Marine usage behavior monitoring method and system based on intelligent identification algorithm

Abstract

The invention discloses a sea usage behavior monitoring method and system based on an intelligent identification algorithm, and relates to the technical field of marine resource management, and the system comprises a behavior detection module which is used for carrying out the snapshot of a sea usage behavior through a camera, and carrying out the recognition of image data through a target detection algorithm; the space positioning module is used for calculating three-dimensional coordinates of pixel points in the image according to the position information of the control points; the offset detection module is used for performing feature point matching on the snapshot image and the preset point image to determine offset generated by the camera; and the visual interaction module is used for presenting the detection data and the early warning information and providing auxiliary decision making. According to the technical scheme of the invention, the method can achieve the precise management and control of the sea usage behavior of the historical reclamation region, builds a complete monitoring system, and has a higher application value.

Description

Technical Field

[0001] The present invention relates to the technical field of marine resource management, and in particular to a method and system for monitoring sea use behavior based on an intelligent recognition algorithm. Background Art

[0002] With the development of the economy, coastal areas have continuously increased their efforts to develop marine resources, which has led to frequent illegal activities such as illegal development, arbitrary reclamation, and illegal mining of sea sand, which has seriously damaged the balance between marine resources and the ecological environment. In order to effectively manage and protect marine resources, it is necessary to increase the monitoring of sea use behaviors in reclamation areas. Traditional marine management methods mainly rely on patrols by law enforcement ships, public reports, radar monitoring, etc. These methods are costly, inefficient, and have serious lags and high missed detection rates. Although the use of satellite remote sensing and drones has improved the efficiency of their management, the work of manual review is still very arduous in the face of massive image data. In summary, the monitoring of sea use behavior faces problems such as many management objects, a wide range of management, a large workload, a single management method, and great difficulty in law enforcement. Therefore, based on this problem, this application proposes a method and system for monitoring sea use behavior based on an intelligent recognition algorithm. Summary of the invention

[0003] Technical Purpose In order to solve the above problems, the purpose of the present invention is to provide a method and system for monitoring sea use behavior based on an intelligent recognition algorithm. By accessing functions such as video intelligent recognition and drone inspection, an auxiliary decision-making tool is established to achieve precise control of different regions, thereby improving the efficiency of sea use behavior control. This method and system have a wide range of application scenarios, and can achieve precise control of sea use behavior in historical reclamation areas, establish a complete monitoring system, and realize automatic early warning of illegal sea use behavior.

[0004] Technical Solution In order to achieve the above-mentioned objectives, the present invention provides a method and system for monitoring sea use behavior based on an intelligent recognition algorithm, which captures the sea use behavior through a camera and identifies the image data through a target detection algorithm, infers the three-dimensional coordinates of the pixel points in the image according to the position information of the control point, and determines the offset caused by the camera by matching the feature points of the captured image and the preset point image, presents the detection data and early warning information, and provides auxiliary decision-making, thereby realizing precise control of the sea use behavior in the historical reclamation area.

[0005] In a first aspect, the present invention provides a sea use behavior monitoring system based on an intelligent recognition algorithm, comprising: Behavior detection module, used to capture sea-use behaviors through cameras and identify image data through target detection algorithms; A spatial positioning module is used to calculate the three-dimensional coordinates of the pixel points in the image based on the position information of the control points; An offset detection module is used to determine the offset of the camera by matching the captured image with the preset point image; Visual interaction module, used to present detection data and warning information and provide auxiliary decision-making.

[0006] Furthermore, the behavior detection module analyzes whether there is any sea use behavior in the captured image through a target detection algorithm, and records the location coordinates of the area. If the sea use behavior occurs in an unauthorized land area, the sea use behavior is determined to be an illegal behavior.

