A target extraction and biasing method suitable for complex illumination background

Abstract

This invention discloses a target extraction and biasing method suitable for complex lighting conditions, belonging to the fields of directed energy optoelectronic countermeasures, optoelectronic detection, and precision tracking and aiming. It includes an optoelectronic imaging detection unit, an illumination layered modeling unit, a multi-feature fusion target extraction unit, an illumination-bias error calculation unit, a servo tracking and aiming drive unit, and a global closed-loop feedback unit. The optoelectronic imaging detection unit simultaneously acquires light intensity and polarization images. This invention constructs an integrated linkage architecture for illumination modeling, target extraction, and bias compensation, establishing data interaction links between units and abandoning the traditional mode of independent module operation and passive error correction. By using layered light field fitting to partition and process multiple types of illumination interference, it replaces globally unified image enhancement / hardware filtering, avoiding image distortion and noise amplification while fully preserving the optical characteristics of UAV targets, significantly reducing the probability of missed and false detections of targets under complex lighting conditions.

Description

Technical Field

[0001] This invention belongs to the field of directed energy optoelectronic countermeasures, optoelectronic detection and precision tracking technology, specifically a target extraction and biasing method suitable for complex lighting backgrounds. Background Technology

[0002] Existing 10,000-watt vehicle-mounted and fixed laser anti-drone systems rely on photoelectric imaging to identify drones and then use servo tracking to complete laser strikes. These systems have extremely high requirements for target extraction accuracy and aiming angle control; even a slight deviation can cause the interception to miss the target. In field combat scenarios, lighting conditions are complex, including direct sunlight during the day, strong backlight, reflections from building surfaces, alternating light and shadow from clouds, and problems such as stray light diffusion, overexposure of local detectors, and continuous light and shadow drift in dawn and dusk.

[0003] The existing technology has the following drawbacks: First, the illumination processing, target recognition, and servo compensation modules are independent of each other and lack closed-loop linkage control; second, all of them use global filtering and global image enhancement to coarsely process illumination, which cannot distinguish between three types of interference: static background, gradual illumination, and instantaneous strong light; third, all bias correction models only calculate the inherent mechanical and lens errors, without taking illumination intensity and gradient as error inputs; fourth, the entire recognition and targeting link is prone to failure in extreme conditions such as strong backlight, UAV swarms, and high-speed vehicle maneuvering, making it difficult to meet the requirements of all-weather, high-precision anti-drone warfare. Summary of the Invention

[0004] The purpose of this invention is to provide a target extraction and biasing method suitable for complex lighting backgrounds, so as to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a target extraction and biasing method suitable for complex lighting backgrounds, the method being executed by a target extraction and biasing system; the system includes a photoelectric imaging detection unit, an illumination layer modeling unit, a multi-feature fusion target extraction unit, an illumination-bias error calculation unit, a servo tracking and aiming drive unit, and a global closed-loop feedback unit;

[0006] The photoelectric imaging detection unit simultaneously acquires light intensity images and polarization images;

[0007] The lighting layering modeling unit constructs a layered light field model that includes a static background, gradient lighting, and instantaneous intense light.

[0008] The multi-feature fusion target extraction unit combines illumination partition data to fuse multiple features to identify targets and dynamically adjusts the polarization channel gain.

[0009] The illumination-bias error calculation unit has a built-in illumination-bias error dynamic mapping model, which uses illumination parameters as model input to calculate the comprehensive bias compensation amount.

[0010] The global closed-loop feedback unit iteratively optimizes the model parameters frame by frame.

[0011] As a further preferred embodiment of this technical solution, the method includes the following data stream identifier processing flow:

[0012] S1. Image Synchronous Acquisition: The photoelectric imaging detection unit synchronously acquires light intensity grayscale images and four-channel polarization images, completes frame synchronization and pixel-level registration, and generates the S1 raw composite image data stream; the photoelectric imaging detection unit includes a broadband imaging camera arranged with the same optical axis and field of view, a time-division four-channel polarization imaging detector, and a front-end optical filter assembly; the front-end optical filter assembly consists of a visible light narrowband filter + an adjustable linear polarizing mirror, which only filters out ultraviolet and infrared stray light, and completely preserves the original lighting information and target optical characteristics on site;

[0013] S2, Illumination layer modeling: The illumination layer modeling unit performs 3×3 Gaussian filtering to denoise the original composite image, divides the illumination region into three types based on the illumination change amplitude of three consecutive frames, and constructs a non-uniform surface light field model for each region using a quadratic surface model, outputting S2 illumination partition marking data and S3 layer light field model parameters.

