A method for automatic detection and tracking of sea surface targets using light-mining linkage

Abstract

The present invention relates to a method for detecting and tracking sea surface targets, and in particular to a method for automatically detecting and tracking sea surface targets in conjunction with a light mine. The method of the present invention comprises: a radar system detects a new target and continuously tracks it for a period of time, and then sends the target information to an optoelectronic system. The optoelectronic system then performs a first-stage system parameter adjustment (horizontal, vertical angle, focal length) to capture the target ship, and then uses a deep learning model to perform target detection to confirm whether the target ship is photographed. If the target ship is not detected, the target ship is searched for by traversing a preset maximum error annular search point diagram. If multiple ships are detected, the target ship is selected using a light mine information fusion candidate target determination algorithm. After the target ship is determined, it is continuously tracked with the help of a target tracking algorithm. During the tracking process, the horizontal and vertical angle adjustment values ​​and the focal length adjustment values ​​of the pan-tilt platform are calculated to perform a two-stage parameter adjustment to achieve self-tracking of the target by the optoelectronic system.

Description

Technical Field

[0001] The present invention relates to a method for detecting and tracking sea surface targets, and in particular to a method for automatically detecting and tracking sea surface targets in conjunction with light and mines. Background Art

[0002] Maritime transportation plays a vital role in social and economic life. Real-time, accurate detection of maritime targets is a crucial foundation for intelligent maritime surveillance systems and a prerequisite for diverse practical applications, such as ensuring efficient and safe navigation for vessels, monitoring critical offshore equipment, combating smuggling, illegal immigration, poaching, and other criminal activities, and protecting the marine environment. Research on precise maritime target detection technology has significant scientific value and practical significance. Existing maritime surveillance systems are mostly based on single sensors, such as radar or optoelectronics. Radar systems (referred to herein as pulse compression radar-based maritime target detection systems) have a relatively long detection range but can only produce binary classification decisions, determining whether a target is present or absent at a specific coordinate location. These systems lack detailed imaging information and therefore cannot accurately determine the target's size, type, or other characteristics. In contrast, optoelectronic systems can obtain high-resolution images of maritime targets. However, these systems are severely affected by rain and fog, have a short range, and exhibit poor nighttime imaging, severely limiting their application. If radar and optoelectronics can be effectively integrated and their advantages can be complemented, it will be possible to achieve high-precision sea surface target detection and reduce false alarms, thus enabling the sea surface surveillance system to truly move towards large-scale application. Summary of the Invention

[0003] The purpose of the present invention is to address the interactive processing requirements of sea surface target detection and tracking, and to propose a method for automatic detection and tracking of sea surface targets by combining light and mine linkage.

[0004] In order to achieve the above-mentioned objectives, the present invention proposes a method for automatic detection and tracking of sea surface targets in conjunction with light and mines. The method mainly includes four parts: radar system data analysis, optoelectronic system command reception, optoelectronic system target detection and tracking, and optoelectronic system parameter adjustment. After the radar system collects the radar echo signal, if the target is detected and can be continuously tracked for a period of time, the target information is sent to the optoelectronic system for further verification of the target. The optoelectronic system then first performs a first-stage optoelectronic system parameter adjustment to capture the target ship, and then performs target detection to confirm whether the target is successfully found. If no target is found, the preset maximum error annular search point diagram is traversed to search for the target ship. If the target ship is found, the tracker is initialized, and in subsequent frames, the second-stage optoelectronic system parameter adjustment is performed based on the tracking results to achieve self-tracking of the target ship by the optoelectronic system.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions.

[0006] The present invention proposes a method for automatically detecting and tracking sea surface targets by combining light and mine linkage, the method comprising:

[0007] The radar system detects and tracks sea surface targets using radar echo data collected by the radar system, and then obtains target information, which is then sent to the optoelectronic system.

[0008] Adjust the optoelectronic system parameters in one stage according to the target information and shoot the target;

[0009] Use a deep learning model to detect targets in images taken by the optoelectronic system, calculate the target detection score, and identify the target based on the detection score;

[0010] If the deep learning model does not detect the target, a maximum error circular search point graph is set up and each search point is traversed to search for the target;

[0011] After determining the target, the optoelectronic system uses a target tracking algorithm to track the target, and during the tracking process continuously adjusts the optoelectronic system parameters in two stages according to the captured target image, so as to realize automatic tracking of the target by the optoelectronic system.