[0007] Furthermore, the steps of the target detection algorithm include: collecting and annotating image data sets, and dividing the data sets into training sets and test sets; converting input images into feature map representations through neural networks; estimating and screening the positions that may contain targets in the image based on feature maps through RPN; adjusting and correcting the positions of candidate boxes so that the candidate boxes can accurately cover the target objects; classifying the regressed target boxes through softmax and fully connected neural networks; performing non-maximum suppression operations on overlapping areas of multiple output target boxes to determine the final target box; training the model through the training set, and completing evaluation and adjustment on the test set. It reduces the need for manual inspections and improves the efficiency of management work through automated image analysis; identifies specific illegal feature elements through intelligent recognition algorithms, improving recognition accuracy; and can obtain image information of the monitoring area in real time and perform instant analysis, realizing the function of real-time warning.

[0008] Furthermore, the target detection algorithm converts the input image into a feature map representation through a neural network, wherein the skeleton network is Swin Transformer. The specific method includes: dividing the input image into multiple image blocks, and sharing some pixels between adjacent image blocks so that the algorithm has a strong semantic understanding ability; each image block is processed by a basic processing unit, and the processing unit includes a convolutional layer and a Swin Transformer structure to extract the feature representation of the image block; the Swin Transformer structure adopts a hierarchical attention mechanism so that the neural network can focus on important information at different levels in feature maps at different levels of abstraction; pre-training is performed on image data sets to improve the generalization ability of the neural network, and the feature representation ability of the neural network is further improved based on a large amount of monitoring materials through a self-supervised learning algorithm.

[0009] Furthermore, the spatial positioning module estimates the position, orientation, field of view and internal parameters of the camera through the least square method through the control points in the image and their position information in the three-dimensional map; the target location is imaged to the center of the image by controlling the horizontal and vertical rotation values ​​of the camera; the latitude and longitude data of the target location are converted into a Cartesian coordinate system, and the camera installation position and orientation are matched based on the altitude and ellipsoid height conversion; the point in the three-dimensional world is projected onto the two-dimensional image plane through the camera through the pinhole imaging method; the radial distortion and tangent distortion of the camera are corrected by the correction method to improve the accuracy of positioning; the optimal estimation of the camera parameters is solved by the numerical optimization method. It realizes high-precision positioning by realizing the conversion between image pixels and latitude and longitude positions.

[0010] Furthermore, the correction formulas for the radial distortion and tangent distortion of the camera are as follows:

[0011]

[0012] in, The horizontal axis coordinate of the distortion point; The vertical axis coordinate of the distortion point; Represents the original horizontal coordinate of the undistorted point; Represents the original vertical coordinate of the undistorted point; Represents the radial distance from the image center to the distortion point; , and Represents the radial distortion coefficient.

[0013] Furthermore, the offset detection module reserves a base map as a reference for offset detection. During the offset detection process, the camera parameters are adjusted to be consistent with the camera parameters in the initial state to capture a test image, and the conversion matrix of the spatial relationship between the reserved base map and the test image is calculated to determine the offset of the camera. It can automatically detect whether the camera is offset and calibrate it, ensuring the accuracy of the monitoring area and the accuracy of image positioning.

[0014] Furthermore, the premise of camera offset is that the camera will only have rotational offset but not translational offset.

[0015] Furthermore, the offset detection module cuts out the portion of the image with the OSD watermark to reduce the probability of matching the OSD watermark in the preset point as a feature point.

[0016] Furthermore, the offset detection module also includes a feature point matching noise removal algorithm, which selects points with higher matching confidence and selects matching points with most consistent change relationships as much as possible during the matching point selection process to facilitate subsequent calculations.

[0017] Furthermore, the visual interaction module presents the detection data of sea use behavior in the reclamation area in the form of an electronic map, generates early warning information by analyzing the detection data, and provides auxiliary decision-making based on the early warning information; users can retrieve detailed data on sea use behavior based on the early warning information, and can remotely view the monitoring screen of the monitoring equipment in real time to conduct remote evidence collection.

[0018] Furthermore, the visual interaction module includes two functional pages: full overview and project inspection. The full overview presents the overall warning situation of the monitored area, including five functional modules: warning overview, violation warning, video monitoring, airport and tools.

[0019] Furthermore, the system also includes a water body separation module, which extracts water body data by suppressing vegetation information through green band and near infrared band based on the spectral characteristic curve of water body in different bands, specifically including: The collected data is preprocessed, and the input image is separated into land and water using the MNDWI index. The MNDWI index calculation formula is as follows:

[0020] in, Indicates the reflectance value of the green band; Indicates the reflectance value in the mid-infrared band; Represents the normalization factor.

