Sea reclamation area illegal behavior supervision method and system

Abstract

The invention discloses a method and system for supervising violation behaviors in a sea reclamation area, and relates to the field of marine resource supervision, and the method comprises the steps: collecting an image of a monitored area in real time through a camera; performing specific target object detection on the monitoring area image through a deep learning model; mapping the image coordinates to geographic coordinates based on a PnP algorithm and a digital elevation model so as to realize spatial positioning of the target object; and generating graded early warning information according to the multi-source data and pushing the graded early warning information to a supervision end. According to the technical scheme of the invention, the method can remarkably improve the recognition speed and response efficiency of the suspected illegal behaviors in the historical reclamation area, and is higher in application value.

Description

Technical Field

[0001] The present invention relates to the field of marine resource supervision, and in particular to a method and system for supervising illegal activities in a reclamation area. Background Art

[0002] As an important means for coastal areas to expand land resources and promote economic development, land reclamation plays an important role in the development of marine resources. However, in recent years, problems such as ecological damage, water quality deterioration, and reduction of biological habitats caused by illegal and irregular land reclamation activities have become increasingly prominent, posing a serious threat to the sustainability of marine ecosystems. In response to such problems, relevant control measures have been introduced one after another, but traditional regulatory measures still face many bottlenecks at the technical level and are difficult to meet the needs of efficient and accurate supervision.

[0003] At present, the supervision of land reclamation mainly relies on two methods: manual inspection and remote sensing image analysis. Manual inspection requires a lot of manpower, and has problems such as limited coverage, poor timeliness, and high cost. It is also easily restricted by bad weather or complex terrain, and it is difficult to achieve all-weather, full-region dynamic monitoring. Although remote sensing imaging technology can expand the monitoring scope, its application is subject to meteorological conditions, and there are obvious deficiencies in target recognition accuracy and real-time performance. Therefore, based on this problem, this application proposes a method and system for supervising illegal behaviors in land reclamation areas. Summary of the invention

[0004] Technical Purpose In order to solve the above problems, the purpose of the present invention is to provide a method and system for supervising illegal behaviors in reclamation areas, aiming to solve the problems of low recognition efficiency, insufficient positioning accuracy, poor dynamic adaptability, etc. caused by reliance on manual inspections and remote sensing images in traditional reclamation supervision. By integrating image recognition and video spatial positioning technology, an intelligent supervision system is constructed to achieve real-time and accurate monitoring of illegal and irregular behaviors in historical reclamation areas, automated early warning and multi-source data linkage analysis, thereby significantly improving the supervision response speed, reducing labor costs, and ensuring the sustainability of marine ecological protection.

[0005] Technical Solution In order to achieve the above-mentioned purpose, the present invention provides a method and system for supervising illegal behaviors in reclamation areas. The method and system adopt the Swin Transformer target recognition model as the basic model, improve the recognition accuracy of the target object through self-supervised comparative learning and historical data fine-tuning, and introduce a feedback incremental learning dynamic optimization model to enhance the system's adaptability to complex scenes; in terms of spatial positioning, the PnP algorithm and the digital elevation model are combined to achieve high-precision mapping of the target's geographic coordinates through camera parameter calibration and automatic offset correction technology; hierarchical warning information is generated based on the linkage analysis of the target type and ownership data, and integrated into the intelligent control ocean platform to support rapid judgment and law enforcement response. Through the collaboration of multiple technologies, this solution realizes a full-process intelligent supervision closed loop from data collection, target recognition, spatial positioning to warning push.

[0006] In a first aspect, the present invention provides a method for supervising illegal activities in a reclamation area, comprising: Collect images of the monitoring area in real time through cameras; Detect specific targets in the image data of the monitored area through deep learning models; Based on the PnP algorithm and digital elevation model, the image coordinates are mapped to geographic coordinates to achieve the spatial positioning of the target object; Generate graded warning information based on multi-source data and push it to the regulatory end.

[0007] Furthermore, the base model of the deep learning model is a Swin Transformer target recognition model, which is pre-trained for self-supervised contrastive learning based on an unlabeled Internet dataset; In the process of contrastive learning, the InfoNCE loss function is used to optimize the sample similarity, and the formula is:

[0008] Where L is the loss function; is an exponential function used to calculate the similarity score in the loss function; and is a positive sample pair; is a sample set including positive sample pairs and K negative samples; is the similarity function; is the temperature parameter, which is used to control the scaling of similarity.

