Seeker target positioning method and system based on image recognition, electronic equipment and storage medium

Abstract

The invention provides a seeker target positioning method and system based on image recognition, electronic equipment and a storage medium, and relates to the technical field of seeker target localization. The method comprises the steps that a target optical image sequence is obtained, and the angular displacement amount and the angular velocity variation amount of a seeker platform are collected; extracting a target contour boundary point set from the continuous frames of the image sequence, and calculating the local curvature variation and the overall shape change rate of the contour; combining the two into a deformation characteristic vector, and establishing a mapping relation to generate a track offset vector; driving the deflection shaft of the fast reflector to perform compensation motion along the reverse direction of the trajectory offset vector, and fusing the angular displacement and the angular velocity variation to generate a real-time correction parameter; the target contour deformation characteristics are dynamically matched with the geometrical relationship of the reflector compensation vector, the stable positioning coordinates are calculated after the compensation vector form is adjusted through real-time correction parameters, and the positioning precision and stability of the high-speed moving target under strong dynamic interference can be improved.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of seeker target positioning, and in particular to a seeker target positioning method and system based on image recognition, an electronic device and a storage medium. BACKGROUND

[0002] In the application scenario of missile intercepting high-speed maneuvering targets, the seeker needs to lock the target profile with severe deformation in real time under strong vibration and high overload environment. Such targets are non-rigidly deformed due to supersonic maneuvering, the traditional image features are easily lost, and the motion trajectory is nonlinearly mutated, so the positioning method needs to solve the two technical requirements of image blur compensation and trajectory prediction at the same time to ensure that stable coordinates are output within a millisecond-level time delay.

[0003] The current mainstream scheme adopts inertial measurement unit and visual tracking fusion technology: the angular motion of the carrier is perceived through the gyroscope, the displacement vector of the feature points between adjacent frames is calculated by using the optical flow method, and the short-time motion trajectory of the target is predicted by combining the Kalman filter. This scheme relies on the stability of the preset feature point space distribution, compensates the influence of carrier jitter on imaging through the motion model, and finally calculates the target position based on the weighted centroid of the feature points.

[0004] This scheme has three defects: first, high-speed maneuvering causes the target profile to be severely distorted, fixed feature points are easily lost or shifted, and the optical flow displacement vector calculation is invalid; second, the asynchronous processing of the inertial unit and the visual data produces timing misalignment, and the carrier jitter compensation and target motion prediction interfere with each other; and finally, the motion model cannot adapt to nonlinear trajectory mutations, and the centroid calculation deviates from the real target center when the profile is deformed. SUMMARY

[0005] The application aims to provide a seeker target positioning method and system based on image recognition, an electronic device and a storage medium, to solve the problem of insufficient positioning accuracy and stability of high-speed moving targets under strong dynamic interference in the prior art.

[0006] To solve the above technical problems, in a first aspect, the application provides a seeker target positioning method based on image recognition, comprising:

[0007] In the scenario where a high-speed moving target causes imaging blur, an optical image sequence of the high-speed moving target is acquired in real time, the angular displacement and angular velocity change of the seeker platform are synchronously collected, and the angular displacement includes dynamic offset data of the pitch axis and the yaw axis;

[0008] Boundary point sets of the target profile are extracted from consecutive frames of the optical image sequence, the local curvature change and the overall shape change rate of the target profile are calculated in real time by comparing the spatial distribution difference between the boundary point sets of adjacent frames;

[0009] combine the local bending degree change amount and the overall shape change rate into a deformation feature vector, establish a mapping relationship between the deformation feature vector and a target motion direction, and generate a trajectory offset vector of the target at a next moment;

[0010] drive a deflection axis of the fast steering mirror to perform a compensation motion in a reverse direction of the trajectory offset vector, and simultaneously fuse the angular displacement amount and the angular velocity change amount to generate a real-time correction parameter of a mirror control loop, wherein the compensation motion is used to offset an image tailing phenomenon caused by high-speed motion;

[0011] In the process of performing the compensation motion by the fast steering mirror, dynamically match a spatial distribution pattern of the target profile deformation feature with a geometric relationship of the mirror compensation vector, after adjusting the compensation vector form in the matching process by the real-time correction parameter, calculate a stable positioning coordinate of the target in an image coordinate system based on the matching result.

[0012] Optionally, the combining the local bending degree change amount and the overall shape change rate into a deformation feature vector, establishing a mapping relationship between the deformation feature vector and a target motion direction, and generating a trajectory offset vector of the target at a next moment, comprises:

[0013] constructing a two-dimensional deformation feature vector according to a fixed order, with the local bending degree change amount as a first dimension element and the overall shape change rate as a second dimension element;

[0014] establishing a mapping model of vector element values and motion angle changes according to a corresponding relationship between the deformation feature vector and an actual motion direction of the target in historical motion data;

[0015] inputting the deformation feature vector of the current frame into the mapping model to predict an offset angle and an offset distance of the target relative to a current motion direction at a next moment, and generating a trajectory offset vector containing an angle component and a distance component.

[0016] Optionally, the driving the deflection axis of the fast steering mirror to perform a compensation motion in a reverse direction of the trajectory offset vector, and simultaneously fusing the angular displacement amount and the angular velocity change amount to generate a real-time correction parameter of a mirror control loop, wherein the compensation motion is used to offset an image tailing phenomenon caused by high-speed motion, comprises:

[0017] decomposing a direction component and a distance component of the trajectory offset vector, calculating a reverse compensation angle and a reverse compensation displacement amount required by the deflection axis of the fast steering mirror to drive the deflection axis to move in the reverse direction of the trajectory offset vector;

[0018] convert the angular displacement amount of the seeker platform into a platform attitude offset vector, and convert the angular velocity change amount into a platform jitter frequency, superimpose and fuse the platform attitude offset vector and the platform jitter frequency with the reverse compensation angle and the reverse compensation displacement amount.

[0019] Generate real-time correction parameters of the mirror control loop based on the superimposed fusion result, the real-time correction parameters include mirror axial correction amplitude and jitter suppression factor, used to control compensation motion to offset image tailing phenomenon.

[0020] Optionally, the target contour boundary point set is extracted from the continuous frames of the optical image sequence, the local curvature change amount and the overall shape change rate of the target contour are calculated in real time by comparing the spatial distribution difference of the boundary point set between adjacent frames, including:

[0021] Identify the target edge contour line from the current frame of the optical image sequence, collect the plane coordinates of all points on the target edge contour line to form a boundary point set, and the boundary point set is arranged in sequence according to the contour direction;

[0022] Compare the boundary point set of the current frame with the boundary point set of the previous frame in position, calculate the coordinate offset of the corresponding point position of the boundary point set, and statistically analyze the displacement direction and average displacement distance of the overall boundary point set according to the distribution of the coordinate offset;

[0023] Based on the displacement direction and average displacement distance, analyze the change amplitude of the contour line curvature, output the local curvature change amount, and measure the stretching ratio of the overall boundary point set, output the overall shape change rate.

