Intelligent sensing navigation system and navigation method for unmanned ship

Abstract

The invention discloses an intelligent sensing navigation system and a navigation method for an unmanned ship, and belongs to the technical field of unmanned ship navigation.According to the method, a visual sensor, a radar and a positioning device are combined, environment information is synchronously collected, data are verified in real time, and the trust weight of the sensor is adjusted; the method comprises the following steps: setting two parallel navigation control units, wherein the first control unit constructs a dynamic decision model based on environment data and sensor trust weight, and generates a navigation instruction; the second control unit generates a basic course instruction only according to the current position and the obstacle position, and when the credibility of the first control unit is reduced due to a sensor abnormal instruction, the unmanned ship is automatically switched to the second control unit and sends an abnormal early warning to the control center; the control center judges task feasibility according to the basic course instruction, plans a return route if the task is not feasible, and splits the instruction into multiple channels for sending; and after receiving the verification, the unmanned surface vehicle combines and executes the homeward voyage task.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned boat navigation, and in particular to an intelligent perception navigation system and navigation method for unmanned boats. Background Art

[0002] The current intelligent navigation system of unmanned boats mainly relies on a multi-sensor data fusion architecture to achieve environmental perception and positioning by integrating heterogeneous sensors such as vision, lidar, millimeter-wave radar, and inertial navigation unit (IMU). However, although the existing solutions in the prior art adopt parallel multi-sensor acquisition, when some sensors deviate due to environmental interference or hardware failures, the system usually adopts a fixed-priority weighted fusion strategy and cannot dynamically eliminate abnormal data sources, resulting in the fusion result deviating from the true state. For example, when relying on pure inertial navigation after GPS signal loss, the cumulative error will increase with the mission duration and may ultimately lead to navigation failure. In addition, when the sensors encounter electromagnetic interference, weather effects, or hardware failures, their performance may drop significantly, easily leading to navigation failure. Moreover, the single control model (such as pre-programmed path or PID control) equipped on unmanned boats lacks dynamic adaptability and cannot effectively cope with sudden obstacles or environmental changes. At the same time, when remote control instructions are transmitted through a single link, they are vulnerable to channel fading or interference, resulting in instruction delay or loss, thus seriously affecting the autonomous navigation ability of unmanned boats in complex scenarios. Summary of the Invention

[0003] The purpose of the present invention is to provide an intelligent perception navigation system and navigation method for unmanned boats to solve the problems raised in the above background art.

[0004] To solve the above technical problems, the present invention provides the following technical solution: An intelligent perception navigation method for unmanned boats, the method comprising: Step S100: Synchronously collect the environmental information around the unmanned boat through a heterogeneous sensor array composed of a vision sensor, a radar, and a positioning device, verify and identify the data output by each sensor in real time, exclude abnormal data, and dynamically adjust the trust weights of each sensor in combination with the historical data of each sensor; Step S200: Set two parallel navigation control units, the navigation control unit including a first control unit and a second control unit; the first control unit constructs a dynamic decision-making model based on the collected environmental data and sensor trust weights to generate real-time navigation control instructions; the second control unit does not rely on real-time sensor data and only generates a basic heading instruction according to the current position of the unmanned boat and the position of the obstacle; when the instruction credibility of the first control unit decreases due to sensor abnormalities, the unmanned boat automatically switches to the second control unit, sails according to the basic heading instruction, and at the same time triggers an abnormal warning signal to be sent to the control center; Step S300: After the control center receives the abnormal warning signal, it determines whether the unmanned boat can go to the designated position to complete the task according to the basic course instruction. If it is determined that it cannot be completed, a return route is planned based on the real-time data; Step S400: Generate corresponding navigation instructions according to the path planning scheme in Step S300, split the instructions into multiple segments, and send them through different channels respectively. After the unmanned boat receives the instruction data, it verifies and combines them to execute the return task.

[0005] Further, the said Step S100 includes: Step S101: Deploy a lidar, a camera, a millimeter wave radar, a sonar and a GPS positioning device on the unmanned boat; the lidar is used to measure the distance and position of surrounding obstacles and generate a three-dimensional point cloud map; the camera is used to identify target objects and analyze the captured images; the millimeter wave radar is used to monitor the speed and distance of objects around the unmanned boat; the sonar is installed at the bottom of the unmanned boat for underwater target monitoring and depth measurement; the GPS positioning device is used to obtain the position information of the unmanned boat in real time; create a separate thread for each sensor, let each thread run simultaneously, and each sensor periodically obtains data according to its own acquisition frequency and stores the data in a temporary buffer; Step S102: After reading the data collected by each sensor from the temporary buffer, perform time synchronization processing, and use the linear interpolation method to align the data in time; compare the measured value data of the same type, calculate the difference between the data, and compare it with the set error threshold; if the difference exceeds the threshold, it is considered that the data of the two sensors at the same moment is inconsistent, marked as a conflict data pair, stored in the abnormal candidate set; at the same time, mark the sensor corresponding to the abnormal data acquisition. Step S103: Further calculate the mean and standard deviation of the historical data of each sensor to identify abnormal data. Set the data collected by the i-th sensor at the past n moments as [x i1 , x i2 ,..., x in , where x in represents the data collected by the i-th sensor at the n-th moment, then the mean μ i = 1 / nΣ n j=1 x ij , x ij represents the data collected by the i-th sensor at the j-th moment, j represents the index of the time series, and the value ranges from 1 to n; the standard deviation σ i = [1 / nΣ n j=1 (x ij - μ i ) 2 1 / 2 ​; Compare the currently collected data with the historical statistical data. If |x ic -μ i |>kσ i , where k is a preset multiple, and x ic represents the data collected in real time by the i-th sensor. If the current data deviates from the mean by more than k times the standard deviation and there is no data anomaly in other sensors, it is determined as a data anomaly, and this abnormal data is screened out from the conflict data pairs for exclusion; Step S104: After excluding the abnormal data, dynamically adjust the trust level of each sensor; initially, assign an initial trust weight w i (0) to each sensor, representing the initial weight of the i-th sensor. According to the error rate E i (t) and accuracy rate A i (t) of the sensor data, calculate the real-time data reliability r i (t)=(1 - E i (t)) * A i (t), where E i (t) represents the error rate of the i-th sensor at time t, A i (t) represents the accuracy rate of the i-th sensor at time t, and r i (t) represents the data reliability of the i-th sensor at time t; for the error rate, calculate by dividing the deviation between the measured value of the sensor and the recorded true value by the true value, and the true value is determined by taking the average value through mutual comparison between sensors; the accuracy rate is calculated by the ratio of the number of correct measurements to the total number of measurements; according to the reliability of each sensor, dynamically adjust the trust weight of the sensor according to the formula: w i (t + 1)=αw i (t)+(1 - α)r i (t); where, w i (t) represents the trust weight of the i-th sensor at time t, w i (t + 1) represents the adjusted trust weight of the i-th sensor at the (t + 1)-th moment, α represents a weight adjustment coefficient, and its value range is (0, 1). At the same time, normalize the adjusted weight.