[0021] The outliers of the MNDWI index are judged and eliminated. The outlier judgment formula is:

[0022] in, represents a small positive number used to define the range of outliers; The outlier elimination formula is:

[0023] in, Represents the MNDWI index value of a single pixel; Indicates a specific coefficient.

[0024] It can accurately identify water bodies, improve monitoring accuracy, and enrich the management methods for illegal use of the sea by realizing real-time monitoring of water bodies to detect whether there is any illegal use of the sea.

[0025] Furthermore, the system also includes an isolation forest module, which randomly divides the data space by constructing multiple isolated trees based on the target detection algorithm, and determines abnormal data by calculating the path length of the data in each isolated tree. It can quickly process large-scale data sets, thereby saving computing resources and time; reduce the overfitting problem that may occur in a single model, and improve the generalization ability of the model; it can effectively control the false alarm rate and reduce unnecessary resource waste and response delays.

[0026] In a second aspect, the present invention further provides a method for monitoring sea use behavior based on an intelligent recognition algorithm, the method being based on the system described in the first aspect, comprising: The sea-use behavior is captured by the camera and the image data is recognized by the target detection algorithm; Calculate the three-dimensional coordinates of the pixel points in the image based on the position information of the control points; The offset of the camera is determined by matching the captured image with the preset point image; Present detection data and warning information and provide decision support.

[0027] In a third aspect, the present invention further provides a computer device, comprising a processor and a memory, wherein the processor is connected to the memory, the memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory, so that the computer device executes the aforementioned method for monitoring sea use behavior based on an intelligent recognition algorithm.

[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 processor, the aforementioned method for monitoring sea use behavior based on an intelligent recognition algorithm is implemented.

[0029] The present invention uses a camera to capture sea use behavior and identifies image data through a target detection algorithm, and infers the three-dimensional coordinates of pixel points in the image based on the position information of the control point; determines the offset generated by the camera by matching the captured image and the preset point image with feature points; extracts water body data by suppressing vegetation information through the green band and near-infrared band based on the spectral characteristic curve of the water body in different bands; and randomly divides the data space by constructing multiple isolated trees based on the target detection algorithm, and determines abnormal data by calculating the path length of the data in each isolated tree. The method and system have a wide range of application scenarios, realizing the precise control of sea use behavior in historical reclamation areas, establishing a complete monitoring system, and realizing automatic early warning of illegal sea use behavior.

[0030] Beneficial Effects By implementing the sea use behavior monitoring method and system based on intelligent recognition algorithm provided by the present invention, the following technical effects are achieved: (1) The camera captures the sea use behavior and the target detection algorithm identifies the image data, and the three-dimensional coordinates of the pixel points in the image are calculated based on the location information of the control point; through automated image analysis, the need for manual inspections is reduced and the efficiency of management work is improved; through intelligent recognition algorithms, specific illegal feature elements are identified, thereby improving the accuracy of recognition; the image information of the monitored area can be obtained in real time and analyzed immediately, realizing the function of real-time warning; by realizing the conversion between image pixels and longitude and latitude positions, high-precision positioning is achieved.

[0031] (2) Feature point matching is performed on the captured image and the preset point image to determine the offset of the camera; it can automatically detect whether the camera is offset and calibrate it, ensuring the accuracy of the monitored area and the accuracy of image positioning.

[0032] (3) Based on the spectral characteristic curves of water bodies in different bands, vegetation information is suppressed through the green band and near-infrared band to extract water body data; it can accurately identify water bodies and improve the accuracy of monitoring. By realizing real-time monitoring of water bodies to detect whether there are any illegal uses of the sea, it enriches the management methods for illegal uses of the sea.

