Offshore commutation platform anti-collision early warning system and method based on remote sensing monitoring and machine learning

Abstract

The invention discloses a marine commutation platform anti-collision early warning system based on remote sensing monitoring and machine learning, and the system comprises a multi-source data collection module which obtains multi-source data and carries out the fusion of the multi-source data to generate a fusion data field; the target recognition module recognizes and extracts the spatial position, length, speed and course of the target based on the fused data field; the relative motion modeling module is used for constructing a relative velocity vector between the platform and the target, calculating a distance change rate between the platform and the target, and judging a potential threat target according to the distance change rate; the risk level evaluation module is used for predicting a future trajectory of the platform and the potential threat target, constructing a collision risk function and obtaining a corresponding risk integral; inputting the target feature vector set of the potential threat target into the trained risk classification model to output a predicted risk level; and fusing the risk integral and the predicted risk level by adopting a fusion rule to obtain a risk level result. Therefore, dynamic monitoring and intelligent early warning of the potential threat target are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of marine engineering safety and remote sensing monitoring, in particular to a marine converter platform anti-collision early warning system and method based on remote sensing monitoring and machine learning. BACKGROUND

[0002] Marine converter platforms (such as floating production systems and deep-sea observation platforms) play an important role in energy development and environmental monitoring. However, due to factors such as severe sea conditions and low visibility, platforms are prone to collisions with floating objects and sailing ships, threatening equipment safety and personnel life. Current conventional radars, sonars and AIS (Automatic Identification System) can partially achieve surrounding perception, but have limited identification range, are easily disturbed by weather, and are isolated systems. SUMMARY

[0003] To overcome the deficiencies of the prior art, the present application provides a marine converter platform anti-collision early warning system and method based on remote sensing monitoring and machine learning, which dynamically monitors and intelligently warns potential threat targets through multi-source remote sensing data fusion and machine learning.

[0004] According to an aspect of the present application, a marine converter platform anti-collision early warning system based on remote sensing monitoring and machine learning is provided, comprising: a multi-source data acquisition module for acquiring multi-source data and fusing the multi-source data to generate a unified fusion data field; a target recognition module for recognizing marine targets based on the fusion data field and extracting the spatial position, length, speed and heading of the targets; a relative motion modeling module for constructing the relative velocity vector between the platform and the targets and calculating the distance change rate between the platform and the targets based on the relative velocity vector, and then determining potential threat targets according to the distance change rate; a risk level assessment module for predicting the future trajectory of the platform and the potential threat targets, constructing a collision risk function based on the future trajectory and obtaining the corresponding risk integral; inputting the target feature vector set of the potential threat targets into a trained risk classification model to output a predicted risk level, and fusing the risk integral and the predicted risk level using a fusion rule to obtain a final risk level result; and an early warning response module for implementing a hierarchical response according to the risk level result of the potential threat targets.

[0005] Further, the multi-source data acquisition module unifies the multi-source data to the same coordinate system and time axis through time synchronization and spatial registration; the multi-source data includes SAR images, optical remote sensing images, unmanned aerial vehicle inspection videos, AIS ship trajectory data and platform state information.

[0006] Further, the target recognition module adopts a deep neural network trained based on remote sensing features, and combines a multi-scale attention mechanism to analyze the fusion data field, to identify and extract the spatial position, length, speed and heading of the target.

[0007] Further, the relative motion modeling module extracts the position, speed and heading of the platform; establishes the relative speed vector and relative position vector of the platform and the target; calculates the distance change rate between the platform and the target based on the relative speed vector and the relative position vector; when the distance change rate is positive and the absolute value of the distance change rate is greater than a preset value, determines that the target is a potential threat target.

[0008] Further, the risk level evaluation module constructs a future position prediction function of the target and the potential threat platform, respectively, based on the current position and speed vector of the platform and the potential threat target; constructs a trajectory intersection distance function of the platform and the potential threat target at any future time based on the future position prediction function of the target and the potential threat platform; constructs a collision risk function by integrating the trajectory intersection distance function within a prediction time window, to obtain a risk integral of the prediction time window; fuses the risk integral and the predicted risk level output by the risk classification model using a fusion rule, to obtain a final risk level result; the fusion rule includes a logical priority rule, a model priority rule or a weighted average scoring fusion rule.