[0024] Optionally, during the compensation motion of the fast mirror, the spatial distribution pattern of the target contour deformation feature is dynamically matched with the geometric relationship of the mirror compensation vector, and after adjusting the compensation vector form of the matching process through the real-time correction parameter, the stable positioning coordinates of the target in the image coordinate system are calculated based on the matching result, including:

[0025] During the compensation motion of the mirror, the spatial distribution pattern of the target contour deformation feature is extracted, and the geometric properties of the mirror compensation vector are obtained;

[0026] The direction angle in the geometric property is input into the axial correction amplitude of the real-time correction parameter for angle deviation calibration, and the vector length in the geometric property is input into the jitter suppression factor of the real-time correction parameter for length fluctuation suppression, to generate a calibrated compensation vector;

[0027] Convert the spatial distribution pattern into a contour feature density map, project the calibrated compensation vector to the contour feature density map for position coincidence calculation, select the maximum coincidence position as the target center point, and calculate the stable positioning coordinates in combination with the origin of the image coordinate system.

[0028] Optionally, the morphological feature vector of the current frame is input into the mapping model to predict an offset angle and an offset distance of the target relative to the current motion direction at the next moment, and generate a trajectory offset vector including an angle component and a distance component, comprising:

[0029] The first dimension element value and the second dimension element value of the morphological feature vector of the current frame are read and input into a pre-constructed mapping model processing unit;

[0030] In the mapping model processing unit, the motion angle offset corresponding to the historical curvature change is queried according to the first dimension element value, and the motion distance offset corresponding to the historical shape change is queried according to the second dimension element value;

[0031] The motion angle offset is taken as an angle component, and the motion distance offset is taken as a distance component, to generate a two-dimensional trajectory offset vector.

[0032] Optionally, the direction component and the distance component of the trajectory offset vector are decomposed to calculate a reverse compensation angle and a reverse compensation displacement required by the deflection shaft of the fast steering mirror, to drive the deflection shaft to move in the opposite direction of the trajectory offset vector, comprising:

[0033] The angle component of the trajectory offset vector is analyzed to obtain a target motion direction angle, and the distance component is analyzed to obtain a target motion displacement length;

[0034] A mirror adjustment angle opposite to the target motion direction angle is calculated, and a mirror displacement compensation amount equal to the target motion displacement length is calculated;

[0035] A deflection shaft control instruction is generated according to the mirror adjustment angle and the mirror displacement compensation amount, to drive the deflection shaft to move in the opposite direction of the trajectory offset vector.

[0036] In a second aspect, the application provides an image recognition-based seeker target positioning system, comprising:

[0037] An acquisition module is configured to acquire an optical image sequence of a high-speed motion target in real time in a scene where the high-speed motion target causes imaging blur, and synchronously collect an angular displacement amount and an angular velocity change amount of a seeker platform, wherein the angular displacement amount includes dynamic offset data of a pitch axis and a yaw axis;

[0038] A calculation module is configured to extract a target contour boundary point set from consecutive frames of the optical image sequence, and calculate a local curvature change amount and an overall shape change rate of the target contour in real time by comparing spatial distribution differences between adjacent frame boundary point sets;

[0039] A combination module is configured to combine the local bending degree change and the overall shape change rate into a deformation feature vector, establish a mapping relationship between the deformation feature vector and a target motion direction, and generate a trajectory offset vector of the target at a next time point;

[0040] A driving module is configured to drive a deflection axis of the fast steering mirror to perform a compensation movement in a reverse direction of the trajectory offset vector, and to generate real-time correction parameters of a mirror control loop by fusing the angular displacement and the angular velocity change;

[0041] A matching module is configured to dynamically match a spatial distribution pattern of the target profile deformation feature and a geometric relationship of the mirror compensation vector during the compensation movement of the fast steering mirror, and to calculate a stable positioning coordinate of the target in an image coordinate system based on a matching result after adjusting a compensation vector form of the matching process by using the real-time correction parameters.

[0042] In a third aspect, the present application provides an electronic device, comprising:

[0043] A memory is configured to store a computer program;

[0044] A processor is configured to implement steps of the method according to the first aspect when executing the computer program.

[0045] In a fourth aspect, the present application provides a computer readable storage medium, wherein the computer readable storage medium stores a computer program, and the computer program is executable by a processor to implement steps of the method according to the first aspect.

[0046] The application provides an image recognition-based seeker target positioning method. In a scene where high-speed motion targets cause imaging blur, the optical image sequence of the high-speed motion target is acquired in real time, and the angular displacement amount and angular velocity change amount of the seeker platform are synchronously collected. The angular displacement amount includes dynamic offset data of the pitch axis and the yaw axis. The optical image sequence of the high-speed motion target and the dynamic motion parameters of the seeker platform can be synchronously acquired, high-time-efficiency data basis is provided for subsequent profile deformation analysis, and the coupling interference problem of image blur and platform jitter in a high-speed scene is solved. The target profile boundary point set is extracted from the continuous frames of the optical image sequence, the local curvature change amount and the overall shape change rate of the target profile are calculated in real time by comparing the spatial distribution difference of the boundary point sets between adjacent frames, the dynamic distortion characteristics of the target profile can be quantified by the spatial distribution difference of the boundary point sets between adjacent frames, and the non-rigid deformation characteristics of the high-speed maneuvering target can be accurately captured. The local curvature change amount and the overall shape change rate are combined into a deformation feature vector, a mapping relationship between the deformation feature vector and the target motion direction is established, a trajectory offset vector of the target at the next moment is generated, short-time trajectory prediction based on the deformation characteristics of the target body is realized, and the limitation of traditional models relying on kinematic assumptions is broken through. The deflection axis of the driven fast mirror is moved in the opposite direction of the trajectory offset vector for compensation, and the real-time correction parameters of the mirror control loop are generated by fusing the angular displacement amount and the angular velocity change amount. The compensation movement is used to offset the image trailing phenomenon caused by high-speed motion, the fast mirror can be driven to move in the opposite direction of the predicted trajectory, the real-time correction parameters can be generated by fusing the platform dynamic parameters, and the image trailing effect caused by target high-speed motion can be actively offset. In the process of executing the compensation movement of the fast mirror, the spatial distribution mode of the target profile deformation characteristics is dynamically matched with the geometric relationship of the mirror compensation vector. After adjusting the compensation vector form in the matching process through the real-time correction parameters, the stable positioning coordinates of the target in the image coordinate system are calculated based on the matching result. The stable positioning coordinates with interference resistance can be calculated through the dynamic matching of the deformation characteristics and the compensation vector and the parameter adjustment, and the target center drift problem in a high-speed scene is solved.

[0047] Further, the local curvature change amount and the overall shape change rate are constructed into a two-dimensional deformation feature vector in a fixed order. A mapping model of the vector element value and the target motion angle / distance offset is established based on historical data. The next-time trajectory offset vector is predicted by inputting the current frame deformation feature vector, and the motion compensation basis containing the angle component and the distance component is output. Through the structured combination of the deformation feature vector and the historical mapping mechanism, high-precision conversion from profile dynamic change to motion vector is realized, the trajectory offset amount with direction and distance information is generated, real-time and quantitative counter-directive instructions are provided for mirror cooperative compensation, and the decoupling bottleneck of traditional trajectory prediction and motion compensation is broken through. BRIEF DESCRIPTION OF DRAWINGS

[0048] In order to more clearly illustrate the technical solutions of the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the description of the embodiments or the prior art. Obviously, the drawings in the following description only constitute some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained from these drawings without creative labor.