[0006] Furthermore, the step S200 includes: Step S201: The first control unit is based on the environmental data after multi-sensor fusion in step S100 and the trust weights w i(t), construct a dynamic decision-making model; the input layer data includes the obstacle position matrix, the angle between the current heading and the target heading, and the remaining energy vector of the sensors, to form an environmental state matrix M; for the obstacle position matrix, the obstacle position information collected by different sensors is fused, and each row of the matrix represents the position of an obstacle; for the remaining energy vector of the sensors, each sensor is equipped with an energy monitoring device, and the remaining energy data collected by each sensor is arranged in the order of the corresponding sensor numbers to form a vector, and each element in the vector corresponds to the remaining energy value of a sensor; construct the trust weights of each sensor as a vector W and embed it into the input layer of the model; generate the heading angle adjustment amount Δθ and the target speed v through a three-layer fully connected neural network, and the calculation formula is: Y = f NN (M⊕W)=[Δθ,v]; where f NN represents the forward calculation of the neural network, and ⊕ represents the matrix splicing operation; Step S202: The second control unit constructs a basic navigation model, takes the recorded position of the obstacle as the center of the circle, and generates a repulsive field with a radius of r. The heading angle offset is: Δθ r =∑ p m=1 [(x - x m ) / d m 2 , d m =[(x - x m ) 2 +(y - y m ) 2 1 / 2 ; where (x, y) is the current position coordinate of the unmanned boat, (x m , y m ) is the obstacle coordinate, and p is the number of obstacles; d m represents the Euclidean distance from the current position of the unmanned boat to the position of the m-th obstacle; set the minimum distance d min , when d m < d min , set d m = d min ; The final heading angle is: θ z = θ c + λΔθ r +(1 - λ)Δθ a , where θ z is the final heading angle, and θ c is the initial heading angle; Δθ a represents the target position attraction direction angle: Δθ a = atan2[(y goal - y) / (x goal ​-x)],(x goal , y goal ) represents the position coordinates of the target point; λ represents the repulsive weight coefficient, λ ∈ [0, 1]; Step S203: Calculate the credibility of the first control unit instruction: C = [∑ k i=1 w i (t) * exp(-|x ic - μ i | / σ i )] / [∑ k i=1 w i (t)], set the credibility threshold C th , when C < C th , automatically disconnect the output of the first control unit, enable the heading instruction generated by the second control unit and record the switching time, and at the same time trigger an early warning signal to be sent to the control center.

[0007] Further, the step S300 includes: Step S301: After receiving the early warning signal, the control center calls the environmental data after multi-sensor fusion in step S100, including the obstacle position, the current position of the unmanned boat, the heading, the remaining energy of the sensor, and the sensor credibility C in step S203; based on the historical environmental data and energy data obtained in step S100, a historical energy consumption model is constructed through a linear regression algorithm to predict the energy consumption of the unmanned boat according to the current environmental conditions and navigation parameters; at the same time, combining the remaining voyage and the current environment input into the model, calculate the energy required to complete the current task; if the total remaining energy of the unmanned boat is less than q times the energy required for the current task, it is determined that the unmanned boat has insufficient energy and cannot complete the current task; where q is a safety factor, and a safety margin is reserved by setting the value of q, q ∈ (1, 1.5]; if the current sensor trust degree C < C th and the duration exceeds the preset threshold T th , it is determined that the sensor cannot support the remaining voyage; if any of the above conditions is met, enter the return path planning stage; otherwise, the unmanned boat continues to execute the current task; Step S302: Calculate the distances from the current position of the unmanned boat to each safe mooring point, and select the mooring point with the shortest distance as the return target point; fuse the three-dimensional point cloud map generated in step S100 with the sonar data to construct a two-dimensional grid map, each grid represents a region, and is marked as passable and impassable according to the number and distribution of obstacles; adopt the heuristic function of the A* algorithm, consider the distance, energy consumption and obstacles of the target point, and combine the grid map to plan the return route.