[0012] As an improvement to the above technical solution, the method also includes: if there are multiple targets detected by the target detection network model or the targets searched for by traversing each search point, the target is determined by using the optical mine information fusion candidate target determination algorithm, specifically: the target ship is determined by comprehensively utilizing three aspects of information: target detection confidence, the offset of the ship from the center of the picture, and the consistency of the target detection frame and signal strength.

[0013] As an improvement of the above technical solution, the optoelectronic system includes a camera and a gimbal; wherein the camera is used to capture sea surface images, and the gimbal is used to support the camera and can freely adjust the horizontal and / or vertical angles to assist in tracking the target.

[0014] As an improvement to the above technical solution, the target information is obtained by detecting and tracking the sea surface target using the radar echo data collected by the radar system, specifically including:

[0015] After collecting radar echo data, the radar system performs matched filtering on the echo data and identifies sea surface targets by setting a detection threshold; the detection threshold is dynamically adjusted based on the clutter in the radar echo data using a constant false alarm rate detection algorithm;

[0016] After detecting the sea surface target, the sea surface target is tracked, and the target information is sent to the optoelectronic system when the target enters the detection range of the optoelectronic system; the target information includes: target number, target horizontal angle, target distance and the ranking of the radar echo signal strength detected by the target among the nearby angle targets.

[0017] As an improvement to the above technical solution, the one-stage adjustment of the optoelectronic system parameters according to the target information and the shooting of the target specifically includes:

[0018] After the optoelectronic system receives the target information sent by the radar system, it calculates the camera focal length f and the vertical angle of the pan / tilt according to the target distance. v , the calculation formulas are:

[0019] f=dis*w p / w

[0020] Where dis is the target distance, w is the actual width of the target, and w p is the width of the target in the image;

[0021]

[0022] Where h represents the height of the photovoltaic system relative to sea level;

[0023] After the calculation is completed, the focal length f and the vertical angle of the gimbal angle v The horizontal angle of the gimbal sent by the radar system is sent to the gimbal and camera for a first-stage system parameter adjustment to shoot the target.

[0024] As an improvement to the above technical solution, the method of using a deep learning model to detect targets in images captured by the optoelectronic system, calculating the detection score of the target, and determining the target based on the detection score specifically includes:

[0025] Use deep learning models to detect targets in images taken by optoelectronic systems, and restrict the target detection results to calculate the detection score. To select the target to be tracked; detection score The calculation formula is:

[0026]

[0027]

[0028]

[0029] in, is the confidence level of target detection of deep learning models, is the distance deviation fraction, is the target detection area and signal intensity consistency score; and Respectively represent the horizontal and vertical coordinates of the center of the target detection frame, and Respectively represent the horizontal and vertical coordinates of the center of the picture, l s With w sRespectively represent the length and width of the picture, s r With s v They represent the signal strength ranking of the tracking target sent by the radar and the size ranking of the target detection box in the window, respectively. y 、k d 、k c The value range of is 0 to 1, indicating the importance of

[0030] Calculate the detection scores of the targets in the window in turn, and take the target with the highest detection score as the target to be tracked.

[0031] As an improvement to the above technical solution, if the deep learning model fails to detect the target, a maximum error annular search point diagram is established, and each search point is traversed to search for the target, specifically including:

[0032] If the deep learning model fails to detect the target, the maximum error ring of the target is obtained according to the target detection accuracy of the radar system, and k points are uniformly sampled on the maximum error ring as the preset search points;

[0033] Search for targets by adjusting the horizontal angle, vertical angle and focal length of the gimbal; the horizontal angle adjustment value Vertical angle adjustment value and focus adjustment value The calculation formulas are:

[0034]

[0035]

[0036]

[0037] Where δ is the target detection accuracy of the radar system, θ is the angle between the search point and the target position and the connecting line of the optoelectronic system, and w′ is the reduced preset target size;

[0038] At most 2k preset search points are traversed until the target is found.