[0009] Furthermore, the implementation step of the self-supervised contrastive learning also includes performing data enhancement on the same image to generate positive sample pairs, and generating negative sample pairs on different images.

[0010] Furthermore, the method uses manual review results as new training samples and adopts a proximal strategy optimization algorithm to balance model parameter updates, dynamically improving the model's adaptability to complex scenarios and recognition accuracy.

[0011] Furthermore, after the self-supervised comparative learning is completed, the model is fine-tuned based on the real data of historical land reclamation to improve the model's accuracy in identifying suspected illegal and irregular behaviors, and combined with feedback-based incremental learning, the model parameters are continuously optimized through manual feedback data.

[0012] This method can quickly and accurately identify suspected illegal and irregular behaviors in historical reclamation areas, greatly shorten the identification cycle, reduce verification manpower, and significantly improve supervision efficiency; through continuous learning and adaptation to new scenarios, the model's recognition ability is constantly improved, which can better meet actual supervision needs and provide strong technical support for intelligent marine supervision.

[0013] Furthermore, the method establishes a world coordinate system based on manually calibrated reference points and DEM data, uses the PnP algorithm to solve the camera extrinsic matrix, and completes the coordinate mapping in combination with the camera intrinsic matrix. The mapping process is shown below:

[0014] In the formula, is the scaling parameter; is the camera intrinsic parameter matrix; is the camera pose parameter; The coordinates of the calibration point are converted into three-dimensional vectors of the ENU coordinate system; is the set of points corresponding to image coordinates.

[0015] This technology ensures that the spatial position information of specific targets can be accurately obtained even in complex terrain and harsh environments, thus improving the accuracy and stability of spatial positioning.

[0016] Furthermore, the method also includes an automatic offset correction algorithm for correcting the camera offset caused by environmental factors, specifically including: calculating the homography matrix by matching feature points , the expression is ; Rotate the matrix Decompose it into horizontal and vertical rotation components; adjust the camera posture by calling the gimbal interface to eliminate the deviation.

[0017] Through the automatic offset correction algorithm, the impact of camera offset on positioning accuracy is effectively avoided, ensuring long-term stable target monitoring and positioning, providing reliable geographic data support for reclamation supervision, effectively improving the scientific nature and effectiveness of supervision, and providing important guarantees for the rational use of marine resources and ecological protection.

[0018] Furthermore, the method divides warning levels according to the relevance between the target type and the behavior, and the warning levels include: At the core level, an advanced warning is generated when construction machinery appears in an unconfirmed area; Important level: when temporary structures or piles of earth and gravel appear, a medium-level warning is generated; At the general level, a low-level warning is generated when construction personnel are active.

[0019] Furthermore, the method determines whether the target object is within the scope of legal sea use by conducting a linkage analysis of sea area ownership data, spatial design layers and historical verification data.

[0020] Furthermore, the method also includes a feature fusion method, which extracts the texture features of the visible light image and the temperature difference features of the infrared image through a dual-branch network, and generates joint features through adaptive weight fusion. The feature fusion formula is:

[0021] In the formula, is the dynamic weight coefficient; is the texture feature vector of the visible light image; is the temperature difference feature vector of the infrared image; is a joint feature.

[0022] The enhanced recognition model based on multi-spectral feature fusion integrates visible light and infrared spectral data, combines dynamic weight fusion and cross-domain feature alignment, significantly improves the target recognition capability in complex environments, effectively enhances the robustness of the system in low-light or meteorological interference scenarios, and provides reliable technical support for all-weather monitoring of illegal and irregular behaviors.

[0023] Furthermore, the method also includes a positioning compensation method, which eliminates the influence of dynamic environmental factors on positioning accuracy by constructing a dynamic error correction function, and the dynamic error correction function formula is:

[0024] In the formula, , and is the calibration factor; is the rate of change of tidal height; is the wind speed; is the camera vibration amplitude; is the dynamic error correction output.

[0025] The real-time positioning compensation algorithm based on dynamic environmental feedback integrates tidal, wind speed and vibration parameter modeling error corrections and embeds a dynamic adjustment mechanism for camera external parameters, which significantly suppresses the impact of environmental interference on positioning accuracy. It achieves high-precision and high-stability real-time spatial positioning, providing key guarantees for law enforcement response and accurate verification.

