Cooperative optimization method, device and equipment of multi-probe sensor and medium

Abstract

The application discloses a kind of multi-probe sensor's cooperative optimization method, device, equipment and medium, comprising: acquisition original data and standardization original data, obtain the data after standardization;Using multi-sensor fusion algorithm is fused to the data after standardization, obtain the fusion trajectory data of multi-sensor;With fusion trajectory data as benchmark, the residual of each sensor is calculated, and multi-dimensional error feature set is constructed based on the residual;Based on the multi-dimensional error feature set, construct weighted minimization objective function, and calculate the optimal parameter adjustment amount of each sensor;According to the optimal parameter adjustment amount generation control instruction is issued to each sensor, and the internal working parameter of sensor is updated.The application solves the problem that only data compensation is made in the rear end fusion algorithm link in the prior art, and the performance of sensor is continuously degraded due to long-term operation, suppresses the original measurement deviation from the data acquisition source, effectively improves the accuracy and reliability of target detection result.

Description

Technical Field

[0001] This invention belongs to the field of target multi-domain detection technology, and specifically relates to a collaborative optimization method, device, equipment and medium for multiple detection sensors. Background Technology

[0002] Multi-sensor fusion systems are a core component of modern intelligent sensing technology. In fields such as anti-drone, autonomous driving, and industrial robots, multiple heterogeneous sensors are typically deployed, such as pulse radar, phased array radar, cameras, spectrum analyzers, and drone protocol parsing systems. Fusion algorithms integrate the data from these sensors to obtain more accurate and robust environmental perception results. However, the influence of complex outdoor weather, urban electromagnetic clutter, and the low-speed, small-target characteristics of drones can lead to significant measurement deviations. The cumulative effect of errors from various sensors can cause a decrease in the accuracy of fused trajectories and the failure of target association.

[0003] Existing technologies for optimizing detection errors are mostly limited to data compensation and correction at the backend fusion algorithm level, resulting in a typical unidirectional relationship between the fusion layer and the sensor layer. The sensor merely acts as a data acquisition end, unidirectionally transmitting raw data to the fusion center. The fusion center passively receives the data and improves the accuracy of the fusion result solely through algorithm-level optimizations (such as Kalman filtering, particle filtering, and graph optimization). In this architecture, the sensor's own operating parameters (such as sampling rate, gain, detection threshold, and beam pointing) remain at their factory calibration state or fixed configuration, unable to be dynamically adjusted according to the actual operating environment. This approach of relying solely on post-processing to mitigate measurement deviations has significant technical limitations: it can only compensate for existing errors with hysteresis, without implementing correction and optimization at the sensor's own source (the data generation point), and cannot suppress original acquisition deviations by dynamically adjusting the sensor's built-in operating parameters.

[0004] This leads to the current technology relying on offline calibration to calibrate sensors. However, long-term operation of sensors causes the underlying acquisition error to accumulate continuously, and the compensation effect of the back-end algorithm gradually fails, resulting in problems such as trajectory distortion and target loss. This significantly reduces the stability of the system for long-term detection. Furthermore, factory calibration cannot cope with changes such as device aging, temperature drift, and mechanical deformation caused by long-term operation, nor can it adapt to the optimal parameter configuration requirements under different environmental factors, which seriously restricts the improvement of target detection and tracking accuracy.

[0005] Currently, addressing issues solely through sensor-based adjustments faces several limitations. First, there are inherent design flaws in the sensor system itself. Traditional detection systems employ a hierarchical architecture where sensors transmit data unidirectionally to a fusion center. Various sensors are treated as independent sensing units with fixed parameters, their internal operating parameters completely encapsulated, only able to output raw detection data via standardized interfaces. The fusion center only handles upper-level computational tasks such as data parsing and trajectory fusion, lacking the ability to adjust sensor operating states in reverse. This rigid logical isolation between layers prevents the establishment of a closed-loop feedback loop where the fusion center issues control commands downwards and corrects sensor operating parameters online. Second, existing technologies utilize sensor data in a somewhat one-sided manner, treating sensors merely as standardized data stream output devices, ignoring the adjustable parameters and local computing power inherent in intelligent sensors. Internal configurations such as sampling rate, filtering parameters, and detection thresholds directly affect the quality of raw data, but current technologies lack dynamic optimization strategies for sensors and fail to leverage sensor configuration interfaces for data source-level quality optimization. Third, existing technologies lack dynamic, high-confidence sensor performance evaluation benchmarks. To achieve accurate closed-loop feedback from the fusion layer to the sensor layer, it is necessary to rely on reliable benchmarks to quantitatively evaluate the detection performance of each sensor; however, in complex dynamic scenarios, it is impossible to obtain the true value of the target's trajectory. Traditional fusion outputs themselves are accompanied by errors and uncertainties, and do not meet the conditions for serving as a unified evaluation benchmark.

[0006] A few methods involving sensor feedback also have shortcomings. For example, patent application 2025106470174 discloses a UAV trajectory tracking system based on multi-sensor fusion. In its trajectory prediction deviation scoring module, it uses a stacked long short-term memory network to build a trajectory prediction model, outputting the predicted 3D trajectory coordinates of the UAV, and generating correction commands based on the predicted trajectory to achieve unidirectional correction of the UAV's flight control by the fused trajectory. However, this scheme does not establish a reverse optimization closed loop from the fusion result to the sensor itself; its correction commands only adjust flight control parameters and cannot improve sensor measurement accuracy from the data source.

[0007] Therefore, researching a target detection method that can realize adaptive collaborative optimization of multiple detection sensors driven by fusion trajectory inversion is of great practical significance for improving the long-term stable detection and tracking capability of targets and improving the target detection system. Summary of the Invention

[0008] The purpose of this invention is to provide a collaborative optimization method, apparatus, device, and medium for multiple detection sensors. By establishing a closed-loop feedback link from the fusion algorithm to the sensor and constructing a unified fusion optimization evaluation benchmark based on multi-sensor cross-verification, this invention solves the problem in the prior art where data compensation is only performed in the back-end fusion algorithm stage without dynamically adjusting the parameters of the sensor itself, leading to continuous degradation of sensor performance over long-term operation. This invention suppresses the original measurement deviation from the data acquisition source and effectively improves the accuracy and reliability of target detection results.

[0009] To achieve the above objectives, the technical solution adopted by this invention is: a collaborative optimization method for multiple detection sensors, comprising the following steps: S1: Collect raw data and standardize the raw data to obtain standardized data, wherein the raw data is obtained by collecting multi-dimensional raw detection information of the target through multi-source heterogeneous sensors according to the current working parameters; S2: The standardized data is fused using a multi-sensor fusion algorithm to obtain fused trajectory data from multiple sensors; S3: Using the fused trajectory data as a reference, calculate the residuals of each sensor in real time, and construct a multi-dimensional error feature set based on the residuals; S4: Based on the multidimensional error feature set, construct a weighted minimization objective function and calculate the optimal parameter adjustment amount for each sensor; S5: Based on the optimal parameter adjustment amount, generate control commands and send them to each sensor to update the sensor's internal operating parameters.