[0049] Figure 1 A flowchart of a target positioning method of a seeker based on image recognition provided by an embodiment of the present application is shown in the figure.

[0050] Figure 2 A specific implementation schematic diagram of a target positioning method of a seeker based on image recognition provided by an embodiment of the present application is shown in the figure.

[0051] Figure 3 A specific implementation schematic diagram of a target positioning method of a seeker based on image recognition provided by an embodiment of the present application is shown in the figure.

[0052] Figure 4 A structure schematic diagram of a target positioning system of a seeker based on image recognition provided by an embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0053] The current high-speed target seeker positioning technology faces three bottlenecks: first, the optical flow tracking relying on preset feature points causes trajectory prediction failure due to feature loss when the target is severely deformed; second, the asynchronous processing of the inertial unit and the vision system causes mutual interference between carrier jitter compensation and target motion prediction; third, the trajectory prediction based on a fixed motion model cannot adapt to nonlinear mutations, and the centroid calculation produces systematic drift when the contour is distorted. These defects are caused by the neglect of the coupling mechanism of the target body deformation characteristics and the carrier motion, which makes it difficult to balance real-time and positioning accuracy in the supersonic maneuvering scene.

[0054] In view of the above limitations, the present application proposes a trajectory prediction positioning method based on contour deformation characteristics and fast mirror cooperative motion compensation. The core is to quantify the local bending and overall deformation characteristics by extracting the spatial distribution difference of the target contour boundary point set, to construct a deformation feature vector and to map a trajectory offset vector; to synchronously drive the fast mirror to move in the opposite direction of the offset vector, to generate real-time correction parameters by fusing platform angular motion data; and finally to calculate the stable coordinates through the dynamic matching of deformation characteristics and compensation vectors. This method breaks through the millisecond-level closed-loop control of target body deformation analysis, motion trajectory prediction and mirror compensation - not only avoids the problem of feature point dependence, but also eliminates the asynchronous interference of data through cooperative compensation, and more uses deformation-driven adaptive prediction to replace rigid motion model, fundamentally solving the positioning drift and image tailing defects in the high-speed maneuvering scene, and significantly improving the target locking accuracy in complex dynamic environments.

[0055] In order to make the person skilled in the art better understand the scheme of the present application, the present application will be further described in detail below in combination with the drawings and specific embodiments. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0056] The core of the present application is to provide an image recognition-based seeker target positioning method, and a flowchart of a specific embodiment of the method is shown in Figure 1 The method comprises:

[0057] SS101, in the scene where a high-speed moving target causes imaging blur, an optical image sequence of the high-speed moving target is acquired in real time, and an angular displacement amount and an angular velocity change amount of a seeker platform are synchronously collected, the angular displacement amount comprising dynamic offset data of a pitch axis and a yaw axis;

[0058] In the above scheme, the high-speed moving target refers to a target such as a flying vehicle or a missile whose speed exceeds the speed of sound, and the motion of the target causes the photographed image to be blurred. The optical image sequence is a plurality of target pictures continuously photographed by a seeker camera, which are arranged in time sequence to form a dynamic picture. The seeker platform is a missile or unmanned aerial vehicle carrier loaded with the camera, which shakes in motion. The angular displacement amount is the change value of the inclination angle of the seeker platform in the pitch / tilt up and yaw / turn left and right directions. The angular velocity change amount refers to the change speed of the inclination angle of the platform per unit time. The dynamic offset data comprises a pitch axis angle fluctuation value and a yaw axis angle fluctuation value.

[0059] In the embodiments of the present application, first, a high-speed camera is used to photograph the moving target at a rate of more than 1000 frames per second, to generate a continuous target image sequence. These images are sorted by time stamp to form a dynamic picture that can reflect the position change of the target. Second, while the images are being photographed, a micro gyroscope sensor installed on the seeker platform is used to measure the inclination motion data of the platform in real time: the pitch axis sensor records the angle value of the tilt up and down of the platform; the yaw axis sensor records the angle value of the turn left and right of the platform; and the angular velocity sensor synchronously records the instantaneous speed of the change of the above angles. Finally, each image is bound with the platform motion data at the corresponding time: a time stamp accurate to microseconds is marked for each image; the angular displacement amount and the angular velocity change amount at the same time are read from the sensor according to the time stamp; and the image sequence and the platform motion data are packaged into a synchronous data set, to ensure that each image is attached with the corresponding platform shaking information in subsequent processing.

[0060] S102, extract a target contour boundary point set from a continuous frame of the optical image sequence, and calculate a local bending degree change amount and an overall shape change rate of the target contour in real time by comparing spatial distribution differences of the boundary point sets between adjacent frames;

[0061] Optionally, step S102 can specifically include the following steps:

[0062] S1021, identify a target edge contour line from a current frame of the optical image sequence, collect planar coordinates of all points on the target edge contour line to form a boundary point set, and arrange the boundary point set in a contour direction order;

[0063] S1022, compare the boundary point set of the current frame with a boundary point set of a previous frame in position, calculate coordinate offset amounts of corresponding points of the boundary point sets, and statistically determine a displacement direction and an average displacement distance of the overall boundary point set according to a distribution of the coordinate offset amounts;

[0064] S1023, analyze a change range of a contour line bending degree based on the displacement direction and the average displacement distance, output a local bending degree change amount, measure a stretching and shrinking ratio of an overall shape circumscribed by the boundary point set, and output an overall shape change rate.

[0065] In the above scheme, the boundary point set is a pixel coordinate set of a target contour edge arranged continuously. The coordinate offset amount is a position movement difference value of a same contour point in two frames of images. The displacement direction is a central trend angle of movement directions of all contour points. The average displacement distance is an arithmetic average value of contour point movement distances. The local bending degree change amount is a change value of a bending angle of a specific contour arc segment. The overall shape change rate is a stretching and shrinking degree of a length-width ratio of a target circumscribed polygon.

[0066] In the embodiment of the application, first, contour extraction is implemented through step S1021, an image gradient analysis method is used to identify a light-dark boundary line of a target and a background, and pixel point coordinates are sampled at a fixed interval along the boundary line. The sampling points are smoothly connected to form a closed contour line, and all point coordinates are recorded in a contour direction order. For example, an edge enhancement process is performed on a fighter image, 200 points are collected along a wing-fuselage-tail path, and a coordinate sequence is [(x1, y1), (x2, y2), …, (x200, y200)].

[0067] Secondly, displacement statistics are completed by step S1022: the current frame point set and the previous frame point set are matched point by point in the recording order, and the coordinate difference (Δx, Δy) of each point is calculated. The average value of Δx of all points is obtained as the horizontal displacement trend, and the average value of Δy is obtained as the vertical displacement trend. The overall displacement direction angle is calculated by combining the two. The average of the displacement distances of the points is taken as the average displacement distance. For example, in 200 points, 150 points have Δx>0 (moving to the right), the average Δx=5 pixels; 180 points have Δy<0 (moving upwards), the average Δy=-3 pixels→the displacement direction is 31 degrees to the right and up. The average of the displacement distances of the points is √(Δx 2 +Δy 2 ) = 5.8 pixels.