[0026] In a second aspect, the present invention further provides a system for supervising illegal activities in reclamation areas. The system is based on the method in the first aspect and comprises: An image recognition module is used to detect specific targets in monitoring images of historical reclamation areas through a deep learning model; the specific targets include construction machinery, temporary structures and construction personnel; Spatial positioning module, used to determine the geographic coordinates of the target object based on the PnP algorithm and digital elevation model; The intelligent early warning module is used to generate graded early warning information based on image recognition results and spatial positioning data, and further push it to the supervision platform.

[0027] In a third aspect, the present invention further provides a computer device, comprising a management platform and a memory, wherein the management platform is connected to the memory, the memory is used to store computer programs, and the management platform is used to execute the computer programs stored in the memory, so that the computer device executes at least one step of the aforementioned method for supervising illegal activities in reclamation areas.

[0028] In a fourth aspect, the present invention further provides a computer-readable storage medium, wherein a computer program is stored in the computer-readable storage medium, and when the computer program is executed by a management platform, at least one step of the aforementioned method for supervising illegal activities in reclamation areas is implemented.

[0029] The present invention adopts the Swin Transformer target recognition model as the basic training model, combines self-supervised contrast learning to pre-train the model, and uses feedback incremental learning technology to enable the model to continuously adapt to new scenes through artificial feedback based on actual supervision data; by establishing a mapping relationship between world coordinates and image coordinates, using DEM elevation measurement data and manually calibrated reference points, combined with the PnP algorithm to estimate the camera's posture, thereby determining the geographical location of a specific target; extracting the texture features of visible light images and the temperature difference features of infrared images through a dual-branch network, and generating joint features through adaptive weight fusion; and eliminating the influence of dynamic environmental factors on positioning accuracy by constructing a dynamic error correction function. The method and system have a wide range of application scenarios, can significantly improve the recognition speed and response efficiency of suspected illegal and irregular behaviors in historical reclamation areas, and have high application value.

[0030] Beneficial Effects By implementing the above-mentioned method and system for supervising illegal behaviors in reclamation areas provided by the present invention, the following technical effects are achieved: (1) The present invention uses the Swin Transformer target recognition model as the basic training model, combines self-supervised comparative learning to pre-train the model, and uses feedback-based incremental learning technology to enable the model to continuously adapt to new scenarios through manual feedback based on actual regulatory data. This method can quickly and accurately identify suspected illegal and irregular behaviors in historical reclamation areas, greatly shorten the identification cycle, reduce verification manpower, and significantly improve regulatory efficiency; through continuous learning and adaptation to new scenarios, the model's recognition ability is continuously improved, which can better meet actual regulatory needs and provide strong technical support for intelligent marine supervision.

[0031] (2) By establishing a mapping relationship between world coordinates and image coordinates, using DEM elevation measurement data and manually calibrated reference points, combined with the PnP algorithm, the camera posture is estimated to determine the geographic location of a specific target. In addition, in order to address the problem that a fixed camera may be offset due to external factors, an automatic offset correction algorithm is designed to ensure the accuracy of video spatial positioning by real-time detection and adjustment of the camera posture. This technology ensures that the spatial position information of a specific target can be accurately obtained in complex terrain and harsh environments, improving the accuracy and stability of spatial positioning; through the automatic offset correction algorithm, the influence of camera offset on positioning accuracy is effectively avoided, ensuring long-term stable target monitoring positioning, providing reliable geographic data support for reclamation supervision, effectively improving the scientificity and effectiveness of supervision, and providing important guarantees for the rational use of marine resources and ecological protection.

[0032] (3) The texture features of visible light images and the temperature difference features of infrared images are extracted through a dual-branch network, and joint features are generated through adaptive weight fusion. The enhanced recognition model based on multi-spectral feature fusion integrates visible light and infrared spectral data, combines dynamic weight fusion and cross-domain feature alignment, significantly improves the target recognition capability in complex environments, effectively enhances the robustness of the system in low-light or meteorological interference scenarios, and provides reliable technical support for all-weather monitoring of illegal and irregular behaviors.