[0008] Further, the step S400: Step S401: According to the path planning scheme in step S300, including the target position coordinates, corresponding course angles, target speeds, and the estimated time to reach each target position, the control center organizes them into a complete navigation instruction in the set JSON format; splits it according to the distance of the voyage segment, and every navigation instruction of L meters is a segment. Mark the segment number, timestamp, and check code for each split instruction segment; where the segment number adopts the format of "current segment sequence number / total number of segments", the timestamp is used to record the segment generation time, and the check code is calculated by performing a hash algorithm on the segment content; the control center selects multiple communication channels equipped on the unmanned boat to distribute the marked instruction segments, and distributes them in a round-robin manner according to the load conditions of the channels; at the same time, the control center monitors the sending status of each segment in real time. If a certain segment fails to be sent, record the segment information and try to resend it. Step S402: The communication device equipped on the unmanned boat receives the instruction segments from different communication channels in real time. For each received instruction segment, first perform the check code verification, recalculate the same hash algorithm for the segment content as that of the control center, and compare it with the check code carried in the segment. If they are inconsistent, request a resend; then check whether the received segments are arranged in order according to the segment numbers. If a segment is found to be missing or out of order, record the relevant information and send a feedback signal to the control center to request the resending of the relevant segments; sort the verified instruction segments according to the segment numbers and merge them into a complete navigation instruction. During the merging process, check whether the information between adjacent segments is coherent. If the data shows a mutation, request the relevant segments from the control center again; after verification, the unmanned boat executes the return mission according to the instruction.

[0009] An intelligent perception navigation system for an unmanned boat, the system includes: A data acquisition module, a decision control module, a path planning module, and an instruction operation module; The data acquisition module synchronously acquires the environmental information around the unmanned boat through a heterogeneous sensor array composed of visual sensors, radars, and positioning devices, verifies and identifies the data output by each sensor in real time and excludes abnormal data, and dynamically adjusts the trust weights of each sensor in combination with the historical data of each sensor; The decision control module is used to set two sets of parallel navigation control strategies, construct a dynamic decision model based on the acquired environmental data and sensor trust weights, and generate real-time navigation control instructions; at the same time, it does not rely on real-time sensor data, and only generates a basic course instruction according to the current position of the unmanned boat and the position of obstacles; when the reliability of the control instruction decreases due to abnormal sensor data, the unmanned boat automatically switches to the basic navigation instruction, and at the same time triggers an abnormal warning signal to be sent to the control center; After the path planning module receives the abnormal warning signal at the control center, it determines whether the unmanned boat can go to the designated position to complete the task according to the basic heading instruction. If it is determined that the task cannot be completed, it plans a return route based on real-time data; The instruction operation module generates corresponding navigation instructions according to the path planning scheme, splits the instructions into multiple segments, and sends them through different channels respectively. After receiving the instruction data, the unmanned boat verifies and combines them to execute the return task.

[0010] Compared with the prior art, the beneficial effects achieved by the present invention are as follows: Through the synchronous acquisition and data verification of the heterogeneous sensor array, the present invention effectively improves the accuracy and reliability of environmental information, and by dynamically adjusting the trust weights of each sensor, the unmanned boat can more intelligently select the sensor data with high reliability in the face of complex environments, so as to generate more accurate navigation control instructions; By setting two sets of parallel navigation control units, the present invention realizes the redundant backup of the navigation system; when the credibility of the instructions of the first control unit decreases due to sensor abnormalities, the unmanned boat can automatically switch to the second control unit and continue to sail according to the basic heading instruction. This adjustment of the strategy significantly improves the emergency handling ability of the unmanned boat in case of sensor failures and ensures the safety and continuity of navigation; After receiving the abnormal warning signal, the present invention quickly judges the task progress of the unmanned boat according to the basic heading instruction. If it is determined that the designated task cannot be completed, it plans a return route based on real-time data and splits and sends the instructions to the unmanned boat through multiple communication channels. This not only improves the reliability of instruction transmission, but also effectively reduces the impact of a single communication channel failure on the integrity of the instructions; this setting of the abnormal handling and return execution strategy can enhance the autonomous response ability of the unmanned boat in complex environments and the reliability of task completion. Description of the Drawings

[0011] The drawings are used to provide a further understanding of the present invention and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the present invention and do not constitute a limitation to the present invention. In the drawings: Figure 1 It is a method flow chart of an intelligent perception navigation method for an unmanned boat. Detailed Embodiments

[0012] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0013] Please refer to Figure 1 , the present invention provides a technical solution: an intelligent perception navigation method for an unmanned boat, the method comprising: Step S100: Synchronously collect the environmental information around the unmanned boat through a heterogeneous sensor array composed of a vision sensor, a radar, and a positioning device, verify and identify the data output by each sensor in real time and exclude abnormal data, and dynamically adjust the trust weights of each sensor in combination with the historical data of each sensor; Step S200: Set two sets of parallel navigation control units, the navigation control unit including a first control unit and a second control unit; the first control unit constructs a dynamic decision-making model based on the collected environmental data and the sensor trust weights, and generates real-time navigation control instructions; the second control unit does not rely on the real-time data of the sensors, and only generates a basic heading instruction according to the current position of the unmanned boat and the position of the obstacles; when the instruction credibility of the first control unit decreases due to sensor anomalies, the unmanned boat automatically switches to the second control unit, sails according to the basic heading instruction, and simultaneously triggers an anomaly warning signal to be sent to the control center; Step S300: After receiving the anomaly warning signal, the control center determines whether the unmanned boat can go to the designated position to complete the task according to the basic heading instruction. If it is determined that it cannot be completed, a return route is planned based on the real-time data; Step S400: Generate corresponding navigation instructions according to the path planning scheme of Step S300, split the instructions into multiple segments, and send them through different channels respectively. After receiving the instruction data, the unmanned boat verifies and combines them to execute the return task.

