Intelligent photoelectric management system and method based on big data

Abstract

The invention relates to the field of photoelectric management, in particular to an intelligent photoelectric management system and method based on big data, and the system comprises a target detection module, an image processing module, a mechanical tracking module, a disturbance compensation module and a visual processing module. The image processing module is used for suppressing a sub-peak value and carrying out boundary truncation, the mechanical tracking module is used for continuously tracking a target, the disturbance compensation module is used for compensating disturbance of a target point, and the visual processing module is used for outputting a non-delay motion equation. The performance of the photoelectric tracking system in executing a tracking task is improved, the mechanical disturbance effect is effectively inhibited, high-precision tracking control over a target object is achieved, the self-adaptive capacity of the photoelectric tracking system is improved, and the stability of target tracking and the accuracy of motion data analysis are ensured.

Description

Technical Field

[0001] This invention relates to the field of optoelectronic management, specifically to an intelligent optoelectronic management system and method based on big data. Background Technology

[0002] An optoelectronic tracking system is a system that uses a photodetector to capture light signals, convert them into electrical signals, and analyzes these signals to detect, identify, lock onto, and continuously track moving targets. A typical optoelectronic tracking system includes a camera, a photodetector, and a servo control system. During operation, the camera analyzes the captured images, identifies and detects the target, locks onto the target, and continuously tracks it to determine its trajectory.

[0003] As a high-precision line-of-sight system, the photoelectric tracking system needs to be mounted on a carrier platform. It is easily affected by factors such as carrier vibration, wind resistance disturbance and friction torque disturbance, which can cause the camera's line of sight to shake, reduce image quality, and cause problems such as poor positioning accuracy and target loss. It is necessary to stabilize the image center by performing disturbance compensation through linkage control with the controller. However, there is a delay in the algorithm execution process, which may cause the target miss amount to lag.

[0004] In addition to mechanical interference, photoelectric tracking systems also face optical interference during scanning and imaging processes in different application scenarios, such as interference from similar targets, background interference, camera blur, and poor lighting. When the target being tracked is in high-speed motion, the photoelectric system may fail to track the target, thus limiting its application effectiveness. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent optoelectronic management system and method based on big data to solve the problems mentioned in the background art.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: an intelligent optoelectronic management system based on big data, comprising: a target detection module, an image processing module, a mechanical tracking module, a disturbance compensation module, and a vision processing module;

[0007] The target detection module is used to input the tracking target features into the optoelectronic device. The optoelectronic device acquires real-time image frames, processes the image frames through convolution and cross-correlation, calculates the similarity between each pixel region and the target features, generates a tracking response map, uses an image segmentation algorithm to separate peak point cluster regions, locks the target position when the average similarity of the regions is higher than the threshold, and observes the changes in the cluster regions in the next frame to locate the target's movement direction.

[0008] The image processing module is used to process the tracking response map using quadratic interpolation and Gaussian fitting to achieve sub-pixel accuracy. It suppresses sub-peaks in the tracking response map and increases the proportion of the main peak by using the pixel position of the main peak as the target center coordinate. When the target is located in the image boundary region, the response value of the boundary region is truncated, and a penalty function is applied from the pixel point in the boundary region inward until the target leaves the image boundary to maintain the main peak coordinate feedback.

[0009] The mechanical tracking module is used to calculate the deviation between the target point coordinates and the image center when the response value truncation of the boundary area is triggered, generate an error signal, control the camera movement direction according to the target's movement direction and deviation, and use a PID controller to adjust the gimbal mounted on the optoelectronic device to adjust the line of sight until the tracked target is located at the center of the captured image.

[0010] The disturbance compensation module is used to construct a sliding mode controller based on the line-of-sight error and a fixed-time integral sliding mode surface to compensate for optical disturbances at the target point. It introduces a reaching law based on Euclidean distance into the sliding mode controller to ensure that the system state reaches the sliding mode surface within a fixed time. The target point coordinates and inertial measurement data are input into the sliding mode controller, and the optical disturbance quantity is output. The optical disturbance is treated as a bounded disturbance, and a PID compensation quantity is output. At the same time, the current target position is corrected through the output of the optical disturbance quantity. The target motion sample is processed using a time series model to fit the kinematic equation of the target point.