[0014] S3. Multi-feature fusion target extraction and sub-pixel centroid calculation: The multi-feature fusion target extraction unit dynamically adjusts the polarization channel gain and feature weights based on the S2 illumination partition markings, fuses light intensity features, polarization features, and temporal motion features to complete joint target identification and eliminate illumination false targets, generating S4 multi-dimensional target feature data; combined with the S3 layered light field model parameters, it corrects the target contour distortion caused by illumination, uses a bilinear interpolation algorithm to solve the sub-pixel level target centroid coordinates, and outputs S5 target region and centroid coordinate data;

[0015] S4. Illumination-related bias error calculation: The illumination-bias error calculation unit reads the S3 layered light field model parameters and the S5 target centroid coordinates, superimposes the lens inherent distortion parameters and mechanical installation deviation parameters stored in the equipment, calls the pre-calibrated illumination-bias error dynamic mapping model, and uses the light intensity, light position, and light gradient as input variables. It calculates the S6 illumination-derived dynamic error through table lookup and linear interpolation, and obtains the S7 comprehensive bias compensation amount by vector superposition of inherent geometric errors.

[0016] S5, Servo Tracking and Aiming Correction and Laser Emission: The servo tracking drive unit receives the S7 comprehensive offset compensation amount and converts it into azimuth and pitch axis control commands. It completes axis angle offset correction through dual closed-loop control of position loop + velocity loop. After alignment, it triggers the matching 10,000-watt laser emission unit to emit laser to destroy the target. The servo unit collects the axis angle position in real time and generates S8 servo position feedback data.

[0017] S6. Global Closed-Loop Iterative Optimization: The global closed-loop feedback unit diverts the S8 servo position feedback data and sends it back to the illumination layer modeling unit and the illumination-bias error calculation unit. The light field fitting coefficient and the illumination-bias error model coefficient are slightly corrected frame by frame, with a single adjustment amplitude of ≤2%, suppressing the accumulation of bias error frame by frame.

[0018] As a further preferred embodiment of this technical solution, the criteria for classifying the three types of lighting areas in S2 are as follows: inter-frame illuminance variation < 5% is judged as a static background area; inter-frame illuminance variation 5%~30% is judged as a gradual lighting area; and inter-frame illuminance variation > 30% is judged as a transient strong light / reflection clutter area.

[0019] As a further preferred embodiment of this technical solution: the rules for dynamically adjusting the polarization channel gain and feature weight in S3 are as follows: in the static background area and the gradual illumination area, the light intensity feature weight is 70% and the polarization feature weight is 30%; in the instantaneous strong light and the global overexposure area, the light intensity feature is turned off and the light intensity feature weight is 0, and the target recognition and centroid solution are completed entirely by relying on the polarization feature; the time-division four-channel polarization imaging detector synchronously outputs four sets of polarization images at 0°, 45°, 90° and 135°, which are used to extract the polarization state difference between the target and the natural background and to distinguish the false targets formed by the reflection of the UAV and the ground and building mirrors.

[0020] As a further preferred embodiment of this technical solution: the entire method runs on an FPGA+ARM heterogeneous embedded processing platform. The FPGA is responsible for image preprocessing, light field surface fitting, and multi-feature extraction calculations, while the ARM is responsible for illumination-bias error model calculation, system logic scheduling, and bus communication interaction. In standard mode, the total delay of a single frame for servo correction from S1 to S5 is ≤15ms. The servo tracking drive unit is a high-precision two-axis servo turntable with an angular positioning accuracy of ≤2″, a system response bandwidth of ≥80Hz, and a single bias angle correction time of ≤10ms. The servo turntable, wide-spectrum camera, polarization imaging detector, and 10,000-watt laser emission unit are coaxially calibrated at the factory.

[0021] As a further preferred embodiment of this technical solution, the calibration method of the dynamic mapping model of illumination-bias error is as follows: multiple working conditions such as weak light, normal light, strong light, and all-angle backlight are simulated in the laboratory, and multiple sets of illumination parameters, pixel offset and angle deviation sample data are collected. The three-dimensional mapping model is fitted and fixed to the local storage of the device as the basis for dynamic compensation calculation; the inherent geometric error is calibrated and stored in one go under standard uniform illumination environment.