[0039] As an improvement to the above technical solution, if the target detection network model fails to detect the target, a maximum error annular search point diagram is established, and each search point is traversed to search for the target, further comprising:

[0040] If no target is found after traversing the 2k preset search points, the radar system will be fed back that the target is a false target.

[0041] As an improvement to the above technical solution, after the target is determined, the photoelectric system tracks the target and adjusts the photoelectric system parameters during the tracking process to achieve automatic tracking of the target by the photoelectric system, specifically including:

[0042] Initialize the tracker using the current frame and the detection frame, so that in subsequent frames, only the tracker is used to match the target's real-time position and track the target.

[0043] When tracking is successful, adjust the photoelectric system parameters, including the horizontal and vertical angles of the PTZ, to ensure that the target is always in the center of the image; the horizontal adjustment angle adj h With vertical adjustment angle adk v The calculation formulas are:

[0044]

[0045]

[0046] Among them, CMOS l With CMOS w Respectively represent the length and width of the photosensitive device;

[0047] Adjust the focus to magnify the target to a suitable size to check the target appearance information; if the target width needs to be adjusted to w e , focus adjustment value adj f The calculation formula is:

[0048]

[0049]

[0050] Among them, f n is the current focal length value, w n is the pixel value occupied by the target;

[0051] Adjust the horizontal angle to adj h , vertical angle adjustment value adj v , focal length adjustment value adj f Divide by the weakening factor m h 、m v 、m f get adj h ′, adj v ′, adj f ′, and sent to the PTZ and camera to dynamically shoot the target, so as to realize automatic tracking of the target by the optoelectronic system.

[0052] As an improvement to the above technical solution, if a failure occurs during the tracking process, we return to using the deep learning model for target detection and reinitialize the tracker.

[0053] The advantages of the present invention are:

[0054] Compared with the prior art, the present invention has the following advantages:

[0055] 1. The present invention combines the advantages of radar and optoelectronic systems to propose a method for automatic detection and tracking of sea surface targets by combining light and mine, achieving high-precision monitoring of sea surface targets at long and short distances. The optoelectronic system can also self-track the target vessel without the need to receive radar data in real time.

[0056] 2. This invention proposes a two-stage dynamic target acquisition algorithm for optical mines. Based on the principle of optical imaging, the first stage of optoelectronic system parameter adjustment is combined with the target information transmitted by the radar system. Then, the second stage of optoelectronic system parameter fine-tuning is continuously carried out in combination with the target information detected by the optoelectronic system. The parameter adjustment range includes: horizontal and vertical angles, and focal length value. This achieves good acquisition of the target ship without affecting the target tracking effect.

[0057] 3. This invention proposes a multi-mode information fusion candidate vessel determination algorithm. If multiple vessels are present within the target detection window, the target vessel to be tracked is determined by comprehensively utilizing target detection confidence, distance deviation, target detection area, and radar signal strength consistency information.

[0058] 4. The present invention proposes a target search algorithm based on a preset maximum error circular search point diagram. When target detection fails to find the target vessel, a maximum error circular search point diagram is established according to the operating characteristics of the radar system and the target vessel is traversed and searched according to the imaging principle of the optoelectronic system. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 This is a flow chart of the automatic detection and tracking method of sea surface targets linked with optical mines;

[0060] Figure 2 It is a ring search point diagram with preset maximum error. DETAILED DESCRIPTION

[0061] The technical solution of the present invention is described in detail below with reference to the accompanying drawings and embodiments.

[0062] The present invention proposes a method for automatic detection and tracking of sea surface targets in conjunction with light and mines. The method mainly includes four parts: radar system data analysis, optoelectronic system command reception, optoelectronic system target detection and tracking, and optoelectronic system parameter adjustment. After the radar system collects the radar echo signal, if a target is detected and can be continuously tracked for a period of time, the target information is sent to the optoelectronic system for further verification of the target. The optoelectronic system then first performs a first-stage optoelectronic system parameter adjustment to capture the target vessel, and then performs target detection to confirm whether the target is successfully found. If no target is found, a preset maximum error annular search point diagram is traversed to search for the target vessel. If the target vessel is found, the tracker is initialized, and in subsequent frames, a second-stage optoelectronic system parameter adjustment is performed based on the tracking results to achieve self-tracking of the target vessel by the optoelectronic system.

