A beidou artificial intelligence unmanned ship monitoring application system and platform

Abstract

The application discloses a Beidou-based artificial intelligence unmanned ship monitoring application system and platform, and relates to the technical field of unmanned ship monitoring application.The Beidou-based artificial intelligence unmanned ship monitoring application system comprises a path generation module for generating a covering monitoring sailing path according to a boundary coordinate point set; a path execution module for controlling the unmanned ship to sail according to the path and acquiring Beidou sailing state data; a Beidou regulation and sensing module for analyzing a path deviation state evaluation index based on the data; a path intelligent adjustment module for executing real-time course correction according to the index; and a data sampling and output module for collecting water surface environment sensing data at path intervals and outputting the data.The application can realize real-time judgment of sailing deviation degree through the path deviation state evaluation index, and can match and adjust measures, so that the application has stronger environmental adaptability and adjustment precision than a traditional fixed threshold method, and can effectively improve sailing stability and path covering continuity in complex water areas.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of unmanned ship monitoring application, in particular to a Beidou-based artificial intelligence unmanned ship monitoring application system and platform. BACKGROUND

[0002] With the acceleration of water resource development and utilization and the improvement of ecological environment protection demand, the intelligent monitoring demand for inland and offshore waters such as rivers, lakes, reservoirs and harbors is increasing. The traditional water surface environment inspection method mainly relies on manual driving of ships for fixed-point sampling and cruise observation. The operation process is greatly affected by factors such as personnel skills, weather conditions and water complexity, and there are problems such as low inspection efficiency, insufficient data coverage, delayed emergency response and the like, which are difficult to meet the current dynamic monitoring demand of water quality safety, floating object cleaning and water surface target identification and the like.

[0003] In recent years, with the development of unmanned ship technology and automatic navigation algorithm, water area inspection based on unmanned platform has gradually become a hot research and engineering application direction. Especially under the promotion of Beidou satellite navigation system and artificial intelligence perception algorithm, the unmanned ship has high-precision autonomous positioning capability and strong environmental adaptability, and can perform long-term, multi-parameter and multi-task autonomous navigation monitoring in a large range, providing a feasible path for improving the efficiency of water surface environment management.

[0004] The limitations of the prior art at least include the following problems. The current water area unmanned monitoring system generally lacks real-time perception and dynamic adjustment mechanism for path deviation state during navigation. The traditional path execution method mainly relies on preset route and single-point navigation reference, and does not introduce real-time quantitative means of systematic path deviation. Especially in complex hydrodynamic environment, such as water flow speed change, water surface disturbance or wind interference, the unmanned ship is prone to dynamic deviation such as heading angle drift and lateral deviation, which seriously affects the navigation stability and path execution accuracy.

[0005] At the same time, the existing scheme generally uses binary logic judgment of whether to deviate in path deviation perception, lacks coupling analysis of multi-dimensional navigation state such as propeller power feedback and lateral angular velocity, and causes problems such as inaccurate deviation identification and lagging adjustment feedback, which easily causes path detour, sparse monitoring data or insufficient regional coverage, and affects the continuity and integrity of water quality monitoring. SUMMARY

[0006] In view of the deficiencies of the prior art, the Beidou-based artificial intelligence unmanned ship monitoring application system and platform are provided, which solves the problem of unstable navigation caused by the lack of path deviation adjustment capability in the prior art.

[0007] To achieve the above object, the application is implemented by the following technical solutions: a Beidou-based artificial intelligence unmanned ship monitoring application system, comprising:

[0008] A path generation module is configured to obtain a set of boundary coordinate points of a water area to be monitored and generate a monitoring navigation path comprising a plurality of monitoring navigation points; a path execution module is configured to control the unmanned ship to navigate along the monitoring navigation path in sequence, and obtain Beidou navigation state data of the unmanned ship in real time during the navigation; a Beidou regulation and perception module is configured to analyze a path deviation state evaluation index of the unmanned ship based on the Beidou navigation state data; a path intelligent adjustment module is configured to perform real-time navigation adjustment processing on the unmanned ship based on the path deviation state evaluation index; and a data sampling and output module is configured to sample water surface environment perception data of the water area to be monitored at intervals according to a set path distance interval during the navigation and perform output processing.

[0009] Further, the set of boundary coordinate points comprises a plurality of boundary two-dimensional coordinate points, and the specific steps of generating the monitoring navigation path are as follows: a boundary box region is constructed based on the set of boundary coordinate points, and covering grid line segments are generated in the boundary box at a preset path interval; the covering grid line segments are subjected to spatial clipping processing with the water area to be monitored, and are connected in sequence to form an initial path trajectory according to a preset order; the initial path trajectory is subjected to turning smoothing processing to obtain a smooth path curve; and a plurality of monitoring navigation points are extracted from the smooth path curve at equal path distance intervals to form the monitoring navigation path.

[0010] Further, the Beidou navigation state data comprises Beidou navigation two-dimensional coordinates, a travel heading angle, a travel lateral angular velocity, and a propeller feedback power.

[0011] Further, the specific steps of analyzing the path deviation state evaluation index of the unmanned ship are as follows: in the monitoring navigation path, the nearest monitoring navigation point to the travel direction of the unmanned ship is extracted, and the lateral deviation distance and the heading angle deviation of the unmanned ship are analyzed; the propeller reference propelling power of the unmanned ship is obtained, and a deviation analysis is performed with the corresponding propeller feedback power to obtain a propeller power deviation amount of the unmanned ship; and the lateral deviation distance, the heading angle deviation, the propeller power deviation amount, and the travel lateral angular velocity of the unmanned ship are comprehensively analyzed to obtain the corresponding path deviation state evaluation index.