[0068] Finally, the deformation characteristics are quantified by step S1023, the key section of the contour is selected, and the average change of the included angle of the connecting line of adjacent points in the section is taken as the change of the curvature. The minimum circumscribed rectangle of the contour point set is constructed, and the change of the length-width ratio of the two frames is compared to obtain the overall shape change rate. For example, the bending angle of the wing section increases from 35° to 50°→the change of the curvature is +15°; the length-width ratio of the fuselage circumscribed rectangle changes from 3.2:1 to 2.9:1→the length shrinks by 9.4%.

[0069] In actual application, it is assumed that a missile seeker tracks a supersonic fighter performing "cobra maneuver". The contrast of the fuzzy image of the fighter in the upward pull-up state is enhanced by step S1021; the fuselage contour is recognized: starting from the top point of the nose radar cover, along the cabin cover→vertical tail→engine nozzle→belly path; 320 contour points are collected at an interval of 2 pixels; and the coordinate sequence is stored in the order of the flight direction.

[0070] The corresponding contour points of the previous frame / flat flight state are matched by step S1022: the nose point A1→A2, and the vertical tail point B1→B2; the calculation of the A point Δx=+15 pixels, moving to the right; Δy=+80 pixels, moving downward; 85% of the 320 points have Δy>0, moving downward, and the average Δy=+72 pixels; the displacement direction angle=arctan(72 / 15)=78°; and the average displacement distance=(15 2 +72 2 )^(1 / 2)=74 pixels.

[0071] By analyzing the belly arc segment in step S1023: the bending angle of the previous frame is 120°→the bending angle of the current frame is 160°, and the compression is 40°; the bending angle of the vertical tail segment is 60°→30°, and the relaxation is 30°→the average output bending degree change is -5°; the length-width ratio of the circumscribed rectangle of the previous frame is 4:1→3:1, and the fuselage is vertical; the length contraction is 25%→the shape change rate is -25%. When the fighter completes the maneuver to restore level flight: the bending angle of the belly arc segment is 160°→130°, and the rebound is 30°; the length-width ratio of the circumscribed rectangle is 3:1→3.8:1, and the length is restored; the real-time output bending degree change is +30°, and the shape change rate is +26.7%.

[0072] The overall scheme of step S102 captures the target structure features by sequentially profiling the point set, separates the overall motion and local deformation by using displacement statistics, combines the bending degree of the key segment and the quantization analysis of the circumscribed shape, and provides a feature description basis for high-speed maneuvering targets against deformation interference, overcoming the failure problem of traditional point feature tracking under extreme attitude.

[0073] S103, combining the local bending degree change and the overall shape change rate into a deformation feature vector, establishing a mapping relationship between the deformation feature vector and the target motion direction, and generating a trajectory offset vector of the target at the next moment;

[0074] Optionally, step S103 can specifically include the following steps:

[0075] S1031, taking the local bending degree change as a first dimension element and the overall shape change rate as a second dimension element, and constructing a two-dimensional deformation feature vector in a fixed order;

[0076] S1032, establishing a mapping model of vector element values and motion angle changes according to the corresponding relationship between the deformation feature vector and the actual motion direction of the target in the historical motion data;

[0077] S1033, inputting the deformation feature vector of the current frame into the mapping model to predict the offset angle and offset distance of the target relative to the current motion direction at the next moment, and generating a trajectory offset vector containing an angle component and a distance component.

[0078] In step S1033, the first dimension element value and the second dimension element value of the deformation feature vector of the current frame can be read and input into a pre-constructed mapping model processing unit; in the mapping model processing unit, the motion angle offset corresponding to the historical bending degree change is queried according to the first dimension element value, and the motion distance offset corresponding to the historical shape change is queried according to the second dimension element value; the motion angle offset is taken as an angle component, the motion distance offset is taken as a distance component, and a two-dimensional trajectory offset vector is generated by combination.

[0079] In the above scheme, the two-dimensional deformation feature vector is an array of two values: the first represents the change in the degree of contour curvature, and the second represents the overall shape scaling ratio. The mapping model is a database that records the relationship between historical deformation data and actual motion offset. The trajectory offset vector is a two-dimensional output containing the predicted direction angle and movement distance.

[0080] In this embodiment, firstly, a feature vector is constructed through step S1031: the two feature values ​​output in step S102, the local curvature change and the overall shape change rate, are filled into predefined positions: the first value is placed in the first dimension of the vector, and the second value is placed in the second dimension, forming a two-dimensional data group with a fixed format. This process ensures that the feature data of different frames are comparable. For example, when a fighter jet performs a high-speed climb maneuver, if the measured wing curvature increases by 20 degrees, the first dimension value is +20, the fuselage length is compressed by 15%, and the second dimension value is -15, then the vector [+20, -15] is constructed.

[0081] Secondly, such as Figure 3 As shown, a mapping model is established through step S1032: All vector records in the historical database that are similar to the current deformation characteristics are retrieved; the actual motion data corresponding to these historical vectors are extracted, including the actual offset angle and movement distance of the target; the numerical correlation between deformation characteristic values ​​and motion data is analyzed: the proportional relationship between the first dimension value / curvature change and angle offset is calculated; the proportional relationship between the second dimension value / shape change rate and distance offset is calculated; the proportional relationship is quantified into mapping coefficients to form a conversion rule of "characteristic value → motion amount". For example, analyzing 100 sets of historical data reveals that every 1 unit increase in the first dimension value corresponds to a 2.1 degree rightward deviation, and every 1% increase in the absolute value of the second dimension corresponds to a 35-meter forward movement, thus generating the coefficient set [2.1 degrees / unit, 35 meters / %].

[0082] Finally, the trajectory offset is generated through step S1033: the first and second dimension values ​​are read from the current deformation feature vector; the first dimension value is multiplied by the angle coefficient to obtain the predicted offset angle; the absolute value of the second dimension value is multiplied by the distance coefficient to obtain the predicted movement distance; the angle and distance values ​​are combined to output a new two-dimensional vector. For example, the input vector is [+20, -15], the angle coefficient is 2.1 degrees / unit, and the distance coefficient is 35 meters / %. The calculated angle offset is 20 × 2.1 = 42 degrees, rightward deviation, and the distance offset is 15 × 35 = 525 meters. The output trajectory offset vector is [42 degrees, 525 meters].

[0083] In practical applications, it is assumed that the air defense missile seeker is tracking a stealth fighter performing a "high-speed spiral dive → sharp turn recovery" maneuver. Through step S1031, when the fighter dives at an 80-degree angle: the wing curvature increases sharply to +25 degrees / first dimension value; the fuselage is compressed and shortened by 12% by the airflow, second dimension value -12; construct the deformation characteristic vector [+25, -12].

[0084] Through step S1032, query the same kind of diving maneuver data in the history database: find 3 groups of similar vectors: [ + 22, - 10] corresponds to 46 degrees of actual right deviation and 400 meters of forward movement, [ + 28, - 15] corresponds to 58 degrees of right deviation and 530 meters of forward movement; Calculate the angle coefficient: 46 / 22≈2.09, 58 / 28≈2.07→take the average 2.08 degrees / unit; Calculate the distance coefficient: 400 / 10=40, 530 / 15≈35.3→take the average 37.7 meters / %.