[0033] (4) Based on the target detection algorithm, multiple isolated trees are constructed to randomly divide the data space, and the path length of the data in each isolated tree is calculated to determine the abnormal data; it can quickly process large-scale data sets, thereby saving computing resources and time; it reduces the overfitting problem that may occur in a single model and improves the generalization ability of the model; it can effectively control the false alarm rate and reduce unnecessary resource waste and response delays. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to make the above-mentioned method and system for monitoring sea use behavior based on intelligent recognition algorithm of the present invention more obvious and easy to understand, the following is a brief introduction to the drawings required for use in the specific implementation of the present invention. 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 a method for monitoring sea use behavior based on an intelligent recognition algorithm; Figure 2 Schematic diagram showing the conversion of longitude and latitude to Cartesian coordinate system; Figure 3 A schematic diagram showing the conversion between altitude and ellipsoid height; Figure 4A schematic diagram showing the conversion between the camera coordinate system and the image coordinate system; Figure 5 A schematic diagram showing the conversion between the image coordinate system and the pixel coordinate system; Figure 6 A schematic diagram showing camera rotation. DETAILED DESCRIPTION

[0036] Embodiment 1: A method and system for monitoring sea use behavior based on an intelligent recognition algorithm are provided. The method and system for monitoring sea use behavior based on an intelligent recognition algorithm are as follows: Figure 1 As shown, the system includes: a behavior detection module, a spatial positioning module, an offset detection module and a visual interaction module, as described in detail below.

[0037] The behavior detection module is used to capture sea-use behaviors through a camera and identify image data through a target detection algorithm.

[0038] The behavior detection module uses a target detection algorithm to analyze whether there is any sea use behavior in the captured image, and records the location coordinates of the area. If the sea use behavior occurs in an unauthorized land area, the sea use behavior is determined to be an illegal behavior.

[0039] The steps of the target detection algorithm include: collecting and annotating image data sets, and dividing the data sets into training sets and test sets; converting the input image into a feature map representation through a neural network; estimating and screening the positions that may contain targets in the image based on the feature map through RPN; adjusting and correcting the positions of candidate boxes so that the candidate boxes can accurately cover the target objects; classifying the regressed target boxes through softmax and fully connected neural networks; determining the final target box by performing non-maximum suppression operations on the overlapping areas of multiple output target boxes; training the model through the training set, and completing evaluation and adjustment on the test set.

[0040] The target detection algorithm converts the input image into a feature map representation through a neural network, and the neural network uses SwinTransformer as the skeleton network. The specific method of converting the input image into a feature map representation through the neural network includes: dividing the input image into multiple image blocks, and sharing some pixels between adjacent image blocks so that the algorithm has a strong semantic understanding ability; each image block is processed by a basic processing unit, and the processing unit includes a convolutional layer and a SwinTransformer structure to extract the feature representation of the image block; the Swin Transformer structure adopts a hierarchical attention mechanism so that the neural network can focus on important information at different levels in feature maps at different abstract levels; pre-training is performed on an image data set to improve the generalization ability of the neural network, and the feature representation ability of the neural network is further improved based on a large amount of monitoring materials through a self-supervised learning algorithm.

[0041] The spatial positioning module is used to calculate the three-dimensional coordinates of the pixel points in the image based on the position information of the control points.

[0042] Furthermore, the spatial positioning module estimates the position, orientation, field of view and internal parameters of the camera through the least square method through the control points in the image and their position information in the three-dimensional map; the target location is imaged to the center of the image by controlling the horizontal and vertical rotation values ​​of the camera; the longitude and latitude are converted to the Cartesian coordinate system as shown in Figure 2 As shown, by converting the latitude and longitude data of the target location into a Cartesian coordinate system, and converting based on the altitude and ellipsoid height to match the installation position and orientation of the camera, the altitude and ellipsoid height conversion is as follows Figure 3 As shown, the installation position and orientation of the camera are matched by rotating and translating the coordinate system; the points in the three-dimensional world are projected onto the two-dimensional image plane through the camera through the pinhole imaging method; the radial distortion and tangent distortion of the camera are corrected by the correction method to improve the accuracy of positioning; the parameters of the camera are estimated by calibrating the points on the image and the longitude and latitude values ​​in the corresponding three-dimensional space, the external parameters include the longitude and latitude of the installation position, azimuth, pitch and roll angle, and the internal parameters include the field of view angle and distortion parameters; the optimal estimation of the camera parameters is solved by the numerical optimization method.