[0011] The vision processing module is used to describe the target motion using a state-space model based on the kinematic equation of the target point, establish a computer vision task model, expand the original state vector into an augmented state vector, describe the measurement delay using an augmented state observer, transform the current task system into a delay-free system, determine the number of iterations of the filtering prediction algorithm based on the measurement delay, predict the future state of the target, classify the target motion state, input the current motion state of the target into the delay-free system, and output the delay-free motion equation of the target point.

[0012] Furthermore, the target detection module includes: a feature recognition unit and a tracking response unit;

[0013] The feature recognition unit is used to process the image sequence captured by the camera, and uses a deep learning model to calculate the similarity of each pixel region in the image to detect the appearance of target features;

[0014] The tracking response unit is used to locate the target bounding box based on anchor point regression and generate a tracking response map according to the feature matching degree within the boundary.

[0015] Furthermore, the image processing module includes: a peak suppression unit and a motion localization unit;

[0016] The peak suppression unit is used to slide a window in the tracking response map, set all values ​​except the maximum value in the window to zero, and use Gaussian convolution to process the response map to suppress minor peaks in the response map.

[0017] The motion positioning unit is used to set a boundary threshold. When the peak value is located in the boundary region, the response value is multiplied by the boundary attenuation factor, and the global maximum peak point of the response map is used as the target coordinate.

[0018] Furthermore, the mechanical tracking module includes: a controller unit and a line-of-sight stabilization unit;

[0019] The controller unit is used to determine the motion control quantity of the gimbal or reflector by adaptively tuning the output parameters of the PID controller.

[0020] The line-of-sight stabilization unit is used to determine the line-of-sight stabilization error and adjust the control parameters to improve tracking accuracy so that the tracking target is located in the center of the image.

[0021] Furthermore, the disturbance compensation module includes: a sliding mode compensation unit, an optical observation unit, and a motion calculation unit;

[0022] The sliding mode compensation unit is used to construct an integral sliding surface, adjust the control flexibility through fractional calculus, and accelerate convergence using an exponential reaching law.

[0023] The optical observation unit is used to add an optical disturbance compensation term to the control law to adjust the current target position;

[0024] The motion measurement unit is used to establish the kinematic equation of the target in the image plane based on the target motion sample sequence.

[0025] Furthermore, the vision processing module includes: an augmentation and conversion unit, a filtering and prediction unit, and a state output unit;

[0026] The augmentation and transformation unit is used to establish a computer vision task model and transform the current optoelectronic system into a delay-free system by augmenting the state vector of the optoelectronic system.

[0027] The filtering prediction unit is used to predict target motion using Kalman filtering or particle filtering, and determines the number of prediction iterations based on the delay time and sampling period.

[0028] The state output unit is used to classify motion states using a machine learning classifier and to adjust the motion equation of the target point with time delay.

[0029] A smart optoelectronic management method based on big data includes the following steps:

[0030] Step S1. Input the tracking target features into the photoelectric device, acquire real-time image frames, calculate the similarity between each pixel region and the target features, generate a tracking response map, separate peak point cluster regions, and determine the target's location and direction of movement;

[0031] Step S2. Suppress the subpeaks in the tracking response map by using the subpeak pooling method, increase the proportion of the main peak, use the pixel position of the main peak as the target center coordinate, and truncate the response value of the boundary region when the target is located in the image boundary region, and output the processed tracking response map.

[0032] Step S3. When boundary truncation is triggered, the camera movement direction is controlled according to the target's movement direction and coordinate deviation, and the PID controller is used to adjust the line of sight of the photoelectric device to track the target;

[0033] Step S4. Construct a sliding mode controller based on the line-of-sight error, input the target point coordinates and inertial measurement data of the next frame into the sliding mode controller, output the optical disturbance quantity, treat the optical disturbance as a bounded disturbance, output the PID compensation quantity, and compensate for the optical disturbance of the target point.

[0034] Step S5. Use a state-space model to describe the target motion. By augmenting the state vector of the photoelectric system, the current photoelectric system is transformed into a delay-free system. The number of iterations of the sampling filter prediction algorithm is determined according to the delay time and sampling period. The target motion state is fitted and input into the delay-free system to obtain the delay-free motion equation of the target point.

