeker with a specific hardware environment specified by a user who has evaluation requirements, a specific hardware environment is determined, and a moving target is simulated in this environment to execute the seeker's neural network tracking algorithm and obtain multiple tracking performance indicators of the seeker for the moving target, including steps S11-S12.

[0085] Step S11: Based on the hardware environment of the seeker to be evaluated, simulate a moving target to test the seeker. During the test, collect the x and y axis position information in the image plane of the seeker and the angle information of the moving target.

[0086] During this process, evaluation test index data within the view coordinate system are collected and acquired. These indexes cover the performance of the algorithm in different scenarios and segments.

[0087] A simulated moving target was used for testing to obtain the tracking performance data of the seeker and the angle of the moving target within the view coordinate system. During the seeker testing before evaluation, the moving target was set within a reasonable range, namely a constant velocity V0, and a rectangular motion plane with length x0, width Y0, and a vertical distance H0 from the seeker to the rectangular motion plane that could cause pitch and azimuth changes in the seeker. These parameters can be freely set according to specific circumstances, and once set, they remain fixed.

[0088] Among them, the tracking performance indicators of the seeker for moving targets within the field of view coordinates include: x-axis steady-state error, y-axis steady-state error, x-axis response time, y-axis response time, x-axis overshoot, and y-axis overshoot within the seeker's imaging plane; these six indicators are collectively referred to as x-axis position information and y-axis position information, respectively.

[0089] x-axis steady-state error, x-axis response time, and x-axis overshoot are collectively referred to as x-axis position information.

[0090] The steady-state error of the y-axis, the response time of the y-axis, and the overshoot of the y-axis are collectively referred to as the y-axis position information.

[0091] The definition of the index is consistent with the conceptual definitions in the principles of automatic control, and is defined as follows:

[0092] (1) x-axis steady-state error: the maximum value of the horizontal projection of the target point on the imaging surface of the seeker after tracking has stabilized;

[0093] (2) y-axis steady-state error: the maximum value of the longitudinal projection of the target point on the imaging surface of the seeker after tracking has stabilized;

[0094] (3) x-axis response time: In the horizontal axis projection value, the fluctuation range is ±5% of the ideal value. The time consumed from the start of tracking to the moment when the tracking value first enters the range and no longer jumps out.

[0095] (4) y-axis response time: The vertical axis projection value fluctuates within ±5% of the ideal value. The time consumed from the start of tracking to the moment when the tracking value first enters the range and no longer jumps out.

[0096] (5) x-axis overshoot: The percentage of the maximum error after the tracking value first crosses the coordinate zero point in the horizontal axis projection value to the initial error at the start of tracking;

[0097] (6) y-axis overshoot: The percentage of the maximum error after the tracking value first crosses the coordinate zero point in the vertical axis projection value to the initial error at the start of tracking.

[0098] The angle of the moving target is the angle between the line connecting the moving target and the center point of the seeker's main reflector and the central axis of the seeker.

[0099] Step S12: Based on the multiple x and y axis position information and the angle information of the moving target, a neural network tracking algorithm is used to obtain multiple tracking performance index data of the seeker for the moving target.

[0100] Based on the x and y axis position information and the angle information of the moving target, the BP (Back Propagation) neural network tracking algorithm is used to convert performance evaluation indicators and obtain tracking signals with x and y axis position information.

[0101] A backpropagation (BP) neural network is a multi-layer feedforward neural network consisting of an input layer, hidden layers, and an output layer. Layers are fully interconnected, but neurons within a single layer are not connected. Its learning method is a supervised learning type. The algorithm structure is as follows: Figure 2 As shown.

[0102] The input layer neurons are (x1, x2, ..., x...). m The system receives x-axis position information, y-axis position information, and angle information of the moving target from the seeker test; the hidden layer neurons are (s1, s2, ..., s...). pThe output layer neurons are located between the input and output layers and are used to process intermediate information; the output layer neurons are (y1, y2, ..., y...). n These neurons provide the output of the neural network, outputting tracking signals with x-axis and y-axis position information. The total number of neurons is x0 = m + p + n. The weights from the input layer to the hidden layer and from the hidden layer to the output layer are ω. ih and ω hj The thresholds for hidden layer neurons and output layer neurons are θ, respectively. i and θ j .

[0103] In the tracking algorithm, the neural network connecting the seeker head has m input neurons, of which at least 3 neurons receive information including the x and y-axis position of the moving target and the angle information of the moving target. Specifically,

[0104] The first neuron node receives x-axis position information in a time sequence;

[0105] The second neuron receives y-axis position information in a time sequence;

[0106] The third neuron receives the angle information of the moving target in sequence.