[0022] As a further preferred embodiment of this technical solution, the method includes four automatically switchable differentiated operating modes: standard mode, backlight enhancement mode, lightweight real-time mode, and multi-target mode.

[0023] The standard mode is the default mode for typical complex lighting conditions in the field, enabling complete quadratic surface light field fitting, three-feature fusion recognition, and standard lighting-bias error mapping model.

[0024] The backlight enhancement mode automatically switches when a sudden change in image illuminance > 60% is detected, completely shuts down light intensity features, and calls a dedicated error coefficient library for strong backlight to replace the conventional mapping model.

[0025] The lightweight real-time mode is adapted to the high-speed target working conditions of vehicle-mounted mobility. It replaces the non-uniform quadratic surface fitting with gridded reduced-order fitting, simplifies the features to light intensity + motion dual features, adds first-order predictive compensation, and compresses the total processing latency of a single frame to ≤8ms.

[0026] The multi-target mode automatically switches when ≥2 targets are detected within the field of view. An instance segmentation algorithm is added to split up clustered targets. For each independent target, the corresponding regional illumination parameters are matched and the offset compensation is calculated independently. The servo unit uses time-division aiming logic to strike targets sequentially according to threat level.

[0027] As a further preferred embodiment of this technical solution: the backlight enhancement mode is a sub-scheme processing logic under extreme backlight / direct sunlight. When the image is overexposed in a large area and the light intensity characteristics are completely ineffective, the true sub-pixel centroid of the target is solved only by the polarization distribution after partition adjustment, and the compensation amount is calculated separately using a backlight-specific error model.

[0028] As a further preferred option of this technical solution: the lightweight real-time mode is adapted to the computing power constraints of the vehicle-mounted mobile platform, equipment vibration, and high-speed target flight scenarios. The gridded reduced-order fitting greatly reduces the computational overhead. The first-order predictive compensation, combined with the illumination change rate and target motion speed, outputs the compensation amount in advance, shortening the tracking response delay.

[0029] As a further preferred embodiment of this technical solution: after the equipment is deployed in the field, only one optical axis coaxial re-inspection needs to be completed. The illumination partitioning judgment rules, feature weight ratios, and illumination-bias error model parameters are all adaptively and iteratively updated by the global closed-loop feedback unit.

[0030] Compared with the prior art, the beneficial effects of the present invention are:

[0031] 1. In this invention, an integrated linkage architecture for illumination modeling, target extraction, and bias compensation is constructed, and the data interaction links between each unit are opened up, abandoning the traditional mode of independent operation of separate modules and only passive error correction; by using hierarchical light field fitting to process multiple types of illumination interference in a partitioned manner, replacing the global unified image enhancement / hardware filtering, it not only avoids the problems of image distortion and noise amplification, but also completely preserves the optical characteristics of UAV targets, significantly reducing the probability of missed detection and false detection of targets under complex illumination.

[0032] 2. In this invention, by incorporating illumination parameters into the bias error mapping model, the compensation amount is no longer calculated solely based on mechanical and lens inherent errors. This allows for real-time quantification of dynamic pixel shifts caused by strong light and backlight. Combined with a polarization partition gain adjustment strategy, the sub-pixel centroid is independently calculated based on polarization information under conditions of large-area overexposure of the image, significantly reducing residual laser aiming errors and suppressing drift accumulation caused by long-term tracking.

[0033] 3. This invention incorporates four adaptive operating modes to adapt differentiated algorithm logic to various scenarios such as strong backlight, high-speed vehicle-mounted systems, and drone swarms. It also features lightweight fitting and instance segmentation optimization methods, which can meet the stable operation requirements under extremely harsh lighting conditions and adapt to the computing power limitations of vehicle-mounted platforms, thus greatly expanding the application scope of laser anti-drone equipment in all-weather and all-scenario operations. Attached Figure Description

[0034] Figure 1 This is a block diagram of the overall architecture of a target extraction and biasing method applicable to complex lighting backgrounds according to the present invention;

[0035] Figure 2 This invention provides a closed-loop data flow and action flowchart for image data processing, target extraction, and dynamic biasing, applicable to target extraction and biasing methods under complex lighting conditions. Detailed Implementation

[0036] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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.

[0037] In the description of this invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0038] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0039] Please see Figures 1-2 As shown, the present invention provides a technical solution: a target extraction and biasing method suitable for complex lighting backgrounds.