[0035] Furthermore, step S1 includes:

[0036] Step S11. Load the feature image of the tracked target into the optoelectronic device, including: color histogram, HOG histogram and depth feature image. The optoelectronic device captures and acquires image sequences in real time and performs convolution and cross-correlation processing.

[0037] Step S12. Use a deep learning model to calculate the similarity of each pixel region in the image to the loaded feature image, and generate a tracking response map. The pixel value of each pixel in the tracking response map is equal to the similarity of the region to which the pixel belongs × the position parameter, where the position parameter represents the distance between the pixel and the center pixel of the region.

[0038] Step S13. Use an image segmentation algorithm to separate peak point cluster regions. When the average similarity of the cluster regions is higher than the threshold, trigger photoelectric device tracking, locate the target bounding box based on anchor point regression, lock the target position, and observe the changes in the cluster regions in the next frame to determine the target movement direction.

[0039] Furthermore, step S2 includes:

[0040] Step S21. The tracking response map is processed by quadratic interpolation and Gaussian fitting to achieve sub-pixel accuracy. The subpeak pooling window is slid in the clustered region of the tracking response map. Except for the maximum value in the window, other values ​​are set to zero. The response map is processed by Gaussian convolution, with the global maximum peak point as the main peak, suppressing the secondary peaks in the tracking response map, increasing the proportion of the main peak, and using the pixel position of the main peak as the target center coordinates.

[0041] Step S22. Set a boundary threshold. When the target center coordinates are located in the boundary region, set a penalty function from the boundary region pixels inward. Input the target center coordinates into the penalty function to obtain the boundary attenuation factor. Multiply the response value with the boundary attenuation factor to trigger boundary truncation and output the processed tracking response map.

[0042] Furthermore, step S3 includes:

[0043] Step S31. When the response value truncation of the boundary region is triggered, calculate the deviation between the target point coordinates and the image center, generate an error signal, and control the camera movement direction according to the target's movement direction and the deviation.

[0044] Step S32. By adaptively tuning the output parameters of the PID controller, determine the motion control quantity of the optoelectronic device's pan-tilt unit or reflector, and adjust the line of sight until the tracked target is located in the center of the captured image.

[0045] Furthermore, step S4 includes:

[0046] Step S41. Based on the line-of-sight error, a sliding mode controller is constructed using a fixed-time integral sliding mode surface to compensate for optical disturbances at the target point. The approach law based on Euclidean distance is introduced into the sliding mode controller to ensure that the system state reaches the sliding mode surface within a fixed time.

[0047] Step S42. Input the target point coordinates and inertial measurement data into the sliding mode controller, output the optical disturbance quantity, model the optical disturbance as a bounded uncertainty function, introduce the disturbance compensation term into the output parameters of the PID controller, determine the PID compensation quantity, and correct the current target position.

[0048] Furthermore, step S5 includes:

[0049] Step S51. Process the target motion samples using a time series model, fit the kinematic equation of the target point, describe the target motion using a state space model based on the kinematic equation of the target point, establish a computer vision task model, and transform the current task system into a delay-free system by expanding the original state vector into an augmented state vector and describing the measurement delay with an augmented state observer.

[0050] Step S52. Predict the target motion using Kalman filtering or particle filtering. Determine the number of prediction iterations based on the delay time and sampling period. Classify the motion state using a machine learning classifier based on the prediction results. Input the current motion state of the target into the delay-free system and output the delay-free motion equation of the target point.

[0051] Compared with the prior art, the beneficial effects achieved by the present invention are:

[0052] 1. This invention identifies the position of the target in an image by using an anchor-based target detection algorithm, generates a tracking response map, and uses subpeak pooling and boundary response truncation to suppress subpeaks in the tracking response map, locate the target point coordinates, enhance optical image contrast, filter environmental noise, dynamically adjust the tracking window, and improve the performance of the photoelectric tracking system when performing tracking tasks.