[0012] Further, the specific formula for calculating the path deviation state evaluation index of the unmanned ship is as follows: Wherein, LpH, PkL, HjP, TgP, XhJ are path deviation state evaluation index, lateral offset distance, heading angle deviation, propeller power deviation amount, travel lateral angular velocity of unmanned ship in turn, ξ1, ξ2 are path offset coupling coefficient, dynamic attitude deviation coefficient stored in database in turn.

[0013] Further, based on the path deviation state evaluation index, the specific steps of performing real-time navigation adjustment processing on the unmanned ship are as follows: the path deviation state evaluation index of the unmanned ship is respectively judged and analyzed with a plurality of preset path deviation state evaluation intervals, and each path deviation state evaluation interval corresponds to a navigation adjustment measure; based on the path deviation state evaluation index being in the navigation adjustment measure corresponding to the preset path deviation state evaluation interval, real-time navigation adjustment processing is performed on the unmanned ship.

[0014] An application platform for monitoring Beidou-based artificial intelligence unmanned ship, comprising: a data receiving unit for receiving water surface environment perception data of a water area to be monitored, including water quality perception data and visual perception data; a state analysis unit for analyzing water surface environment perception state index sets of the water area to be monitored based on the water surface environment perception data, including water quality perception state index and visual perception state index; an abnormality analysis unit for analyzing water surface environment comprehensive evaluation index of the water area to be monitored based on the water surface environment perception state index sets, and judging and analyzing with a preset water surface environment abnormality evaluation interval, and considering the water surface environment abnormal when the water surface environment comprehensive evaluation index is within the preset water surface environment abnormality evaluation interval, and sending a water surface environment abnormality alarm.

[0015] Further, the specific formula for calculating the water surface environment comprehensive evaluation index of the water area to be monitored is as follows: Wherein, ShP, SzG, SjG are water surface environment comprehensive evaluation index, water quality perception state index, visual perception state index of the water area to be monitored in turn, δ1, δ2 are water quality influence weight coefficient, visual influence weight coefficient stored in database in turn.

[0016] Further, the water quality perception data includes water temperature value, conductivity value, dissolved oxygen value, PH value, water suspended particulate matter density value, oxidation-reduction potential, ammonia nitrogen concentration value, blue algae fluorescence intensity value, water surface oil film conductivity disturbance value, and the specific steps of analyzing the water quality perception state index of the water area to be monitored are as follows: obtaining water quality perception reference data of the water area to be monitored, including water temperature reference value, conductivity reference value, dissolved oxygen reference value, PH reference value, water suspended particulate matter density reference value, oxidation-reduction reference potential, ammonia nitrogen concentration reference value, blue algae fluorescence intensity reference value, water surface oil film conductivity disturbance reference value; combining the water quality perception data of the water area to be monitored with the corresponding water quality perception reference data for comprehensive analysis to obtain the water quality perception state index of the water area to be monitored.

[0017] Further, the visual perception data is specifically water surface image data, including pixel values of a plurality of water surface pixel points, and the specific steps of analyzing the visual perception state index of the water area to be monitored are as follows: reading the water surface image data of the water area to be monitored, and inputting the water surface image data into a pre-trained visual analysis model for feature extraction to obtain a visual perception feature set of the water area to be monitored, including a heterochromatic spot block proportion value, a water surface reflectivity uniformity value, an image texture disturbance value, a boundary fragmentation degree value and a target detection quantity value; and based on the visual perception feature set, analyzing the visual perception state index of the water area to be monitored.

[0018] The application has the following beneficial effects:

[0019] (1) The Beidou-based artificial intelligence unmanned ship monitoring application system can extract the navigation deviation amount in real time and judge whether it needs to be adjusted during the navigation of the unmanned ship along the preset path, generate a path deviation state evaluation index by calculating parameters such as lateral offset distance, heading angle deviation, thruster power deviation amount and lateral angular velocity, match the index with a plurality of preset state intervals, and the system can select the corresponding adjustment measures accordingly to realize dynamic response to the actual navigation state. Compared with the traditional adjustment method relying on fixed threshold, this method combines multiple parameter characteristics, has stronger environmental adaptability and adjustment flexibility, effectively reduces the path deviation risk, improves the steady-state execution capability of the system in complex situations such as water flow disturbance and thruster abnormality, thereby ensuring the coverage continuity and accuracy of the data acquisition path.

[0020] (2) The Beidou-based artificial intelligence unmanned ship monitoring application system constructs a boundary box based on the water area boundary coordinate point set, generates a coverage grid line segment according to the set path distance, then performs spatial clipping on the line segment and the water area boundary and sequentially connects them to form an initial path trajectory, performs turning smoothing processing on the trajectory to generate a more adaptive smooth flight path, and finally extracts monitoring navigation points at equal path distance intervals to form a structured monitoring path. This process not only ensures that the overall coverage range of the flight path matches the shape of the water area, but also improves the representativeness and uniformity of subsequent data sampling by designing evenly spaced monitoring points, avoiding data deviation or coverage blind spots caused by uneven point density. This structure is suitable for various irregularly shaped water areas and can effectively support the deployment and execution of automated unmanned ships in actual monitoring tasks.

[0021] (3), the application platform for monitoring based on the Beidou artificial intelligence unmanned ship, through setting data receiving unit, state analysis unit and abnormal analysis unit, can respectively carry out structured processing to water quality perception data and visual perception data, extracts key perception parameters and generates corresponding state index, and calculates water surface environment comprehensive evaluation index, and compares with preset abnormal interval, realizes real-time identification and automatic early warning of water surface abnormal state, avoids the limitation that only single data is relied on in the prior art and lacks comprehensive judgment mechanism, effectively improves the judgment accuracy and processing timeliness of pollution, water bloom and other water surface environment abnormalities, has definite practical value.