[0040] The method described in this invention relies on a heterogeneous embedded FPGA+ARM platform for real-time computation. The hardware carrier includes a photoelectric imaging detection unit, an illumination layer modeling unit, a multi-feature fusion target extraction unit, an illumination-bias error calculation unit, a servo tracking and aiming drive unit, a global closed-loop feedback unit, and a 10,000-watt laser emission unit. The entire device continuously executes the six-step closed-loop processing flow of data stream identifiers S1 to S8 at a constant frame rate of 50fps.

[0041] Example 1: Typical Complex Lighting Conditions in the Field: Standard Mode

[0042] Hardware configuration: 1280×1024 resolution, 50fps wide-spectrum CMOS camera with a response band of 400~700nm; time-division four-channel polarization imaging detector, frame-synchronized and pixel-level registered with the wide-spectrum camera; front-end configuration includes a visible light narrowband filter and an adjustable linear polarizing mirror; two-axis servo turntable with angular positioning accuracy ≤2″ and response bandwidth ≥80Hz; equipped with a 10,000-watt fiber laser, with factory-calibrated optical axis coaxiality.

[0043] S1. The photoelectric imaging detection unit simultaneously acquires light intensity grayscale images and four-channel polarization images at 0°, 45°, 90°, and 135°, and generates the S1 original composite image data stream after pixel registration is completed.

[0044] S2. The illumination layering modeling unit uses a 3×3 Gaussian filter to remove salt-and-pepper noise. Based on the illumination changes of three consecutive frames, it divides the illumination regions into three types. It constructs a quadratic surface non-uniform light field model for each region and outputs the S2 illumination partition marker and S3 layered light field model parameters.

[0045] S3. The light intensity feature weight of the static background area and the gradient lighting area is 70%, and the polarization feature weight is 30%. The light intensity contour, polarization state difference, and inter-frame motion features are fused to remove false targets such as specular reflection and cloud spotlights, generating S4 multi-dimensional target feature data. Combined with the S3 layered light field parameters, the lighting distortion is corrected, and the sub-pixel centroid with a precision of 0.1 pixels is obtained through bilinear interpolation. The target area and centroid coordinate data are output in S5.

[0046] S4. The illumination-bias error calculation unit reads the S3 layered light field parameters, S5 target centroid, and factory-stored geometric errors, calls the three-dimensional illumination-bias mapping model to look up the table and interpolate to calculate the S6 illumination-derived dynamic error, and vector superimposes the mechanical and lens inherent errors to obtain the S7 comprehensive bias compensation amount.

[0047] S5. The servo tracking and aiming drive unit receives the S7 comprehensive offset compensation amount and completes the azimuth and pitch axis correction within 10ms through the position loop + velocity loop dual closed-loop control. After aiming at the target, it triggers the 10,000-watt laser to emit damage. The servo unit collects the axis angular position in real time to generate S8 servo position feedback data.

[0048] S6. The global closed-loop feedback unit diverts the S8 servo position feedback data and sends it back to the illumination layer modeling unit and the illumination-bias error calculation unit. The light field fitting coefficient and the illumination-bias error model coefficient are slightly corrected every frame, with a single adjustment amplitude of ≤2%, continuously suppressing long-term tracking drift.

[0049] Performance results: The total latency for a single frame is 12ms, the target extraction accuracy under complex gradient lighting is improved by more than 80%, and the aiming offset error is reduced by an order of magnitude.

[0050] Example 2: Extreme backlighting conditions / direct sunlight: Backlight enhancement mode

[0051] The system automatically switches to this mode when the sudden change in global image illuminance exceeds 60% in real time. The hardware structure remains unchanged and the S1~S8 data flow links are retained; only the algorithm logic is adjusted.

[0052] The S2 illumination layering modeling unit enhances the accuracy of instantaneous strong light and overexposed pixel recognition, accurately segments bright backlight areas, and outputs S2 and S3 data streams normally.

[0053] The S3 target extraction unit resets the light intensity feature weight to 0, solves the sub-pixel centroid solely based on polarization distribution, and outputs S4 and S5 data streams.

[0054] The S4 error calculation unit replaces the conventional mapping model, calls the strong backlight-specific coefficient library to calculate the S6 illumination-derived error, and outputs the S7 comprehensive offset compensation amount adapted to backlight scenes.