[0107] The neural network has p hidden layer neurons and n output layer neurons, of which at least two neurons contain tracking signals for the position information of the target point trace on the x and y axes.

[0108] Step S1 obtains the tracking signals of the x-axis position information and y-axis position information output by the output neuron, and takes the values ​​of the tracking signals according to the time series to obtain multiple tracking performance index data of the seeker for the moving target, which are used for specific tracking position information.

[0109] Step S2, specifically.

[0110] To transform each tracking performance metric into a single-metric score result using linear scoring and weighted fusion.

[0111] Let u be the tracking performance index data value and v be the dynamic score value. Then the corresponding relationship of the smooth linear scoring method is shown in formulas (1)-(2):

[0112] v = HI(u) = au + b (1)

[0113]

[0114] Where a and b are real constants, v max The maximum value u of the tracking performance index data determined after repeated experiments. max The corresponding rating, v* For normalized dynamic scoring.

[0115] After scoring using a linear scoring mechanism, the individual indicator scores obtained from repeated experiments are arrays rather than single numerical values. Therefore, it is necessary to convert the score array into individual scores, ultimately forming the individual indicator score F. i Take the normalized dynamic score v * The mean is J v,i The variance is σ v,i .

[0116] Single indicator score F i The result is obtained through weighted fusion calculation, as shown in formula (3).

[0117] F i =w(J v,i )J v,i +w(σ v,i )σ v,i (3)

[0118] Among them, w(J) v,i ) and w(σ v,i () is the normalized dynamic score v * The weights of the mean and variance are i = 1, 2, ... Q, where Q is the index of the tracking performance index data and Q satisfies Q ≤ 5.

[0119] Step S3, specifically.

[0120] Step S31: Preset the threshold range for single-index scoring and the lower threshold for the comprehensive performance score of the guide head neural network tracking index;

[0121] For each tracking performance indicator, a single indicator score is set, and a single indicator score threshold range is preset. The single indicator score threshold range includes an upper threshold and a lower threshold. A lower threshold for the comprehensive performance score of the guide head neural network tracking indicators is also preset.

[0122] Step S32: Determine whether there are any single indicator scores below the lower threshold of the single indicator score. If there are, the overall performance evaluation result is 0; if not, the single indicator scores above the upper threshold of the single indicator score are judged as full marks.

[0123] The essence of the single-indicator arbitration mechanism lies in introducing direct promotion and veto systems into the weighted nodes of tracking performance indicators.

[0124] (1) For each single indicator score, if it is lower than the lower threshold of the single indicator threshold range, the evaluation is terminated and the comprehensive performance score of the guide head neural network tracking algorithm is directly judged to be zero.

[0125] (4) If the score of a single indicator is not lower than the lower threshold of the threshold range of the single indicator, determine whether it is higher than the upper threshold of the threshold space of the single indicator.

[0126] (5) If a single indicator score is higher than the upper threshold, the single indicator score is directly judged as full score. Perform the weighted fusion in step S33.

[0127] Step S33: After judging the single indicator score results that are higher than the upper threshold of the single indicator score as full marks, the comprehensive performance score is obtained by weighted fusion calculation with the single indicator score results that do not meet the arbitration conditions.

[0128] If the score is below the upper threshold of the single indicator threshold space and also below the lower threshold of the single indicator threshold space, meaning that the single indicator score does not meet the arbitration conditions, then the grey relational analysis method is used to assign weights to each indicator and calculate the comprehensive performance score.

[0129] To ensure the objectivity and rationality of the weight design in the weighted fusion process, the grey relational analysis method was introduced. The grey relational analysis method considers the following when assigning weights to each indicator score:

[0130] The correlation information between indicators reflects, to some extent, the relative importance of different indicators. In order to construct an indicator weight model, it is necessary to quantify the intrinsic correlation between indicators, hence the concept of grey relational depth coefficient is proposed.

[0131] γ(u i ,u j Let q be the grey relational coefficient between the i-th and j-th tracking performance metrics, and let q be the corresponding grey relational depth coefficient function. ij The specific definition is shown in formula (4):

[0132]

[0133] Among them, γ(u i ,u j Let ) be the i-th tracking performance metric u i With the j-th tracking performance metric u j The grey relational coefficient between them, where Q is the sequence number of the tracking performance index data.

[0134] The larger the grey relational coefficient, the higher the degree of correlation between the two tracking performance indicators.