[0022] Of course, implementing any product of the present application does not necessarily require achieving all the advantages described above at the same time. BRIEF DESCRIPTION OF DRAWINGS

[0023] Figure 1 A system block diagram of the application based on the Beidou artificial intelligence unmanned ship monitoring application system.

[0024] Figure 2 A specific step flow chart of analyzing the path deviation state evaluation index of the unmanned ship in the application based on the Beidou artificial intelligence unmanned ship monitoring application system.

[0025] Figure 3 A platform block diagram of the application based on the Beidou artificial intelligence unmanned ship monitoring application platform. DETAILED DESCRIPTION

[0026] Please refer to Figure 1 The embodiment of the application provides a technical scheme: an application system for monitoring based on the Beidou artificial intelligence unmanned ship, comprising: a path generation module, configured to obtain a boundary coordinate point set of a water area to be monitored, and generate a monitoring navigation path, including a plurality of monitoring navigation points; a path execution module, configured to control the unmanned ship to navigate according to the monitoring navigation path, and obtain Beidou navigation state data of the unmanned ship in real time during the navigation process; a Beidou regulation and perception module, configured to analyze a path deviation state evaluation index of the unmanned ship based on the Beidou navigation state data; a path intelligent adjustment module, configured to perform real-time navigation adjustment processing on the unmanned ship based on the path deviation state evaluation index; and a data sampling and output module, configured to sample water surface environment perception data of the water area to be monitored at a set path distance interval during the navigation process, and perform output processing.

[0027] Specifically, the boundary coordinate point set includes a plurality of boundary two-dimensional coordinate points, and the specific steps of generating the monitoring navigation path are as follows:

[0028] The boundary box region is constructed based on the boundary coordinate point set, and a covering grid line segment is generated in the boundary box according to a preset path interval. Specifically, the input boundary coordinate point set is projected into a two-dimensional plane coordinate system, and a minimum circumscribed rectangle containing all boundary points is calculated as the boundary box region. In the boundary box region, parallel line segment grids are generated by cutting at equal intervals along the horizontal direction or the vertical direction according to a preset path interval. The path interval is preset according to the unmanned ship size, water area sampling resolution and track coverage requirement, so as to ensure that adjacent grid line segments have coverage redundancy.

[0029] The covering grid line segment is spatially cut with the water area to be monitored, and is sequentially connected in a preset order to form an initial path trajectory. Specifically, the covering grid line segment is subjected to a Boolean intersection operation with a closed water area polygon composed of the boundary coordinate point set, and only the part of the line segment inside the water area boundary is retained. The retained line segment is sequentially sorted, and is connected in a zigzag (sawtooth) or snake shape to ensure that the trajectory reciprocally covers the water area without repeated segments. A transition connection segment is introduced between the end points of line segments in different directions to avoid the jump of corners causing an unnavigable path.

[0030] The initial path trajectory is subjected to turning smoothing processing to obtain a smooth path curve. Specifically, turning points with direction mutations in the initial path trajectory are extracted, and are subjected to angle identification analysis. Spline interpolation (such as cubic B-spline, Catmull-Rom spline) or Bezier curve algorithm is used to smooth and reconstruct the adjacent path segments of the turning points. The smoothing processing can set a maximum curvature threshold and a lower limit of turning radius to ensure that the unmanned ship does not have a sharp turning problem when running on the trajectory.

[0031] A plurality of monitoring navigation points are extracted from the smooth path curve at equal path distance intervals, and constitute a monitoring navigation path. Specifically, the smooth path curve is resampled according to the path arc length parameter, and equal-interval point extraction is performed along the path curve at a set sampling distance interval. Each extracted point is taken as a monitoring navigation point, and its two-dimensional coordinate and heading angle information are recorded. The sampling distance interval is set according to the ship speed, data sampling period and spatial resolution requirement. All monitoring navigation points constitute a final monitoring navigation path in the path order, which is used for sequential control and navigation of the unmanned ship.

[0032] In this embodiment, the multiple requirements of path continuity, coverage range and heading controllability of unmanned ship in real water area can be met by converting the boundary coordinate point set into a boundary box and generating an equidistant grid, and combining spatial clipping and turning smoothing processing, avoiding the problems of path breakage, excessive turning or region omission in the conventional method, and the path density is flexibly set according to the ship size and perception demand, so that the finally generated navigation points are uniformly distributed in space, which is convenient for automatic navigation control, and also facilitates subsequent data sampling and environmental analysis work, and improves the execution efficiency and water area adaptation ability of the whole system.

[0033] Specifically, the Beidou navigation state data includes Beidou navigation two-dimensional coordinates, heading angle, lateral angular velocity, and propeller feedback power.

[0034] As shown in Figure 2 The specific steps of analyzing the path deviation state evaluation index of the unmanned ship are as follows: in the monitoring navigation path, the nearest monitoring navigation point in the direction of the unmanned ship is extracted, and the lateral deviation distance of the unmanned ship (i.e. the projection value of the current Beidou navigation two-dimensional coordinates of the unmanned ship in the direction perpendicular to the tangent direction of the nearest monitoring navigation point path, minus the projection value of the monitoring navigation point in the same direction), the heading angle deviation (i.e. the heading angle of the current unmanned ship minus the tangent direction angle of the nearest monitoring navigation point path); the propeller reference propelling power of the unmanned ship is obtained, and the deviation analysis is carried out with the corresponding propeller feedback power to obtain the propeller power deviation amount (i.e. the propeller reference propelling power minus the propeller feedback power) of the unmanned ship; the lateral deviation distance, the heading angle deviation, the propeller power deviation amount and the lateral angular velocity of the unmanned ship are comprehensively analyzed to obtain the corresponding path deviation state evaluation index.

[0035] Wherein, in the calculation of the path deviation state evaluation index, the lateral deviation distance, the heading angle deviation, the propeller power deviation amount and the lateral angular velocity are unit-free.