[0014] Further, the Step S100 includes: Step S101: Deploy a lidar, a camera, a millimeter wave radar, a sonar, and a GPS positioning device on the unmanned boat; the lidar is used to measure the distance and position of surrounding obstacles and generate a three-dimensional point cloud map; the camera is used to identify target objects and analyze the captured images; the millimeter wave radar is used to monitor the speed and distance of objects around the unmanned boat; the sonar is installed at the bottom of the unmanned boat and is used for underwater target monitoring and depth measurement; the GPS positioning device is used to obtain the position information of the unmanned boat in real time; create a thread for each sensor, let each thread run simultaneously, and each sensor periodically obtains data according to its own acquisition frequency and stores the data in a temporary buffer; Step S102: After reading the data collected by each sensor from the temporary buffer, perform time synchronization processing, and use linear interpolation method to align the data in time; compare the measured value data of the same type, calculate the difference between the data, and compare it with the set error threshold; if the difference exceeds the threshold, it is considered that the data of the two sensors at the same moment is inconsistent, marked as a conflict data pair, and stored in the abnormal candidate set; at the same time, mark the sensors corresponding to the abnormal collected data. Step S103: Further calculate the mean and standard deviation of the historical data of each sensor to identify abnormal data. Set the data collected by the i-th sensor in the past n moments as [x i1 ,x i2 ,...,x in , where x in represents the data collected by the i-th sensor at the n-th moment, then the mean μ i = 1 / nΣ n j=1 x ij ,x ij represents the data collected by the i-th sensor at the j-th moment, j represents the index of the time series, and the value ranges from 1 to n; the standard deviation σ i = [1 / nΣ n j=1 (x ij - μ i ) 2 1 / 2 ; compare the currently collected data with the historical statistical data. If |x ic - μ i | > kσ i , where k is a preset multiple, x ic represents the data collected by the i-th sensor in real time. If the current data deviates from the mean by more than k times the standard deviation and no data anomaly occurs in other sensors, it is determined as data anomaly, and this abnormal data is screened out from the conflict data pair for exclusion. Step S104: After excluding abnormal data, dynamically adjust the trust level of each sensor; initially, assign an initial trust weight w i (0) to each sensor, representing the initial weight of the i-th sensor. According to the error rate E i (t) and accuracy rate A i (t) of the sensor data, calculate the real-time data reliability r i (t) = (1 - E i (t)) * A i (t), where E i (t) represents the error rate of the i-th sensor at the t-th moment, A i (t) represents the accuracy rate of the i-th sensor at the t-th moment, r​i (t) represents the data reliability of the i-th sensor at time t; the error rate is calculated by dividing the deviation between the measured value of the sensor and the recorded true value by the true value, and the true value is determined by taking the average through mutual comparison among the sensors; the accuracy rate is calculated by the proportion of the number of correct measurements to the total number of measurements; according to the reliability of each sensor, the trust weight of the sensor is dynamically adjusted according to the formula: w i (t + 1)=αw i (t)+(1 - α)r i (t); where w i (t) represents the trust weight of the i-th sensor at time t, w i (t + 1) represents the adjusted trust weight of the i-th sensor at the (t + 1)-th moment, α represents a weight adjustment coefficient, and its value range is (0, 1), and at the same time, the adjusted weight is normalized.

[0015] Further, the step S200 includes: Step S201: The first control unit constructs a dynamic decision-making model based on the environmental data after multi-sensor fusion and the trust weights w i (t) of each sensor in step S100; the input layer data includes the obstacle position matrix, the angle between the current heading and the target heading, and the sensor remaining energy vector, forming an environmental state matrix M; for the obstacle position matrix, the obstacle position information collected by different sensors is fused, and each row of the matrix represents the position of an obstacle; for the sensor remaining energy vector, each sensor is equipped with an energy monitoring device, and the remaining energy data collected by each sensor is arranged in the order of the sensor corresponding numbers to form a vector, and each element in the vector corresponds to the remaining energy value of a sensor; the trust weights of each sensor are constructed as a vector W and embedded into the input layer of the model; the heading angle adjustment amount Δθ and the target speed v are generated through a three-layer fully connected neural network, and the calculation formula is: Y = f NN (M⊕W)=[Δθ, v]; where f NN represents the forward calculation of the neural network, and ⊕ represents the matrix splicing operation; Step S202: The second control unit constructs a basic navigation model, generates a repulsive field with a radius of r centered on the recorded position of the obstacle, r is a preset safety distance, and the heading angle offset is: Δθ r =∑ p m=1 [(x - x m ) / d m 2 , d m =[(x - x m ) 2 +(y - ym ) 2 1 / 2 ; where (x, y) are the current position coordinates of the unmanned boat, (x m , y m ) are the obstacle coordinates, and p is the number of obstacles; d m represents the Euclidean distance from the current position of the unmanned boat to the position of the m-th obstacle; set the minimum distance d min , when d m < d min , set d m = d min ; The final heading angle is: θ z = θ c + λΔθ r + (1 - λ)Δθ a , where θ z is the final heading angle, and θ c is the initial heading angle; Δθ a represents the attracting direction angle of the target position: Δθ a = atan2[(y goal - y) / (x goal - x)], (x goal , y goal ) represents the position coordinates of the target point; λ represents the repulsion weight coefficient, λ ∈ [0, 1]; Step S203: Calculate the credibility of the first control unit instruction: C = [∑ k i=1 w i (t)*exp(-|x ic - μ i | / σ i )] / [∑ k i=1 w i (t)], set the credibility threshold C th , when C < C th , automatically disconnect the output of the first control unit, enable the heading instruction generated by the second control unit and record the switching time, and at the same time trigger a warning signal to be sent to the control center.