[0010] This invention fuses data collected from various sensors, using a high-confidence fused trajectory as a unified benchmark for sensor feedback parameter tuning, thus solving the problem of existing technologies lacking a reliable benchmark for reverse adjustment of sensor parameters. Based on the fusion result and the original observations of individual sensors, the measurement residuals are solved to construct a multi-dimensional error feature set. The optimal parameter adjustment amount for collaborative optimization of each sensor is solved by a preset weighted minimization objective function, clarifying the standardized adjustment method for sensor parameters and solving the problem of existing technologies lacking a quantitative parameter tuning scheme. The adjustment amount is then sent to the corresponding sensors to complete adaptive parameter correction, thereby constructing a complete closed-loop feedback link from the data fusion layer to the bottom-level sensors, solving the technical problems in existing technologies that are limited to back-end algorithm compensation and whose error compensation effect continuously decays over long-term system operation.

[0011] Furthermore, S2 specifically includes the following steps: S2.1: Based on the standardized data, calculate the association probability between each sensor and the existing track. ; S2.2: Estimating the trajectory l Based on k- The state obtained at time 1k Predicted state estimate at time 1 ; S2.3: Based on the aforementioned correlation probability and predicted state estimate, the fused trajectory data of the multi-sensor system is calculated according to the following formula. : ; in, l Indicates the flight path. i Indicates the sensor index. k Indicates time, F The state transition matrix is ​​calculated in advance. For process noise, Indicates the flight path l The corresponding target utilizes the first to k The result obtained after fusing and updating all sensor measurements at time -1 k The optimal posterior state estimate at time -1; Indicates the flight path l Based on k The state obtained at time -1 k The prior predicted state estimate at time t; It is the first i Each sensor at time k Output measurement values; Kalman gain; ; Indicates sensor i The measurement function.

[0012] This fusion method uses dynamic weighted multi-sensor observations with measurement-track correlation probability, combined with motion model temporal state prediction, to adaptively distinguish the reliability of each sensor measurement and output a high-precision unified fused track. This provides a highly reliable benchmark for subsequent sensor bias assessment and underlying parameter calibration, ensuring the stable operation of the overall collaborative optimization closed loop.

[0013] Furthermore, the multi-source heterogeneous sensor includes a radar detection sensor, a spectrum detection sensor, and a photoelectric detection sensor. S3 specifically includes the following steps: S3.1: Using the fused trajectory data Based on this, calculate the measurement values ​​of each sensor. Residuals between the fusion benchmark and the fusion benchmark The residuals of the radar detection sensor, spectrum detection sensor, and photoelectric detection sensor are calculated based on spatial motion parameters, radio frequency feature positioning parameters, and optical angle parameters, respectively. S3.2: Calculate the system bias based on the residual sequence within a preset sliding window. b i Random noise variance With confidence weight w i Thus, a multidimensional error feature set is obtained; in, i Indicates the sensor index. k Indicates time, Indicates sensor i Measurement function; This represents the Euclidean norm.

[0014] This invention calculates the residuals of each heterogeneous sensor by classifying them based on the fused trajectory, and extracts multidimensional error features through a sliding window. It provides a specific implementation method for quantifying sensor errors, provides a complete error basis for solving the optimal parameter tuning, and is suitable for the collaborative optimization of multiple sensors such as radar, spectrum, and photoelectric sensors.

[0015] Furthermore, S4 specifically includes constructing a set of all sensor parameter adjustment values ​​based on the multidimensional error feature set. To optimize the weighted minimization objective function of variables : ; in, i,j Indicates the sensor index. S ref This represents the set of sensor references with a confidence level higher than a preset threshold. , These represent the sensors. i,j The amount of parameter adjustment; w i As the confidence level weight, b i λ is the system bias; μ is the regularization coefficient; C is the coupling constraint regularization coefficient; ij For sensors i and sensors j The coupling functions between them.

[0016] By constructing a weighted minimization objective function, the first term represents the weighted sum of squared system biases. Optimization priorities are assigned based on confidence levels, prioritizing bias correction for high-confidence sensors to effectively reduce systematic measurement errors. The second term is a penalty term for parameter adjustment magnitude, preventing drastic parameter changes from causing system instability. The third term ensures the consistency of coordinated adjustments. This function can solve for the optimal parameter adjustments for all sensors in one go, balancing single-device error suppression with multi-sensor coordinated consistency, thus clarifying a complete computational implementation scheme for quantitative sensor parameter tuning.

[0017] Furthermore, calculating the optimal parameter adjustment for each sensor specifically includes the following steps: S4.1: Calculate the independent optimization cycle for each sensor in real time based on the confidence level of each sensor.T i For each sensor, if the cumulative runtime of the sensor reaches the corresponding independent optimization cycle, proceed to S4.1.1 to calculate the initial preprocessing adjustment value of the sensor; otherwise, reuse the historical initial preprocessing adjustment value cached by the sensor. S4.1.1: Establishing a deviation model Based on the aforementioned deviation model, the optimal parameter estimation algorithm is used to calculate the parameters that satisfy... Given the sensor's θ value under certain conditions, an estimated value of θ is obtained. ; Calculate the sensor value based on the estimated value. i Theoretical compensation amount ; S4.1.2: Calculate the maximum allowable adjustment step size of the sensor in a single operation. : ; S4.1.3: Combine the theoretical compensation amount with the maximum allowable adjustment step size per cycle to calculate the initial adjustment value for preprocessing. The initial adjustment value satisfies: ; S4.2: Input the preprocessed initial adjustment values ​​corresponding to all sensors into the weighted minimization objective function for global joint optimization calculation, and obtain the optimal parameter adjustment amount of all sensors after iterative convergence; in, i Indicates sensor index, via Calculate the independent optimization period, where α is the scene adaptation coefficient. θ is the physical parameter of the sensor bias to be estimated, g(θ) is a function of the physical parameter, and ε is the random error; S ref This represents a set of sensor references with a confidence level higher than a preset threshold. w i As the confidence level weight, T base The pre-set base period; δ is the minimum value to prevent division by zero; and , P i For sensors i The original parameters, For sensors i The parameter adjustment amount, , This indicates the maximum and minimum values ​​of the sensor's internal operating parameters; This represents the maximum allowable adjustment amount for a single parameter adjustment.

[0018] This invention designs a complete hierarchical solution scheme from residual error feature extraction to the output of the globally optimal parameter tuning: First, the residuals of various sensors are calculated based on the fused trajectory to extract multi-dimensional error features; then, a weighted minimization objective function that takes into account deviation suppression, amplitude modulation constraints, and multi-sensor coupling is constructed; during hierarchical solution, differentiated optimization cycles are set according to confidence level, and the theoretical compensation amount is solved by the deviation model and the initial value is obtained by amplitude limiting preprocessing, and then substituted into the objective function for global joint optimization, iteratively outputting the optimal parameter tuning. The invention provides the benchmark, error quantization, optimization model, and step-by-step solution logic for reverse parameter tuning, solving the problems of existing technologies such as lack of parameter tuning benchmark, lack of quantification adjustment methods, and long-term algorithm compensation failure, and realizing synchronous and collaborative tuning of multiple sensors.

[0019] Furthermore, S5 also includes: combining real-time detection status information from multiple sensors, scheduling multi-source heterogeneous sensors to cooperate in joint detection, and dynamically adjusting the internal operating parameters of each sensor.