[0085] Through step S1033, input the current vector [ + 25, - 12]; Angle offset = 25 × 2.08 = 52 degrees (right deviation); Distance offset = 12 × 37.7≈452 meters; Output trajectory offset vector [52 degrees, 452 meters]. Assume that the warplane suddenly changes to dive right and turns sharply: new deformation characteristics: wing curvature rebounds to -8 degrees, fuselage stretches +18%→vector [-8,+18]; Call the history turning rule: the first dimension coefficient is -1.5 degrees / unit (left deviation), and the distance coefficient is 42 meters / %; Prediction: angle offset = -8 × (-1.5) = 12 degrees (left deviation), distance offset = 18 × 42 = 756 meters; Actual movement: left deviation 15 degrees, forward movement 780 meters, error less than 8%.

[0086] The overall scheme of the above step S103 realizes real-time quantitative prediction from target physical deformation to motion trajectory by structuring and packaging deformation characteristics and associating historical motion rules, provides millisecond-level trajectory prediction capability for high-speed maneuvering targets, and significantly improves the accuracy and timeliness of the mirror cooperative compensation.

[0087] S104, drive the deflection axis of the fast steering mirror to perform compensation movement in the opposite direction of the trajectory offset vector, while fusing the angle displacement and the angle velocity change to generate real-time correction parameters of the mirror control loop, wherein the compensation movement is used to offset the image tailing phenomenon caused by high-speed movement;

[0088] Optionally, step S104 can specifically include the following steps:

[0089] S1041, decompose the direction component and the distance component of the trajectory offset vector, calculate the reverse compensation angle and the reverse compensation displacement required for the deflection axis of the fast steering mirror, and drive the deflection axis to move in the opposite direction of the trajectory offset vector;

[0090] The step S1041 can specifically include the following processes: obtaining a target motion direction angle by analyzing the angle component of the trajectory offset vector, and obtaining a target motion displacement length by analyzing the distance component; calculating a mirror adjustment angle opposite to the target motion direction angle, and simultaneously calculating a mirror displacement compensation amount equal to the target motion displacement length; generating a deflection shaft control instruction according to the mirror adjustment angle and the mirror displacement compensation amount, and driving the deflection shaft to move in the opposite direction of the trajectory offset vector.

[0091] S1042, converting the angular displacement amount of the seeker platform into a platform attitude offset vector and converting the angular velocity change amount into a platform jitter frequency, and superimposing and fusing the platform attitude offset vector and the platform jitter frequency with the reverse compensation angle and the reverse compensation displacement amount;

[0092] S1043, generating a real-time correction parameter of a mirror control loop based on the superimposed and fused result, the real-time correction parameter including a mirror axial correction amplitude and a jitter suppression factor, which are used to control the compensation motion to offset the image tailing phenomenon.

[0093] In the above scheme, the reverse compensation angle is the angle that the mirror needs to adjust in the opposite direction of the predicted direction. The reverse compensation displacement amount is the reverse displacement that the mirror needs to move, which is equal to the predicted distance. The platform attitude offset vector is a description of the platform tilt direction composed of the pitch angle and the yaw angle. The platform jitter frequency is a quantitative value of the speed of the platform angle change. The axial correction amplitude is the final angle value that the mirror needs to adjust. The jitter suppression factor is a control strength parameter for suppressing platform jitter.

[0094] In the embodiments of the present application, first, the compensation parameters are analyzed by step S1041: the angle data and the distance data are separated from the trajectory offset vector. The opposite direction of the angle value is taken as the mirror compensation angle, and the original distance value is directly taken as the compensation displacement amount that the mirror needs to move. This process converts the prediction instruction into the physical action reference of the mirror. For example, the trajectory offset vector [left 40°, 500 meters]→ compensation angle = right turn 40°, compensation displacement amount = 500 meters.

[0095] Secondly, the platform dynamic is fused by step S1042: the angular displacement is decomposed into three-dimensional space components: pitch angle change → vertical direction offset value; yaw angle change → horizontal direction offset value; the absolute value of angular velocity change is taken as the platform jitter frequency; the platform data and the compensation parameter are dynamically superimposed: horizontal compensation angle + yaw offset value = comprehensive horizontal adjustment; vertical compensation displacement x (1 + pitch offset value / 100) = displacement compensation correction. For example, platform yaw -5°, pitch +8°, angular velocity 10° / s → horizontal offset -5°, vertical offset +8%, jitter frequency 10 Hz. Superimposed with the compensation parameter [right turn 40°, 500 meters]: comprehensive horizontal adjustment = 40° + (-5°) = 35° right turn, displacement correction = 500 x (1 + 0.08) = 540 meters.

[0096] Finally, the control parameter is generated by step S1043: comprehensive horizontal adjustment value → mirror axial correction amplitude; the reciprocal of jitter frequency → jitter suppression factor, the higher the value, the smaller the value; the output parameter format: correction amplitude + suppression factor. For example, axial correction amplitude 35°, jitter suppression factor 1 / 10 = 0.1 seconds → control instruction "right turn 35°, displacement 540 meters, suppression strength 0.1 seconds".

[0097] In actual application, it is assumed that the air defense missile seeker tracks the stealth fighter performing "high-speed snake maneuver + barrel roll". Through step S1041, when the missile seeker tracks the stealth fighter performing high-speed sharp turn, the trajectory offset vector is [left offset 50°, 800 meters]. The system decomposes the direction component left offset 50° and the distance component 800 meters of the vector, and calculates that the mirror needs to right turn 50° in the opposite direction of the direction component, and directly uses the distance component as the reverse compensation displacement 800 meters. This process converts the predicted target left movement into a right compensation instruction of the mirror.

[0098] Through step S1042, when the fighter enters barrel roll maneuver, the platform sensor monitors the pitch angle oscillation (+6° to -4°) and the yaw angle mutation (+12° right turn) in real time, and the peak value of angular velocity is 15° / s. The system first converts the angular displacement into the attitude offset vector (horizontal +12°, vertical average +1%), and converts the absolute value of angular velocity into the jitter frequency 15 Hz. Then, the data is fused with the output of step S1041 (right turn 50°, displacement 800 meters): the horizontal compensation angle is superimposed with the platform yaw value to obtain the comprehensive right turn 62° (50°+12°), and the displacement is corrected to 800 x 1.01 = 808 meters in combination with the pitch coefficient.

[0099] Based on the fusion result [right turn 62°, displacement 808 meters, dithering frequency 15 Hz] by step S1043, the system directly takes the integrated angle value as the axial correction amplitude 62°, and takes the dithering frequency reciprocal 1 / 15≈0.067 seconds to generate a dithering suppression factor. The final output control parameter is: "right turn 62°, displacement 808 meters, suppression strength 0.067 seconds", which drives the mirror to precisely offset the combined interference of target displacement and platform dithering in the barrel roll maneuver.

[0100] The overall scheme of the above step S104 generates a mirror basic compensation instruction by reverse conversion of trajectory prediction data, dynamically fuses platform motion disturbance for real-time correction, and finally outputs accurate optical compensation control parameters, effectively eliminates image tailing phenomenon in high-speed maneuvering scenarios, and significantly improves target locking accuracy and imaging stability.