[0043] The conversion between the camera coordinate system and the image coordinate system is as follows: Figure 4As shown, after the camera forms an image, it is assumed that the camera coordinate system takes the camera focus C as the origin, the plane parallel to the image sensor is the XY axis, and the plane perpendicular to the image sensor is the Z axis; the image coordinate system is assumed to be a two-dimensional rectangular coordinate system with the center point P of the image sensor as the origin, the width as the X axis, and the height as the Y axis. The length of the origin CP of the two coordinate systems is the focal length zf, and the position of point A in the camera coordinate system (XA, YA, ZA) can be calculated by converting the Cartesian coordinate system to the camera coordinate system; through the geometric algorithm, CP: ZA is known, and the position of point a in the image coordinate system (xa, ya) is calculated, that is, the longitude and latitude position is converted to image pixels; conversely, assuming that a is known Point coordinates (xa, ya), draw a straight line from point C through point a, and intersect with the terrain DEM at point A, that is, the pixel coordinates are converted to longitude and latitude positions; since the distance of CP is zf, the position of point a in the camera coordinate system is (xa, ya, zf), and the coordinates of each unit distance increase in the direction of the straight line Ca are calculated through the geometric relationship. To calculate the coordinates of point A in the camera coordinate system (XA, YA, ZA), it is necessary to continuously increase the unit distance from point C along the straight line Ca direction to obtain the coordinates, and calculate whether the obtained Y coordinate value is equal to the elevation value of the terrain DEM where the point is located. If they are equal, the point is the intersection point A.

[0044] The conversion between image coordinate system and pixel coordinate system is as follows: Figure 5 As shown, the image coordinate system is based on the imaging sensor, that is, the unit is the actual physical size. Since the unit of the image screen is pixel, it is necessary to convert between the physical size and the pixel, that is, the conversion between the image coordinate system and the pixel coordinate system; the image coordinate system takes the upper left corner of the image screen as the origin (0,0), assuming that the camera sensor width is w, the height is h, the pixel width is u, and the height is v, then the pixel size corresponding to each physical unit is (pw / w, ph / h), so the coordinates of any point (x, y) converted to pixels are (x*u / w+(u / w)*u / 2, y*v / h+(v / h)*v / 2), and conversely, the physical size of the image coordinates calculated from the pixel coordinates is obtained.

[0045] The Cartesian coordinate system takes the center of the sphere as the origin, the x-axis from the center of the sphere to the equator 0 longitude, the y-axis from the center of the sphere to the equator 90 longitude, and the z-axis from the center of the sphere to the true North Pole. Given the radius of the Earth's equator R and the radius of the Earth's North and South Poles r, any point on the Earth can be converted into Cartesian coordinates after determining its longitude and latitude and altitude. The longitude and latitude of the camera and point A and their altitude are obtained through measurement. Suppose the converted Cartesian coordinates of the camera are (x1, y1, z1), and the Cartesian coordinates of point A are (x2, y2, z2), in the matrix operation, the coordinates of point A minus the camera coordinates get the three-dimensional rectangular coordinate system with the camera as the origin, that is, the Cartesian coordinate system with the center of the earth as the origin is translated to the three-dimensional rectangular coordinate system with the camera as the origin, and then the longitude and latitude of the camera is rotated to transform it into a three-dimensional rectangular coordinate system with the camera as the origin and the camera installation position as the orientation, and then the current horizontal angle, pitch angle and roll angle of the camera are rotated to transform it into a three-dimensional rectangular coordinate system with the camera as the origin and the current orientation of the camera as the orientation, that is, the camera coordinate system.

[0046] The principle of pinhole imaging is Figure 6 As shown, the image of the object being photographed in the real world is focused onto the sensor through the camera lens. Any point in the real world mapped to the sensor is on the same straight line as the focus. When the camera is rotated, the lens forms an image in another direction. At this time, the imaging content is determined by the horizontal angle azimuth and pitch angle during rotation, and the tilt angle of the lens when it is installed, that is, the roll angle, as well as the image size, that is, the horizontal field of view angle and the vertical field of view angle. The horizontal field of view angle and the vertical field of view angle are determined by the physical length, width and focal length of the imaging sensor. The physical length and width of the imaging sensor are fixed, and the focal length is proportional to the magnification. Suppose the focal length at 1x is f, the magnification is z, then the focal length is zf.