[0053] 2. This invention can adjust the camera and stabilize the line of sight by means of a fractional sliding mode compensator, introduce the sliding mode approach law with Euclidean distance into the fixed-time integral sliding mode surface, and compensate for the optical disturbance of the target point by adjusting the angle of the servo motor, thereby more effectively suppressing the mechanical disturbance effect, obtaining a clear and stable tracking image, solving the image rotation problem of photoelectric tracking system, and realizing high-precision tracking control of target objects.

[0054] 3. This invention can establish a computer vision task model, transform the current optoelectronic system into a delay-free augmented system by augmenting the state vector of the optoelectronic system, process the target motion samples using a big data model, and output the delay-free motion equation of the target point. The prediction algorithm avoids the interference of computing power delay on the tracking results, improves the adaptive capability and operational reliability of the optoelectronic tracking system, and ensures the stability of target tracking and the accuracy of motion data analysis. Attached Figure Description

[0055] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0056] Figure 1 This is a schematic diagram of the structure of an intelligent optoelectronic management system based on big data according to the present invention;

[0057] Figure 2 This is a schematic diagram illustrating the steps of an intelligent optoelectronic management method based on big data according to the present invention. Detailed Implementation

[0058] 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.

[0059] Please see Figure 1 The present invention provides a technical solution: an intelligent optoelectronic management system based on big data, comprising: a target detection module, an image processing module, a mechanical tracking module, a disturbance compensation module, and a vision processing module;

[0060] The target detection module is used to input the tracking target features into the optoelectronic device. The optoelectronic device acquires real-time image frames, processes the image frames through convolution and cross-correlation, calculates the similarity between each pixel region and the target features, generates a tracking response map, uses an image segmentation algorithm to separate peak point cluster regions, locks the target position when the average similarity of the regions is higher than the threshold, and observes the changes in the cluster regions in the next frame to locate the target's movement direction.

[0061] The target detection module includes: a feature recognition unit and a tracking response unit;

[0062] The feature recognition unit is used to process the image sequence captured by the camera, and uses a deep learning model to calculate the similarity of each pixel region in the image to detect the appearance of target features;

[0063] The tracking response unit is used to locate the target bounding box based on anchor point regression and generate a tracking response map according to the feature matching degree within the boundary.

[0064] The image processing module is used to process the tracking response map using quadratic interpolation and Gaussian fitting to achieve sub-pixel accuracy. It suppresses sub-peaks in the tracking response map and increases the proportion of the main peak by using the pixel position of the main peak as the target center coordinate. When the target is located in the image boundary region, the response value of the boundary region is truncated, and a penalty function is applied from the pixel point in the boundary region inward until the target leaves the image boundary to maintain the main peak coordinate feedback.

[0065] The image processing module includes: a peak suppression unit and a motion localization unit;

[0066] The peak suppression unit is used to slide a window in the tracking response map, set all values ​​except the maximum value in the window to zero, and use Gaussian convolution to process the response map to suppress minor peaks in the response map.

[0067] The motion positioning unit is used to set a boundary threshold. When the peak value is located in the boundary region, the response value is multiplied by the boundary attenuation factor, and the global maximum peak point of the response map is used as the target coordinate.

[0068] The mechanical tracking module is used to calculate the deviation between the target point coordinates and the image center when the response value truncation of the boundary area is triggered, generate an error signal, control the camera movement direction according to the target's movement direction and deviation, and use a PID controller to adjust the gimbal mounted on the optoelectronic device to adjust the line of sight until the tracked target is located at the center of the captured image.

[0069] The mechanical tracking module includes: a controller unit and a line-of-sight stabilization unit;

[0070] The controller unit is used to determine the motion control quantity of the gimbal or reflector by adaptively tuning the output parameters of the PID controller.

[0071] The line-of-sight stabilization unit is used to determine the line-of-sight stabilization error and adjust the control parameters to improve tracking accuracy so that the tracking target is located in the center of the image.

[0072] The disturbance compensation module is used to construct a sliding mode controller based on the line-of-sight error and a fixed-time integral sliding mode surface to compensate for optical disturbances at the target point. It introduces a reaching law based on Euclidean distance into the sliding mode controller to ensure that the system state reaches the sliding mode surface within a fixed time. The target point coordinates and inertial measurement data are input into the sliding mode controller, and the optical disturbance quantity is output. The optical disturbance is treated as a bounded disturbance, and a PID compensation quantity is output. At the same time, the current target position is corrected through the output of the optical disturbance quantity. The target motion sample is processed using a time series model to fit the kinematic equation of the target point.