[0033] (4) The influence of dynamic environmental factors on positioning accuracy is eliminated by constructing a dynamic error correction function. The real-time positioning compensation algorithm based on dynamic environmental feedback integrates the error correction value of tide, wind speed and vibration parameters, and embeds the dynamic adjustment mechanism of camera external parameters, which significantly suppresses the influence of environmental interference on positioning accuracy and realizes high-precision and high-stability real-time spatial positioning, providing key guarantees for law enforcement response and accurate verification. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to make the above-mentioned reclamation area illegal behavior supervision method and system of the present invention more obvious and easy to understand, the drawings required for use in the specific implementation of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0035] Figure 1 A schematic diagram showing the flow of the method for supervising illegal activities in the reclamation area; Figure 2 A schematic diagram showing the technical route of the present invention; Figure 3 A schematic diagram showing the image recognition process; Figure 4 A schematic diagram showing the topographic DEM point A to the image coordinate a; Figure 5 Schematic diagram showing coordinate changes. DETAILED DESCRIPTION

[0036] Embodiment 1: A method and system for supervising illegal activities in a reclamation area is provided. The process of the method for supervising illegal activities in a reclamation area is as follows: Figure 1 As shown, it includes: collecting images of the monitoring area in real time through cameras; detecting specific targets in the monitoring area image data through deep learning models; mapping image coordinates to geographic coordinates based on the PnP algorithm and digital elevation model to achieve spatial positioning of the target; generating graded warning information based on multi-source data and pushing it to the supervision end. The specific steps are as follows: a method and system for supervising illegal behaviors in reclamation areas.

[0037] The technical route of the present invention is as follows Figure 2 shown.

[0038] The basic model of the deep learning model is the Swin Transformer target recognition model, which is pre-trained for self-supervised contrastive learning based on a large-scale Internet unlabeled dataset; In the process of contrastive learning, the InfoNCE loss function is used to optimize the similarity relationship between samples. By maximizing the similarity between positive sample pairs and pushing the similarity between negative sample pairs, the model can learn the subtle but important differences between images. The formula is:

[0039] Where L is the loss function; is an exponential function used to calculate the similarity score in the loss function; and is a positive sample pair; is a sample set including positive sample pairs and K negative samples; is the similarity function; is the temperature parameter used to control the scaling of similarity. The application of the InfoNCE loss function makes the model better at identifying small changes in the reclamation area, such as soil changes and water level changes, which are easily ignored in traditional methods. Verification shows that the model is suitable for detecting early violations and can even warn at a stage that is difficult for humans to detect. Therefore, the supervision system of this application can significantly reduce the harm of potential violations in the reclamation area.

[0040] The implementation steps of the self-supervised contrastive learning also include performing data enhancement on the same image to generate positive sample pairs and generating negative sample pairs from different images, the purpose of which is to maximize the similarity of the positive sample pairs while minimizing the similarity of the negative sample pairs, thereby obtaining a robust feature representation without the need for labeled data.

[0041] The method uses manual review results as new training samples and adopts a proximal strategy optimization algorithm to balance model parameter updates, dynamically improving the model's adaptability to complex scenarios and recognition accuracy.

[0042] After the self-supervised comparative learning is completed, the model is fine-tuned based on the real data of historical land reclamation to improve the model's recognition accuracy of suspected illegal and irregular behaviors. Combined with feedback-based incremental learning, the model parameters are continuously optimized through manual feedback data, so that the model can continuously adapt to new scenarios and continuously improve recognition accuracy.

[0043] The process of the method for detecting a specific target is as follows Figure 3 As shown in the figure, it specifically includes: slicing the monitoring image in advance, dividing the image into several image blocks, and using the convolution layer and the hierarchical attention mechanism to extract the features of each image block; after extracting the image features, generating candidate boxes through the region proposal network, these candidate boxes represent the areas where specific targets may exist in the image, and then further adjusting and regressing the positions of these candidate boxes to ensure that they accurately cover the specific targets; after the candidate boxes are generated, the specific targets are identified through the classification network, and the objects in the candidate boxes are classified into multiple categories using the Softmax classifier, so as to clarify the types of activities suspected of illegal and irregular behaviors in the monitoring area.

[0044] The non-maximum suppression algorithm is used to filter out overlapping or redundant candidate boxes to improve the accuracy of the recognition results.