[0135] The weighting model based on the grey relational depth coefficient is as follows:

[0136] Constraints on the expected range of change in indicator weights. A grey relational depth coefficient is introduced to determine the importance of different indicators based on the significance of their changes, thereby determining the weights of each indicator. Based on the above theory, this can be achieved using the grey relational depth coefficient q.ij To determine the weight value range of the indicators. The weight value change range constraint is shown in formula (5):

[0137]

[0138] The constraint on the fluctuation range of the variance of the indicator weight values, that is, the fluctuation range of the indicator weight values is also determined by the grey correlation depth coefficient. The constraint condition of introducing the weight value variance is shown in formula (6):

[0139]

[0140] where D(q0) is the variance set of the grey correlation depth coefficient set q0 of all indicators, and Q * is to satisfy Q * <Q represents any summation quantity less than the total number of indicators, and D(q0) represents the fluctuation range of the indicator weight values, as shown in formula (7).

[0141]

[0142] After obtaining the weight values based on the above formulas (5)-(7), continue to perform weighted fusion on the comprehensive performance scores of single indicators, and calculate the comprehensive performance score of the seeker neural network tracking algorithm through weighted fusion. Weighted fusion is for the weighting based on the comprehensive performance scores of single performance indicators, and the expression of the fusion algorithm is shown in formula (8).

[0143]

[0144] where W e,i is the overall weight value of indicator i, and Q is the serial number of the tracking performance index data. F i is the single indicator score, and F is the comprehensive performance evaluation result.

[0145] Step S34, determine whether the comprehensive performance score is lower than the lower threshold of the comprehensive performance score of the seeker neural network tracking index. If it is lower, directly reset the comprehensive performance score of the seeker neural network tracking index to zero; if not, retain the comprehensive performance score.

[0146] From step S3, obtain the comprehensive performance score results of the tracking performance indicators, including:

[0147] (1) The zero value when the single indicator score with arbitration conditions is lower than the lower threshold of the single indicator threshold interval;

[0148] (2) The zero value of the comprehensive performance score with arbitration conditions;

[0149] (3) The comprehensive performance score without arbitration conditions.

[0150] Step S4, specifically.

[0151] The overall performance score is F, and the comments according to the score range are as shown in (9):

[0152]

[0153] The tracking performance index data collected in step S1, the individual index scores in step S2, and the comprehensive performance score results in step S3 are stored and displayed. This can help users to have a more comprehensive understanding of the guide head's comprehensive performance evaluation in different scenarios and provide important basis for decision-making.

[0154] These four steps provide a clearer overview of the entire performance evaluation process, from data collection to the final presentation of comprehensive performance evaluation results, offering users a systematic and comprehensive performance evaluation method for seeker neural network tracking algorithms.

[0155] Figures 3(a)-(b) Figure 4 and Figure 5 This is the verification result of this solution. Figure 3a , Figure 3b The horizontal axis represents the number of simulation steps, i.e., the time point, and the vertical axis represents the tracking error; Figure 4 The horizontal axis represents performance indicators: number of nodes, x-axis steady-state error, y-axis steady-state error, x-axis response time, y-axis response time, x-axis overshoot, and y-axis overshoot. The vertical axis represents the weight of each indicator. Figure 5 The horizontal axis represents the performance indicators, including the number of nodes, x-axis steady-state error, y-axis steady-state error, x-axis response time, y-axis response time, x-axis overshoot, y-axis overshoot, and the textual description of the weighted fusion calculation. The vertical axis represents the scores of each individual indicator and the overall performance score.

[0156] Example 2:

[0157] Another embodiment of the present invention discloses a performance evaluation system for a seeker neural network tracking algorithm, thereby implementing the performance evaluation method for a seeker neural network tracking algorithm in Embodiment 1. The specific implementation of each module is described in the corresponding description in Embodiment 1.

[0158] like Figure 6 As shown, the system includes a data acquisition module M1, a single-index scoring and evaluation module M2, and a comprehensive performance scoring and evaluation module M3, which are as follows:

[0159] The acquisition module M1 is used to simulate a moving target to test the seeker, and acquires multiple tracking performance index data of the seeker against the moving target based on a neural network tracking algorithm;

[0160] The single-index scoring and evaluation module M2 is used to convert each tracking performance index data into a single-index scoring result based on the multiple tracking performance index data through linear scoring and weighted fusion.