[0073] The disturbance compensation module includes: a sliding mode compensation unit, an optical observation unit, and a motion calculation unit;

[0074] The sliding mode compensation unit is used to construct an integral sliding surface, adjust the control flexibility through fractional calculus, and accelerate convergence using an exponential reaching law.

[0075] The optical observation unit is used to add an optical disturbance compensation term to the control law to adjust the current target position;

[0076] The motion measurement unit is used to establish the kinematic equation of the target in the image plane based on the target motion sample sequence.

[0077] The vision processing module is used to describe the target motion using a state-space model based on the kinematic equation of the target point, establish a computer vision task model, expand the original state vector into an augmented state vector, describe the measurement delay using an augmented state observer, transform the current task system into a delay-free system, determine the number of iterations of the filtering prediction algorithm based on the measurement delay, predict the future state of the target, classify the target motion state, input the current motion state of the target into the delay-free system, and output the delay-free motion equation of the target point.

[0078] The vision processing module includes: an augmentation and conversion unit, a filtering and prediction unit, and a state output unit;

[0079] The augmentation and transformation unit is used to establish a computer vision task model and transform the current optoelectronic system into a delay-free system by augmenting the state vector of the optoelectronic system.

[0080] The filtering prediction unit is used to predict target motion using Kalman filtering or particle filtering, and determines the number of prediction iterations based on the delay time and sampling period.

[0081] The state output unit is used to classify motion states using a machine learning classifier and to adjust the motion equation of the target point with time delay.

[0082] like Figure 2 As shown, a smart optoelectronic management method based on big data includes the following steps:

[0083] Step S1. Input the tracking target features into the photoelectric device, acquire real-time image frames, calculate the similarity between each pixel region and the target features, generate a tracking response map, separate peak point cluster regions, and determine the target's location and direction of movement;

[0084] Step S1 includes:

[0085] Step S11. Load the feature image of the tracked target into the optoelectronic device, including: color histogram, HOG histogram and depth feature image. The optoelectronic device captures and acquires image sequences in real time and performs convolution and cross-correlation processing.

[0086] Step S12. Use a deep learning model to calculate the similarity of each pixel region in the image to the loaded feature image, and generate a tracking response map. The pixel value of each pixel in the tracking response map is equal to the similarity of the region to which the pixel belongs × the position parameter, where the position parameter represents the distance between the pixel and the center pixel of the region.

[0087] Step S13. Use an image segmentation algorithm to separate peak point cluster regions. When the average similarity of the cluster regions is higher than the threshold, trigger photoelectric device tracking, locate the target bounding box based on anchor point regression, lock the target position, and observe the changes in the cluster regions in the next frame to determine the target movement direction.

[0088] Step S2. Suppress the subpeaks in the tracking response map by using the subpeak pooling method, increase the proportion of the main peak, use the pixel position of the main peak as the target center coordinate, and truncate the response value of the boundary region when the target is located in the image boundary region, and output the processed tracking response map.

[0089] Step S2 includes:

[0090] Step S21. The tracking response map is processed by quadratic interpolation and Gaussian fitting to achieve sub-pixel accuracy. The subpeak pooling window is slid in the clustered region of the tracking response map. Except for the maximum value in the window, other values ​​are set to zero. The response map is processed by Gaussian convolution, with the global maximum peak point as the main peak, suppressing the secondary peaks in the tracking response map, increasing the proportion of the main peak, and using the pixel position of the main peak as the target center coordinates.

[0091] Step S22. Set a boundary threshold. When the target center coordinates are located in the boundary region, set a penalty function from the boundary region pixels inward. Input the target center coordinates into the penalty function to obtain the boundary attenuation factor. Multiply the response value with the boundary attenuation factor to trigger boundary truncation and output the processed tracking response map.

[0092] Step S3. When boundary truncation is triggered, the camera movement direction is controlled according to the target's movement direction and coordinate deviation, and the PID controller is used to adjust the line of sight of the photoelectric device to track the target;

[0093] Step S3 includes:

[0094] Step S31. When the response value truncation of the boundary region is triggered, calculate the deviation between the target point coordinates and the image center, generate an error signal, and control the camera movement direction according to the target's movement direction and the deviation.