[0055] The S5 and S6 processes remain unchanged, continuously generating S8 servo position feedback data and completing closed-loop iteration.

[0056] Applicable scenarios: Extreme combat environments such as direct midday sun, strong backlighting on the front line, large areas of overexposure in the image, and failure of conventional light intensity recognition algorithms, to stably output the true centroid of the target and avoid laser strikes missing the target.

[0057] Example 3: Vehicle-mounted mobility + high-speed UAV operating condition: lightweight real-time mode

[0058] The system automatically switches to lightweight mode to adapt to scenarios with limited onboard computing power, equipment vibration, and high-speed target flight, while retaining the S1~S8 data stream links and simplifying the computing logic.

[0059] S2 replaces quadratic surface fitting with meshed reduced-order fitting, reducing computational load and outputting S2 and S3 data streams normally.

[0060] S3 feature fusion is simplified to light intensity + temporal motion dual features, simplifying the logic of generating S4 multi-dimensional target feature data;

[0061] S4 adds first-order predictive compensation, which combines the rate of change of illumination and the target velocity to pre-generate S6 and S7 compensation data;

[0062] The S5 and S6 processes remain unchanged, while continuously outputting S8 servo position feedback data; the total delay of single-frame processing + servo correction is compressed to 7ms after optimization, meeting the real-time requirements of high-speed tracking.

[0063] Example 4: Drone Swarm + Complex Lighting Conditions: Multi-Target Mode

[0064] When two or more drones are detected clustered together within the field of view, the multi-target mode is automatically activated while retaining the basic data streams S1 to S8, and multi-target parallel processing logic is added:

[0065] S3 adds an instance segmentation algorithm to split together sticky targets and outputs multiple sets of S4 and S5 target data streams in parallel.

[0066] S4 matches the corresponding S2 and S3 illumination parameters for each independent target and independently calculates multiple sets of S6 and S7 offset compensation amounts.

[0067] The S5 servo unit uses time-division aiming logic, reads the S7 compensation values ​​of each group according to the threat level to complete the correction strike, and outputs the S8 servo position feedback data stream in a unified manner.

[0068] S6 independently iterates and updates the model coefficients corresponding to each target, eliminating aiming deviations caused by global uniform compensation.

[0069] Working principle or structural principle: First, simultaneously acquire two coaxial images of light intensity and polarization to fully preserve the original lighting and target optical characteristics, avoiding the loss of effective information caused by excessive hardware filtering; then, distinguish three types of areas based on the changes in image illuminance: static background, gradual lighting, and instantaneous strong light, and establish a non-uniform quadratic surface light field model for each area to classify and handle different lighting interferences, abandoning the coarse mode of global uniform processing.

[0070] By dynamically adjusting the weights of light and polarization features based on the partitioned illumination data, the two features are used to identify targets in conventional scenes. In strong light overexposure scenes, the target outline is analyzed by relying solely on polarization information. Then, motion features are combined to eliminate reflective and spot-like false targets. Combined with light field distortion correction, high-precision sub-pixel centroid positioning is achieved, solving the problems of target omission and positioning offset under complex illumination.

[0071] Illumination parameters are introduced into the offset calculation. The pixel and angle offset caused by illumination are calculated through the pre-calibrated three-dimensional mapping relationship. The final compensation amount is obtained by superimposing the inherent mechanical and lens errors of the device, which makes up for the shortcomings of the existing compensation model that ignores the interference of illumination. The servo mechanism receives the compensation command, completes the angle correction and triggers the laser strike, and outputs position feedback data synchronously.

[0072] Feedback information flows back to the two core modules of illumination modeling and error calculation, updating the model coefficients frame by frame to continuously suppress error drift caused by long-term tracking. At the same time, the system has four built-in adaptive operation logics that can automatically adapt to four working conditions: normal field, strong backlight, vehicle-mounted high speed, and drone swarm, and adjust the fitting algorithm, feature combination, error database and multi-target control strategy respectively.

[0073] The system forms a complete link encompassing image acquisition, layered illumination modeling, multi-feature target recognition, illumination-related offset compensation, servo tracking and aiming strike, and closed-loop adaptive optimization. It simultaneously addresses two major issues—target extraction failure caused by illumination and aiming offset deviation—from the source of interference, significantly improving the recognition accuracy and laser strike stability of UAVs under various lighting conditions, and adapting to the all-weather combat needs of 10,000-watt laser anti-UAV operations.