[0036] The specific formula for calculating the path deviation state evaluation index of the unmanned ship is as follows: Wherein, LpH is the path deviation state evaluation index of the unmanned ship, PkL is the lateral deviation distance of the unmanned ship, HjP is the heading angle deviation of the unmanned ship, ξ1 is the path deviation coupling coefficient stored in the database, TgP is the propeller power deviation amount of the unmanned ship, XhJ is the lateral angular velocity of the unmanned ship, and ξ2 is the dynamic attitude deviation coefficient stored in the database.

[0037] It should be explained that the specific steps of obtaining the path deviation coupling coefficient ξ1 and the dynamic attitude deviation coefficient ξ2 stored in the database are as follows:

[0038] For the path offset coupling coefficient: first, construct several representative monitoring water areas (including straight sections, turning sections, winding sections, etc.), and plan a standard monitoring navigation path in each area. Control the unmanned ship to navigate along the planned path, continuously collect the current position coordinates and the corresponding heading angle during the entire process. The system calculates the lateral offset distance between the current coordinates and the nearest navigation point on the path, as well as the heading angle deviation between the current heading angle and the tangent direction of the path, and multiplies the two values to obtain the composite offset sample at the current time. Then, each offset sample is associated with the adjustment operation triggered at that time (no adjustment, slight adjustment, severe adjustment) to form a dataset with adjustment level labels. A linear regression model is established for this dataset, with the offset product as the input feature and the adjustment level code (e.g. 0, 1, 2) as the output. The least squares method is used for fitting and training. After training, the absolute value of the fitted regression coefficient is taken as the path offset coupling coefficient. The residual sum of squares of all experimental samples is calculated, and the experimental coefficient with the smallest error is selected as the final value, i.e. the path offset coupling coefficient.

[0039] For the dynamic attitude deviation coefficient: in the same experiment, the propeller reference power (set by the control system) and the propeller feedback power (measured by the sensor) of the unmanned ship at each time point are continuously collected, and the difference between the two is calculated as the propeller power deviation. The lateral angular velocity at that time point is also collected, which is the rotation speed of the ship body around the vertical axis. Considering the different units of propeller power and angular velocity, to ensure the dimensional consistency of the combined variable, first normalize the two parameters to the value range [0, 1], then take their equal-weighted average to obtain the joint dynamic attitude disturbance variable. Then, take this joint variable as the input feature and the adjustment behavior level recorded by the system (e.g. 0 for no adjustment, 1 for adjustment) as the output variable, construct a binary Logistic regression model for fitting, and automatically learn the nonlinear response relationship between the disturbance variable and the adjustment behavior during the training process. Finally, extract the dominant weight coefficient in the model, which is defined as the dynamic attitude deviation coefficient.

[0040] Wherein, the specific implementation example of calculating the path deviation state evaluation index of the unmanned ship is as follows, the existing parameters (after unit processing) are as follows:

[0041] The lateral offset distance of the unmanned ship is about -0.103.

[0042] The heading angle deviation of the unmanned ship is about 0.220.

[0043] The path offset coupling coefficient stored in the database is about 0.863.

[0044] The propeller power deviation amount of the unmanned ship is about 0.344.

[0045] The travel lateral angular velocity of the unmanned ship is about 0.658.

[0046] The dynamic attitude deviation coefficient stored in the database is about 0.729.

[0047] The above data are respectively substituted into the specific formula for calculating the path deviation state evaluation index of the unmanned ship to obtain:

[0048] The path deviation state evaluation index of the unmanned ship is ((|-0.103|*0.220)^0.863)+((0.344*0.658)^0.729)≈0.377.

[0049] In the embodiment, by combining the Beidou navigation data and the propeller operation data, a coupling model between the lateral deviation and the heading deviation, the power deviation and the angular velocity is constructed, which can accurately reflect the attitude deviation degree and the dynamic stability change of the unmanned ship in actual navigation. The path deviation coupling coefficient and the dynamic attitude deviation coefficient are introduced to quantize the influence of the static deviation and the dynamic disturbance on the deviation state respectively, which improves the sensitive recognition ability of the attitude change in the complex navigation environment. The method fully utilizes the real experimental samples for fitting optimization in design, which is helpful to significantly improve the path stability and the regulation response efficiency of the unmanned ship in the long-range autonomous task.

[0050] Specifically, based on the path deviation state evaluation index, the specific steps of performing real-time navigation regulation processing on the unmanned ship are as follows: the path deviation state evaluation index of the unmanned ship is judged and analyzed with a plurality of preset path deviation state evaluation intervals respectively, and each path deviation state evaluation interval corresponds to a navigation regulation measure; based on the navigation regulation measure corresponding to the path deviation state evaluation interval in which the path deviation state evaluation index is located, the real-time navigation regulation processing is performed on the unmanned ship, including but not limited to the following examples:

[0051] Path deviation state evaluation interval I: [0.00, 0.80), representing a stable navigation state, and the corresponding navigation regulation measure is to maintain the current propeller control parameter without adjustment;

[0052] Path deviation state evaluation interval II: [0.80, 1.30), representing a slight deviation state, and the corresponding navigation regulation measure is to make a micro correction of ±3° or less to the heading angle to keep the track consistent;

[0053] Path deviation state evaluation interval III: [1.30, 1.90), representing a moderate deviation state, and the corresponding navigation regulation measure is to implement a direction adjustment between ±3° and ±8° and dynamically adjust the differential ratio of the left and right propellers.

[0054] Path yaw state evaluation interval IV: [1.90, +∞), representing a serious yaw state, and the corresponding navigation adjustment measure is: performing rapid correction action, adjusting the amplitude more than ±8°, and synchronously reducing the propulsion power, and entering the deceleration correction mode.