[0095] Step S32. By adaptively tuning the output parameters of the PID controller, determine the motion control quantity of the optoelectronic device's pan-tilt unit or reflector, and adjust the line of sight until the tracked target is located in the center of the captured image.

[0096] Step S4. Construct a sliding mode controller based on the line-of-sight error, input the target point coordinates and inertial measurement data of the next frame into the sliding mode controller, output the optical disturbance quantity, treat the optical disturbance as a bounded disturbance, output the PID compensation quantity, and compensate for the optical disturbance of the target point.

[0097] Step S4 includes:

[0098] Step S41. Based on the line-of-sight error, a sliding mode controller is constructed using a fixed-time integral sliding mode surface to compensate for optical disturbances at the target point. The approach law based on Euclidean distance is introduced into the sliding mode controller to ensure that the system state reaches the sliding mode surface within a fixed time.

[0099] Step S42. Input the target point coordinates and inertial measurement data into the sliding mode controller, output the optical disturbance quantity, model the optical disturbance as a bounded uncertainty function, introduce the disturbance compensation term into the output parameters of the PID controller, determine the PID compensation quantity, and correct the current target position.

[0100] Step S5. Use a state-space model to describe the target motion. By augmenting the state vector of the photoelectric system, the current photoelectric system is transformed into a delay-free system. The number of iterations of the sampling filter prediction algorithm is determined according to the delay time and sampling period. The target motion state is fitted and input into the delay-free system to obtain the delay-free motion equation of the target point.

[0101] Step S5 includes:

[0102] Step S51. Process the target motion samples using a time series model, fit the kinematic equation of the target point, describe the target motion using a state space model based on the kinematic equation of the target point, establish a computer vision task model, and transform the current task system into a delay-free system by expanding the original state vector into an augmented state vector and describing the measurement delay with an augmented state observer.

[0103] Step S52. Predict the target motion using Kalman filtering or particle filtering. Determine the number of prediction iterations based on the delay time and sampling period. Classify the motion state using a machine learning classifier based on the prediction results. Input the current motion state of the target into the delay-free system and output the delay-free motion equation of the target point.

[0104] Example: The optoelectronic device captures an image of the surrounding environment, inputs the image into the target detection module, outputs the target point coordinates, calibrates the coordinates through a sliding mode controller, generates a control signal based on the boundary truncation state and the target point coordinates, inputs it into a PID controller to adjust the line of sight, and simultaneously performs motion prediction on the target point coordinate sequence. Through delay compensation, it outputs a delay-free motion equation, performs feedforward control based on the delay-free motion equation, and improves tracking accuracy.

[0105] 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.