[0020] By simultaneously adjusting internal operating parameters through joint detection, the detection advantages of different sensors such as radar, spectrum, and photoelectric can be fully utilized to compensate for the blind spots and performance defects of single devices, and continuously improve the overall target detection coverage and tracking continuity.

[0021] Furthermore, the method also includes performing a consistency constraint check on the sensor measurements based on the collected raw data. When a sensor measurement value is detected to violate the consistency constraint, a forced calibration of the sensor is triggered. The consistency constraint check includes a kinematic consistency constraint check, a geometric consistency constraint check, and a target feature continuity constraint check.

[0022] By conducting multi-dimensional consistency constraint verification on the raw sensor measurements, abnormal measurements are identified from three dimensions: motion, geometry, and target features. Once data deviation from the constraint conditions is detected, sensor forced calibration is triggered. This can screen out failed detection data in advance, correct sensor offset faults in a timely manner, ensure the reliability of data in the input fusion process, and avoid abnormal observations interfering with subsequent tuning and target tracking accuracy.

[0023] Based on the same concept, the present invention also provides a collaborative optimization device for multiple detection sensors, which is used to implement the above method, the device comprising: The data acquisition module is used to collect raw data and standardize the raw data to obtain standardized data. The raw data is obtained by collecting multi-dimensional raw detection information of the target through multi-source heterogeneous sensors according to the current operating parameters. The fusion trajectory data generation module is used to fuse standardized data using a multi-sensor fusion algorithm to obtain fusion trajectory data from multiple sensors. A multidimensional error feature set construction module is used to calculate the residuals of each sensor in real time based on the fused trajectory data, and construct a multidimensional error feature set based on the residuals; The optimal parameter adjustment calculation module is used to construct a weighted minimization objective function based on the multidimensional error feature set, and to calculate the optimal parameter adjustment for each sensor. The sensor internal operating parameter update module is used to generate control commands based on the optimal parameter adjustment amount and send them to each sensor to update the sensor's internal operating parameters.

[0024] Based on the same concept, the present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method described above.

[0025] Based on the same concept, the present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method described above.

[0026] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention constructs a complete software closed-loop optimization link for multi-source heterogeneous sensors, from data acquisition and fusion calculation to parameter back-feedback correction. This solves the problem that existing technologies only perform data compensation at the back-end fusion algorithm level and do not dynamically correct the sensor's intrinsic parameters. This effectively addresses the technical problem of performance degradation and parameter mismatch caused by long-term operation of sensors due to factors such as device aging and environmental interference. Simultaneously, this invention uses high-confidence fusion trajectories as a unified performance evaluation benchmark for sensor feedback parameter tuning. By fully utilizing fused data and designing a multi-constraint weighted minimization objective function, it provides a quantitative calculation method for multi-sensor collaborative optimization. This provides a reliable and standardized implementation scheme for sensor adaptive parameter tuning and multi-device collaborative correction, significantly improving the long-term stability and detection accuracy of the target detection system. Attached Figure Description

[0027] To more clearly illustrate the technical method of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only one embodiment of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 This is a flowchart of the collaborative optimization method for multiple detection sensors in an embodiment of the present invention; Figure 2 This is a system architecture diagram of the collaborative optimization method for multiple detection sensors in this embodiment of the invention. Detailed Implementation

[0029] The present invention will now be described in detail with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described herein can be combined with each other. For ease of description, the terms "upper," "lower," "left," and "right" used below only indicate that they correspond to the upper, lower, left, and right directions in the accompanying drawings and do not limit the structure.

[0030] The technical methods of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0031] The technical methods of this application will be described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0032] Figure 1 A flowchart of the collaborative optimization method for multiple detection sensors provided in an embodiment of the present invention is shown, as follows: Figure 1 As shown, the method includes the following steps: S1: Collect raw data and standardize the raw data to obtain standardized data. The raw data is obtained by collecting multi-dimensional raw detection information of the target using a multi-source heterogeneous sensor based on current operating parameters. The multi-source heterogeneous sensor includes radar detection sensors, spectrum detection sensors, and photoelectric detection sensors. For example, radar detection sensors include pulse Doppler radar, whose operating parameters include pulse repetition frequency, transmit power, receiver gain, and detection threshold; and phased array radar, whose operating parameters include beam pointing angle, beamwidth, dwell time, and scanning mode. Spectrum detection sensors include radio spectrum analyzers, whose operating parameters include sampling rate, bandwidth, resolution bandwidth, and detection method. Photoelectric detection sensors include infrared thermal imagers, whose operating parameters include integration time, gain coefficient, non-uniformity correction parameters, and temperature drift compensation coefficient; and visible light cameras, whose operating parameters include exposure time, aperture, gain, and automatic white balance parameters.

[0033] Various sensors collect raw detection data according to their current operating parameters. Radar detection data is obtained by transmitting pulse signals through pulse Doppler radar, receiving target echoes, and performing pulse compression, Moving Target Indication (MTI), Moving Target Detection (MTD), and Constant False Alarm Rate (CFAR) detection to extract target range, velocity, angle, and radar cross section (RCS). Phased array radar scans a designated airspace based on current beam pointing parameters to acquire multi-dimensional target information. Spectrum detection data is obtained by performing frequency sweeping or FFT analysis within a set frequency band using a radio spectrum analyzer to detect radiation source signals and extract parameters such as center frequency, pulse width, pulse repetition interval, and angle of arrival. Electro-optical detection data is obtained by acquiring infrared radiation images of the target through an infrared thermal imager, performing non-uniformity correction and image enhancement to extract target position and infrared features. A visible light camera acquires visible light images of the target, performing automatic exposure adjustment and target detection and recognition.

[0034] The preprocessed measurement data is converted into a standard format under a unified coordinate system and sent to the fusion engine through the communication interface.

[0035] S2: The standardized data is fused using a multi-sensor fusion algorithm to obtain fused trajectory data from multiple sensors.

[0036] S2.1: Specifically, the JPDA algorithm is used to calculate the association probability between each measurement and existing tracks, addressing the uncertainty of measurement sources in multi-target environments. Association Probability Indicates measurement i Belongs to the track l The probability satisfies ; S2.2: Predict the state based on the target motion model , in, F To incorporate the system sampling period, the state transition matrix is ​​calculated using the discretized target motion model. This represents process noise, used to update the prediction covariance matrix.

[0037] S2.3: State update, using valid measurements from all sensors to update the state: ; in, The Kalman gain is calculated by incorporating the summation and inversion of the global covariance matrix. For the track l On kThe target state estimate is continuously updated and fused. For the track l Based on k- The predicted state estimate at time k is obtained from the state at time 1; Indicates the flight path l The corresponding target utilizes the first to k The result obtained after fusing and updating all sensor measurements at time -1 k The optimal posterior state estimate at time -1; It is the first i Each sensor at time k The output measurement value, For the first i The nonlinear measurement mapping function corresponding to each sensor (i.e., the sensor) i The measurement function is used to map the state space to the observation space.