[0045] The method establishes a world coordinate system based on manually calibrated reference points and DEM data, uses the PnP algorithm to solve the camera extrinsic matrix, and completes the coordinate mapping in combination with the camera intrinsic matrix. The conversion process from terrain DEM point A to image coordinate a is as follows: Figure 4 As shown, the specific algorithm is as follows: Calibration points , is the three-dimensional vector of the calibration point coordinates converted to the ENU coordinate system; let the camera extrinsic parameter be , which is one of is a 3x3 rotation matrix, is a 3x1 displacement vector; These points correspond to the points on the image coordinates; the mapping process is as follows:

[0046] In the formula, is the scaling parameter; is the camera intrinsic parameter matrix; is the camera's pose parameter; The coordinates of the calibration point are converted into three-dimensional vectors of the ENU coordinate system; is the set of points corresponding to the image coordinates.

[0047] The method also includes an automatic offset correction algorithm for correcting the camera offset caused by environmental factors. The coordinate change process is as follows: Figure 5 As shown, it specifically includes: calculating the homography matrix by matching feature points , the expression is ; Rotate the matrix Decompose it into horizontal and vertical rotation components; adjust the camera posture by calling the gimbal interface to eliminate the deviation.

[0048] The method divides warning levels according to the relevance between the target type and the behavior, and the warning levels include: At the core level, an advanced warning is generated when construction machinery appears in an unconfirmed area; Important level: when temporary structures or piles of earth and gravel appear, a medium-level warning is generated; At the general level, a low-level warning is generated when construction personnel are active.

[0049] The method determines whether the target object is within the legal sea use range by linking and analyzing sea area ownership data, spatial design layers and historical verification data.

[0050] Embodiment 2: This embodiment uses a certain district in a certain province as the research area to provide a method for supervising illegal behaviors in reclamation areas. In recent years, there have been frequent marine development activities in a certain district, mainly involving port expansion and tourism facility development, which have exerted great pressure on the local marine ecological environment. Due to its complex geographical environment and high-intensity development conditions, it has become a key area for reclamation supervision.

[0051] The accuracy of image recognition is a key indicator of the intelligent supervision method for land reclamation. Through the analysis and verification of the background data of the intelligent recognition and supervision application of a certain island in a certain district for one month, the results show that with the support of deep learning algorithms, the image recognition technology can achieve an average accuracy of more than 90% in identifying specific targets in historical land reclamation areas, as shown in Table 1.

[0052] Table 1. Statistical table of recognition accuracy of specific targets in historical reclamation areas

[0053] The accuracy of video spatial positioning is an important basis for intelligent supervision and inspection of land reclamation. The video positioning data of three points in a certain district and an island for intelligent recognition and supervision were analyzed and verified for one month. The results show that the error of video spatial positioning based on deep learning image recognition is about 3%, as shown in Table 2.

[0054] Table 2. Statistics of spatial positioning errors of specific targets in historical reclamation areas Test points Camera magnification Camera Angle Measuring point distance Measurement error Error rate Azimuth point 1 1 349,349 467m 13.9 m 3.0% Azimuth point 1 2 349,349 467m 10.2 m 2.2% Azimuth point 1 5 349,349 467m 8.2 m 1.8% Azimuth point 1 1 346,350 467m 13.8 m 3.0% Azimuth point 1 position 2 1 349,349 259m 14.1 m 5.4% Azimuth point 1 position 2 2 349,349 259m 15.3 m 5.9% Azimuth point 2 position 3 1 115,356 777 meters 12.2 m 1.6% Azimuth point 2 position 3 2 115,356 777 meters 18.4 m 2.4% Azimuth point 2 position 3 18 115,356 777 meters 11.1 m 1.4% total — — — — 3.0% The intelligent supervision application developed based on the intelligent marine control system can realize the early warning push of suspected illegal and irregular behaviors. The monitoring equipment detects specific targets through deep learning image recognition technology and pushes early warning information to the intelligent marine control system within 10 seconds of detecting the target. The discovery of suspected illegal and irregular behaviors is shortened from 15 days to 1 day, the response time is reduced from 7 days to 1 day, and the manpower for verification work is reduced from 5 man-days to 0.5 man-days, as shown in Table 3.