[0106] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A big data based intelligent opto-electric management method, characterized in that, The method comprises the following steps: Step S1. Input the tracking target feature into the optoelectronic device, collect real-time image frames, calculate the similarity of each pixel region to the target feature, generate a tracking response map, separate the peak point cluster region, and determine the target region and moving direction; Step S2. Suppress sub-peak values in the tracking response map by sub-peak pooling method to improve the proportion of main peaks, use the pixel position of the main peak as the target center coordinate, and when the target is located in the boundary region of the image, cut off the response value of the boundary region, and output the processed tracking response map; Step S3. When the boundary cut-off is triggered, control the moving direction of the camera according to the moving direction and coordinate deviation of the target, and use the PID controller to adjust the boresight direction of the optoelectronic device to track the target; Step S4. Based on the boresight error, a sliding mode controller is constructed, the next frame target point coordinate and inertial measurement data are input into the sliding mode controller, and the optical disturbance is output as a bounded disturbance, and a PID compensation quantity is output to compensate for the optical disturbance of the target point; Step S5. Use a state space model to describe the target motion, convert the current optoelectronic system into a non-delay system by augmenting the state vector of the optoelectronic system, determine the iteration number of the sampling filter prediction algorithm according to the delay time and the sampling period, fit the target motion state, input the motion state into the non-delay system, and obtain the non-delay motion equation of the target point. 2.The big data-based intelligent photoelectric management method according to claim 1, characterized in that: Step S1 comprises: Step S11. Load the feature image of the tracking target into the optoelectronic device, including: color histogram image, HOG histogram image and depth feature image, the optoelectronic device real-time shoots and collects image sequences and performs convolution and cross-correlation processing; Step S12. Calculate the similarity of each pixel region in the image in the loaded feature image using a deep learning model to generate a tracking response map, wherein the pixel value of each pixel point in the tracking response map = the similarity of the region to which the pixel point belongs × position parameter, wherein the position parameter represents the distance between the pixel point and the center pixel point of the region; Step S13. Use an image segmentation algorithm to separate the peak point cluster region, when the average similarity of the cluster region is higher than the threshold value, trigger the optoelectronic device tracking, locate the target boundary box based on anchor point regression, lock the target position, and observe the change of the cluster region in the next frame to determine the target moving direction. 3.The big data based intelligent opto-electronic management method according to claim 2, characterized in that: Step S2 comprises: Step S21. Use quadratic interpolation and Gaussian fitting processing to achieve sub-pixel accuracy, slide the sub-peak pooling window in the cluster region of the tracking response map, set other values to zero except the maximum value in the window, use Gaussian convolution processing to respond to the map, use the global maximum peak point as the main peak, suppress the secondary peak in the tracking response map, and improve the proportion of the main peak, and use the pixel position of the main peak as the target center coordinate; Step S22. Set a boundary threshold, when the target center coordinate is located in the boundary region, set a penalty function inward from the boundary region pixel point, input the target center coordinate into the penalty function to obtain a boundary attenuation factor, multiply the response value by the boundary attenuation factor, trigger the boundary cut-off, and output the processed tracking response map. 4.The big data-based intelligent optoelectronic management method according to claim 3, characterized in that: Step S3 comprises: Step S31. When the response value truncation of the boundary region is triggered, the deviation of the target point coordinates from the image center is calculated, an error signal is generated, and the moving direction of the camera is controlled according to the moving direction of the target and the deviation; Step S32. The action control amount of the photoelectric device gimbal or mirror is determined by self-adaptively setting the output parameters of the PID controller, and the line-of-sight pointing is adjusted until the tracking target is located at the center of the shooting image; Step S4 includes: Step S41. A fixed-time integral sliding mode surface is used to construct a sliding mode controller according to the line-of-sight error, and the target point optical disturbance is compensated. A reaching law based on Euclidean distance is introduced into the sliding mode controller to ensure that the system state reaches the sliding mode surface within a fixed time; Step S42. The target point coordinates and inertial measurement data are input into the sliding mode controller, and the optical disturbance amount is output. The optical disturbance is modeled as a bounded uncertainty function, and a disturbance compensation term is introduced into the output parameters of the PID controller to determine the PID compensation amount and correct the current target position. 5.The big data based intelligent opto-electronic management method according to claim 4, characterized in that: Step S5 includes: Step S51. The target motion sample is processed using a time series model, and the kinematics equation of the target point is fitted. According to the kinematics equation of the target point, a state space model is used to describe the target motion, a computer vision task model is established, and the original state vector is expanded into an augmented state vector to describe the measurement delay with an augmented state observer. The current task system is converted into a non-delay system; Step S52. The target motion is predicted using Kalman filtering or particle filtering. The prediction iteration number is determined according to the delay time and the sampling period. The motion state is classified using a machine learning classifier according to the prediction result. The current motion state of the target is input into the non-delay system, and the non-delay motion equation of the target point is output.