[0009] Further, the early warning response module implements a hierarchical response according to the risk level result, triggers automatic broadcasting and obstacle avoidance control for a potential threat target with a high risk level result, triggers manual monitoring prompts for a potential threat target with a medium risk level result, and triggers log archiving processing for a potential threat target with a low risk level result.

[0010] According to an aspect of the present application, a marine converter platform anti-collision early warning method based on remote sensing monitoring and machine learning is provided, which is implemented by using a marine converter platform anti-collision early warning system based on remote sensing monitoring and machine learning. The anti-collision early warning method comprises: acquiring multi-source data and fusing the multi-source data to generate a unified fused data field; identifying a marine target based on the fused data field and extracting a spatial position, a length, a speed and a heading of the target; constructing a relative velocity vector between the platform and the target, calculating a distance change rate between the platform and the target based on the relative velocity vector, and further determining a potential threat target according to the distance change rate; predicting a future trajectory of the platform and the potential threat target, constructing a collision risk function based on the future trajectory and obtaining a corresponding risk integral; inputting a target feature vector set of the potential threat target into a trained risk classification model to output a predicted risk level; fusing the risk integral and the predicted risk level by using a fusion rule to obtain a final risk level result; and implementing a hierarchical response according to the risk level result of the potential threat target.

[0011] Further, the relative velocity vector between the platform and the target is constructed, the distance change rate between the platform and the target is calculated based on the relative velocity vector, and the potential threat target is determined according to the distance change rate, which comprises: extracting a position, a speed and a heading of the platform, and establishing a relative velocity vector and a relative position vector of the platform and the target; calculating a distance change rate between the platform and the target based on the relative velocity vector and the relative position vector; when the distance change rate is positive and the absolute value of the distance change rate is greater than a preset value, determining that the target is a potential threat target.

[0012] Further, the future trajectory of the platform and the potential threat target is predicted, a collision risk function is constructed based on the future trajectory and a corresponding risk integral is obtained, which comprises: constructing a future position prediction function of the target and the platform based on the current position and the speed vector of the platform and the potential threat target; constructing a trajectory intersection distance function of the platform and the potential threat target at any time in the future based on the future position prediction function; constructing a collision risk function by integrating the trajectory intersection distance function in a prediction time window, and further obtaining a risk integral of the prediction time window.

[0013] Further, the hierarchical response is implemented according to the risk level result of the potential threat target, which comprises: for a potential threat target with a high risk level result, triggering an automatic broadcast and obstacle avoidance control; for a potential threat target with a medium risk level result, triggering an artificial monitoring prompt; and for a potential threat target with a low risk level result, triggering a log archiving process.

[0014] The technical scheme realizes wide-area and all-weather target monitoring by using multi-source remote sensing data, and solves the problem of insufficient monitoring range of the existing platform anti-collision monitoring.

[0015] Compared with the prior art, the present application has the following advantages:

[0016] (1) The present application realizes wide-area and all-weather target monitoring by using multi-source remote sensing data, and solves the problem of insufficient monitoring range of the existing platform anti-collision monitoring.

[0017] (2) The detection model of the present application uses a deep neural network trained based on remote sensing features, combined with a multi-scale attention mechanism, to improve the detection capability of low-resolution, small-size and weak-contrast targets, and solves the problem of incomplete target identification of the existing platform.

[0018] (3) The present application not only uses a risk assessment function for risk assessment, but also uses a risk classification model to improve the classification stability, significantly enhancing the warning accuracy in non-regular motion, abnormal target behavior and other scenes. At the same time, the present application executes a differentiated linkage strategy for different levels of risk targets, realizes continuous monitoring and adaptive response to dynamic threat targets, and solves the problem of untimely warning of the existing platform. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 A structure diagram of a sea converter platform anti-collision warning system based on remote sensing monitoring and machine learning is provided for the embodiments of the present application.

[0020] Figure 2 A result diagram of collision risk assessment and trajectory prediction is provided for the embodiments of the present application.