[0074] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0075] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

[0076] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A target extraction and biasing method suitable for complex lighting backgrounds, characterized in that: This method relies on a target extraction and biasing system; the system includes a photoelectric imaging detection unit, an illumination layer modeling unit, a multi-feature fusion target extraction unit, an illumination-bias error calculation unit, a servo tracking and aiming drive unit, and a global closed-loop feedback unit. The photoelectric imaging detection unit simultaneously acquires light intensity images and polarization images; The lighting layering modeling unit constructs a layered light field model that includes a static background, gradient lighting, and instantaneous intense light. The multi-feature fusion target extraction unit combines illumination partition data to fuse multiple features to identify targets and dynamically adjusts the polarization channel gain. The illumination-bias error calculation unit has a built-in illumination-bias error dynamic mapping model, which uses illumination parameters as model input to calculate the comprehensive bias compensation amount. The global closed-loop feedback unit iteratively optimizes the model parameters frame by frame.

2. The target extraction and biasing method suitable for complex lighting backgrounds according to claim 1, characterized in that: The method includes the following data stream identifier processing flow: S1. Image synchronous acquisition: The photoelectric imaging detection unit synchronously acquires light intensity grayscale images and four-channel polarization images, completes frame synchronization and pixel-level registration, and generates the S1 original composite image data stream. The photoelectric imaging detection unit includes a broadband imaging camera arranged with a common optical axis and common field of view, a time-division four-channel polarization imaging detector, and a front-end optical filter assembly. The front-end optical filter assembly consists of a visible light narrowband filter and an adjustable linear polarizing mirror, which only filters out ultraviolet and infrared stray light, and completely preserves the original lighting information and target optical characteristics of the scene. S2, Illumination layer modeling: The illumination layer modeling unit performs 3×3 Gaussian filtering to denoise the original composite image, divides the illumination region into three types based on the illumination change amplitude of three consecutive frames, and constructs a non-uniform surface light field model for each region using a quadratic surface model, outputting S2 illumination partition marking data and S3 layer light field model parameters. S3. Multi-feature fusion target extraction and sub-pixel centroid calculation: The multi-feature fusion target extraction unit dynamically adjusts the polarization channel gain and feature weights based on the S2 illumination partition markings, fuses light intensity features, polarization features, and temporal motion features to complete joint target identification and eliminate illumination false targets, generating S4 multi-dimensional target feature data; combined with the S3 layered light field model parameters, it corrects the target contour distortion caused by illumination, uses a bilinear interpolation algorithm to solve the sub-pixel level target centroid coordinates, and outputs S5 target region and centroid coordinate data; S4. Illumination-related bias error calculation: The illumination-bias error calculation unit reads the S3 layered light field model parameters and the S5 target centroid coordinates, superimposes the lens inherent distortion parameters and mechanical installation deviation parameters stored in the equipment, calls the pre-calibrated illumination-bias error dynamic mapping model, and uses the light intensity, light position, and light gradient as input variables. It calculates the S6 illumination-derived dynamic error through table lookup and linear interpolation, and obtains the S7 comprehensive bias compensation amount by vector superposition of inherent geometric errors. S5, Servo Tracking and Aiming Correction and Laser Emission: The servo tracking drive unit receives the S7 comprehensive offset compensation amount and converts it into azimuth and pitch axis control commands. It completes axis angle offset correction through dual closed-loop control of position loop + velocity loop. After alignment, it triggers the matching 10,000-watt laser emission unit to emit laser to destroy the target. The servo unit collects the axis angle position in real time and generates S8 servo position feedback data. S6. Global Closed-Loop Iterative Optimization: The global closed-loop feedback unit diverts the S8 servo position feedback data and sends it back to the illumination layer modeling unit and the illumination-bias error calculation unit. The light field fitting coefficient and the illumination-bias error model coefficient are slightly corrected frame by frame, with a single adjustment amplitude of ≤2%, suppressing the accumulation of bias error frame by frame.

3. The target extraction and biasing method suitable for complex lighting backgrounds according to claim 2, characterized in that: The criteria for classifying the three types of lighting areas in S2 are as follows: inter-frame illuminance variation < 5% is classified as a static background area; inter-frame illuminance variation 5% to 30% is classified as a gradual lighting area; and inter-frame illuminance variation > 30% is classified as a transient strong light / reflection clutter area.