[0063] The process of the automatic detection and tracking method of sea surface targets linked with light mines is as follows: Figure 1 As shown, the following steps are included:

[0064] The radar system receives the radar target echo signal, obtains target information based on the received signal, and detects and tracks the target. After tracking the target, it determines whether the target enters the detection range of the optoelectronic system. If not, it continues tracking. If so, it sends the target information (including target number, horizontal angle, distance, and target radar signal strength ranking based on the water surface angle) to the optoelectronic system.

[0065] According to the imaging principle of the optoelectronic system and the target information sent by the radar system, the parameters of the optoelectronic system (including horizontal angle, vertical angle and focal length) are adjusted in one stage to shoot the target;

[0066] The images captured by the optoelectronic system are detected using a trained deep learning target detection model. If no target is detected, a target search algorithm based on a preset maximum error annular search point diagram is used to search for the target. If the target is still not found, the target is considered a false target.

[0067] After detecting the target, it is determined whether there is only one target in the image window. If not, a target is determined using the optical mine information fusion candidate target determination algorithm. If so, the optoelectronic system tracker is initialized and updated to determine whether the target is tracked successfully:

[0068] If not, the system returns to the image captured by the optoelectronic system and uses the trained deep learning target detection model for detection. If tracking is successful, the target's pixel offset from the center of the screen and the target's screen ratio are calculated based on the tracking results. The system also determines whether the target is close to the edge of the screen:

[0069] If not, return to update the optoelectronic system tracker to determine whether the target is tracked successfully; if so, perform a two-stage adjustment of the optoelectronic system based on the optical imaging principle and the target information detected by the optoelectronic system; and determine whether the required number of video frames required for evidence collection has been collected:

[0070] If not, return to update the optoelectronic system tracker to determine whether the target is tracked successfully; if so, the target detection and tracking is completed.

[0071] More specifically, an embodiment of the method of the present invention comprises the following steps:

[0072] Step 1) Radar system data analysis. If new radar data is received, matched filtering is first performed to improve the radar signal signal-to-noise ratio to better distinguish the target from noise, clutter and interference. Subsequently, a detection threshold is set to identify sea targets. However, since the interference of noise on the radar signal cannot be completely eliminated, if the threshold value is set too low during actual detection, a large number of false alarms will be caused, and if it is set too high, some small targets may be missed, greatly affecting the credibility of the detection result. Therefore, the present invention uses constant false alarm rate detection (CFAR) to dynamically adjust the detection threshold according to radar clutter data, maximizing the target detection probability when the false alarm probability remains unchanged. After detecting the sea target, the sea target is first tracked by an α-β filter radar tracking algorithm. If it can be continuously tracked for a period of time, the target information is sent to the optoelectronic system for further evidence collection, wherein the target information includes: target ID, target horizontal angle, target distance and the radar echo signal intensity ranking of the target detected by the vessel in its vicinity.

[0073] Step 2) Receiving the command from the optoelectronic system. After receiving the target information sent by the radar system, in order to ensure the effective evidence collection of the sea surface target, it is necessary to accurately calculate the camera focal length f and the vertical angle a of the pan / tilt head according to the distance from the ship. v To improve the target capture effect. As for the focal length value, according to the camera imaging principle, it is easy to know that the calculation process of the focal length value f is:

[0074] f=dis*w p / w

[0075] Where dis is the target distance, w is the actual width of the target, and w p is the width of the target in the image. In actual estimation, the actual width of the target w and the width of the target in the image w are required. p Assume that the vertical angle is a right triangle formed by the sea level, the target ship, and the optoelectronic system. The vertical angle calculation formula is as follows:

[0076]

[0077] Where dis represents the target distance, and h represents the height of the optoelectronic system relative to sea level. After the calculation is complete, the control parameters are sent to the gimbal and camera to adjust the system parameters in the first stage to effectively capture the target vessel. Furthermore, the horizontal angle sent is derived from target detection by the radar system.