[0055] Table 3. Comparison of early warning response time for suspected illegal and irregular behaviors in historical reclamation areas Regulatory measures Research and judgment cycle Response time (days) On-site verification (man-days) Manual investigation ≥ 1 month ≤15 ≤5 Remote sensing images ≥ half a month ≤7 ≤5 Intelligent early warning ≤1 day ≤1 ≤0.5 This embodiment constructs an intelligent supervision method for reclamation based on deep learning image recognition and video spatial positioning technology, and relies on the intelligent control ocean system to conduct regional verification of the intelligent supervision method for reclamation. Supervisors can quickly judge based on image recognition, spatial positioning, and combined with video playback and other functions, and conduct timely and effective verification of suspected illegal and irregular behaviors, which significantly improves the supervision efficiency. The analysis and verification results show that the recognition accuracy of this method is more than 90%, and the spatial positioning accuracy error rate is about 3%. It has certain feasibility and reliability in the supervision of historical reclamation, and can effectively respond to various reclamation supervision needs. It provides a new idea and model for sea area supervision, realizes intelligent supervision of suspected illegal and irregular behaviors in historical reclamation areas, and has certain reference value for the intelligentization and modernization of sea use supervision.

[0056] Embodiment 3: Based on the above-mentioned embodiment, the method also includes a feature fusion method, which constructs a multimodal feature fusion network by integrating multispectral imaging technology.

[0057] Multi-spectral cameras are deployed to synchronously acquire multi-band images of the target area.

[0058] The texture features of the visible light image and the temperature difference features of the infrared image are extracted through a dual-branch network, and the joint features are generated through adaptive weight fusion. The feature fusion formula is:

[0059] In the formula, is a dynamic weight coefficient, which is automatically generated by the attention mechanism according to the ambient light conditions. Its value range is [0,1]. It is close to 1 when there is sufficient light and close to 0 when there is low light or fog. is the texture feature vector of the visible light image, extracted through the convolution layer; is the temperature difference feature vector of the infrared image, extracted through the convolution layer; is a joint feature.

[0060] A spectral feature adaptation layer is added at the end of the Swin Transformer model, and the multi-spectral feature alignment is optimized by contrast loss function. The loss function formula is:

[0061] In the formula, It is a feature mapping function, which is implemented using a fully connected network and is used to map infrared features to visible light feature space to achieve cross-spectral domain alignment; is the visible light vector of the i-th group of samples; is the infrared feature vector of the i-th group of samples; is the output of the loss function.

[0062] Verification shows that under the condition of obtaining an average error similar to that of the above embodiment, in foggy or night scenes, the target recognition accuracy rate is increased from 90% to 94%, the false alarm rate is reduced by 25%, and the missed detection rate of temporary structures is reduced from 10% to 6%, indicating that the application of feature fusion methods helps to enhance the robustness of the system in complex environments. The addition of a spectral feature adaptation layer can also solve nighttime violations that are difficult to detect in real time and are highly concealed, thereby strengthening the shortcomings of relying on traditional manual inspections, satellite remote sensing, drones and other technical applications, and greatly improving the efficiency of nighttime supervision of violations.

[0063] Embodiment 4: On the basis of the above-mentioned embodiments, the method proposes a real-time compensation model for the influence of dynamic environmental factors such as tide and wind speed on positioning accuracy.

[0064] The sensors collect tide height, wind speed and camera vibration amplitude in real time.

[0065] Construct a dynamic error correction function and integrate environmental parameters and historical positioning data. The formula is as follows:

[0066] In the formula, , and To calibrate the coefficients, ridge regression is used to learn from historical data, corresponding to the influence weights of tide, wind speed, and vibration; is the tidal height change rate, reflecting the impact of dynamic changes in water level on positioning; It is the real-time wind speed, which affects the camera stability; is the camera vibration amplitude, measured by the acceleration sensor; is the dynamic error correction output.

[0067] The dynamic error correction function is embedded in the PnP algorithm to correct the camera extrinsic matrix and displacement vector. The formula is as follows:

[0068] In the formula, and are attenuation factors, which are the correction amplitude control parameters of the rotation matrix and displacement vector, respectively, and their value range is [0,0.05] to prevent over-correction; is the rotation matrix; It is the output after the dynamic error correction function is embedded in the PnP algorithm.

[0069] For example, suppose the environmental parameters at a certain moment are: Tidal rate of change ; Wind speed ; Vibration amplitude ;Calibration coefficient , , ; Attenuation factor , .