[0021] Figure 3 A flowchart of a sea converter platform anti-collision warning method based on remote sensing monitoring and machine learning is provided for the embodiments of the present application.

[0022] Figure 4 A first sub-flowchart of a sea converter platform anti-collision warning method based on remote sensing monitoring and machine learning is provided for the embodiments of the present application.

[0023] Figure 5 A second sub-flowchart of a sea converter platform anti-collision warning method based on remote sensing monitoring and machine learning is provided for the embodiments of the present application.

[0024] Figure 6 A third sub-flowchart of a sea converter platform anti-collision warning method based on remote sensing monitoring and machine learning is provided for the embodiments of the present application.

[0025] In the figure: 1, multi-source data acquisition module; 2, target identification module; 3, relative motion modeling module; 4, risk level assessment module; 5, early warning response module; 100, anti-collision warning system; 200, platform. DETAILED DESCRIPTION

[0026] The technical solutions of the embodiments of the present application will be described below in conjunction with the drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of the present application.

[0027] In the description of the present application, it should be understood that the terms "center", "longitudinal", "transverse", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise" and the like indicate the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation of the present application.

[0028] In the description of the present application, it should be noted that unless otherwise explicitly specified and limited, the terms "mounting", "connection", "connection" should be understood broadly, for example, it can be fixed connection, or detachable connection, or integral connection; it can be mechanical connection, or electrical connection or can communicate with each other; it can be directly connected, or indirectly connected through an intermediate medium, or the internal communication of two elements or the interaction relationship between two elements. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0029] In the present application, unless otherwise explicitly specified and limited, the first feature "on" or "under" the second feature can include the first and second features in direct contact, or the first and second features not in direct contact but in contact through another feature between them. Moreover, the first feature "on", "above" and "on" the second feature includes the first feature directly above and obliquely above the second feature, or only indicates that the horizontal height of the first feature is higher than that of the second feature. The first feature "under", "below" and "under" the second feature includes the first feature directly below and obliquely below the second feature, or only indicates that the horizontal height of the first feature is less than that of the second feature.

[0030] With the development of remote sensing technology and artificial intelligence, the use of multi-source remote sensing data (such as SAR, optical satellite, high-resolution unmanned aerial image) to realize wide-area, all-weather target monitoring provides a new possibility for platform collision avoidance. Based on this, the present application provides a sea converter platform collision avoidance early warning system 100 based on remote sensing monitoring and machine learning. The collision avoidance early warning system 100 fuses multi-source remote sensing images, platform body perception data and ship behavior information, constructs an intelligent platform safety protection system integrating data perception, dynamic modeling, collision risk judgment and early warning linkage, significantly improves the autonomous safety protection capability of the platform 200 in complex environment, and solves the problems of insufficient collision monitoring range, incomplete target identification and untimely warning of the existing platform 200 in severe sea conditions.

[0031] It should be noted that the collision avoidance early warning system 100 is deployed in an edge computing device (for example, an industrial computer, an embedded hardware, etc., which is not limited here) and has continuous operation capability at sea. The present application is suitable for deep-sea operation platform, especially in the area where communication is not smooth, sea conditions are complex or visibility is poor, and has significant practical value and promotion prospect.

[0032] Please refer to the accompanying drawings Figures 1-2 The sea converter platform collision avoidance early warning system 100 based on remote sensing monitoring and machine learning provided by the present application comprises a multi-source data acquisition module 1, a target identification module 2, a relative motion modeling module 3, a risk level evaluation module 4 and an early warning response module 5. The multi-source data acquisition module 1 is used to acquire multi-source data and fuse the multi-source data to generate a unified fusion data field. The target identification module 2 is used to identify sea targets based on the fusion data field and extract the spatial position, length, speed and heading of the target. The relative motion modeling module 3 is used to construct the relative velocity vector between the platform and the target, and calculate the distance change rate between the platform and the target based on the relative velocity vector, and then determine the potential threat target according to the distance change rate. The risk level evaluation module 4 is used to predict the future trajectory of the platform and the potential threat target, construct a collision risk function based on the future trajectory and obtain the corresponding risk integral; input the target feature vector set of the potential threat target into the trained risk classification model to output the predicted risk level, fuse the risk integral and the predicted risk level by using the fusion rule to obtain the final risk level result. The early warning response module 5 is used to implement graded response according to the risk level result of the potential threat target.