[0016] Further, the step S300 includes: ​Step S301: After the control center receives the warning signal, it calls the environmental data after multi-sensor fusion in Step S100, including the obstacle position, the current position of the unmanned boat, the heading, the remaining energy of the sensor, and the sensor credibility C in Step S203; based on the historical environmental data and energy data obtained in Step S100, a historical energy consumption model is constructed through a linear regression algorithm to predict the energy consumption of the unmanned boat according to the current environmental conditions and navigation parameters; at the same time, combining the remaining voyage and the current environment and inputting them into the model, the energy required to complete the current task is calculated; if the total remaining energy of the unmanned boat is less than q times the energy required for the current task, it is determined that the unmanned boat has insufficient energy and cannot complete the current task; where q is a safety factor, and a safety margin is reserved by setting the value of q, q ∈ (1, 1.5]; if the current sensor trust degree C < C th and the duration exceeds the preset threshold T th , it is determined that the sensor cannot support the remaining voyage; if any of the above conditions is met, it enters the return path planning stage; otherwise, the unmanned boat continues to execute the current task; Step S302: Calculate the distances from the current position of the unmanned boat to each safe mooring point, and select the mooring point with the shortest distance as the return target point; fuse the 3D point cloud map generated in Step S100 with the sonar data to construct a 2D grid map, where each grid represents an area, and is marked as passable or impassable according to the quantity and distribution of obstacles; adopt the heuristic function of the A* algorithm, consider the distance, energy consumption and obstacles of the target point, and combine the grid map to plan the return route.

[0017] Further, the said Step S400: Step S401: The control center sorts out the complete navigation instructions in the JSON format according to the path planning scheme in Step S300, including the target position coordinates, the corresponding heading angle, the target speed, and the time expected to reach each target position; split them according to the distance of the voyage segment, and each navigation instruction every L meters is a segment, and mark each split instruction segment with a segment number, a timestamp and a check code; where the segment number adopts the format of "current segment serial number / total segment quantity", the timestamp is used to record the segment generation time, and the check code is calculated by performing a hash algorithm on the segment content; the control center selects a variety of communication channels equipped on the unmanned boat to distribute the split and marked instruction segments, and adopts a round-robin distribution method according to the load conditions of the channels; at the same time, the control center monitors the sending status of each segment in real time, and if a certain segment fails to be sent, record the segment information and try to resend it; Step S402: The communication device equipped on the unmanned boat receives instruction fragments from different communication channels in real time. For each received instruction fragment, first perform checksum verification. Recalculate the hash algorithm on the fragment content in the same way as the control center, and compare it with the checksum carried in the fragment. If they are inconsistent, request retransmission. Then, check whether the received fragments are arranged in order according to the fragment numbers. If a fragment is found to be missing or out of order, record the relevant information and send a feedback signal to the control center to request retransmission of the relevant fragments. Sort the verified instruction fragments according to the fragment numbers and merge them into a complete navigation instruction. During the merging process, check whether the information between adjacent fragments is coherent. If the data shows a sudden change, request the relevant fragments from the control center again. After verification, the unmanned boat executes the return mission according to the instruction.

[0018] An intelligent perception navigation system for an unmanned boat, the system comprising: A data acquisition module, a decision control module, a path planning module, and an instruction operation module; The data acquisition module synchronously acquires the environmental information around the unmanned boat through a heterogeneous sensor array composed of visual sensors, radars, and positioning devices, verifies and identifies the data output by each sensor in real time and excludes abnormal data, and dynamically adjusts the trust weights of each sensor in combination with the historical data of each sensor; The decision control module is used to set two sets of parallel navigation control strategies, construct a dynamic decision model based on the acquired environmental data and sensor trust weights, and generate real-time navigation control instructions; at the same time, without relying on real-time sensor data, generate a basic heading instruction only according to the current position of the unmanned boat and the position of obstacles; when the credibility of the control instruction decreases due to abnormal sensor data, the unmanned boat automatically switches to the basic navigation instruction and triggers an abnormal warning signal to be sent to the control center; After the control center receives the abnormal warning signal, the path planning module determines whether the unmanned boat can reach the specified position to complete the task according to the basic heading instruction. If it is determined that it cannot be completed, it plans a return route based on real-time data; The instruction operation module generates corresponding navigation instructions according to the path planning scheme, splits the instructions into multiple fragments, and sends them through different channels respectively. After receiving the instruction data, the unmanned boat verifies and merges them to execute the return mission.