[0101] S105, in the process of executing compensation motion by the fast mirror, dynamically match the spatial distribution pattern of the target profile deformation feature with the geometric relationship of the mirror compensation vector, and after adjusting the compensation vector form in the matching process through the real-time correction parameter, calculate the stable positioning coordinates of the target in the image coordinate system based on the matching result.

[0102] Optionally, step S104 can specifically include the following steps:

[0103] S1051, during the compensation motion of the mirror, extract the spatial distribution pattern of the target profile deformation feature, and simultaneously acquire the geometric properties of the mirror compensation vector;

[0104] S1052, input the direction angle in the geometric properties into the axial correction amplitude of the real-time correction parameter for angle deviation calibration, input the vector length in the geometric properties into the dithering suppression factor of the real-time correction parameter for length fluctuation suppression, and generate a calibrated compensation vector;

[0105] S1053, convert the spatial distribution pattern into a profile feature density map, project the calibrated compensation vector to the profile feature density map for position coincidence calculation, select the maximum coincidence position as the target center point, and calculate the stable positioning coordinates in combination with the origin of the image coordinate system.

[0106] In the above scheme, the spatial distribution pattern is the bending density distribution of different regions of the target profile. The geometric properties are the physical characteristics of the compensation vector, the direction angle refers to the offset direction, and the vector length refers to the movement distance. The profile feature density map is a two-dimensional heat map that maps the profile deformation intensity. The position coincidence degree is the matching degree of the projection point of the compensation vector and the profile deformation region.

[0107] In the embodiments of the present application, first, the matching elements are extracted through step S1051: during the mirror movement process, the target profile is scanned in real time and divided into several continuous segments, such as the nose segment, the left wing segment, the fuselage segment, and the right wing segment. The bending intensity value of each segment is calculated: the maximum vertical distance between the connecting line of the head and tail of the segment and the middle point is measured, and the greater the distance, the higher the bending intensity. The bending intensity values of all segments are arranged in spatial order to form a spatial distribution pattern. The real-time output parameters of the mirror control system are synchronously read: the direction angle of the compensation vector [the mirror deflection azimuth] and the vector length [the mirror displacement amount], as the geometric attributes. For example, the profile of a high-speed maneuvering warplane is divided into 6 segments, and the bending intensity sequence [0.2, 0.8, 0.3, 0.7, 0.1, 0.6] is measured; the compensation vector parameters are the direction angle of 35° and the length of 750 meters.

[0108] Secondly, dynamic calibration is performed through step S1052: angle calibration: the original direction angle of the compensation vector is algebraically added to the axial correction amplitude in the real-time correction parameter. If the correction amplitude is positive, the deflection angle is increased, and if the correction amplitude is negative, the angle is decreased. Length suppression: the original vector length is multiplied by the jitter suppression factor, and the smaller the suppression factor, the stronger the suppression of jitter. New vector generation: the calibrated angle and length values are combined to form an anti-interference calibrated compensation vector. For example, the original direction angle of 35° is superimposed with a correction amplitude of +5° to obtain 40°; the original length of 750 meters is multiplied by the suppression factor of 0.12 to obtain 90 meters; and the output calibrated vector is [40°, 90 meters].

[0109] Finally, precise positioning is realized through step S1053: thermal map generation: taking the image plane as the base map, the bending intensity values of the spatial distribution pattern are mapped to the pixel gray value, the higher the intensity, the brighter, and the profile feature density map is generated. Vector projection: starting from the image center point, a straight line is drawn according to the direction angle of the calibrated vector, and the length of the straight line is equal to the length of the calibrated vector, forming a projection path. Matching calculation: the cumulative pixel number of the projection path passing through the high-light area with a gray value >0.7 is counted as the position coincidence degree. Center positioning: the midpoint of the path segment with the highest coincidence degree is selected as the target center point, and the absolute coordinates of the image coordinate system origin are combined to convert the target positioning coordinates. For example, the projection path extends 90 meters in the 40° direction and passes through the left wing high-light area, and the path with a gray value of 0.8 accounts for 85%, and the midpoint coordinates (120, 180) are taken; the image origin (0, 0)→target coordinates (120, 180).

[0110] In practical applications, when the missile seeker tracks the stealth fighter performing a 90-degree sharp turn, the mirror is performing left turn compensation through step S1051. The system divides the target profile into 8 segments: nose, left wing leading edge, left wing trailing edge, fuselage midsection, right wing leading edge, right wing trailing edge, vertical tail, and belly fin. The bending intensity of each segment is calculated: the left wing trailing edge has a high intensity of 0.85 due to airflow compression, while the nose has a low intensity of 0.25, forming a distribution sequence [0.25, 0.75, 0.85, 0.3, 0.6, 0.4, 0.5, 0.2]. The mirror compensation vector is obtained synchronously: direction angle 50°, left turn, length 800 meters.

[0111] When the fighter enters a low-altitude turbulent area, the platform shakes violently. The real-time correction parameters are: axial correction amplitude +8°, enhanced left turn, and shaking suppression factor 0.1. The calibration process is: direction angle 50°+8°=58°; vector length 800 meters x 0.1=80 meters. The calibration vector [58°, 80 meters] is generated. This calibration compresses the original displacement by 90%, significantly suppressing the path deviation caused by shaking.

[0112] Finally, the bending intensity distribution is converted into a heat map through step S1053: left wing trailing edge area, coordinate range X: 50-150, Y: 200-300, gray level 0.85 / highlight, and gray level <0.6 for the remaining areas. From the center of the image (100, 200), an 80-meter projection path is drawn along the 58° direction, with the equation: Y=1.6X-160. The length of the path passing through the highlight area is calculated: 42 meters overlap in the Y=200-300 interval with the highlight area, which is 52.5% of the total length of 80 meters, with a much higher coincidence degree than other areas. The midpoint (110, 250) of the overlapping segment is taken as the target center. The fighter suddenly performs an S-shaped evasion maneuver, and the profile is severely distorted. The traditional method has a positioning drift of 35 pixels, while the proposed solution matches the heat map through dynamic calibration: the projection path of the calibration vector [62°, 85 meters] still passes through the newly deformed highlight area, with a bending intensity of 0.82, and the center point coordinates fluctuate only ±3 pixels. The final output stable coordinates (135, 265) ensure that the missile continues to lock the target.

[0113] The overall scheme of step S105 quantifies the bending intensity distribution and dynamically projects the mirror vector, achieving precise target center locking in high-speed maneuvering scenarios, effectively overcoming positioning drift caused by image shaking and profile deformation, and significantly improving the target tracking stability of the seeker in extreme environments.

[0114] The following is a complete embodiment for steps S101-S105:

[0115] As Figure 2As shown, in the scene of missile seeker tracking high-speed evading fighter, through step S101, when the missile seeker tracks the supersonic fighter with "S-type evading" maneuver, in the scene of target speed causing image blur: through the high-speed infrared imaging system, the sequence of optical images of the fighter is acquired at a rate of 1200 frames per second, the synchronous MEMS gyroscope mounted on the seeker platform collects the angular displacement of the platform, the dynamic offset of the pitch axis is +6.3°, the dynamic offset of the yaw axis is-4.8°, and the angular velocity variation is 28° / s for the pitch axis and 32° / s for the yaw axis; the hardware trigger signal is used to bind each frame of image with the platform motion data at the corresponding moment to form a time-aligned data packet for subsequent processing.