[0047] The correction formulas for radial distortion and tangential distortion of the lens are as follows:

[0048]

[0049] in, The horizontal axis coordinate of the distortion point; The vertical axis coordinate of the distortion point; Represents the original horizontal coordinate of the undistorted point; Represents the original vertical coordinate of the undistorted point; Represents the radial distance from the image center to the distortion point; , and Represents the radial distortion coefficient.

[0050] Based on the pinhole imaging and coordinate system conversion principles, a conversion relationship model between image pixels and geographic space coordinates is established through calibration, so that any point on the image has spatial coordinate information. The main applications are: converting image pixel coordinates to longitude and latitude positions to achieve alarm positioning; converting image pixel coordinates through longitude and latitude positions, and fusing geographic space vector data with image real scenes.

[0051] Vector data in geographic space, such as the basic farmland layer, is dynamically drawn in real time onto the video screen and integrated with the video for real-life scene fusion, specifically including: converting image pixels to longitude and latitude positions and calculating the longitude and latitude positions of the four corners of the current screen, then merging the four corners into a polygon for spatial calculation with the basic farmland layer, calculating the intersecting plots, and obtaining the longitude and latitude position set of the plots; converting the plot coordinate set and the camera coordinates into Cartesian coordinates, taking point A in the plot coordinate set as an example, translating the Cartesian coordinates of point A relative to the camera Cartesian coordinates and rotating them into camera coordinate system coordinates, converting the camera coordinate system coordinates of point A into image coordinate system coordinates of point a, converting the image coordinates of point a into pixel coordinates, drawing points and connecting lines based on pixel coordinates on the video screen, thereby realizing the fusion of geographic space vector data and real-life video scenes.

[0052] The offset detection module is used to determine the offset caused by the camera by matching the captured image with the preset point image.

[0053] The offset detection module reserves a base map as a reference for offset detection. During the offset detection process, the camera parameters are adjusted to be consistent with the camera parameters in the initial state to capture a test image, and the offset generated by the camera is determined by calculating the transformation matrix of the spatial relationship between the reserved base map and the test image.

[0054] The premise of camera offset is that the camera will only have rotational offset, not translational offset.

[0055] The method for offset detection by the offset detection module specifically includes: inputting the horizontal and vertical field of view angles pov_w and pov_h of the camera, inputting the reference base map t1 and the captured image t2 that may have an offset under the same PTZ parameters, and inputting the height h and width w of the image; obtaining the transformation matrix from t2 to t1 by calculating and matching the features of t1 and t2; calculating the coordinates of the four corners of the image t2 after being converted to the coordinates of t1, and further calculating the average deviations offset_w and offset_h in the horizontal and vertical directions, thereby obtaining the horizontal angular deviation (offset_w / w) * pov_w and the vertical angular deviation (offset_h / h) * pov_h.

[0056] Since the rotational position change of the camera is certain, the transformation relationship between the two positions of other points in the physical space in different lenses is also certain. This transformation relationship is a 3×3 homography matrix. The following transformation relationship is calculated by the pinhole imaging principle:

[0057] The changes of the lens are determined by the homography matrix. Assuming that the lens only undergoes rotational changes, the transformation relationship of the image points is obtained by the pinhole imaging principle and the intrinsic parameters of the intrinsic reference camera as follows:

[0058] in Is the rotation matrix, which can be decomposed into three directions of rotation Pan, Tilt and Roll, that is, the rotation matrix , , , which has the following relationship:

[0059] Assuming that the lens does not rotate in the Roll direction, the rotation in the Pan and Tilt directions is expanded as follows:

[0060]

[0061] Calculate the change in the lens's posture through the unit matrix, that is, solve the 8 unknowns of the homography matrix; Match the points of two images through feature point detection and feature point matching algorithm.

[0062] The offset detection module cuts out the part with the OSD watermark in the image to reduce the probability of matching the OSD watermark in the preset point as a feature point.