[0161] The comprehensive performance scoring and evaluation module M3 is used to determine whether there are any single indicator scores below the lower threshold of the single indicator score. If so, the comprehensive performance evaluation result is 0; if not, the single indicator scores above the upper threshold of the single indicator score are judged as full marks, and then weighted and merged with the single indicator scores that do not meet the arbitration conditions to obtain the comprehensive performance evaluation result; it is determined whether the comprehensive performance score meets the arbitration conditions. If not, the comprehensive performance evaluation result is set to zero; otherwise, it is retained to obtain the comprehensive performance evaluation result of the tracking performance indicator.

[0162] Since the system in this embodiment and the method in Embodiment 1 are related and can be referenced from each other, this description is redundant and will not be repeated here. Because this system embodiment shares the same principle as the above method embodiment, it also possesses the corresponding technical effects of the above method embodiment.

[0163] The contents not described in detail in this specification are common knowledge to those skilled in the art.

[0164] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware, and the program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.

[0165] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A performance evaluation method for a seeker neural network tracking algorithm, characterized in that, Includes the following steps: Step S1: Simulate a moving target to test the seeker, and obtain multiple tracking performance index data of the seeker on the moving target based on the neural network tracking algorithm; Step S2: Based on the multiple tracking performance index data, each tracking performance index data is converted into a single index score result through linear scoring and weighted fusion; Step S3: Determine whether there are any single indicator scores below the lower threshold of the single indicator score. If so, the comprehensive performance evaluation result is 0. If not, the single indicator scores above the upper threshold of the single indicator score are judged as full marks, and then weighted and merged with the single indicator scores that do not meet the arbitration conditions to obtain the comprehensive performance evaluation result. Determine whether the comprehensive performance score meets the arbitration conditions. If not, the comprehensive performance evaluation result is set to zero; otherwise, it is retained to obtain the comprehensive performance evaluation result of the tracking performance indicator. The process of converting each tracking performance index data into a single index score result through linear scoring and weighted fusion includes: Linear scoring transforms each performance tracking metric into a normalized dynamic score, specifically: v = HI(u) = au + b Where u is the tracking performance index data value, v is the dynamic score value, a and b are real constants, and v max The maximum value u of the tracking performance index data determined after repeated experiments. max The corresponding rating, v * For normalized dynamic scoring; After linear scoring transformation, the array of single-indicator score values ​​is converted into numerical values ​​using weighted fusion. Step S3 includes: Step S31: Preset the threshold range for single-index scoring and the lower threshold for the comprehensive performance score of the guide head neural network tracking index; Step S32: Determine whether there are any single indicator scores below the lower threshold of the single indicator score. If there are, the overall performance evaluation result is 0; if not, the single indicator scores above the upper threshold of the single indicator score are judged as full marks. Step S33: After judging the single indicator score results that are higher than the upper threshold of the single indicator score as full marks, the comprehensive performance score is obtained by weighted fusion calculation with the single indicator score results that do not meet the arbitration conditions. Step S34: Determine whether the overall performance score is lower than the lower threshold of the overall performance score of the seeker neural network tracking index. If it is lower, the overall performance score of the seeker neural network tracking index is directly set to zero. If not, the overall performance score is retained. Step S33 includes: After judging the single indicator score results that are higher than the upper threshold of the single indicator score as full marks, the grey relational analysis method is used to perform weighted fusion calculation on each single indicator score with the single indicator score results that do not meet the arbitration conditions to obtain the comprehensive performance score. γ(u i ,u j Let q be the grey relational coefficient between the i-th and j-th tracking performance metrics, and let q be the corresponding grey relational depth coefficient function. ij for: Among them, γ(u i ,u j Let ) be the i-th tracking performance metric u i With the j-th tracking performance metric u j The grey relational coefficient between them, where Q is the sequence number of the tracking performance index data; The weight variation range is constrained as follows: The constraint condition for the variance of the weights is: Among them, D(q0) is the variance set of the full-index grey correlation depth coefficient set q0, and Q * is to satisfy Q * <Q represents any summation quantity less than the total number of indicators, and D(q0) represents the fluctuation range of the indicator weights as:

2. The performance evaluation method according to claim 1, characterized in that, During testing, the x and y axis position information within the seeker's imaging plane and the angle information of the moving target are collected; the multiple tracking performance index data of the seeker for the moving target obtained based on the neural network tracking algorithm include: The three neurons in the tracking algorithm are input with the x and y axis position information and the angle information of the moving target according to the time sequence; There are n output neurons, two of which contain tracking signals containing the position information of the target point on the x and y axes. The tracking signal values ​​are taken according to the time series to obtain multiple tracking performance index data of the seeker for the moving target. The angle of the moving target is the angle between the line connecting the moving target and the center point of the seeker's main reflective surface and the central axis of the seeker.