6. A big data based intelligent opto-electric management system, characterized in that, The system includes the following modules: a target detection module, an image processing module, a mechanical tracking module, a disturbance compensation module, and a vision processing module; The target detection module is used to input the tracking target feature into the photoelectric device, and the photoelectric device collects real-time image frames. The image frames are processed by convolution and cross-correlation to calculate the similarity of each pixel region and the target feature, generate a tracking response map, and separate the peak point cluster region using an image segmentation algorithm. When the regional average similarity is higher than the threshold, the target position is locked, and the cluster region changes are observed in the next frame to locate the target moving direction; The image processing module is used to process the tracking response map using quadratic interpolation and Gaussian fitting to achieve sub-pixel accuracy. The sub-peak value in the tracking response map is suppressed by the sub-peak pooling method to improve the proportion of the main peak. The pixel position of the main peak is used as the target center coordinates. When the target is located in the boundary region of the image, the response value of the boundary region is truncated, and a penalty function is applied from the boundary region pixel point inward until the main peak coordinates are fed back before the target leaves the image boundary; The mechanical tracking module is used to calculate the deviation of the target point coordinates from the image center when the response value truncation of the boundary region is triggered, generate an error signal, control the moving direction of the camera according to the moving direction of the target and the deviation, adjust the line-of-sight pointing by using a PID controller to adjust the photoelectric device mounted gimbal, and adjust the line-of-sight pointing until the tracking target is located at the center of the shooting image. The disturbance compensation module is used for constructing a sliding mode controller by using a fixed time integral sliding surface according to the boresight error, compensating for optical disturbance of the target point, introducing an approaching law based on an Euclidean distance into the sliding mode controller to ensure that system states reach a sliding surface within a fixed time, inputting target point coordinates and inertial measurement data into the sliding mode controller, and outputting an optical disturbance quantity; the optical disturbance is taken as a bounded disturbance, and a PID compensation quantity is outputted; meanwhile, the optical disturbance quantity is used to correct a current target position; a time series model is used to process target motion samples to fit a kinematic equation of the target point; The vision processing module is used for describing target motion by using a state space model according to the kinematic equation of the target point, establishing a computer vision task model, extending an original state vector into an augmented state vector to describe measurement delay by using an augmented state observer, converting a current task system into a delay-free system, and determining an iteration number of a filtering prediction algorithm according to the measurement delay to predict a future state of the target; the target motion state is classified, and a current target motion state is inputted into the delay-free system to output a delay-free motion equation of the target point.

7. The intelligent opto-electronic management system based on big data according to claim 6, characterized in that: The target detection module comprises a feature recognition unit and a tracking response unit. The feature recognition unit is used for processing image sequences collected by a camera, calculating a similarity of each pixel region in the image by using a deep learning model, and detecting target appearance features. The tracking response unit is used for positioning a target bounding box based on anchor point regression, and generating a tracking response map according to feature matching degrees in the bounding box. The image processing module comprises a peak suppression unit and a motion positioning unit. The peak suppression unit is used for sliding a window in the tracking response map, setting other values in the window to zero except for a maximum value in the window, and processing the response map by using Gaussian convolution to suppress secondary peaks in the response map. The motion positioning unit is used for setting a boundary threshold, multiplying a response value by a boundary attenuation factor when a peak value is located in a boundary region, and taking a global maximum peak point of the response map as a target coordinate.

8. The intelligent opto-electronic management system based on big data according to claim 7, characterized in that: The mechanical tracking module comprises a controller unit and a boresight stabilization unit. The controller unit is used for determining a motion control quantity of a pan-tilt head or a mirror by outputting parameters of a self-adaptive PID controller. The boresight stabilization unit is used for determining a boresight stabilization error, adjusting control parameters to improve tracking precision, and making a tracking target located in a center of an image.

9. The intelligent opto-electronic management system based on big data according to claim 8, characterized in that: The disturbance compensation module comprises a sliding mode compensation unit, an optical observation unit, and a motion calculation unit. The sliding mode compensation unit is used for constructing an integral sliding surface, adjusting control flexibility by using fractional calculus, and accelerating convergence by using an exponential approaching law. The optical observation unit is used for adding an optical disturbance compensation term in a control law to adjust a current target position. The motion calculation unit is used for establishing a kinematic equation of a target in an image plane according to a target motion sample sequence.

10. The intelligent opto-electronic management system based on big data according to claim 9, characterized in that: The vision processing module comprises an augmented conversion unit, a filtering prediction unit, and a state output unit. The augmented conversion unit is used for establishing a computer vision task model, and converting a current optoelectronic system into a delay-free system by extending a state vector of the optoelectronic system. The filter prediction unit is used for target motion prediction by Kalman filtering or particle filtering, and the number of prediction iterations is determined according to a delay time and a sampling period. The state output unit is used for motion state classification by using a machine learning classifier, and time delay adjustment is performed on the target point motion equation.

Images