[0070] Calculate the environmental error correction:

[0071] Calculate the corrected camera extrinsics:

[0072]

[0073] The effect of the positioning compensation method is shown in Table 4.

[0074] Table 4. Summary of the effects of positioning compensation methods

[0075] According to the experimental table, under dynamic environments such as tides and strong winds, the positioning error rate dropped from an average of 4.2% to 1.4%, a decrease of 66.7%, and the target tracking continuity increased from 75% to 98%, meeting the real-time law enforcement needs. The system response time was shortened by 20%, supporting rapid early warning, and the dynamic environment feedback compensation algorithm significantly suppressed the influence of interference factors such as tides and wind speed on positioning accuracy. The error rate was reduced to 1.4%, and the stability and response speed were improved simultaneously, providing a highly reliable spatial positioning guarantee for reclamation supervision.

[0076] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable non-transient storage media containing computer-usable program code.

[0077] The present invention can provide computer program instructions to a management platform of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the management platform of the computer or other programmable data processing device produce a device for implementing the system.

[0078] These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory produce a product including an instruction device that implements the functions of the system.

[0079] These computer program instructions may also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, whereby the instructions executed on the computer or other programmable device provide steps for implementing the functions of the system.

Claims

1. A method for supervising illegal activities in reclamation areas, characterized in that: include: Collect images of the monitoring area in real time through cameras; Detect specific targets in the monitored area images through deep learning models; Based on the PnP algorithm and digital elevation model, the image coordinates are mapped to geographic coordinates to achieve the spatial positioning of the target object; Generate graded warning information based on multi-source data and push it to the regulatory end.

2. The method according to claim 1, characterized in that: The basic model of the deep learning model is the Swin Transformer target recognition model, which performs self-supervised comparative learning based on an unlabeled Internet dataset; In the contrastive learning process, the InfoNCE loss function is used to optimize the sample similarity, and the formula is: Where L is the loss function; is an exponential function; and is a positive sample pair; is a sample set including positive sample pairs and K negative samples; is the similarity function; is the temperature parameter.

3. The method according to claim 2, characterized in that: The method uses manual review results as new training samples and adopts a proximal strategy optimization algorithm to balance model parameter updates.

4. The method according to claim 1, characterized in that: The method establishes a world coordinate system based on manually calibrated reference points and DEM data, uses the PnP algorithm to solve the camera extrinsic matrix, and completes coordinate mapping in combination with the camera intrinsic matrix.

5. The method according to claim 1, characterized in that: The method determines whether the target object is within the legal sea use range by linking and analyzing sea area ownership data, spatial design layers and historical verification data.

6. The method according to any one of claims 1 to 5, characterized in that: The method also includes a feature fusion method, which extracts the texture features of the visible light image and the temperature difference features of the infrared image through a dual-branch network, and generates a joint feature through adaptive weight fusion. The feature fusion formula is: In the formula, is the dynamic weight coefficient; is the texture feature vector of the visible light image; is the temperature difference feature vector of the infrared image; is a joint feature.

7. The method according to any one of claims 1 to 5, characterized in that: The method also includes a positioning compensation method, which eliminates the influence of dynamic environmental factors on positioning accuracy by constructing a dynamic error correction function, and the dynamic error correction function formula is: In the formula, , and is the calibration factor; is the rate of change of tidal height; is the wind speed; is the camera vibration amplitude; is the dynamic error correction output.

8. A system for supervising illegal activities in reclamation areas, characterized by: The implementation of the system is based on the method described in any one of claims 1 to 7: The system comprises: An image recognition module is used to detect specific targets in images of historical reclamation areas using a deep learning model; Spatial positioning module, used to determine the geographic coordinates of the target object based on the PnP algorithm and digital elevation model; The intelligent warning module is used to generate graded warning information based on image recognition results and spatial positioning data.

9. A computer device, comprising a management platform and a memory, wherein the management platform is connected to the memory, and the memory is used to store a computer program, characterized in that: The management platform is used to execute the computer program stored in the memory, so that the computer device performs at least one step in the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, wherein a computer program is stored in the computer-readable storage medium, characterized in that: When the computer program is executed, at least one step of the method according to any one of claims 1 to 7 is implemented.

Images