[0019] Embodiments of the present invention: Step S100: Install lidar, camera, millimeter-wave radar, sonar, and GPS positioning device on the unmanned boat. Create a thread for each sensor, let each thread run simultaneously, and each sensor periodically obtains data according to its own acquisition frequency and stores the data in a temporary buffer. Taking lidar as an example, the lidar collects data every 0.1 seconds and stores it in a temporary buffer with a capacity of 1000 data points. After reading the data collected by each sensor from the temporary buffer, perform time synchronization processing. Use the linear interpolation method to align the data in time, and set the time window to 0.1 seconds. For example, for the data of lidar and millimeter-wave radar, if the acquisition times of the two differ by 0.15 seconds, calculate the approximate data values at the same time point through linear interpolation. Compare the measured value data of the same type, calculate the difference between the data, and compare it with the set error threshold. For obstacle distance measurement, the error thresholds of lidar and millimeter-wave radar are set to 1 meter. If the difference exceeds the threshold, it is considered that the data of the two sensors at the same moment is inconsistent, and it is marked as a conflict data pair and stored in the abnormal candidate set. At the same time, mark the sensor corresponding to the abnormal data acquisition. Calculate the mean and standard deviation of the historical data of each sensor to identify abnormal data. Set the data collected by the i-th sensor at the past n = 100 moments as x i1 , x i2 ,..., x i100 , then the mean μ i = 1 / 100Σ n j=1 x ij , and the standard deviation σ i = [1 / 100Σ n j=1 (x ij - μ i ) 2 1 / 2 ; Compare the currently collected data with the historical statistical data. If |x ic - μ i | > kσ i , where k = 3 and the current data deviates from the mean by more than 3 times the standard deviation, then it is determined that the data is abnormal, and this abnormal data is screened out from the conflict data pairs for exclusion. Initially, assign an initial trust weight w i (0) to each sensor. For lidar, w1(0) = 0.3, for camera, w2(0) = 0.2, for millimeter-wave radar, w3(0) = 0.2, for sonar, w4(0) = 0.1, and for GPS positioning device, w5(0) = 0.2; Calculate the error rate E i (t) and accuracy rate A i ​(t), the error rate is calculated to be E1(t) = 0.1; the accuracy rate is calculated to be A1(t) = 0.9, then the real-time data reliability r1(t) = (1 - 0.1) × 0.9 = 0.81; according to the reliability of each sensor, according to the formula w i (t + 1) = αw i (t) + (1 - α)r i (t), dynamically adjust the trust weights of the sensors, and normalize the adjusted weights; Step S200: Fuse the obstacle position information collected by different sensors. Each row of the matrix represents the position of an obstacle. Currently, 3 obstacles are detected, and their positions are (10, 20), (30, 40), and (50, 60) respectively, to construct an obstacle matrix; it is calculated through the GPS positioning device and the target point information that the angle between the current heading and the target heading is 30°, which is converted to radians as π / 6; each sensor is equipped with an energy monitoring device, and the remaining energy data collected by each sensor is arranged in the order of the sensor corresponding numbers to form a vector: the remaining energy of the lidar is 80%, the remaining energy of the camera is 70%, the remaining energy of the millimeter-wave radar is 75%, the remaining energy of the sonar is 60%, and the remaining energy of the GPS positioning device is 85%, then the sensor remaining energy vector Q = [0.8, 0.7, 0.75, 0.6, 0.85]; construct the trust weights of each sensor into a vector W, and combine each data to form an environmental state matrix M; generate a heading angle adjustment amount Δθ and a target speed v through a three-layer fully connected neural network. The calculation formula is Y = f NN (M⊕W) = [Δθ, v]; the input layer of the neural network has 12 neurons, including 6 elements of the obstacle position matrix, 1 element of the angle, and 5 elements of the sensor remaining energy vector. The hidden layer has 20 neurons, and the output layer has 2 neurons. After the forward calculation of the neural network, the heading angle adjustment amount Δθ = 0.1 radian, and the target speed v = 2 m / s; The second control unit constructs a basic navigation model, generates a repulsive field with a radius of r = 5 m centered on the recorded position of the obstacle. The current position coordinates of the unmanned boat are (x, y) = (20, 30), and the obstacle coordinates are (x1, y1) = (15, 25), (x1, y1) = (25, 35), and the number of obstacles p = 2, then d1 = 7.07 m, d2 = 7.07 m; Δθ r = 0; no angle adjustment is required; the position coordinates of the target point are (x goal , y goal ) = (50, 60); then Δθ a = π / 4 radian; the initial heading angle in radians θ c = π / 3, and the repulsive weight coefficient λ = 0.6, then the final heading angle θ z = 13π / 30 radian; Calculate the credibility of the first control unit's instructions: k = 5, indicating 5 sensors. Through calculation, C = 0.6 is obtained. Set the credibility threshold C th = 0.7. Since C < C th , the output of the first control unit is automatically disconnected, the heading instruction generated by the second control unit is enabled, and the switching time is recorded. At the same time, a warning signal is triggered and sent to the control center; Step S300: After receiving the warning signal, the control center calls the environmental data after multi-sensor fusion in step S100, including the obstacle position, the current position of the unmanned boat, the heading, the remaining energy of the sensors, and the sensor credibility C = 0.6 in step S203; Based on the historical environmental data and energy data, a historical energy consumption model D = 0.1v 2 + 0.05f + 0.02h is constructed by the linear regression algorithm, where v represents the sailing speed of 2 m / s, f represents the wind speed of 5 m / s, and h represents the wave height of 1 m; then D = 0.67 unit of energy; The total remaining energy of the unmanned boat is 5 units of energy, the energy required for the remaining voyage is estimated to be 4 units of energy, and the safety factor q = 1.2. Then q times the energy required for the current task is 1.2 * 4 = 4.8 units of energy; Since the total remaining energy of the unmanned boat 5 > 4.8, but the current sensor trust degree C = 0.6 < C th = 0.7 and the duration exceeds the preset threshold T th = 60 seconds, it is determined that the sensor cannot support the remaining voyage, and enter the return path planning stage; Calculate the distances from the current position of the unmanned boat to each safe mooring point. The coordinates of the 3 safe mooring points are (100, 200), (150, 250), and (200, 300) respectively. The current position of the unmanned boat is (50, 60). Calculate the distances, and select the nearest mooring point (100, 200) as the return target point; Integrate the three-dimensional point cloud map generated in step S100 with the sonar data to construct a two-dimensional grid map. Each grid represents an area, and the grid size is 1 m × 1 m; According to the number and distribution of obstacles, mark as passable and impassable; Use the heuristic function of the A* algorithm to obtain an optimal path from the current position of the unmanned boat to the return target point; Step S400: The control center sorts the path planning solution in step S300, including the target position coordinates, corresponding heading angles, target speeds, and the time expected to reach each target position, into a complete navigation instruction in the set JSON format; splits it according to the distance of the voyage section, and each navigation instruction every L = 20 meters is a segment; if the above path planning solution contains a 100-meter voyage section, it is split into 5 segments, and each split instruction segment is marked with a segment number, a timestamp, and a check code; the segment numbers are "1 / 5", "2 / 5", etc.; the check code is obtained by calculating the hash algorithm for the segment content; the control center selects multiple communication channels equipped on the unmanned boat, including satellite communication, 4G communication, and shortwave communication, to allocate the split and marked instruction segments, and uses a round-robin allocation method according to the load conditions of the channels for allocation. The first segment is sent via satellite communication, the second segment is sent via 4G communication, the third segment is sent via shortwave communication, and so on; at the same time, the control center monitors the sending status of each segment in real time. If a certain segment fails to be sent, the segment information is recorded and an attempt is made to resend it; the communication equipment equipped on the unmanned boat receives the instruction segments from different communication channels in real time; and checks each received instruction segment. After verification, the unmanned boat executes the return mission according to the instruction.