[0078] Step 3) Target detection and tracking by optoelectronic system. Target detection is performed using a deep learning model. The deep learning detection model can be YOLO, SSD series, or Faster RCNN series models. In this embodiment, the YOLO model is used. Target detection is first performed using the YOLO model. During the detection process, to avoid interference from other ships within the imaging window, the target detection results must be restricted and the detection score calculated. To select the vessel to be tested. The detection score consists of three parts: YOLO target detection confidence Distance deviation score and target detection area, signal intensity consistency score Among them, the YOLO target detection confidence can reflect the quality of the recognition effect of different ships at the current focal length. Since the camera focal length has been adjusted according to the ship distance data returned by the radar, the higher the detection confidence of the target detection model, the more likely it is that the ship is the ship to be tracked. The distance deviation score can measure the degree to which the ship deviates from the center of the picture. Since the horizontal and vertical angles of the gimbal have been adjusted according to the ship information returned by the radar, the closer the ship is to the center of the picture, the more likely it is the target ship to be tracked. In addition, since the radar echo signal strength is proportional to the size of the ship, the ship whose detection box area ranking of the target detected in the window is close to the radar signal strength ranking is more likely to be the target ship. The specific calculation process is as follows:

[0079]

[0080]

[0081]

[0082] in and Respectively represent the horizontal and vertical coordinates of the center of the ship detection frame, and Respectively represent the horizontal and vertical coordinates of the center of the picture, l s With w s Respectively represent the length and width of the picture, s r With s v They represent the strength ranking of the tracking target signal sent by the radar and the size ranking of the ship detection box in the window, respectively. y 、k d 、k c Respectively The importance of k is determined according to the actual experimental results. In this embodiment, the actual value is: y is 1, k d is 0.1, k c 0.1. The detection scores of each vessel in the view window are calculated sequentially, and the vessel with the highest score is selected as the target vessel. Furthermore, due to the influence of sea clutter or other interference, the target's horizontal angle or distance information obtained by the radar may exhibit slight deviations. Furthermore, because the target is far away, even small angular deviations at a large focal length can prevent the camera from capturing the vessel to be tracked. Therefore, if the YOLO model fails to detect a vessel, the horizontal and vertical angles of the gimbal and the focal length must be adjusted to search for the vessel. Assuming the radar system's target detection accuracy is δ, the maximum error ring for the target vessel can be obtained. Eight points are evenly sampled on the maximum error ring as preset search points. Figure 2 The figure shows the preset maximum error circular search point diagram. Assume that the angle between the search point and the target ship position and the optical-electronic system connection line is θ, then the horizontal angle adjustment value is Vertical angle adjustment value The calculation process is:

[0083]

[0084]

[0085] In addition, since the actual size of the target ship w needs to be assumed in advance during the calculation of the focal length value, if the actual length of the ship is less than w, the target may be too small to be effectively detected. Therefore, the preset target ship size needs to be reduced to w′, and the focal length adjustment value is The calculation process is:

[0086]

[0087] A maximum of 16 preset search points are traversed. If the target vessel is still not found after the traversal is completed, the radar system will be fed back that the target is a false target.

[0088] After determining the target, the optoelectronic system uses a target tracking algorithm to track the target, wherein the target tracking algorithm can be a KCF (Kernel Correlation Filter) algorithm, a TLD (Tracking-Learning-Detection) algorithm, etc. In this embodiment, the KCF algorithm is used; if the target ship is detected, the KCF tracker is initialized using the current frame image and the detection frame, so that the optoelectronic system can self-track by matching the target's real-time position with the help of the tracker in subsequent frames. If the tracking is successful during the tracking process, the two-stage system parameter adjustment is performed. The specific process is as follows: Calculate the distance between the center of the ship and the center of the picture and then adjust the horizontal and vertical angles of the pan-tilt head to ensure that the target is always in the center of the picture. The specific process is as follows: Calculate the horizontal coordinate pixel value difference P between the center of the ship and the center of the picture respectively h The difference P between the vertical coordinate pixel value v , and further calculate the actual deviation distance of the target ship on the photosensitive device compared to the center of the device, and finally calculate the horizontal adjustment angle adj according to the current focal length value h Adjust the angle with vertical angle adj v .