[0033] Please refer to the accompanying drawings Figure 3 The present application also provides a sea converter platform collision avoidance early warning method based on remote sensing monitoring and machine learning, which is realized by using the sea converter platform collision avoidance early warning system 100 based on remote sensing monitoring and machine learning as described above, and comprises the following steps (steps S101-S106):

[0034] Step S101 : acquiring multi-source data and fusing the multi-source data to generate a unified fused data field.

[0035] In step S101, multi-source data includes, but is not limited to, SAR imagery, optical remote sensing imagery, drone inspection videos, AIS vessel trajectory data, and platform status information. In this embodiment, the collision avoidance warning system 100 acquires SAR and optical remote sensing images in real time via a satellite remote sensing receiver, and acquires near-field, high-resolution drone inspection video images via scheduled drone flights. Furthermore, the collision avoidance warning system 100 incorporates an integrated AIS information receiving antenna, enabling it to collect the real-time position, speed, and heading of surrounding vessels. Furthermore, the platform 200 itself is equipped with a GNSS device and an inertial navigation unit to collect precise position and attitude information.

[0036] In this embodiment, after acquiring multi-source data, the data is first unified into the same coordinate system and time axis through time synchronization and spatial registration processes, and then the multi-source data with the unified coordinate system and time axis are fused into a fused data field. It can be understood that unifying multi-source data into the same coordinate system and time axis can eliminate errors caused by differences in sampling intervals and spatial resolutions of different devices, providing a data foundation for subsequent data fusion, target recognition, and dynamic modeling. Fused data field (i.e., fused image) as follows: .in, 、 、 They are SAR images, optical remote sensing images, and drone inspection videos. is the AIS ship track data, is the platform status information, represents the spatiotemporal registration function.

[0037] Step S102: Identify the maritime target based on the fused data field and extract the spatial position, length, speed and heading of the maritime target.

[0038] In step S102, after completing the fusion of multi-source data, the anti-collision warning system 100 uses a deep neural network based on remote sensing feature training, combined with a multi-scale attention mechanism, based on the fused image Detect and identify maritime targets, including but not limited to small vessels, unpowered floating objects, and iceberg fragments. This improves the recognition of blurred targets and irregularly shaped objects, making it particularly suitable for low-resolution, low-contrast remote sensing images.

[0039] In this embodiment, for any identified target , the collision avoidance warning system 100 extracts the target spatial location , length estimation , instantaneous velocity , orientation angle , etc. to constitute the complete state vector of the target: , wherein the instantaneous velocity is calculated by the displacement difference between consecutive frames (i.e. consecutive fused images ): . Further, the present application continuously updates the attribute information of the target when the target moves continuously, and the complete state vector obtained is used to provide key input for the target individual behavior model of the collision avoidance warning system 100, which is directly called in subsequent relative motion modeling, trajectory prediction, risk assessment and machine learning training, and is one of the core foundations of the entire collision avoidance system logic chain.

[0040] Step S103, the relative velocity vector between the platform and the target is constructed, and the distance change rate between the platform and the target is calculated based on the relative velocity vector, and then the potential threat target is determined according to the distance change rate.

[0041] Please refer to Figure 4 , the following will further introduce step S103 (including steps S1031-S1035).

[0042] Step S1031, the position, velocity and heading of the platform are extracted, and the relative velocity vector and the relative position vector between the platform and the target are established. In this embodiment, while obtaining the state of each target, the platform 200 extracts the position , velocity and heading of the platform 200 at the current time through the GNSS device and the inertial navigation device, based on which the collision avoidance warning system 100 further establishes the relative motion relationship between the platform 200 and each monitoring target. That is, the relative velocity vector between the platform 200 and each monitoring target is constructed: . The relative position vector between the platform 200 and each monitoring target is constructed, and the length of the relative position vector is .