[0116] Through step S102, the fighter contour of the continuous frame is extracted from the image sequence: first, the current frame is subjected to edge enhancement processing, and 320 boundary points are collected along the path of nose-left wing-tail-right wing to form an ordered point set; the point set of the current frame is compared with the point set of the frame 0.01 seconds ago point by point, and it is calculated that the coordinates of the wing tip point move from (150, 300) to (142, 292), Δx=-8 pixels, Δy=-8 pixels, 85% of the 320 points move to the lower left, the average displacement is 11.3 pixels, and the overall displacement direction is 45° to the lower left; based on the displacement distribution, it is analyzed that the left wing segment bending angle increases from 30° to 48°, the bending degree changes by +18°, and the length-width ratio of the circumscribed rectangle of the fuselage changes from 3.2:1 to 2.8:1, and the overall shape change rate is-12.5%.

[0117] Through step S103, the bending degree change +18° is taken as the first dimension element, and the shape change rate-12.5% is taken as the second dimension element, and a deformation feature vector [18,-12.5] is constructed; 20 similar vectors in the historical database are queried, such as [15,-10] corresponding to actual right deviation 35° and advancing 400 meters, a mapping rule is established: the first dimension value×2.33 obtains the angle deviation, and the second dimension absolute value×32 obtains the distance deviation; input the current vector to calculate: angle deviation=18×2.33≈42°, right deviation, distance deviation=12.5×32=400 meters, and a trajectory deviation vector [42°, 400m] is generated.

[0118] Through step S104, the direction component right deviation 42° and the distance component 400 meters are obtained by decomposing the trajectory deviation vector, and the mirror needs to be turned left by 42° [in the opposite direction] and displaced by 400 meters; the real-time angular displacement of the platform [pitch +7°, yaw-5°] is converted into an attitude vector [vertical +7%, horizontal-5°], and the angular velocity 36° / s is converted into a jitter frequency 36Hz; the horizontal compensation angle left turn 42° is superimposed with the platform yaw-5° to obtain the comprehensive left turn 47°, and the displacement amount 400 meters×(1+7%)=428 meters; based on the fusion result, real-time correction parameters are generated: the axial correction amplitude is 47°, the jitter suppression factor is 1 / 36≈0.028 seconds, and the driving mirror executes anti-tailing compensation.

[0119] Through step S105, the warplane profile is calculated for the bending strength in 6 zones during the mirror compensation: left wing 0.85, nose 0.25, right wing 0.6, forming a spatial distribution pattern; at the same time, the original geometric properties of the compensation vector are obtained, the direction angle is left turn 47°, the length is 428 meters; the real-time correction parameter is calibrated: the direction angle is 47°+correction amplitude 0°=47°, no angle adjustment, the length is 428 meters x suppression factor 0.028≈12 meters; the distribution pattern is converted into a heat map, the left wing coordinates X: 50-150, Y: 200-300, gray scale 0.85; the 12-meter path is projected from the image center (100, 200) along the 47° direction, the equation Y=1.07X-107, the coincidence degree of the path passing through the left wing highlight area is calculated as 92%; the intersection area center (120, 180) is taken as the target center point, and the stable positioning coordinates (120, 180) are calculated in combination with the image lower left corner origin (0, 0).

[0120] The image recognition-based seeker target positioning method provided in the application realizes stable target positioning in a supersonic evasion maneuvering scene, effectively overcomes the tracking failure problem caused by image trailing and profile deformation, and significantly improves the locking precision and anti-interference ability of the missile seeker in an extreme dynamic environment.

[0121] Figure 4 A specific implementation structure diagram of a kind of image recognition-based seeker target positioning system provided for the embodiment of the application, refer to Figure 4 The system can include:

[0122] The acquisition module 41 is used to acquire the optical image sequence of the high-speed moving target in real time in the scene where the high-speed moving target causes imaging blur, and synchronously collect the angular displacement and angular velocity variation of the seeker platform, wherein the angular displacement includes the dynamic offset data of the pitch axis and the yaw axis.

[0123] The calculation module 42 is used to extract the target profile boundary point set from the continuous frames of the optical image sequence, and calculate the local bending degree variation and overall shape change rate of the target profile in real time by comparing the spatial distribution difference between adjacent frames.

[0124] The combination module 43 is used to combine the local bending degree variation and overall shape change rate into a deformation feature vector, establish a mapping relationship between the deformation feature vector and the target motion direction, and generate a trajectory offset vector of the target at the next moment.

[0125] The driving module 44 is configured to drive the deflection axis of the fast steering mirror to perform a compensation movement in the opposite direction of the trajectory deflection vector, and to generate real-time correction parameters of a mirror control loop by fusing the angular displacement and the angular velocity variation.

[0126] The matching module 45 is configured to dynamically match the spatial distribution mode of the target profile deformation feature with the geometric relationship of the mirror compensation vector during the compensation movement of the fast steering mirror, and to calculate the stable positioning coordinates of the target in the image coordinate system based on the matching result after adjusting the compensation vector mode of the matching process by using the real-time correction parameters.

[0127] The image recognition-based seeker target positioning system according to the embodiments of the present application is used to implement the image recognition-based seeker target positioning method described above, and the specific implementation of the image recognition-based seeker target positioning system can be seen from the foregoing embodiment part of the image recognition-based seeker target positioning method, and the specific implementation can be referred to the description of the corresponding embodiment part, which will not be repeated here.

[0128] The present application further provides an electronic device, which comprises a memory for storing a computer program and a processor for executing the computer program to implement the steps of the image recognition-based seeker target positioning method described above.

[0129] The present application further provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the steps of the image recognition-based seeker target positioning method described above.

[0130] In an exemplary embodiment, the computer readable storage medium described above can include, but is not limited to, a U disk, a read-only memory, a random access memory, a mobile hard disk, a magnetic disk or an optical disk, and various media that can store computer programs.

[0131] The embodiments of the present application further provide a computer program product, which comprises a computer program, and the computer program is executed by a processor to implement the steps of the image recognition-based seeker target positioning method described above.