4. The target extraction and biasing method suitable for complex lighting backgrounds according to claim 2, characterized in that: The rules for dynamically adjusting the polarization channel gain and feature weight in S3 are as follows: in static background areas and gradually changing illumination areas, the light intensity feature weight is 70% and the polarization feature weight is 30%; in instantaneous strong light and overexposed areas, the light intensity feature is turned off and the light intensity feature weight is 0, relying entirely on the polarization feature to complete target recognition and centroid solving; the time-division four-channel polarization imaging detector simultaneously outputs four sets of polarization images at 0°, 45°, 90°, and 135°, which are used to extract the polarization state difference between the target and the natural background, and to distinguish between drones and illumination false targets formed by ground and building mirror reflections.

5. The target extraction and biasing method suitable for complex lighting backgrounds according to claim 2, characterized in that: The entire method runs on an FPGA+ARM heterogeneous embedded processing platform. The FPGA is responsible for image preprocessing, light field surface fitting, and multi-feature extraction calculations, while the ARM is responsible for solving the illumination-bias error model, system logic scheduling, and bus communication interaction. In standard mode, the total delay of a single frame for servo correction from S1 to S5 is ≤15ms. The servo tracking drive unit is a high-precision two-axis servo turntable with an angular positioning accuracy of ≤2″, a system response bandwidth of ≥80Hz, and a single bias angle correction time of ≤10ms. The servo turntable, broadband camera, polarization imaging detector, and 10,000-watt laser emission unit are coaxially calibrated at the factory.

6. The target extraction and biasing method suitable for complex lighting backgrounds according to claim 2, characterized in that: The calibration method for the dynamic mapping model of illumination-bias error is as follows: multiple working conditions such as weak light, normal light, strong light, and all-angle backlight are simulated in the laboratory, and multiple sets of illumination parameters, pixel offset and angle deviation sample data are collected. The three-dimensional mapping model is fitted and fixed to the local storage of the device as the basis for dynamic compensation calculation; the inherent geometric error is calibrated and stored in a one-time manner under standard uniform illumination environment.

7. The target extraction and biasing method suitable for complex lighting backgrounds according to claim 1, characterized in that: The method includes four automatically switchable differentiated operating modes: standard mode, backlight enhancement mode, lightweight real-time mode, and multi-target mode; The standard mode is the default mode for typical complex lighting conditions in the field, enabling complete quadratic surface light field fitting, three-feature fusion recognition, and standard lighting-bias error mapping model. The backlight enhancement mode automatically switches when a sudden change in image illuminance > 60% is detected, completely shuts down light intensity features, and calls a dedicated error coefficient library for strong backlight to replace the conventional mapping model. The lightweight real-time mode is adapted to the high-speed target working conditions of vehicle-mounted mobility. It replaces the non-uniform quadratic surface fitting with gridded reduced-order fitting, simplifies the features to light intensity + motion dual features, adds first-order predictive compensation, and compresses the total processing latency of a single frame to ≤8ms. The multi-target mode automatically switches when ≥2 targets are detected within the field of view. An instance segmentation algorithm is added to split up clustered targets. For each independent target, the corresponding regional illumination parameters are matched and the offset compensation is calculated independently. The servo unit uses time-division aiming logic to strike targets sequentially according to threat level.

8. The target extraction and biasing method suitable for complex lighting backgrounds according to claim 7, characterized in that: The backlight enhancement mode is a sub-scheme processing logic under extreme backlight / direct sunlight. When the image is overexposed in large areas and the light intensity features are completely lost, it only relies on the polarization distribution after partition adjustment to solve the true sub-pixel centroid of the target, and uses a backlight-specific error model to calculate the compensation amount.

9. A target extraction and biasing method suitable for complex lighting backgrounds according to claim 7, characterized in that: The lightweight real-time mode is adapted to scenarios with computing power constraints on vehicle-mounted mobile platforms, equipment vibration, and high-speed target flight. The gridded reduced-order fitting significantly reduces computational overhead. The first-order predictive compensation, combined with the rate of change of illumination and the target's motion speed, outputs the compensation amount in advance, shortening the tracking response latency.

10. A target extraction and biasing method suitable for complex lighting backgrounds according to claim 2, characterized in that: After the equipment is deployed in the field, only one optical axis coaxial re-inspection needs to be completed. The illumination zoning judgment rules, feature weight ratios, and illumination-bias error model parameters are all adaptively and iteratively updated by the global closed-loop feedback unit.

Images