[0043] Step S1033, the distance change rate between the platform and the target is calculated based on the relative velocity vector and the relative position vector. In this embodiment, the distance change rate (i.e. the approach speed) can be calculated by the dot product of the relative position vector and the relative velocity vector divided by the length of the relative position vector, and the specific formula is as follows: ​The formula not only considers the target approaching speed, but also takes into account the direction difference, and realizes accurate modeling of the dynamic behavior of multiple targets. Understandably, the modeling process can accurately identify targets that do not turn on AIS, especially when the platform 200 is in night or wind wave conditions, which can better play the advantages of remote sensing and autonomous modeling, and make up for the blind area of the traditional radar and AIS system.

[0044] Step S1035, when the distance change rate is positive and the absolute value of the distance change rate is greater than the preset value, the target is determined as a potential threat target. Specifically, the target can be judged to approach or move away from the platform according to the positive and negative of the distance change rate, and the speed of the target approaching or moving away from the platform can be judged according to the absolute value of the distance change rate. When the distance change rate is positive and the absolute value of the distance change rate is greater than the preset value, it means that the target is moving towards the platform 200 and the speed is fast, so the target is determined as a potential threat target. The preset value can be set according to the actual scene, which is not limited here.

[0045] Step S104, predicting the future trajectory of the platform and the potential threat target, and constructing a collision risk function based on the future trajectory and obtaining the corresponding risk integral.

[0046] In step S104, the collision avoidance warning system 100 performs trajectory prediction for a short time window for each potential threat target, calculates the minimum distance and relative direction between each potential threat target and the platform 200 in a future period of time. If each potential threat target has a trajectory intersection with the platform 200 within the prediction time window, and the intersection point distance is lower than the safety threshold, the collision avoidance warning system 100 will mark the target as a high-risk target. This evaluation process not only considers the current position of each potential threat target, but also refers to its historical motion trend, speed change characteristics and the movement of the platform 200 itself, to form a dynamic risk level. Further, please refer to Figure 5 , which will be further described below.

[0047] Step S1041, based on the current position and speed vector of the platform and the potential threat target, a future position prediction function of the target and the potential threat platform is constructed. In this embodiment, in order to judge the potential collision risk in the future period in advance, the collision avoidance warning system 100 performs trajectory extrapolation prediction for the platform 200 and the potential threat target. Based on the current position and speed vector of the platform 200 and the potential threat target, the future position prediction function of the platform 200 and the potential threat target is constructed respectively, and the formula is as follows: , , wherein i represents the i-th identified sea target. represents the time interval for predicting the future (i.e., the prediction time window in step S1045), i.e., a certain length of time extrapolated from the current time t to the future, which is a continuously changing variable with units of seconds (s) or minutes (min). , represents the horizontal and vertical coordinates of the platform 200 at the current time t, , represents the horizontal and vertical coordinates of the platform 200 at the time t+ , represents the speed of the platform 200. , represents the horizontal and vertical coordinates of the i-th maritime target at the current time t, , represents the horizontal and vertical coordinates of the i-th maritime target at the time t+ , represents the speed of the i-th maritime target.

[0048] Step S1043, constructing a trajectory intersection distance function of the platform and the potential threat target at any future time based on the future position prediction function. In the present embodiment, the trajectory intersection distance function of the platform 200 and the potential threat target at any future time is constructed according to the future position prediction functions of the two, and the formula is as follows: .

[0049] Step S1045, constructing a collision risk function by integrating the trajectory intersection distance function within the prediction time window, and then obtaining the risk integral of the prediction time window. In the present embodiment, the collision risk function is constructed by integrating the trajectory intersection distance function within the prediction time window, and the formula is as follows: , wherein represents the perception parameter dynamically adjusted according to the target size and radar detection error. represents the maximum value of the variable . represents the numerical value of the collision risk function (i.e., the risk integral). It can be understood that the higher the risk integral , the greater the possibility of future intersection, and the higher the risk level.