[0063] The offset detection module also includes a feature point matching noise removal algorithm, which selects points with higher matching confidence and selects matching points with the most consistent change relationships as much as possible during the matching point selection process to facilitate subsequent calculations.

[0064] The visualization interaction module presents the detection data of sea use behavior in the reclamation area in the form of an electronic map, generates early warning information by analyzing the detection data, and provides auxiliary decision-making based on the early warning information; users can retrieve detailed data on sea use behavior based on the early warning information, and can remotely view the monitoring screen of the monitoring equipment in real time to conduct remote evidence collection.

[0065] The visual interaction module includes two functional pages: full overview and project inspection. The full overview presents the overall warning situation of the monitored area, including five functional modules: warning overview, violation warning, video monitoring, airport and tools.

[0066] The system also includes a water body separation module, which extracts water body data by suppressing vegetation information through green band and near-infrared band based on the spectral characteristic curve of water body in different bands, and can accurately identify water body, improve the monitoring accuracy, and realize real-time monitoring of water body to detect whether there is any illegal use of sea, thus enriching the management means of illegal use of sea, including: The collected data is preprocessed, and the input image is separated into land and water using the MNDWI index. The MNDWI index calculation formula is as follows:

[0067] in, Indicates the reflectance value of the green band; Indicates the reflectance value in the mid-infrared band; Represents the normalization factor.

[0068] The outliers of the MNDWI index are judged and eliminated. The outlier judgment formula is:

[0069] in, represents a small positive number used to define the range of outliers; The outlier elimination formula is:

[0070] in, Indicates the MNDWI index value of a single pixel.

[0071] Embodiment 2: On the basis of the above-mentioned embodiment, the system adds an isolation forest module, which randomly divides the data space by constructing multiple isolation trees based on the target detection algorithm, and determines the abnormal data by calculating the path length of the data in each isolation tree.

[0072] First, the collected data needs to be cleaned and preprocessed to ensure data quality, and the image data is identified through the target detection algorithm.

[0073] By constructing multiple isolated trees to randomly divide the data space, the number of abnormal data is small and the difference from normal data is large, so fewer steps are required to isolate them. The formula for calculating the abnormal score is as follows:

[0074] in, represents a harmonic number; represents the average path length of the tree; represents the number of trees in the forest; represents the anomaly score calculation function; Indicates the number of samples used when building a subtree.

[0075] By calculating the path length of the data in each isolated tree and calculating the average path length, we can determine which data points are abnormal. The anomaly score formula is as follows:

[0076] in, Indicates that the sample is Anomaly score in the tree.

[0077] The accuracy of the recognition result is determined by the confidence interval. The confidence interval formula is as follows:

[0078] in, Representation parameters confidence interval of ; represents the sample statistic; represents the critical value of the standard normal distribution; Represents the population standard deviation.

[0079] For example, suppose you have the following dataset: .

[0080] The data is split by randomly selecting a feature and randomly choosing a split point between the maximum and minimum values ​​of the feature.

[0081] First split: the maximum value is 250, the minimum value is 1, and the randomly selected split point is 50; the data is split into two groups, and the data less than 50 is , data greater than or equal to 50 is .

[0082] Second segmentation: For the first group (less than 50), randomly select a segmentation point, the maximum value is 3, the minimum value is 1, and the randomly selected segmentation point is 2. The segmentation result: the data less than 2 is , data greater than or equal to 2 is ; For the second group (greater than or equal to 50), randomly select a split point, the maximum value is 250, the minimum value is 100, and the randomly selected split point is 150. The split result: the data less than 150 is , data greater than or equal to 150 is .

[0083] Compute the path length for each data point: , the path length is equal to 2; , the path length is equal to 2; , the path length is equal to 2; , the path length is equal to 2; , the path length is equal to 2; , the path length is equal to 2; , the path length is equal to 1; , the path length is equal to 1; , the path length is equal to 1; , the path length is equal to 1.

[0084] Assume that A tree, , ; calculate :

[0085] calculate :

[0086] Calculating anomaly score :

[0087] It follows that , , and have shorter path lengths, so they have lower anomaly scores and are likely to be outliers.