3. The performance evaluation method according to claim 1, characterized in that, The process of converting an array of single-indicator score values ​​into numerical values ​​using weighted fusion includes: F i =w(J v,i )J v,i +w(σ v,i )s v,i Among them, F i For a single indicator score, w(J) v,i ) and w(σ v,i () is the normalized dynamic score v * The weights of the mean and variance are i = 1, 2, ... Q, where Q is the sequence number of the tracking performance index data.

4. The performance evaluation method according to claim 1, characterized in that, The single-index scoring threshold range includes: The single-index scoring for each tracking performance indicator is preset with a single-index scoring threshold range; The single-index scoring threshold range includes an upper threshold and a lower threshold.

5. The performance evaluation method according to claim 1, characterized in that, The overall performance evaluation results of the seeker neural network tracking algorithm are as follows: Where w e,i denoted by , i represents the overall weight of index i, and F represents the comprehensive performance evaluation result.

6. The performance evaluation method according to any one of claims 1-5, characterized in that, include: The tracking performance metrics include x-axis position information and y-axis position information; The x-axis position information includes x-axis steady-state error, x-axis response time, and x-axis overshoot. The y-axis position information includes the y-axis steady-state error, y-axis response time, and y-axis overshoot.

7. A performance evaluation system for a seeker neural network tracking algorithm, characterized in that, include: The acquisition module M1 is used to simulate a moving target to test the seeker and acquire multiple tracking performance index data of the seeker to the moving target based on a neural network tracking algorithm. The single-index scoring and evaluation module M2 is used to convert each tracking performance index data into a single-index scoring result based on the multiple tracking performance index data through linear scoring and weighted fusion. The comprehensive performance scoring and evaluation module M3 is used to determine whether there are any single indicator scores below the lower threshold of the single indicator score. If so, the comprehensive performance evaluation result is 0. If not, the single indicator scores above the upper threshold of the single indicator score are judged as full marks and then weighted and merged with the single indicator scores that do not meet the arbitration conditions to obtain the comprehensive performance evaluation result. It is also determined whether the comprehensive performance score meets the arbitration conditions. If not, the comprehensive performance evaluation result is set to zero; otherwise, it is retained to obtain the comprehensive performance evaluation result of the tracking performance indicator. The process of converting each tracking performance index data into a single index score result through linear scoring and weighted fusion includes: Linear scoring transforms each performance tracking metric into a normalized dynamic score, specifically: v = HI(u) = au + b Where u is the tracking performance index data value, v is the dynamic score value, a and b are real constants, and v max The maximum value u of the tracking performance index data determined after repeated experiments. max The corresponding rating, v * For normalized dynamic scoring; After linear scoring transformation, the array of single-indicator score values ​​is converted into numerical values ​​using weighted fusion. The comprehensive performance scoring and evaluation module M3 includes: Step S31: Preset the threshold range for single-index scoring and the lower threshold for the comprehensive performance score of the guide head neural network tracking index; Step S32: Determine whether there are any single indicator scores below the lower threshold of the single indicator score. If there are, the overall performance evaluation result is 0; if not, the single indicator scores above the upper threshold of the single indicator score are judged as full marks. Step S33: After judging the single indicator score results that are higher than the upper threshold of the single indicator score as full marks, the comprehensive performance score is obtained by weighted fusion calculation with the single indicator score results that do not meet the arbitration conditions. Step S34: Determine whether the overall performance score is lower than the lower threshold of the overall performance score of the seeker neural network tracking index. If it is lower, the overall performance score of the seeker neural network tracking index is directly set to zero. If not, the overall performance score is retained. Step S33 includes: After judging the single indicator score results that are higher than the upper threshold of the single indicator score as full marks, the grey relational analysis method is used to perform weighted fusion calculation on each single indicator score with the single indicator score results that do not meet the arbitration conditions to obtain the comprehensive performance score. γ(u i ,u j Let q be the grey relational coefficient between the i-th and j-th tracking performance metrics, and let q be the corresponding grey relational depth coefficient function. ij for: Among them, γ(u i ,u j Let ) be the i-th tracking performance metric u i With the j-th tracking performance metric u j The grey relational coefficient between them, where Q is the sequence number of the tracking performance index data; The weight variation range is constrained as follows: The constraint condition for the variance of the weights is: Among them, D(q0) is the variance set of the full-index grey correlation depth coefficient set q0, and Q * is to satisfy Q * <Q represents any summation quantity less than the total index quantity, and D(q0) represents the fluctuation range of the index weight as:

Images