[0055] In the embodiment, the dynamic identification and responsive control of the navigation state of the unmanned ship are realized, the track stability and navigation safety are effectively improved, by dividing the continuous yaw index into multiple intervals and setting corresponding navigation adjustment measures for each interval, the system can take differentiated processing strategies under different yaw degrees, avoiding the problems of excessive intervention or response lag caused by the one-size-fits-all control mode, and only making fine adjustment when the yaw is slight, ensuring the continuity of the heading; when the yaw is moderate, the differential regulation of the propeller is introduced to enhance the lateral correction ability; when the yaw is serious, active deceleration and large-scale correction are triggered to ensure that the track quickly returns to the reasonable interval, the method realizes the real-time closed loop from perception to control, so that the unmanned ship has adaptive, hierarchical and fine navigation control capability, and is suitable for continuous monitoring tasks in complex water environment.

[0056] Referring to Figure 3 The embodiment of the present application provides a technical scheme: an application platform for monitoring based on Beidou artificial intelligence unmanned ship, comprising: a data receiving unit for receiving water surface environment perception data of a to-be-monitored water area, including water quality perception data and visual perception data; a state analysis unit for analyzing water surface environment perception state index sets of the to-be-monitored water area based on the water surface environment perception data, including water quality perception state index and visual perception state index; an abnormality analysis unit for analyzing the water surface environment comprehensive evaluation index of the to-be-monitored water area based on the water surface environment perception state index set, and judging and analyzing with the preset water surface environment abnormality evaluation interval, and regarding the water surface environment as abnormal when the water surface environment comprehensive evaluation index is within the preset water surface environment abnormality evaluation interval, and sending a water surface environment abnormality alarm to relevant staff.

[0057] The specific formula for calculating the water surface environment comprehensive evaluation index of the to-be-monitored water area is as follows: Wherein, ShP is the water surface environment comprehensive evaluation index of the to-be-monitored water area, SzG is the water quality perception state index of the to-be-monitored water area, δ1 is the water quality influence weight coefficient stored in the database, SjG is the visual perception state index of the to-be-monitored water area, and δ2 is the visual influence weight coefficient stored in the database.

[0058] It should be explained that the specific acquisition steps of the water quality influence weight coefficient δ1 and the visual influence weight coefficient δ2 stored in the database are as follows:

[0059] First, a dataset is constructed including a plurality of typical water area abnormal event samples, each sample recording the water quality perception state index, the visual perception state index and the result label of the water area environment being judged as abnormal or normal during the corresponding period of the event, and the samples cover various abnormal types such as water pollution, floating object aggregation, oil film coverage and different situations to ensure the representativeness of the samples.

[0060] Secondly, the fusion formula structure (such as linear weighted form) is set by using the calculated water quality perception state index and visual perception state index in each sample, and an optimization target is set: the judgment error between the integrated evaluation index of the water surface environment calculated after fusion and the sample label (i.e. whether abnormal) is minimized, and the target can use the maximum classification accuracy, the minimum mean square error or the AUC promotion as the evaluation index.

[0061] Thirdly, the training algorithm (such as logistic regression fitting, weighted least squares or improved gradient descent) is used to learn all samples, dynamically adjust the weight values of the water quality perception state index and the visual perception state index in the fusion model until the optimization target converges, and the two weight values finally obtained are the water quality influence weight coefficient and the visual influence weight coefficient respectively.

[0062] Specifically, the water quality perception data includes water temperature value, conductivity value, dissolved oxygen value, PH value, water suspended particle density value, oxidation reduction potential, ammonia nitrogen concentration value, cyanobacterial fluorescence intensity value, and water surface oil film conductivity disturbance value. The specific steps of analyzing the water quality perception state index of the water area to be monitored are as follows: obtaining water quality perception reference data of the water area to be monitored, including water temperature reference value, conductivity reference value, dissolved oxygen reference value, PH reference value, water suspended particle density reference value, oxidation reduction reference potential, ammonia nitrogen concentration reference value, cyanobacterial fluorescence intensity reference value, and water surface oil film conductivity disturbance reference value; and comprehensively analyzing the water quality perception data of the water area to be monitored in combination with the corresponding water quality perception reference data to obtain the water quality perception state index of the water area to be monitored.

[0063] The specific formula for calculating the water quality perception state index of the water area to be monitored is as follows: Wherein, SzG is the water quality perception state index of the water area to be monitored, CsZ i is the i-th water quality perception parameter (including water temperature value, conductivity value, dissolved oxygen value, PH value, water suspended particle density value, oxidation reduction potential, ammonia nitrogen concentration value, cyanobacterial fluorescence intensity value, and water surface oil film conductivity disturbance value) of the water area to be monitored, and CsC ia weight coefficient of the i-th water quality sensing parameter, i = 1, 2, 3, …, i0, i0 being the number of water quality sensing parameters in the water quality sensing data. i a weight coefficient of the i-th water quality sensing parameter, i = 1, 2, 3, …, i0, i0 being the number of water quality sensing parameters in the water quality sensing data.

[0064] It should be explained that the weight coefficient λ i of the i-th water quality sensing parameter is obtained by the following steps: first, a historical sample library containing different types of typical water quality abnormal events is established, the water quality abnormal events including but not limited to eutrophication events, organic pollution events, algal outbreak events, heavy metal leakage events, etc., the sample recording the occurrence period, water body location and historical value sequence of the water quality sensing parameters corresponding to each abnormal event, second, the change values of each water quality sensing parameter in the continuous period before and after the occurrence of each abnormal event are extracted from the record of each abnormal event, the relative change rate of each parameter in the occurrence process of the abnormal event is calculated as the response strength of the parameter to the abnormal event, third, the response strength of each parameter in multiple abnormal events is weighted and averaged to obtain the average sensitivity index value of the parameter; then, the average sensitivity index values of all parameters are normalized to map the numerical range to the interval of 0 to 1, the normalized sensitivity index values are nonlinearly transformed (the normalized sensitivity index value of each parameter is taken as the base, a constant greater than 1 is set as the index (such as the index value is set to 2 or 3), and the power operation is performed), and the transformation result is normalized to obtain the weight coefficient λ i of the i-th water quality sensing parameter, which is uniformly stored in the database for subsequent calculation of the water quality sensing state index.