[0038] Confidence calculation: Confidence of the fusion result Based on the state covariance matrix calculate: ; in, For matrix trace operations; P k for k The covariance matrix corresponding to the fused state at each time step is recursively calculated through two steps: time update and measurement update using extended Kalman filtering. The smaller the trace value, the higher the fusion confidence, representing the overall uncertainty of the target state estimation. C k Higher; fusion trajectory (Right now It includes the target's three-dimensional position, velocity, acceleration, and radiation characteristic parameters.

[0039] The fusion result is essentially a high-confidence estimate obtained after cross-verification and correction of multi-source information, containing real-time evaluation information of the performance of each sensor. However, in existing systems, this valuable information is only used to output to upper-layer applications and is not used to improve the data quality of front-end sensors, forming a missing closed loop in information utilization. In the subsequent steps of this invention, the fused trajectory data is used as a unified benchmark to mine the error evaluation value, thereby forming a feedback loop from back-end fusion to front-end sensing control.

[0040] S3: Using the fused trajectory data as a reference, calculate the residuals of each sensor in real time, and construct a multi-dimensional error feature set based on the residuals. S3 specifically includes the following steps: S3.1: Using the fused trajectory data Based on this, calculate the measurement values ​​of each sensor. Residuals between the fusion benchmark and the fusion benchmark For different types of sensors, the residuals have different physical meanings. Specifically, the residuals of radar detection sensors, spectrum detection sensors, and photoelectric detection sensors are calculated based on spatial motion parameters, radio frequency feature positioning parameters, and optical angle parameters, respectively. For example, radar residuals can be deviations in attributes such as range, angle, and Doppler velocity; spectrum residuals can be deviations in attributes such as frequency, pulse width, angle of arrival, and multi-spectral positioning range; and photoelectric residuals can be deviations in attributes such as azimuth and elevation angles.

[0041] S3.2: Based on the residual sequence within a preset sliding window, establish the real-time error distribution and calculate the system deviation. b i Random noise variance With confidence weight w i This yields a multidimensional error feature set, for example, a sliding window length M = 100 frames; where systematic bias is used to reflect the systematic drift of the sensor, such as radar antenna installation deviation, infrared focal plane non-uniformity, etc.; random noise variance is used to reflect the noise level of the sensor; confidence weights are dynamically calculated based on the current residual size and are used for weighting in subsequent fusion and decision-making.

[0042] in, i Indicates the sensor index. k Indicates time, Indicates sensor i Measurement function; This represents the Euclidean norm.

[0043] S4: Based on the multi-dimensional error feature set, a weighted minimization objective function is constructed, and the optimal parameter adjustment amount for each sensor is calculated. Specifically, based on the error distribution of each sensor, a multi-objective optimization method is used to generate the optimal parameter adjustment scheme, and a set of parameter adjustment amounts for all sensors is constructed. To optimize the weighted minimization objective function of variables : ; in, i,j Indicates the sensor index. S ref This represents the set of sensor references with a confidence level higher than a preset threshold. , These represent the sensors. i,j The parameter adjustment amount, such as the radar detection threshold adjustment amount, pulse repetition frequency adjustment amount, etc.; w i As the confidence level weight, b i For sensors iThe systematic bias reflects the systematic drift of the sensor; λ is the regularization coefficient, which controls the weight of the adjustment magnitude penalty. The larger λ is, the heavier the penalty for drastic parameter changes, and the more the optimization result tends to be a small adjustment; μ is the coupling constraint regularization coefficient, used to control the weight of the sensor coupling constraint term in the overall optimization objective. The larger μ is, the more emphasis is placed on maintaining cooperative consistency; C ij For sensors i and sensors j The coupling function between them describes the mutual influence of parameter adjustments.

[0044] Among them, the first item This represents the weighted sum of squared systematic biases, used to minimize systematic biases and aim to reduce systematic errors in sensors. The weights are based on confidence levels. w i Sensors with higher confidence levels have greater bias weights and are optimized first; the second item This indicates a penalty for parameter adjustment, limiting the range of parameter adjustment to prevent drastic parameter changes from causing system instability; the third item... Used to ensure consistency in coordinated adjustments, such as the coverage consistency of beam pointing to overlapping areas, and to account for sensor coupling constraints.

[0045] Calculating the optimal parameter adjustments for each sensor involves the following steps: S4.1: Calculate the independent optimization cycle for each sensor in real time based on the confidence level of each sensor. T i For each sensor, if the cumulative runtime of the sensor reaches the corresponding independent optimization cycle, proceed to S4.1.1 to calculate the initial preprocessing adjustment value of the sensor; otherwise, reuse the historical initial preprocessing adjustment value cached by the sensor. S4.1.1: Establishing a deviation model Based on the aforementioned deviation model, the optimal parameter estimation algorithm is used to calculate the parameters that satisfy... Given the sensor's θ value under certain conditions, an estimated value of θ is obtained. The systematic errors of low-confidence sensors can be inverted using information from high-confidence sensors, for example, a reference set. S ref Select confidence weights The high confidence threshold can be adjusted according to system requirements. The higher the threshold, the more reliable the selected reference sensor, but the fewer available reference sources may be. Through an error mutual compensation mechanism, when the sensor i When significant systematic bias occurs, utilize high-confidence sensors. jThe causes of measurement inversion biases are investigated. For radar ranging bias, a benchmark is established using optical sensor angle measurement and laser ranging to invert the radar delay error. For infrared pointing bias, the infrared optical axis offset is inverted using the radar target position. For frequency measurement bias, the local oscillator drift is inverted using the known radiation source frequency. The inversion employs extended Kalman filtering or particle filtering algorithms to estimate the bias parameters in real time. .

[0046] Calculate the sensor value based on the estimated value. i Theoretical compensation amount .

[0047] This deviation model aims to incorporate the sensor i Systematic bias b i The model is a function g(θ) of a certain physical parameter θ plus a random error ε, based on measurements from a reference sensor, and estimated using least squares or extended Kalman filtering. Calculation sensor i Theoretical compensation amount , The deviation mapping function g(θ) on physical parameters The partial derivatives (Jacobi matrix) reflect the change of θ. b i Sensitivity to changes.

[0048] Existing technologies suffer from the inability to detect and correct parameter drift in sensors during actual operation in real time, resulting in the system operating in a sub-optimal state for extended periods. This leads to high maintenance costs and poor timeliness. Regular shutdowns for manual calibration are required, which not only increases maintenance costs but also compromises system performance during the calibration intervals.

[0049] This invention utilizes information from high-confidence sensors to invert physical deviation parameters (such as delay time, local oscillator drift, and installation errors) of low-confidence sensors, generating compensation commands to eliminate deviations at their physical source. When a systematic deviation is detected in a sensor, other sensors with higher current confidence are automatically selected as reference benchmarks, and a deviation parameterization model is established. The observed systematic deviation is mapped to physically compensable underlying parameters, generating targeted compensation commands that directly adjust the corresponding parameters of the sensor's internal signal processing module, eliminating deviations at their physical source, rather than merely correcting data in post-processing.