[0132] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0133] The foregoing has provided a detailed description of the image recognition-based target localization method, system, electronic device, and storage medium provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and its core ideas. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. An image recognition-based seeker target positioning method, characterized in that, The application relates to a method for tracking a high-speed moving target in a real-time imaging system. In a scene where a high-speed moving target causes imaging blur, an optical image sequence of the high-speed moving target is acquired in real time, and the angular displacement amount and the angular velocity variation amount of a seeker platform are synchronously collected, wherein the angular displacement amount comprises dynamic offset data of a pitch axis and a yaw axis; Target contour boundary point sets are extracted from continuous frames of the optical image sequence, the local curvature variation amount and the overall shape change rate of the target contour are calculated in real time by comparing the spatial distribution differences of the boundary point sets between adjacent frames, the local curvature variation amount and the overall shape change rate are combined into a deformation feature vector, a mapping relationship between the deformation feature vector and the target moving direction is established, and a trajectory offset vector of the target at the next moment is generated; A deflection axis of a fast reflector is driven to perform compensation movement in the reverse direction of the trajectory offset vector, and real-time correction parameters of a reflector control loop are generated by fusing the angular displacement amount and the angular velocity variation amount, wherein the compensation movement is used to offset the image tailing phenomenon caused by the high-speed movement; In the process of executing the compensation movement of the fast reflector, the spatial distribution mode of the target contour deformation feature is dynamically matched with the geometric relationship of the reflector compensation vector, after the compensation vector form of the matching process is adjusted by the real-time correction parameters, the stable positioning coordinates of the target in an image coordinate system are calculated based on the matching result. The method for tracking a high-speed moving target in a real-time imaging system comprises the following steps:

2. The method of claim 1, wherein, The local curvature variation amount is taken as a first dimension element, the overall shape change rate is taken as a second dimension element, and a two-dimensional deformation feature vector is constructed in a fixed order; A mapping model of vector element values and motion angle changes is established according to the corresponding relationship between the deformation feature vector and the actual moving direction of the target in historical motion data; The deformation feature vector of the current frame is input into the mapping model, the offset angle and the offset distance of the target relative to the current moving direction at the next moment are predicted, and a trajectory offset vector containing an angle component and a distance component is generated. The method for tracking a high-speed moving target in a real-time imaging system comprises the following steps:

3. The method of claim 1, wherein, The direction component and the distance component of the trajectory offset vector are decomposed, the reverse compensation angle and the reverse compensation displacement amount required by the deflection axis of the fast reflector are calculated, and the deflection axis is driven to move in the reverse direction of the trajectory offset vector; The angular displacement amount of the seeker platform is converted into a platform posture offset vector, the angular velocity variation amount is converted into a platform jitter frequency, the platform posture offset vector and the platform jitter frequency are superimposed and fused with the reverse compensation angle and the reverse compensation displacement amount; Real-time correction parameters of a reflector control loop are generated based on the superimposed and fused result, the real-time correction parameters contain a reflector axial correction amplitude and a jitter suppression factor, and are used to control the compensation movement to offset the image tailing phenomenon. ​ 4. The method of claim 1, wherein, The target contour boundary point set is extracted from the continuous frames of the optical image sequence, the local bending degree change and the overall shape change rate of the target contour are calculated in real time by comparing the spatial distribution difference of the boundary point set between adjacent frames, and the method comprises the following steps: A target edge contour line is identified from a current frame of the optical image sequence, a boundary point set of all points on the target edge contour line is collected, and the boundary point set is arranged in sequence according to the contour direction; The boundary point set of the current frame is compared with the boundary point set of the previous frame in position, the coordinate offset of the corresponding points of the boundary point set is calculated, and the displacement direction and the average displacement distance of the overall boundary point set are calculated according to the distribution of the coordinate offset; The change amplitude of the contour line bending degree is analyzed based on the displacement direction and the average displacement distance, the local bending degree change is output, and the stretching ratio of the overall shape of the boundary point set is measured, and the overall shape change rate is output.

5. The method of claim 1, wherein, The spatial distribution pattern of the target contour deformation feature is dynamically matched with the geometric relationship of the mirror compensation vector during the compensation movement of the fast mirror, and after the compensation vector form of the matching process is adjusted by the real-time correction parameter, the stable positioning coordinates of the target in the image coordinate system are calculated based on the matching result, and the method comprises the following steps: During the compensation movement of the mirror, the spatial distribution pattern of the target contour deformation feature is extracted, and the geometric properties of the mirror compensation vector are obtained; The direction angle in the geometric properties is input into the axial correction amplitude of the real-time correction parameter for angle deviation calibration, and the vector length in the geometric properties is input into the jitter suppression factor of the real-time correction parameter for length fluctuation suppression, to generate a calibrated compensation vector; The spatial distribution pattern is converted into a contour feature density map, the calibrated compensation vector is projected onto the contour feature density map for position coincidence calculation, the maximum coincidence position is selected as the target center point, and the stable positioning coordinates are calculated in combination with the origin of the image coordinate system.

6. The method of claim 2, wherein, The deformation feature vector of the current frame is input into the mapping model to predict the offset angle and the offset distance of the target relative to the current motion direction at the next moment, to generate a trajectory offset vector containing an angle component and a distance component, and the method comprises the following steps: The first dimension element value and the second dimension element value of the deformation feature vector of the current frame are read and input into a pre-constructed mapping model processing unit; In the mapping model processing unit, the motion angle offset corresponding to the historical bending degree change is queried according to the first dimension element value, and the motion distance offset corresponding to the historical shape change is queried according to the second dimension element value; The motion angle offset is taken as an angle component, the motion distance offset is taken as a distance component, and a two-dimensional trajectory offset vector is generated by combination.

7. The method of claim 3, wherein, The direction component and the distance component of the trajectory offset vector are decomposed, the reverse compensation angle and the reverse compensation displacement required by the deflection shaft of the fast mirror are calculated, and the deflection shaft is driven to move in the opposite direction of the trajectory offset vector, and the method comprises the following steps: The angle component of the trajectory offset vector is analyzed to obtain the target motion direction angle, and the distance component is analyzed to obtain the target motion displacement length; calculating a mirror adjustment angle opposite to the target motion direction angle, and calculating a mirror displacement compensation amount equal to the target motion displacement length; generating a deflection shaft control instruction according to the mirror adjustment angle and the mirror displacement compensation amount, and driving the deflection shaft to move in the opposite direction of the trajectory offset vector.

8. An image recognition based seeker target positioning system, characterized by, The method comprises the steps of: an acquisition module, configured to acquire an optical image sequence of a high-speed motion target in real time in a scenario where the high-speed motion target causes imaging blur, and synchronously collect an angular displacement amount and an angular velocity variation amount of a seeker platform, the angular displacement amount including dynamic offset data of a pitch axis and a yaw axis; a calculation module, configured to extract a target contour boundary point set from consecutive frames of the optical image sequence, and calculate a local curvature variation amount and an overall shape change rate of the target contour in real time by comparing spatial distribution differences between boundary point sets of adjacent frames; a combination module, configured to combine the local curvature variation amount and the overall shape change rate into a deformation feature vector, establish a mapping relationship between the deformation feature vector and a target motion direction, and generate a trajectory offset vector of the target at a next time point; a driving module, configured to drive a deflection shaft of a fast mirror to perform compensation motion in the opposite direction of the trajectory offset vector, and generate real-time correction parameters of a mirror control loop by fusing the angular displacement amount and the angular velocity variation amount, wherein the compensation motion is used to offset image trailing caused by high-speed motion; a matching module, configured to dynamically match a spatial distribution mode of a target contour deformation feature with a geometric relationship of a mirror compensation vector during the compensation motion of the fast mirror, and calculate a stable positioning coordinate of the target in an image coordinate system based on a matching result after adjusting a compensation vector form in the matching process by using the real-time correction parameters.

9. An electronic device, comprising: The method comprises the steps of: a memory, configured to store a computer program; a processor, configured to execute the computer program to implement steps of the image recognition-based seeker target positioning method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer program is stored in the computer readable storage medium, and the computer program is executed by the processor to implement the image recognition-based seeker target positioning method according to any one of claims 1 to 7.

Images