[0065] In the embodiment, by collecting water temperature, conductivity, dissolved oxygen, pH value, suspended particle density and other water quality sensing data, and combining the corresponding water quality reference value for difference analysis, a quantifiable water quality sensing state index is constructed, which effectively realizes the overall evaluation of the degree of water quality change. Unlike the traditional single-index judgment method, this method can take into account multiple water quality factors at the same time, avoiding the problem of false judgment caused by single parameter abnormality. By archiving and modeling the historical water quality abnormal events, the change law of each parameter in the actual emergency scenario is mined, and the weight coefficient is further generated, which can highlight the parameter item that plays a key role in the abnormal precursor, making the evaluation result more targeted and identifiable. In addition, the data parameters used by this method can be obtained in real time by conventional water quality sensors, which has good on-site deployment conditions and is stable for operation on the unmanned ship platform, forming a closed-loop processing flow of automatic sensing-real-time evaluation-intelligent warning, and improving the response efficiency and early intervention ability of the system to water quality fluctuations.

[0066] Specifically, the visual perception data is specifically water surface image data, including pixel values of a plurality of water surface pixel points, and the specific steps of analyzing the visual perception state index of the water area to be monitored are as follows: reading the water surface image data of the water area to be monitored, and inputting the water surface image data into the pre-trained visual analysis model for feature extraction to obtain the visual perception feature set of the water area to be monitored, including the proportion of abnormal color patches, the water surface reflectivity uniformity value, the image texture disturbance value, the boundary fragmentation value and the target detection quantity value; based on the visual perception feature set, the visual perception state index of the water area to be monitored is analyzed, and the calculation formula is as follows: Wherein, SjG is the visual perception state index of the water area to be monitored, YsZ is the proportion of abnormal color patches of the water area to be monitored, μ1 is the abnormal color patch influence coefficient stored in the database, SfJ is the water surface reflectivity uniformity value of the water area to be monitored, μ2 is the reflectivity disturbance influence coefficient stored in the database, TwR is the image texture disturbance value of the water area to be monitored, μ3 is the texture disturbance influence coefficient stored in the database, BjP is the boundary fragmentation value of the water area to be monitored, μ4 is the boundary fragmentation influence coefficient stored in the database, MjS is the target detection quantity value of the water area to be monitored, and μ5 is the target recognition influence coefficient stored in the database.

[0067] It needs to be explained that the specific acquisition steps of the abnormal color patch influence coefficient μ1, the reflectivity disturbance influence coefficient μ2, the texture disturbance influence coefficient μ3, the boundary fragmentation influence coefficient μ4 and the target recognition influence coefficient μ5 stored in the database are as follows: image samples under a plurality of actual water area scenes are collected, and each group of images is labeled with a corresponding visual anomaly level label by manual or expert system, for example, normal, mild abnormality, moderate abnormality or severe abnormality, then, based on the pre-trained visual analysis model, the above-mentioned feature parameters are extracted from each group of image data, and all feature parameters are uniformly normalized to form a feature parameter set, the normalized feature parameters are paired with the manually labeled abnormality level labels to form a sample training set. An optimization target is set, that is, the error between the visual perception state index value calculated by the feature weight in all samples and the actual abnormality level label thereof is minimized, based on the target, a fitting algorithm (such as gradient descent or least squares fitting) in supervised learning is used for model training, so as to determine the optimal weight proportion of the five feature parameters in the overall index evaluation, and finally the five weight values are obtained as μ1, μ2, μ3, μ4 and μ5.

[0068] The visual analysis model includes:

[0069] The image input layer receives the original water surface image collected by the unmanned ship, the image size is HxW, the image data is in RGB three-channel format, and the original image is input into the model after uniform scaling and normalization.

[0070] Color perception convolution layer: used to extract color distribution information in the image, and the output feature map is used to calculate the heterochromatic patch proportion value. The specific steps are as follows: the convolution kernel with a kernel size of 3*3 is used for sliding operation to extract color block regions, and the output color feature map is clustered and segmented (for example, using K-means algorithm), the regions with color distribution significantly deviating from the background color are identified, and the pixel proportion of the deviated regions in the whole image is counted to obtain the heterochromatic patch proportion value.

[0071] Reflectivity analysis layer: used to estimate the uniformity of water surface reflected light, corresponding to extract the water surface reflectivity uniform value. The specific steps are as follows: the input image is processed by gray scale, and the brightness mean value of each region is calculated. The local brightness variance is used as a reflection disturbance index, and the smaller the variance, the more uniform the reflection. The average value of the local variance of the whole image is normalized to obtain the reflectivity uniform value.

[0072] Texture disturbance extraction layer: used to extract water surface texture changes, corresponding to extract the image texture disturbance value. The specific steps are as follows: a multi-scale Gabor filter bank (such as 5 frequencies x 4 directions) is used to analyze the response of the image, and the energy spectrum of the response result is extracted to represent the distribution characteristics of different texture directions and frequencies in the image. The variance of the energy spectrum is calculated as a texture disturbance measure, and the larger the value, the more significant the water surface disturbance.

[0073] Edge structure extraction layer: performs structure boundary identification on the image, corresponding to extract the boundary fragmentation value. The specific steps are as follows: the Canny operator is used to extract the edge contour of the image, and the contour connectivity analysis is performed on the edge map. The average contour length and the number of broken segments are calculated, and the boundary fragmentation index is formed by the number of edge broken segments / the number of continuous edge segments. The larger the value, the higher the degree of water surface fragmentation.