[0050] It should be noted that when the anti-collision warning system 100 is in the initial deployment stage or the risk classification model has not been enabled, the risk integral can be directly used as the basis for the warning level, and a static level division threshold is set to judge the risk level result. In the present embodiment, the risk level result , wherein represents the first risk threshold, represents the second risk threshold. That is, when the risk score is greater than the first risk threshold, the risk level result is determined as high risk; when the risk score is less than or equal to the first risk threshold and greater than the second risk threshold, the risk level result is determined as medium risk; and when the risk score is less than or equal to the second risk threshold, the risk level result is determined as low risk. Understandably, the first risk threshold and the second risk threshold may be adaptively adjusted according to the traffic density and historical accident risk of the sea area where the platform 200 is located.

[0051] In step S105, the target feature vector set of the potential threat target is input into the trained risk classification model to output a predicted risk level, and the risk score and the predicted risk level are fused by using a fusion rule to obtain a final risk level result.

[0052] In step S105, it can be understood that the present application not only uses the collision risk function to evaluate the risk level, but also uses the risk classification model to improve the stability of risk classification, which significantly enhances the warning accuracy of the anti-collision warning system 100 in the scene of irregular motion and abnormal target behavior. Further, please refer to Figure 6 , which will be further introduced below.

[0053] In step S1051, a risk classification model is trained based on a machine learning model and a historical risk sample set. In this embodiment, in order to further improve the accuracy of risk level evaluation and the intelligent level of response, the anti-collision warning system 100 introduces a machine learning model, which uses a random forest (Random Forest) or a lightweight neural network model, has good real-time response capability, can dynamically adjust the model parameters according to different regions or sea areas, and improves the adaptability and generalization ability of the model. The present application uses a machine learning model to construct a risk classification model , and trains the risk classification model based on a risk sample set. The training process is as follows: a historical risk sample set is constructed: , wherein, is the feature vector set of the ith sample, which includes but is not limited to the current relative speed module , the relative angle difference , the predicted minimum intersection distance , the future risk score within 5 seconds. is the actual risk level label marked by a person or a historical accident, . The historical risk sample set is input into the risk classification model , and the training target is:​ After continuous training, the risk classification model Output predicted risk level Continuously approaching the actual risk level.

[0054] Step S1053: Input the target feature vector set of the potential threat target into the trained risk classification model to output the predicted risk level. In this embodiment, after the training is completed, the target feature vector set of the potential threat target is input into the trained risk classification model to predict the risk level of the potential threat target based on the target feature vector set of the potential threat target, and output the predicted risk level. .

[0055] Step S1055: Based on the predicted risk level and risk score, the final risk level result is obtained by using the fusion rule. Specifically, the risk score and machine learning output Joint calibration, adjust the final warning level result by the following methods : FusionRule is a specific fusion rule. Depending on the actual deployment, it can be set to any of the following strategies: logic priority rule, model priority rule, or weighted average scoring fusion rule. As can be seen, fusing the predicted risk level with the risk score can avoid misjudgments of risk level based on a single rule.

[0056] It should be noted that the logical priority rule refers to the output of machine learning and risk score If any of the two parties indicates a high risk, the final warning level result will be The model priority rule means that you can fully trust the output of the machine learning model. , which is used as the basis for determining the final warning level. The weighted average scoring fusion rule refers to setting the risk score based on credibility. and machine learning outputs The final warning level score is calculated by weighted average, and then the warning level is determined based on the score. The weight is w1, and the machine learning output is The weight is w2, and w1+w2=1. Therefore, for each sample, according to the risk score and machine learning outputs The final score is calculated by weight: S=w1× +w2× Finally, the warning level result is determined based on the final score.

[0057] Step S106, implement a hierarchical response according to the risk level result of the potential threat target.

[0058] In step S106, it is necessary to point out that, in order to realize the grading and differentiated response control of the risk target, the collision avoidance warning system 100 divides the warning levels according to different strategies determination. Among them, the collision avoidance warning system 100 supports two types of warning level determination modes: (1) Basic determination mode: separate division based on risk function value. (2) Fusion determination mode: introduce machine learning and fusion strategy. The basic determination mode has strong universality and stability, and the collision avoidance warning system 100 can run independently in the absence of samples or without machine learning model support, ensuring the basic collision avoidance capability of the collision avoidance warning system 100. When the machine learning model is trained, the fusion determination mode is adopted. This fusion determination mode not only uses the risk function value provided by the physical model to judge, but also introduces the risk classification model to accurately identify complex behaviors, significantly improving the warning accuracy and system intelligence level in non-regular motion or abnormal target behavior scenarios.