[0089]

[0090]

[0091] Among them, CMOS l With CMOS w Represent the length and width of the photosensitive device, respectively. In addition, since the optoelectronic system is far away from the target vessel, the focal length needs to be adjusted to magnify the target to an appropriate size to better inspect the target vessel's appearance. Based on the principle of camera imaging, it is easy to see that the focal length is proportional to the width of the target in the image:

[0092]

[0093] Therefore, if the focal length is f n , the pixel value of the target is w n If you need to adjust the target width to w e , then the focal length adjustment value adj f The calculation formula is:

[0094]

[0095] In addition, if the tracking continues successfully for a long time, the target detection should be repeated and the tracker should be initialized to eliminate the phenomenon of gradual decline in tracker accuracy. If a failure occurs during the tracking process, the target detection should be repeated and the tracker should be reinitialized.

[0096] Step 4) Photoelectric system parameter adjustment. When the horizontal angle adjustment value adj is obtained h , vertical angle adjustment value adj v , focal length adjustment value adj f When h , m v , m f get adj h , adj v , adj f To avoid the deterioration of target tracking effect caused by drastic adjustment of the above parameters, m h With m v is 10, m f is 30. Subsequently, the modified values ​​of the above system parameters are sent to the PTZ and camera for dynamic shooting.

[0097] Finally, it should be noted that the above embodiments are intended only to illustrate the technical solutions 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 embodiments, it should be understood by those skilled in the art that modifications or equivalent substitutions to the technical solutions of the present invention do not depart from the spirit and scope of the technical solutions of the present invention and are intended to be encompassed by the claims of the present invention.

Claims

1. A method for automatically detecting and tracking sea surface targets using a combination of light and mine, the method comprising: The radar system detects and tracks sea surface targets using radar echo data collected by the radar system, and then obtains target information, which is then sent to the optoelectronic system. Adjust the optoelectronic system parameters in one stage according to the target information and shoot the target; Use a deep learning model to detect targets in images taken by the optoelectronic system, calculate the target detection score, and identify the target based on the detection score; If the deep learning model does not detect the target, a maximum error circular search point graph is set up and each search point is traversed to search for the target; After determining the target, the optoelectronic system uses a target tracking algorithm to track the target, and during the tracking process continuously adjusts the optoelectronic system parameters in two stages according to the captured target image, so as to realize automatic tracking of the target by the optoelectronic system.

2. The method for automatic detection and tracking of sea surface targets using light-mining linkage according to claim 1 is characterized in that: The method also includes: if there are multiple targets detected by the deep learning model or targets searched for by traversing each search point, then using the light mine information fusion candidate target determination algorithm to determine the target, specifically: comprehensively utilizing three aspects of information: target detection confidence, the offset of the ship from the center of the picture, and the consistency of the target detection frame and signal strength to determine the target ship.

3. The method for automatic detection and tracking of sea surface targets using light-mining linkage according to claim 1 is characterized in that: The optoelectronic system includes a camera and a pan-tilt platform; wherein the camera is used to capture sea surface images, and the pan-tilt platform is used to support the camera and can freely adjust the horizontal and / or vertical angles to assist in tracking targets.

4. The method for automatic detection and tracking of sea surface targets using light-mine linkage according to claim 3 is characterized in that: The target information is obtained by detecting and tracking the sea surface target through the radar echo data collected by the radar system, specifically including: After collecting radar echo data, the radar system performs matched filtering on the echo data and identifies sea surface targets by setting a detection threshold; the detection threshold is dynamically adjusted based on the clutter in the radar echo data using a constant false alarm rate detection algorithm; After detecting the sea surface target, the sea surface target is tracked, and the target information is sent to the optoelectronic system when the target enters the detection range of the optoelectronic system; the target information includes: target number, target horizontal angle, target distance and the ranking of the radar echo signal strength detected by the target among the nearby angle targets.