[0020] Finally, it should be noted that the above are only the preferred embodiments of the present invention and are not used to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, for those skilled in the art, they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. An intelligent perception navigation method for an unmanned boat, characterized in that: The method comprises: Step S100: synchronously collect environmental information around the unmanned boat through a heterogeneous sensor array composed of visual sensors, radars and positioning devices, verify and identify the data output by each sensor in real time and exclude abnormal data, and dynamically adjust the trust weight of each sensor based on the historical data of each sensor; Step S200: two sets of parallel navigation control units are set, the navigation control units include a first control unit and a second control unit; the first control unit builds a dynamic decision model based on the collected environmental data and the sensor trust weight to generate real-time navigation control instructions; the second control unit does not rely on the real-time sensor data, and only generates basic heading instructions according to the current position of the unmanned boat and the position of obstacles; when the first control unit causes the command credibility to decrease due to sensor abnormality, the unmanned boat automatically switches to the second control unit, navigates according to the basic heading instructions, and triggers an abnormal warning signal to be sent to the control center; Step S300: After receiving the abnormal warning signal, the control center determines whether the unmanned boat can reach the designated location to complete the mission according to the basic heading instruction. If it is determined that it cannot be completed, the return route is planned based on the real-time data; Step S400: Generate corresponding navigation instructions according to the path planning scheme of step S300, split the instructions into multiple segments, and send them through different channels respectively. After receiving the instruction data, the unmanned boat verifies it and merges it to execute the return mission.

2. The intelligent sensing navigation method for an unmanned boat according to claim 1, characterized in that: The step S100 includes: Step S101: deploying laser radar, camera, millimeter wave radar, sonar and GPS positioning device on the unmanned boat; the laser radar is used to measure the distance and position of surrounding obstacles and generate a three-dimensional point cloud map; the camera is used to identify the target object and analyze the captured image; the millimeter wave radar is used to monitor the speed and distance of objects around the unmanned boat; the sonar is installed on the bottom of the unmanned boat for monitoring and depth measurement of underwater targets; the GPS positioning device is used to obtain the position information of the unmanned boat in real time; create a thread for each sensor, let each thread run simultaneously, each sensor periodically acquires data according to its own acquisition frequency, and stores the data in a temporary buffer; Step S102: After reading the data collected by each sensor from the temporary buffer, perform time synchronization processing and use the linear interpolation method to time align the data; compare the same type of measurement data, calculate the difference between the data, and compare it with the set error threshold; if the difference exceeds the threshold, it is considered that the data of the two sensors at the same time are inconsistent, marked as conflicting data pairs, and stored in the abnormal candidate set; at the same time, mark the sensor corresponding to the abnormal data; Step S103: further calculate the mean and standard deviation of each sensor's historical data to identify abnormal data, and set the data collected by the i-th sensor at the past n moments as [x i1 ,x i2 ,...,x in ], where x in represents the data collected by the i-th sensor at the n-th time, then the mean μ i =1 / nΣ n j=1 x ij , x ij represents the data collected by the i-th sensor at the j-th time, j represents the index of the time series, ranging from 1 to n; standard deviation σ i =[1 / nΣ n j=1 (x ij -μ i ) 2 ] 1 / 2 ; Compare the currently collected data with the historical statistical data. If |x ic -μ i |>kσ i , where k is a preset multiple, x ic It represents the data collected by the i-th sensor in real time. If the current data deviates from the mean by more than k times the standard deviation, and no data anomalies occur in other sensors, it is considered as data anomaly, and this abnormal data is screened out from the conflicting data pairs for exclusion; Step S104: After eliminating abnormal data, dynamically adjust the trust level of each sensor; initially, assign an initial trust weight w to each sensor. i (0), represents the initial weight of the i-th sensor, according to the error rate E of the sensor data i (t), accuracy A i (t), calculate the real-time data reliability r of each sensor i (t)=(1-E i (t))*A i (t), where E i (t) represents the error rate of the i-th sensor at time t, A i (t) represents the accuracy of the i-th sensor at time t, r i (t) represents the data reliability of the i-th sensor at time t; the error rate is calculated by dividing the deviation between the sensor's measured value and the recorded true value by the true value, and the true value is determined by comparing each sensor and taking the average value; the accuracy is calculated by the ratio of the number of correct measurements to the total number of measurements; according to the reliability of each sensor, the trust weight of the sensor is dynamically adjusted according to the formula: w i (t+1)=αw i (t)+(1-α)r i (t); where w i (t) represents the trust weight of the i-th sensor at time t, w i (t+1) represents the adjusted trust weight of the i-th sensor at the t+1-th time, α represents a weight adjustment coefficient with a value range of (0,1), and the adjusted weight is normalized.