[0059] Further, after the risk level result is determined, the collision avoidance warning system 100 sets up a "high risk", "medium risk", "low risk" three-grade response mechanism according to the final evaluated risk level result. For the potential threat target with high risk level result, the collision avoidance warning system 100 immediately triggers the platform broadcast system to perform voice and light warning, and sends control commands to the platform controller to enter the obstacle avoidance mode; for the potential threat target with medium risk level result, the collision avoidance warning system 100 will pop up a window on the monitoring terminal to notify the duty personnel for manual confirmation. For the potential threat target with low risk level result, the collision avoidance warning system 100 will automatically record its trajectory and attribute information for subsequent playback and model optimization. Understandably, the collision avoidance warning system 100 forms a closed-loop control process from "automatic perception → intelligent judgment → rapid response" in operation, which significantly improves the autonomous protection capability of the platform 200 in complex and uncontrollable environments.

[0060] In summary, the present application relates to a collision warning system and method for offshore converter platform based on remote sensing monitoring and machine learning. The collision warning system fuses multi-source remote sensing data, ship AIS trajectory data and platform state information, and constructs a fusion data field through unified space-time registration. A deep neural network is used to extract the position, size, speed and orientation parameters of the offshore target, and a relative motion model between the platform and the target is established. Based on trajectory prediction, the future intersection distance is calculated and a collision risk function is constructed, and the risk level result is preliminarily judged using the collision risk function. A risk classification model is further introduced, and the trained risk classification model is used to obtain the predicted risk level, and the risk level result is optimized and adjusted according to the fusion rule. Finally, the risk level result triggers a hierarchical response. The collision warning method has the characteristics of high-precision perception, intelligent judgment and response closed loop, and is suitable for all-weather platform safety protection under complex sea conditions.

[0061] Finally, it should be noted that: the above examples are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing examples, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing examples, or make equivalent replacement for part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the technical solutions of the present application.

Claims

1. A collision avoidance warning system for offshore converter platforms based on remote sensing monitoring and machine learning, characterized in that: include: A multi-source data acquisition module, used to acquire multi-source data and fuse the multi-source data to generate a unified fused data field; a target recognition module, configured to recognize maritime targets based on the fused data field and extract the spatial position, length, speed, and heading of the targets; A relative motion modeling module is used to construct a relative velocity vector between the platform and the target, and calculate the rate of change of the distance between the platform and the target based on the relative velocity vector, and then determine the potential threat target according to the rate of change of the distance; A risk level assessment module is used to predict the future trajectory of the platform and the potential threat target, and to construct a collision risk function based on the future trajectory and obtain a corresponding risk score; Input the target feature vector set of potential threat targets into the trained risk classification model to output the predicted risk level. Use the fusion rule to fuse the risk score and the predicted risk level to obtain the final risk level result. The early warning response module is used to implement graded responses based on the risk level of potential threat targets.

2. The offshore converter platform anti-collision warning system based on remote sensing monitoring and machine learning according to claim 1, characterized in that: The multi-source data acquisition module unifies the multi-source data into the same coordinate system and time axis through time synchronization and spatial registration; the multi-source data includes SAR images, optical remote sensing images, drone inspection videos, AIS ship trajectory data and platform status information.

3. The offshore converter platform anti-collision warning system based on remote sensing monitoring and machine learning according to claim 1, characterized in that: The target recognition module uses a deep neural network trained based on remote sensing features and combines it with a multi-scale attention mechanism to analyze the fused data field to identify and extract the spatial position, length, speed and heading of the target.