[0050] For example, in radar ranging bias inversion, radar ranging bias Typically determined by delay time Caused by, its for Utilizing the high-precision distance measurement capabilities of laser rangefinders As a benchmark, construct the optimization objective. ,estimate Delay compensation commands are then generated. In frequency measurement deviation inversion, the frequency measurement deviation is usually caused by the local oscillator frequency drift. Caused by... using the nominal frequency of a known radiation source. By detecting frequency deviation Inversion of local oscillator drift This generates a local oscillator correction voltage command. In infrared pointing deviation inversion, the optical axis pointing deviation is caused by installation error. , Caused by. Target position determined by radar ( , Construct a baseline line-of-sight angle ( , ), and infrared measurement angle ( , By comparing and reversing the installation errors, an optical axis adjustment command is generated; Where c is the speed of light, Z laser Z serves as the high-precision ranging value for laser rangefinders, acting as an error-free reference value; radar These are the original range measurement observations output by the radar. f measured This represents the signal frequency currently measured by the receiver.

[0051] S4.1.2: Dynamically adjust and optimize the intensity and frequency based on the sensor confidence level, and calculate the maximum allowable adjustment step size of the sensor in a single operation. (i.e., the adjustment range): ; The confidence-weighted feedback strategy is based on confidence weights. w i Determine the adjustment priority and magnitude, and the confidence level. w i The higher the confidence level, the smaller the adjustment range, maintaining stability; the lower the confidence level, the larger the adjustment range, resulting in rapid correction.

[0052] S4.1.3: Combine the theoretical compensation amount with the maximum allowable adjustment step size per cycle to calculate the initial adjustment value for preprocessing. The initial adjustment value satisfies: ; S4.2: Input the preprocessed initial adjustment values ​​corresponding to all sensors into the weighted minimization objective function for global joint optimization calculation. After iterative convergence, the optimal parameter adjustment amount of all sensors is obtained. The adjustment direction is adjusted along the gradient descent direction of the objective function to ensure that the system performance is improved with each adjustment. in, i Indicates sensor index, via Calculate the independent optimization period; sensors with low confidence levels use shorter optimization periods to achieve rapid convergence; α is the scene adaptation coefficient. In a strong interference environment, the value is increased to improve the response speed, and in a stable environment, it is decreased to maintain stability; θ is the physical parameter of the sensor deviation to be estimated (e.g., sensor mounting angle deviation, delay time, local oscillator drift rate), g(θ) is a function of the physical parameter, and ε is the random error; S ref This represents a set of sensor references with a confidence level higher than a preset threshold. w i As the confidence level weight, T base The pre-set base period, for example, 1 second; δ is a minimum value to prevent division by zero; and , Used to limit the magnitude of a single adjustment, ensuring system stability; The parameter values ​​must be within the limits allowed by the sensor hardware; P i For sensors i The original parameters, For sensors i The parameter adjustment amount, , This indicates the maximum and minimum values ​​of the sensor's internal operating parameters; This represents the maximum allowable adjustment amount for a single parameter adjustment.

[0053] While some existing technologies have introduced the concept of feedback, their feedback is limited to parameter adjustments within the algorithm (such as adjusting the noise matrix of the Kalman filter or the data correlation threshold), without addressing the underlying physical sensors. This allows for post-processing after the data enters the fusion engine, failing to improve the quality of the original data. The upper limit of optimization effectiveness is also limited. Algorithm-level optimization cannot compensate for distortions in the source data; it can only achieve temporary corrections to surface parameters and cannot eradicate systemic biases at the sensor's core. Therefore, there is a significant upper limit to the improvement in system detection performance.

[0054] This invention constructs a weighted multi-objective optimization function, simultaneously considering system bias minimization, parameter adjustment smoothness, and inter-sensor coupling constraints. It also designs specialized collaborative optimization strategies for the characteristics of heterogeneous sensors to achieve optimal overall sensing performance. The core of this approach lies in constructing a multi-objective optimization function that pursues three objectives: minimizing the weighted system bias of each sensor, limiting parameter adjustment amplitude to prevent drastic system fluctuations, and satisfying inter-sensor coupling constraints to ensure collaborative consistency. Furthermore, physical constraints on parameter adjustment are set to ensure that the optimized parameters remain within the limits allowed by the sensor hardware, guaranteeing system stability and security. Specialized collaborative optimization strategies are designed for different sensor combinations to achieve optimized resource allocation and performance complementarity.

[0055] S5: Integrate the optimal parameter adjustment vectors obtained from S4 for each sensor. Based on the optimal parameter adjustment amount, control commands are generated and sent to each sensor to update the sensor's internal operating parameters.

[0056] The optimization decisions are translated into specific sensor control commands and executed. Command generation involves converting the parameter adjustment scheme into a standard command frame format, such as "|synchronization head|sensor ID|parameter type|adjustment value|effective time|checksum|". Here, the parameter type is encoded as follows: 01-pulse repetition frequency, 02-beam pointing angle, 03-detection threshold, 04-integration time, 05-gain coefficient, 06-center frequency, etc. The adjustment value refers to the quantized parameter value or increment, and the effective time can be immediate or timed.

[0057] Commands are issued via a feedback control link (fiber optic / CAN bus / wireless) by broadcasting or unicasting to the target sensor. Examples of adjusting internal sensor operating parameters include: for pulse Doppler radar, adjusting the pulse repetition frequency to resolve range ambiguity and adjusting the CFAR detection threshold to adapt to clutter environments; and / or for phased array radar, adjusting the beam pointing angle to track key targets and adjusting the dwell time to improve the detection signal-to-noise ratio; for infrared thermal imagers, adjusting the integration time to adapt to target radiation intensity and updating non-uniformity correction coefficients to compensate for temperature drift; and / or for radio spectrum analyzers, adjusting the center frequency to track target radiation sources and adjusting resolution bandwidth to optimize the signal-to-noise ratio.

[0058] After the sensor completes the parameter adjustment, it sends an acknowledgment frame to provide feedback on the actual adjustment results and prepares for the next data acquisition cycle.

[0059] In one embodiment, S5 further includes: combining the real-time detection status information of multiple sensors, scheduling the multi-source heterogeneous sensors to cooperate with each other to complete joint detection, and dynamically adjusting the internal operating parameters of each sensor.

[0060] Specifically, achieving coordinated operation among multiple sensor types includes radar-optoelectronic coordination, spectrum-radar coordination, multi-sensor coverage optimization, and anti-interference coordination optimization. Dynamically adjusting the internal operating parameters of each sensor includes: the trigger condition for the radar-optoelectronic coordination strategy is when the confidence level of the optoelectronic sensor... w i If the value is below a preset threshold (e.g., 0.7), it is presumed to be caused by insufficient lighting or adverse weather conditions. Cooperative actions can be performed such as adjusting the radar beam to point towards the current field of view of the photoelectric sensor to ensure that the area is still effectively detected; and / or adjusting the radar detection threshold to improve sensitivity to small / low signal-to-noise ratio targets; and / or using the target position detected by the radar to generate guidance commands to drive the photoelectric servo system to point towards the target.