[0074] Target detection layer: based on YOLOv5 (or other lightweight target detection framework), the number of abnormal objects in the image is extracted, corresponding to extract the target detection number value. The specific steps are as follows: the input image is subjected to an anchor-based candidate box extraction mechanism, and abnormal objects such as floating objects, garbage, plastic bottles, etc. are identified through classification and confidence threshold, and the number of detection targets with recognition confidence exceeding a certain threshold (such as 0.6) is counted as the target detection number value.

[0075] The pre-training steps of the visual analysis model are as follows:

[0076] First, a sample dataset containing multiple types of typical water surface environment images is established. The image samples are collected by an unmanned ship in multiple actual water areas, covering normal clean water surface, water surface containing floating objects, oil film covered water surface, algae outbreak area, and turbid water area, etc. For each image, a professional in the field of water quality monitoring or an expert system is organized for manual labeling, and the corresponding visual environment level label is set, including "normal", "mild abnormality", "moderate abnormality" and "serious abnormality". Each image sample records the shooting location, time and weather conditions, etc. to form a standardized training input.

[0077] The constructed sample set is input into a multi-layer visual analysis model, which includes a color perception convolution layer, a reflectivity analysis layer, a texture disturbance extraction layer, an edge structure extraction layer and a target detection layer. Each layer corresponds to extract an environmental perception feature parameter (such as the proportion of color spots, the uniform value of water surface reflectivity, etc.) in the image. The model outputs the feature values as a feature vector, which is combined to form a visual perception state index. The training target is set to minimize the error between the state index extracted from each image and its manually labeled environment level value by optimizing the parameter weight in the feature extraction process. The mean square error is used as the loss function to fit the approximate relationship between the perception index and the expert scoring result.

[0078] A supervised learning mechanism is used to iteratively train the model on the constructed training sample set. The Adam optimization algorithm is used, with an initial learning rate of 0.001 and a batch size of 32. The model calculates the validation set error after each iteration and adjusts the learning rate using an adaptive learning rate strategy (such as ReduceLROnPlateau). The training is terminated when the error converges or reaches a set number of rounds (such as 100 rounds).

[0079] In this embodiment, the visual perception analysis method can automatically extract multiple key feature values based on the images collected by the unmanned ship, comprehensively judge the water surface environment state, and effectively realize real-time identification and grading of abnormal water areas. By extracting multiple features such as color spots, reflection disturbances, texture fluctuations, boundary fragmentation and abnormal target quantities through a layered model structure, the blind spots and lag problems existing in traditional manual inspection or single indicator detection are avoided. The weight coefficients are supervised trained based on expert labeled samples, so that the model result has good matching degree with the actual environment. The overall processing flow supports automatic conversion from image to indicator, and can be directly applied to edge device end operation, suitable for intelligent monitoring of dynamic water environment.

[0080] Although the preferred embodiments of the present application have been described, those skilled in the art can make additional changes and modifications to these embodiments once they know the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0081] Obviously, many modifications and variations of the present application are possible in light of the above teachings. It is, therefore, to be understood that within the scope of the appended claims and their equivalents, the application can be practiced otherwise than as specifically described.

Claims

1. A Beidou-based artificial intelligence unmanned ship monitoring application system, characterized in that, The application relates to a path generation module, a path execution module and a Beidou regulation and sensing module. The path generation module is used for acquiring boundary coordinate point sets of a water area to be monitored and generating a monitoring navigation path, the monitoring navigation path comprising a plurality of monitoring navigation points. The path execution module is used for controlling the unmanned ship to navigate along the monitoring navigation path in sequence, and acquiring Beidou navigation state data of the unmanned ship in real time during the navigation process, the Beidou navigation state data comprising Beidou navigation two-dimensional coordinates, a navigation heading angle, a lateral angular velocity and a propeller feedback power. The Beidou regulation and sensing module is used for analyzing a path deviation state evaluation index of the unmanned ship based on the Beidou navigation state data, wherein the path deviation state evaluation index is obtained by extracting the nearest monitoring navigation point to the direction in which the unmanned ship travels in the monitoring navigation path, analyzing a lateral deviation distance and a heading angle deviation of the unmanned ship, acquiring a reference propeller propelling power of the unmanned ship, and analyzing a deviation between the reference propeller propelling power and a corresponding propeller feedback power to obtain a propeller power deviation amount of the unmanned ship, and comprehensively analyzing the lateral deviation distance, the heading angle deviation, the propeller power deviation amount and a lateral angular velocity to obtain the path deviation state evaluation index. ； Wherein, 、 、 、 、 The path deviation state evaluation index of the unmanned ship, the lateral offset distance, the heading angle deviation, the propeller power deviation amount and the travel lateral angular velocity are in turn, 、 The path offset coupling coefficient and the dynamic posture deviation coefficient stored in the database are in turn, and the specific acquisition steps are as follows: For the path deviation coupling coefficient, a plurality of representative monitoring water areas are constructed, a standard monitoring navigation path is planned in each area, the unmanned ship is controlled to navigate along the planned path, the current position coordinates and the corresponding heading angle are continuously collected during the whole process, the system calculates the lateral deviation distance between the current coordinates and the nearest navigation point on the path and the heading angle deviation between the current heading angle and the tangent direction of the path in real time, and the two values are multiplied to obtain a composite deviation sample at the current time, each deviation sample is associated with whether a heading regulation operation is triggered at the time to form a data set with regulation level labels, a linear regression model is established for the data set, the deviation product is taken as the input feature, the regulation level code value is taken as the output, the least square method is used for fitting training, after the training is completed, the absolute value of the fitting regression coefficient is extracted as the path deviation coupling coefficient, and the residual sum of squares of all experimental samples is statistically analyzed to select a group of experimental coefficients with the minimum error as the final value, that is, the path deviation coupling coefficient. For the dynamic attitude deviation coefficient, in the same experiment, the reference propeller propelling power and the propeller feedback power of the unmanned ship at each time point are continuously collected, and the difference between the two is calculated as the propeller power deviation amount, and the lateral angular velocity at the time point, that is, the rotation speed of the ship body around the vertical axis, is collected, considering that the units of the propeller power and the angular velocity are different, in order to ensure the dimensionality of the synthesized variable, the two parameters are first normalized to the value range of [0, 1], and then the two are weighted and averaged with equal weights to obtain a joint dynamic attitude disturbance variable, the joint dynamic attitude disturbance variable is taken as the input feature, the regulation behavior level recorded by the system is taken as the output variable, a binary Logistic regression model is constructed for fitting, and the nonlinear response relationship between the disturbance variable and the regulation behavior is automatically learned during the training process, the dominant weight coefficient in the model is extracted and defined as the dynamic attitude deviation coefficient. The path intelligent adjustment module is used for performing real-time navigation adjustment processing on the unmanned ship based on the path yaw state evaluation index, and specifically comprises the following steps: A first measure, corresponding is in the first interval, the instruction to maintain the current thruster parameters is executed; The second measure, corresponding In the second interval, a micro correction of ±3° in the heading angle is performed, maintaining the command to keep the track adherent. Third measure, corresponding In the third interval, a directional adjustment of between ±3° to ±8° to the heading angle is performed, and the command for the differential ratio of the left and right thrusters is dynamically adjusted. The fourth measure corresponds to The fourth interval corresponds to the instruction of performing a rapid correction action on the heading angle, adjusting the amplitude to more than ±8°, and simultaneously reducing the propulsion power to enter the deceleration correction mode. The data sampling output module is used for sampling the water surface environment perception data of the to-be-monitored water area at intervals according to the set path distance interval and performing output processing.