4. The offshore converter platform anti-collision warning system based on remote sensing monitoring and machine learning according to claim 1, characterized in that: The relative motion modeling module extracts the position, velocity, and heading of the platform; establishes a relative velocity vector and a relative position vector between the platform and the target; calculates the rate of change of the distance between the platform and the target based on the relative velocity vector and the relative position vector; and determines that the target is a potential threat target when the rate of change of the distance is positive and the absolute value of the rate of change of the distance is greater than a preset value.

5. The offshore converter platform anti-collision warning system based on remote sensing monitoring and machine learning according to claim 1, characterized in that: The risk level assessment module constructs future position prediction functions for the target and the potential threat platform based on the current position and velocity vectors of the platform and the potential threat target, respectively; constructs a trajectory intersection distance function for the platform and the potential threat target at any future time based on the future position prediction functions of the target and the potential threat platform; and constructs a collision risk function by integrating the trajectory intersection distance function within a prediction time window, thereby obtaining a risk integral for the prediction time window; The risk score and the predicted risk level output by the risk classification model are fused using a fusion rule to obtain a final risk level result; The fusion rule includes a logic priority rule, a model priority rule or a weighted average scoring fusion rule.

6. The offshore converter platform anti-collision warning system based on remote sensing monitoring and machine learning according to claim 5, characterized in that: The early warning response module implements a graded response based on the risk level results. For potential threat targets with a high risk level result, automatic broadcasting and obstacle avoidance control are triggered accordingly; For potential threat targets with a medium risk level, manual monitoring prompts are triggered; for potential threat targets with a low risk level, log archiving is triggered.

7. A collision avoidance warning method for offshore converter platforms based on remote sensing monitoring and machine learning, characterized in that: The anti-collision warning system for offshore converter platforms based on remote sensing monitoring and machine learning according to any one of claims 1 to 6 is implemented, and the anti-collision warning method includes: Acquiring multi-source data and fusing the multi-source data to generate a unified fused data field; Identify maritime targets based on the fused data field and extract the spatial position, length, speed and heading of the targets; Constructing a relative velocity vector between the platform and the target, and calculating a rate of change of the distance between the platform and the target based on the relative velocity vector, and then determining a potential threat target based on the rate of change of the distance; Predicting the future trajectories of the platform and potential threat targets, and constructing a collision risk function based on the future trajectories and obtaining a corresponding risk score; Input the target feature vector set of potential threat targets into the trained risk classification model to output the predicted risk level; and use the fusion rule to fuse the risk score and the predicted risk level to obtain the final risk level result; Implement graded responses based on the risk level of potential threat targets.

8. The method for anti-collision warning of an offshore converter platform based on remote sensing monitoring and machine learning according to claim 7, characterized in that: Constructing a relative velocity vector between the platform and the target, calculating a rate of change of the distance between the platform and the target based on the relative velocity vector, and then determining a potential threat target based on the rate of change of the distance, including: Extract the platform's position, velocity, and heading, and establish the relative velocity vector and relative position vector between the platform and the target; Calculating a rate of change of a distance between the platform and the target based on the relative velocity vector and the relative position vector; When the distance change rate is positive and the absolute value of the distance change rate is greater than a preset value, the target is determined to be a potential threat target.

9. The method for anti-collision warning of an offshore converter platform based on remote sensing monitoring and machine learning according to claim 7, characterized in that: Predicting the future trajectory of the platform and the potential threat target, and constructing a collision risk function based on the future trajectory and obtaining the corresponding risk score include: Based on the current position and velocity vectors of the platform and potential threat target, a function is constructed to predict the future position of the target and the platform. Constructing a trajectory intersection distance function between the platform and the potential threat target at any time in the future based on the future position prediction function; A collision risk function is constructed by integrating the trajectory intersection distance function within the prediction time window, thereby obtaining a risk integral for the prediction time window.

10. The method for anti-collision warning of an offshore converter platform based on remote sensing monitoring and machine learning according to claim 7, characterized in that: Implementing a graded response based on the risk level of potential threats to the target includes: For potential threat targets with a high risk level, automatic broadcasting and obstacle avoidance control will be triggered accordingly; For potential threat targets with a medium risk level, manual monitoring prompts will be triggered accordingly; For potential threat targets with a low risk level, log archiving processing is triggered accordingly.

Images