[0061] The trigger condition for the spectrum-radar cooperative strategy is that the spectrum equipment system detects a new radiation source signal, and / or the deviation of the radiation source parameters (frequency, pulse width, pulse repetition interval, angle of arrival, etc.) from the historical average exceeds a preset threshold. Cooperative actions that can be performed include: optimizing the radar waveform design (e.g., selecting the corresponding frequency point, adjusting the pulse width) based on parameters such as radiation source frequency, pulse width, and PRI; and / or guiding the phased array radar beam towards the radiation source direction based on direction finding results, increasing the dwell time; and / or using radar target echoes to verify the correspondence between the radiation source and the target, assisting in spectrum signal sorting and identification. For example, when a new radiation source is detected in the spectrum, the phased array radar is guided to stay in the corresponding azimuth based on the radiation source frequency and pulse width characteristics, increasing the interception probability; simultaneously, the radar target echo is used to verify the correspondence between the radiation source and the target.

[0062] This includes a multi-sensor coverage optimization strategy. The optimization goal is to maximize detection coverage under resource constraints while reducing unnecessary overlap and redundancy. It can perform cooperative actions such as coordinating the adjustment of beam pointing and scanning range of each sensor to eliminate detection blind spots; and / or optimizing sensor energy allocation, allocating more resources to key areas (such as threat directions); and / or dynamically adjusting sensor operating modes, such as switching between long-range surveillance mode and short-range precision tracking mode. This is used to coordinate the adjustment of the detection range and beam pointing of each sensor, eliminate detection blind spots, reduce overlapping coverage areas, and achieve optimal allocation of airspace energy resources.

[0063] Anti-jamming collaborative processing steps: Real-time detection of active interference or false target injection; when active interference or false target injection is detected, multi-sensor collaborative anti-jamming actions are executed. The trigger condition for the anti-jamming collaborative strategy is the detection of active interference (such as suppression interference, deception interference) or false target injection. Collaborative actions can be executed, such as radar equipment activating frequency agility / frequency hopping mode, adjusting pulse waveforms, and increasing processing gain to suppress interference; and / or infrared cameras switching spectral filtering bands, adjusting non-uniform correction parameters, and suppressing laser blinding interference; and / or spectrum equipment adjusting pulse width thresholds and angle of arrival consistency thresholds to filter out interference pulses; and / or multi-sensor cross-verification, performing multi-sensor joint confirmation of suspicious targets, identifying and eliminating false targets generated by interference. When active interference is detected, the radar frequency hopping mode, infrared spectral filtering parameters, and spectral pulse width thresholds are collaboratively adjusted to form a comprehensive anti-jamming strategy.

[0064] In one embodiment, the method further includes performing a consistency constraint check on sensor measurements based on the acquired raw data. When a sensor measurement value is detected to violate the consistency constraint, a forced calibration of the sensor is triggered. The consistency constraint check includes kinematic consistency constraint checks, geometric consistency constraint checks, and target feature continuity constraint checks. The kinematic consistency constraint check primarily detects that the target acceleration must not exceed the platform's maneuverability. Where a is the target acceleration value calculated based on sensor data, a max The maximum allowable maneuvering acceleration of the target motion platform is predetermined by the physical limits of the target carrier. The geometric consistency constraint test mainly examines the geometric reachability of the target to ensure that the target position must be within the detectable airspace; the target feature continuity constraint test mainly examines whether the changes in characteristics such as RCS and infrared radiation intensity conform to physical laws.

[0065] This mandatory calibration step addresses root cause physical parameter deviations. Regular adjustments (such as step S4) only modify ordinary surface parameters based on optimization decisions; however, mandatory calibration relies on continuous real-time monitoring. Once anomalies violating physical constraints are detected in sensor observation data (such as target velocity, position, or acceleration exceeding limits), deep-level corrections are performed to specifically eliminate systematic physical errors within the sensor. This mandatory calibration has the highest execution priority: if observations continuously violate physical constraints, it indicates that the sensor's output data is unreliable and root cause calibration must be performed first.

[0066] Figure 2 A system architecture diagram illustrating the method for implementing embodiments of the present invention is shown, demonstrating the core modules and data flow of an online collaborative optimization system for multiple types of sensors, including radar, spectrum, and optoelectronic sensors. The entire system consists of three main layers: the sensor layer, the fusion engine, and the evaluation and decision layer.

[0067] The sensor layer comprises three types of heterogeneous detection sensors, each with a dynamically configurable operating parameter interface supporting online adjustment. The sensor layer is responsible for acquiring target detection data, performing necessary preprocessing, and then sending the measurement data to the fusion engine. The fusion engine receives the measurement data from each sensor and, in one embodiment, employs a multi-sensor fusion algorithm combining Joint Probabilistic Data Association (JPDA) and Extended Kalman Filter (EKF).

[0068] The fusion engine performs the following functions: initiation, maintenance, and termination of multi-target tracks; correlation and matching of heterogeneous sensor measurements and tracks; fusion state estimation to generate fused trajectories (including target position, velocity, acceleration, radar cross section, etc.); and calculation of the confidence level of the fusion results based on the covariance matrix and consistency test.

[0069] The fusion result is essentially a high-confidence estimate obtained after cross-verification of multi-source information, implicitly containing the current performance status of each sensor (such as deviation magnitude and noise level). However, existing systems only output the fusion result to upper-layer applications (such as planning and control modules) without transmitting it back as a feedback signal. This results in the direct discarding of information containing real-time error characteristics of the sensors, and the system cannot perceive the performance degradation of individual sensors (such as radar ranging drift and spectral angle changes). The systematic errors of the sensors cannot be compensated in a timely manner, and as the operating time increases, the errors gradually accumulate, eventually leading to a decrease in fusion accuracy and even causing sensing failure. Even if a few solutions attempt to issue control commands, there is a lack of a unified benchmark, making it difficult to achieve cross-system sensor collaborative adjustment. This invention calculates the measurement residuals of each sensor based on the fusion trajectory, establishes a three-dimensional error feature set including systematic deviation, random noise variance, and confidence weight through sliding window statistical analysis, and achieves differentiated feedback adjustment based on the confidence weight. Using the high-confidence fusion trajectory as the evaluation benchmark, the residual between the measurement data of each sensor and the fusion benchmark is calculated. This residual reflects the measurement error of the sensor at the current moment. Through statistical analysis of the residual sequence within the sliding window, the systematic bias (reflecting the long-term drift trend) and random noise characteristics (reflecting the measurement fluctuation level) of the sensor are extracted. A confidence-weighted feedback strategy is adopted, and sensors with low confidence obtain a larger parameter adjustment range and a shorter optimization cycle, thereby achieving rapid convergence and differentiated optimization scheduling.

[0070] The evaluation decision layer consists of three core modules: an online sensor performance evaluation module, which calculates the measurement residuals of each sensor based on the fused trajectory, and establishes a real-time error distribution, including system bias, random noise characteristics, and confidence weights; a collaborative optimization decision module, which solves a multi-objective optimization problem based on the error distribution, employing an error mutual compensation mechanism and a confidence weighting strategy to generate optimal operating parameter adjustment schemes for each sensor; and a feedback control command generation module, which transforms the optimization decisions into specific parameter control commands, performing format encapsulation and timing scheduling.

[0071] The feedback control link is the physical channel connecting the evaluation and decision-making layer and the sensor layer. Depending on the application scenario, it can use fiber optic communication (high bandwidth, low latency), Controller Area Network (CAN) bus (high reliability), or wireless communication (distributed deployment) to transmit control commands to each sensor.