2. The Beidou-based artificial intelligence unmanned ship monitoring application system according to claim 1, characterized in that, The boundary coordinate point set comprises a plurality of boundary two-dimensional coordinate points, and the specific steps of generating the monitoring navigation path are as follows: A boundary box region is constructed based on the boundary coordinate point set, and covering grid line segments are generated in the boundary box according to a preset path interval; The covering grid line segments are subjected to spatial clipping processing with the to-be-monitored water area, and are sequentially connected in a preset order to form an initial path trajectory; The initial path trajectory is subjected to turning smoothing processing to obtain a smooth path curve; A plurality of monitoring navigation points are extracted from the smooth path curve at equal path distance intervals to form the monitoring navigation path.

3. A Beidou-based artificial intelligence unmanned ship monitoring application platform, applying the Beidou-based artificial intelligence unmanned ship monitoring application system of any one of claims 1-2, characterized in that, The data receiving unit is configured to receive water surface environment perception data of the to-be-monitored water area, including water quality perception data and visual perception data. The state analysis unit is configured to analyze a water surface environment perception state index set of the to-be-monitored water area based on the water surface environment perception data, including a water quality perception state index and a visual perception state index. The abnormality analysis unit is configured to analyze a water surface environment comprehensive evaluation index of the to-be-monitored water area based on the water surface environment perception state index set, and judge and analyze the water surface environment comprehensive evaluation index with a preset water surface environment abnormality evaluation interval. When the water surface environment comprehensive evaluation index is within the preset water surface environment abnormality evaluation interval, the water surface environment is considered to be abnormal, and a water surface environment abnormality alarm is sent. The specific formula for calculating the water surface environment comprehensive evaluation index of the to-be-monitored water area is as follows:

4. The Beidou-based artificial intelligence unmanned ship monitoring application platform according to claim 3, characterized in that, The water quality perception data includes water temperature value, conductivity value, dissolved oxygen value, PH value, water suspended particle density value, oxidation-reduction potential, ammonia nitrogen concentration value, blue algae fluorescence intensity value, and water surface oil film conductivity disturbance value. The specific steps for analyzing the water quality perception state index of the to-be-monitored water area are as follows: ； Wherein, , , The water surface environment comprehensive evaluation index, the water quality perception state index, and the visual perception state index of the water area to be monitored are sequentially, , The water quality influence weight coefficient and the visual influence weight coefficient stored in the database are sequentially. 5.The Beidou artificial intelligence unmanned ship monitoring application platform based on Beidou according to claim 3, characterized in that, Obtain water quality perception reference data of the to-be-monitored water area, including water temperature reference value, conductivity reference value, dissolved oxygen reference value, PH reference value, water suspended particle density reference value, oxidation-reduction reference potential, ammonia nitrogen concentration reference value, blue algae fluorescence intensity reference value, and water surface oil film conductivity disturbance reference value. The water quality perception data of the to-be-monitored water area is combined with the corresponding water quality perception reference data for comprehensive analysis to obtain the water quality perception state index of the to-be-monitored water area. The visual perception data is specifically water surface image data, including pixel values of a plurality of water surface pixels. The specific steps for analyzing the visual perception state index of the to-be-monitored water area are as follows: 6.The Beidou artificial intelligence unmanned ship monitoring application platform based on Beidou according to claim 3, characterized in that, ​ The water surface image data of the water area to be monitored is read and input into a pre-trained visual analysis model for feature extraction to obtain a visual perception feature set of the water area to be monitored, including a heterochromatic spot block proportion value, a water surface reflectivity uniformity value, an image texture disturbance value, a boundary fragmentation degree value and a target detection quantity value; Based on the visual perception feature set, the visual perception state index of the water area to be monitored is analyzed.

Images