[0088] The effect of the isolation forest module is shown in Table 1: Table 1. Summary of the effects of the Isolation Forest module

[0089] As shown in Table 1, the method optimized by the isolation forest module has improved both the recall rate and the precision rate; this indicates that the isolation forest module can effectively improve the recognition accuracy of the sea use behavior monitoring system for illegal sea use behaviors.

[0090] 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.

[0091] The present invention can provide computer program instructions to a processor 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 processor of the computer or other programmable data processing device produce means for implementing the system.

[0092] 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.

[0093] 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 sea use behavior monitoring system based on intelligent recognition algorithm, characterized by: include: Behavior detection module, used to capture sea-use behaviors through cameras and identify image data through target detection algorithms; A spatial positioning module is used to calculate the three-dimensional coordinates of the pixel points in the image based on the position information of the control points; An offset detection module is used to determine the offset of the camera by matching the captured image with the preset point image; Visual interaction module, used to present detection data and warning information and provide auxiliary decision-making.

2. The sea use behavior monitoring system based on intelligent recognition algorithm according to claim 1 is characterized by: The target detection algorithm converts an image into a feature map representation through a neural network, and the neural network Swin Transformer is a skeleton network. The target detection algorithm converts an image into a feature map representation through a neural network, specifically including: dividing the input image into multiple image blocks, and sharing some pixels between adjacent image blocks; each image block is processed by a basic processing unit, and the processing unit includes a convolution layer and a Swin Transformer structure, and the Swin Transformer structure adopts a hierarchical attention mechanism.

3. The sea use behavior monitoring system based on intelligent recognition algorithm according to claim 1 is characterized in that: The spatial positioning module estimates the parameters of the camera through the control points in the image and their position information in the three-dimensional map, and through the least squares method; the target location is imaged to the center of the image by controlling the horizontal and vertical rotation values ​​of the camera; the latitude and longitude data of the target location are converted into a Cartesian coordinate system, and converted based on the altitude and ellipsoid height to match the installation position and orientation of the camera.

4. The sea use behavior monitoring system based on intelligent recognition algorithm according to claim 1 is characterized by: The offset detection module reserves a base map as a reference for offset detection. During the offset detection process, the camera parameters are adjusted to be consistent with the camera parameters in the initial state to capture a test image, and the offset generated by the camera is determined by calculating the transformation matrix of the spatial relationship between the reserved base map and the test image.

5. The sea use behavior monitoring system based on intelligent recognition algorithm according to claim 1 is characterized by: The visualization interaction module presents the detection data of the sea use behavior in the reclamation area in the form of an electronic map, generates early warning information by analyzing the detection data, and provides auxiliary decision-making based on the early warning information.

6. The sea use behavior monitoring system based on intelligent recognition algorithm according to any one of claims 1 to 5, characterized in that: The system also includes a water body separation module, which extracts water body data by suppressing vegetation information through a green band and a near infrared band based on spectral characteristic curves of the water body in different bands.

7. The sea use behavior monitoring system based on intelligent recognition algorithm according to any one of claims 1 to 5, characterized in that: The system also includes an isolation forest module, which randomly divides the data space by constructing multiple isolation trees based on the target detection algorithm, and determines abnormal data by calculating the path length of the data in each isolation tree.

8. A method for monitoring sea use behavior based on an intelligent recognition algorithm, characterized in that: The method is implemented based on the system according to any one of claims 1 to 7: The method comprises: The sea-use behavior is captured by the camera and the image data is recognized by the target detection algorithm; Calculate the three-dimensional coordinates of the pixel points in the image based on the position information of the control points; The offset of the camera is determined by matching the captured image with the preset point image; Present detection data and warning information and provide decision support.

9. A computer device comprising a processor and a memory, wherein the processor is connected to the memory, and the memory is used to store a computer program, wherein: The processor is used to execute the computer program stored in the memory so that the computer device executes the sea use behavior monitoring method based on the intelligent recognition algorithm as claimed in claim 8.

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, the method for monitoring sea use behavior based on intelligent recognition algorithm as described in claim 8 is implemented.

Images