[0072] Traditional sensor systems are designed from the outset with sensors defined as non-interventional data sources. Only a data upload interface is provided, without a downlink control channel from the fusion center to the sensor's operating parameters. This directly results in the sensors always operating with fixed factory-calibrated parameters (such as fixed sampling rate and fixed detection threshold). Even when performance degrades due to changes in lighting, severe weather, or signal attenuation, no adjustment commands are received. The system cannot dynamically optimize sensor configuration based on real-time operating conditions, leading to unstable data quality in complex dynamic environments and insufficient robustness of the fusion results.

[0073] This invention, through the design of a feedback control link with supporting hardware and software, connects the evaluation and decision-making layer and the sensor layer in reverse. It feeds back commands generated from fused data information to the sensors for parameter adjustment, establishing a reverse control link from radar-spectrum fusion traces to the signal processing algorithms within the radar and spectrum equipment, forming a closed loop of "equipment detection → trace fusion → equipment optimization." The high-confidence fused trace generated by fusing radar and spectrum traces serves as the feedback benchmark, rather than the external truth value used in traditional methods. Simultaneously, the feedback target is the internal signal processing algorithm parameters of the equipment layer (such as radar CFAR detection threshold, pulse compression delay compensation, local oscillator correction voltage, etc.), rather than the fusion algorithm parameters. This achieves improved detection quality from the data source, rather than merely optimizing the post-processing algorithm.

[0074] The method of this invention breaks down the unidirectional data flow barrier between the fusion layer and the sensor layer in a multi-sensor system, solving the problem that in traditional systems, sensors only act as passive data sources and cannot dynamically adjust their own operating parameters based on the fusion results, leading to the inability to continuously optimize the overall system performance. It overcomes the limitations of static calibration in adapting to dynamic environmental changes, addressing the performance degradation and parameter mismatch issues caused by offline calibration, such as device aging, temperature drift, and environmental changes during operation. The fusion trajectory is used as an evaluation benchmark for the optimization of front-end sensors, constructing a closed-loop information flow of "perception-fusion-evaluation-optimization." Through multi-sensor mutual verification and fusion algorithm optimization, a fusion trajectory with higher confidence than any single sensor is generated. This trajectory is then used as a benchmark to evaluate and optimize individual sensors, solving the problem of wasted high-confidence fusion results and improving information utilization efficiency. It compensates for the shortcomings of existing feedback mechanisms that are limited to optimization within the algorithm layer, extending the scope of feedback optimization from within the fusion algorithm to the front-end physical sensors, achieving systematic and fundamental optimization from the data source to the fusion output. Table 1 below details the differences between the method of this invention and traditional methods in the prior art.

[0075] Table 1. Comparison of the method of the present invention with the traditional method This invention overcomes the technical bias of traditional unidirectional architectures by establishing a bidirectional closed loop. For the first time, it achieves reverse control from the fusion layer to the sensor layer, breaking the long-standing unidirectional relationship between fusion and perception in the field and forming a positive reinforcement loop for perception performance. It systematically improves the quality of raw data by dynamically optimizing sensor parameters, fundamentally improving the accuracy and reliability of the raw data. Compared to simply optimizing the fusion algorithm, this results in a more fundamental performance improvement. Sensor parameters can be automatically adjusted according to the actual operating environment (such as changes in lighting, weather effects, and scene complexity) without manual intervention, improving the system's robustness and adaptability, and enabling adaptive adaptation to environmental changes. It also reduces system maintenance costs by decreasing reliance on periodic offline calibration, thus lowering the human and material costs of system maintenance.

[0076] The following details an embodiment of the present invention, employing a pulse Doppler radar (angle error 0.5 degrees), a spectrum analyzer (direction finding error 1 degree, frequency measurement error 50 kHz), and an infrared thermal imager (pointing deviation 0.3 degrees). Around the 20th second, temperature drift introduced by the infrared sensor causes the pointing error to gradually increase to 1.2 degrees; from the 40th second onwards, the radar experiences interference suppression, resulting in a 6 dB decrease in signal-to-noise ratio; local oscillator drift of the spectrum analyzer leads to an increase in frequency measurement error. The target performs an S-shaped maneuver at a constant speed of 10 m / s for 60 seconds. Table 2 below shows the sensor control effects output by the traditional collaborative control method and the method of this embodiment, with comparison indicators including the root mean square error of the fused trajectory (RMSE), target interception rate, number of track interruptions, and target false alarm rate.

[0077] Table 2 Comparison of Results between Traditional Cooperative Regulation Methods and the Method of the Embodiments of the Present Invention RMSE, or Root Mean Square Error, is a commonly used metric to measure the deviation between an estimated value and the true value. The calculation formula is as follows: For the location estimation of the fused trajectory, assuming at time... k The target 3D position output by fusion is The corresponding real location is Where x, y, and z are the horizontal, pitch, and height coordinates, respectively, and the statistical time period is N. The formula is defined as follows: ; When considering only a two-dimensional plane, the z-component can be omitted. The smaller this value, the closer the fused trajectory is to the actual target motion, and the higher the positioning accuracy.

[0078] Closed-loop feedback reduces fusion position error by 50% because the system dynamically compensates for deviations caused by infrared temperature drift and radar interference; by guiding the radar beam through the spectrum (step S4), the target acquisition probability increases from 72% to 94%; due to real-time parameter optimization, the number of track interruptions is reduced by 74%, effectively improving track continuity and maintaining stable tracking; the collaborative anti-jamming strategy (frequency hopping + threshold adjustment + cross-validation) reduces false tracks under interference by 78.2%, significantly enhancing the system's robustness.

[0079] In this embodiment, a centralized fusion engine and evaluation decision layer are used. All sensor data is aggregated at the central node for processing, and optimization instructions are uniformly generated and issued by the center. In another embodiment, the sensor system adopts a distributed architecture, where each sensor node has local fusion and decision-making capabilities, forming a distributed collaborative optimization network.

[0080] Table 3 below provides an explanation of the terms used in the embodiments of the present invention.

[0081] Table 3. Terminology Explanation Example 2 Based on the same concept, embodiments of the present invention also provide a collaborative optimization device for multiple detection sensors, the device comprising: The data acquisition module is used to collect raw data and standardize the raw data to obtain standardized data. The raw data is obtained by collecting multi-dimensional raw detection information of the target through multi-source heterogeneous sensors according to the current operating parameters. The fusion trajectory data generation module is used to fuse standardized data using a multi-sensor fusion algorithm to obtain fusion trajectory data from multiple sensors. A multidimensional error feature set construction module is used to calculate the residuals of each sensor in real time based on the fused trajectory data, and construct a multidimensional error feature set based on the residuals; The optimal parameter adjustment calculation module is used to construct a weighted minimization objective function based on the multidimensional error feature set, and to calculate the optimal parameter adjustment for each sensor. The sensor internal operating parameter update module is used to generate control commands based on the optimal parameter adjustment amount and send them to each sensor to update the sensor's internal operating parameters.

[0082] Example 3 Based on the same concept, embodiments of the present invention also provide a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method described in Embodiment 1 above.

[0083] Example 4 Based on the same concept, embodiments of the present invention also provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the method described in Embodiment 1 above.