5. The method for automatic detection and tracking of sea surface targets using light-mining linkage according to claim 4 is characterized in that: The first-stage adjustment of the optoelectronic system parameters according to the target information and the shooting of the target specifically includes: After the optoelectronic system receives the target information sent by the radar system, it calculates the camera focal length f and the vertical angle of the pan / tilt according to the target distance. v , the calculation formulas are: f=dis*w p / w Where dis is the target distance, w is the actual width of the target, and w p is the width of the target in the image; Where h represents the height of the photovoltaic system relative to sea level; After the calculation is completed, the focal length f and the vertical angle of the gimbal angle v The horizontal angle of the gimbal sent by the radar system is sent to the gimbal and camera for a first-stage system parameter adjustment to shoot the target.

6. The method for automatic detection and tracking of sea surface targets using light and mine linkage according to claim 5 is characterized in that: The method of using a deep learning model to detect targets in images captured by the optoelectronic system, calculating the detection score of the target, and determining the target based on the detection score specifically includes: Use deep learning models to detect targets in images taken by optoelectronic systems, and restrict the target detection results to calculate the detection score. To select the target to be tracked; detection score The calculation formula is: in, is the confidence level of target detection of deep learning models, is the distance deviation fraction, is the target detection area and signal intensity consistency score; and Respectively represent the horizontal and vertical coordinates of the center of the target detection frame, and Respectively represent the horizontal and vertical coordinates of the center of the picture, l s With w s Respectively represent the length and width of the picture, s r With s v They represent the signal strength ranking of the tracking target sent by the radar and the size ranking of the target detection box in the window, respectively. y 、k d 、k c The value range of is 0 to 1, indicating the importance of Calculate the detection scores of the targets in the window in turn, and take the target with the highest detection score as the target to be tracked.

7. The method for automatic detection and tracking of sea surface targets using light-mining linkage according to claim 6 is characterized in that: If the deep learning model fails to detect the target, a maximum error annular search point graph is established, and each search point is traversed to search for the target, specifically including: If the deep learning model fails to detect the target, the maximum error ring of the target is obtained according to the target detection accuracy of the radar system, and k points are uniformly sampled on the maximum error ring as the preset search points; Search for targets by adjusting the horizontal angle, vertical angle and focal length of the gimbal; the horizontal angle adjustment value Vertical angle adjustment value and focus adjustment value The calculation formulas are: Among them, δ is the target detection accuracy of the radar system, θ is the angle between the search point and the target position and the connecting line of the optoelectronic system, and w ′ The preset target size is reduced; At most 2k preset search points are traversed until the target is found.

8. The method for automatic detection and tracking of sea surface targets using light-mining linkage according to claim 7 is characterized in that: If the deep learning model does not detect the target, a maximum error annular search point diagram is established, and each search point is traversed to search for the target, further comprising: If no target is found after traversing the 2k preset search points, the radar system will be fed back that the target is a false target.

9. The method for automatic detection and tracking of sea surface targets using light-mining linkage according to claim 7 or 8, characterized in that: After the target is determined, the optoelectronic system tracks the target and adjusts the optoelectronic system parameters during the tracking process to achieve automatic tracking of the target by the optoelectronic system, specifically including: Initialize the tracker using the current frame and the detection frame, so that in subsequent frames, only the tracker is used to match the target's real-time position and track the target. When tracking is successful, adjust the photoelectric system parameters, including the horizontal and vertical angles of the PTZ, to ensure that the target is always in the center of the image; the horizontal adjustment angle adj h Adjust the angle with vertical angle adj v The calculation formulas are: Among them, CMOS l With CMOS w Respectively represent the length and width of the photosensitive device; Adjust the focus to magnify the target to a suitable size to check the target appearance information; if the target width needs to be adjusted to w e , focus adjustment value adj f The calculation formula is: Among them, f n is the current focal length value, w n is the pixel value occupied by the target; Adjust the horizontal angle to adj h , vertical angle adjustment value adj v , focal length adjustment value adj f Divide by the weakening factor m h 、m v 、m f get adj h ′ ,adj v ′ ,adj f ′ , and sent to the pan-tilt head and camera to dynamically shoot the target, so as to realize automatic tracking of the target by the optoelectronic system.

10. The method for automatic detection and tracking of sea surface targets using light-mining linkage according to claim 9 is characterized in that: If a failure occurs during tracking, we return to using the deep learning model for target detection and reinitialize the tracker.

Images