[0084] In the embodiments disclosed in this application, a computer storage medium may be a tangible medium that may contain or store programs for use by or in conjunction with an instruction execution system, apparatus, or device. The computer storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of computer storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0085] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0086] The above embodiments should be understood as being used only to illustrate the present invention more clearly, and not to limit the scope of the present invention. After reading the present invention, any modifications of the present invention in various equivalent forms by those skilled in the art fall within the scope defined by the appended claims.

Claims

1. A collaborative optimization method for multiple detection sensors, characterized in that, Includes the following steps: S1: Collect raw data and standardize the raw data to obtain standardized data, wherein the raw data is obtained by collecting multi-dimensional raw detection information of the target through multi-source heterogeneous sensors according to the current working parameters; S2: The standardized data is fused using a multi-sensor fusion algorithm to obtain fused trajectory data from multiple sensors; S3: Using the fused trajectory data as a reference, calculate the residuals of each sensor in real time, and construct a multi-dimensional error feature set based on the residuals; S4: Based on the multidimensional error feature set, construct a weighted minimization objective function and calculate the optimal parameter adjustment amount for each sensor; S5: Based on the optimal parameter adjustment amount, generate control commands and send them to each sensor to update the sensor's internal operating parameters.

2. The collaborative optimization method for multiple detection sensors according to claim 1, characterized in that, S2 specifically includes the following steps: S2.1: Based on the standardized data, calculate the association probability between each sensor and the existing track. ; Estimated track l Based on k- The state obtained at time 1 k Predicted state estimate at time 1 ; S2.2: Based on the aforementioned correlation probability and predicted state estimate, the fused trajectory data from multiple sensors is calculated according to the following formula. : ； in, l Indicates the flight path. i Indicates the sensor index. k Indicates time, F The state transition matrix is ​​calculated in advance. For process noise, Indicates the flight path l The corresponding target utilizes the first to k The result obtained after fusing and updating all sensor measurements at time -1 k The optimal posterior state estimate at time -1; Indicates the flight path l Based on k The state obtained at time -1 k The prior predicted state estimate at time t; It is the first i Each sensor at time k Output measurement values; Kalman gain; ; Indicates sensor i The measurement function.

3. The collaborative optimization method for multiple detection sensors according to claim 1, characterized in that, The multi-source heterogeneous sensor includes a radar detection sensor, a spectrum detection sensor, and a photoelectric detection sensor. S3 specifically includes the following steps: S3.1: Using the fused trajectory data Based on this, calculate the measurement values ​​of each sensor. Residuals between the fusion benchmark and the fusion benchmark The residuals of the radar detection sensor, spectrum detection sensor, and photoelectric detection sensor are calculated based on spatial motion parameters, radio frequency feature positioning parameters, and optical angle parameters, respectively. S3.2: Calculate the system bias based on the residual sequence within a preset sliding window. b i Random noise variance With confidence weight w i Thus, a multidimensional error feature set is obtained; in, i Indicates the sensor index. k Indicates time, Indicates sensor i Measurement function; This represents the Euclidean norm.

4. The collaborative optimization method for multiple detection sensors according to claim 1, characterized in that, S4 specifically includes constructing a set of all sensor parameter adjustment values ​​based on the multidimensional error feature set. To optimize the weighted minimization objective function of variables : ； in, i,j Indicates the sensor index. S ref This represents the set of sensor references with a confidence level higher than a preset threshold. , These represent the sensors. i,j The amount of parameter adjustment; w i As the confidence level weight, b i λ is the system bias; μ is the regularization coefficient; C is the coupling constraint regularization coefficient; ij For sensors i and sensors j The coupling functions between them.

5. The collaborative optimization method for multiple detection sensors according to claim 4, characterized in that, Calculating the optimal parameter adjustments for each sensor involves the following steps: S4.1: Calculate the independent optimization cycle for each sensor in real time based on the confidence level of each sensor. T i For each sensor, if the cumulative runtime of the sensor reaches the corresponding independent optimization cycle, proceed to S4.1.1 to calculate the initial preprocessing adjustment value of the sensor; otherwise, reuse the historical initial preprocessing adjustment value cached by the sensor. S4.1.1: Establishing a deviation model Based on the aforementioned deviation model, the optimal parameter estimation algorithm is used to calculate the parameters that satisfy... Given the sensor's θ value under certain conditions, an estimated value of θ is obtained. ; Calculate the sensor value based on the estimated value. i Theoretical compensation amount ; S4.1.2: Calculate the maximum allowable adjustment step size of the sensor in a single operation. : ； S4.1.3: Combine the theoretical compensation amount with the maximum allowable adjustment step size per cycle to calculate the initial adjustment value for preprocessing. The initial adjustment value satisfies: ； S4.2: Input the preprocessed initial adjustment values ​​corresponding to all sensors into the weighted minimization objective function for global joint optimization calculation, and obtain the optimal parameter adjustment amount of all sensors after iterative convergence; in, i Indicates sensor index, via Calculate the independent optimization period, where α is the scene adaptation coefficient. θ is the physical parameter of the sensor bias to be estimated, g(θ) is a function of the physical parameter, and ε is the random error; S ref This represents a set of sensor references with a confidence level higher than a preset threshold. w i As the confidence level weight, T base The pre-set base period; δ is the minimum value to prevent division by zero; and , P i For sensors i The original parameters, For sensors i The parameter adjustment amount, , This indicates the maximum and minimum values ​​of the sensor's internal operating parameters; This represents the maximum allowable adjustment amount for a single parameter adjustment.

6. The collaborative optimization method for multiple detection sensors according to claim 1, characterized in that, S5 also includes: combining real-time detection status information from multiple sensors, scheduling multi-source heterogeneous sensors to cooperate in joint detection, and dynamically adjusting the internal operating parameters of each sensor.

7. The collaborative optimization method for multiple detection sensors according to claim 1, characterized in that, The method further includes performing consistency constraint checks on sensor measurements based on the collected raw data. When a sensor measurement value is detected to violate the consistency constraint, the sensor is triggered to undergo mandatory calibration. The consistency constraint checks include kinematic consistency constraint checks, geometric consistency constraint checks, and target feature continuity constraint checks.

8. A collaborative optimization device for multiple detection sensors, used to implement the method according to any one of claims 1 to 7, characterized in that, The device includes: The data acquisition module is used to collect raw data and standardize the raw data to obtain standardized data. The raw data is obtained by collecting multi-dimensional raw detection information of the target through multi-source heterogeneous sensors according to the current operating parameters. The fusion trajectory data generation module is used to fuse standardized data using a multi-sensor fusion algorithm to obtain fusion trajectory data from multiple sensors. A multidimensional error feature set construction module is used to calculate the residuals of each sensor in real time based on the fused trajectory data, and construct a multidimensional error feature set based on the residuals; The optimal parameter adjustment calculation module is used to construct a weighted minimization objective function based on the multidimensional error feature set, and to calculate the optimal parameter adjustment for each sensor. The sensor internal operating parameter update module is used to generate control commands based on the optimal parameter adjustment amount and send them to each sensor to update the sensor's internal operating parameters.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the steps of the method according to any one of claims 1 to 7.

Images