3. The intelligent sensing navigation method for an unmanned boat according to claim 1, characterized in that: The step S200 includes: Step S201: The first control unit calculates the environmental data after multi-sensor fusion in step S100 and the trust weight w of each sensor. i (t), construct a dynamic decision model; the input layer data includes the obstacle position matrix, the angle between the current heading and the target heading, and the sensor residual energy vector, forming an environmental state matrix M; the obstacle position matrix, the obstacle position information collected by different sensors is integrated, and each row of the matrix represents the position of an obstacle; the sensor residual energy vector, each sensor is equipped with an energy monitoring device, and the residual energy data collected by each sensor is arranged in the order of the corresponding number of the sensor to form a vector, and each element in the vector corresponds to the residual energy value of a sensor; the trust weight of each sensor is constructed as a vector W embedded in the model input layer; the heading angle adjustment Δθ and the target speed v are generated through a three-layer fully connected neural network, and the calculation formula is: Y=f NN (M⊕W)=[Δθ,v]; where f NN Represents the forward calculation of the neural network, and ⊕ represents the matrix concatenation operation; Step S202: The second control unit constructs a basic navigation model, takes the obstacle record position as the center, generates a repulsion field with a radius of r, where r is a preset safety distance, and the heading angle offset is: Δθ r =∑ p m=1 [(xx m ) / d m 2 ],d m =[(xx m ) 2 +(yy m ) 2 ] 1 / 2 ; (x, y) is the current position coordinate of the unmanned boat, (x m ,y m ) is the obstacle coordinate, p is the number of obstacles; d m Indicates the Euclidean distance from the current position of the unmanned boat to the position of the mth obstacle; sets the minimum distance d min , when d m <d min When d m =d min ; The final heading angle is: θ z =θ c +λΔθ r +(1-λ)Δθ a , where θ z is the final heading angle, θ c is the initial heading angle; Δθ a Indicates the target position attraction angle: Δθ a =atan2[(y goal -y) / (x goal -x)],(x goal ,y goal ) represents the position coordinates of the target point; λ represents the rejection weight coefficient, λ∈[0,1]; Step S203: Calculate the credibility of the first control unit instruction: C=[∑ k i=1 w i (t)*exp(-|x ic -μ i | / σ i )] / [∑ k i=1 w i (t)], set the credibility threshold C th , when C <C th When the switch is on, the output of the first control unit is automatically disconnected, the heading command generated by the second control unit is enabled and the switching time is recorded, and an early warning signal is triggered and sent to the control center.

4. The intelligent sensing navigation method for unmanned boat according to claim 1, characterized in that: The step S300 includes: Step S301: After receiving the warning signal, the control center calls the environmental data after multi-sensor fusion in step S100, including the obstacle position, the current position of the unmanned boat, the heading, the remaining energy of the sensor, and the sensor credibility C in step S203; based on the historical environmental data and energy data obtained in step S100, a historical energy consumption model is constructed through a linear regression algorithm to predict the energy consumption of the unmanned boat according to the current environmental conditions and navigation parameters; at the same time, the remaining range and the current environmental input model are combined to calculate the energy required to complete the current task; if the total remaining energy of the unmanned boat is less than q times the energy required for the current task, it is determined that the unmanned boat has insufficient energy and cannot complete the current task; where q is a safety factor, and a safety margin is reserved by setting the value of q, q∈(1,1.5]; if the current sensor credibility C <C th And the duration exceeds the preset threshold T th , it is determined that the sensor cannot support the remaining navigation; if any of the above conditions is met, the return path planning stage is entered; otherwise, the unmanned boat continues to perform the current mission; Step S302: Calculate the distance from the current position of the unmanned boat to each safe anchorage point, and select the nearest anchorage point as the return target point; fuse the three-dimensional point cloud map generated in step S100 with the sonar data to construct a two-dimensional grid map, where each grid represents an area, and marks it as passable or impassable according to the number and distribution of obstacles; use the heuristic function of the A* algorithm, consider the distance to the target point, energy consumption and obstacles, and plan the return route in combination with the grid map.

5. The intelligent sensing navigation method for unmanned boat according to claim 1, characterized in that: The step S400: Step S401: The control center compiles the path planning scheme of step S300, including the target position coordinates, the corresponding heading angle, the target speed and the estimated time to reach each target position, into a complete navigation instruction in the set JSON format; splits the instruction according to the distance of the segment, and the navigation instruction of every L meters is a segment, and marks the segment number, timestamp and check code for each segmented instruction segment; the segment number adopts the format of "current segment sequence number / total number of segments", the timestamp is used to record the segment generation time, and the check code is calculated by hashing the segment content; the control center selects multiple communication channels equipped by the unmanned boat to distribute the split and marked instruction segments, and distributes them in a rotating manner according to the load of the channel; at the same time, the control center monitors the sending status of each segment in real time. If a segment fails to be sent, the segment information is recorded and tried to be resent; Step S402: The communication equipment equipped by the unmanned boat receives instruction fragments from different communication channels in real time. For each instruction fragment received, the check code is first verified, and the fragment content is recalculated with the same hash algorithm as the control center, and compared with the check code carried in the fragment. If it is inconsistent, it is requested to resend; then check whether the received fragments are arranged in order according to the fragment number. If it is found that the fragment is missing or the order is wrong, record the relevant information and send a feedback signal to the control center to request the resending of the relevant fragment; sort the verified instruction fragments according to the fragment number and merge them into a complete navigation instruction. During the merging process, check whether the information between adjacent fragments is coherent. If there is a mutation in the data, request the control center for the relevant fragment again; after verification, the unmanned boat performs the return mission according to the instruction.

6. An intelligent sensing navigation system for an unmanned boat, characterized in that: The system comprises: Data acquisition module, decision control module, path planning module and instruction operation module; The data acquisition module synchronously collects environmental information around the unmanned boat through a heterogeneous sensor array composed of visual sensors, radars and positioning devices, verifies and identifies the data output by each sensor in real time and eliminates abnormal data, and dynamically adjusts the trust weight of each sensor based on the historical data of each sensor; The decision control module is used to set up two sets of parallel navigation control strategies, build a dynamic decision model based on the collected environmental data and sensor trust weights, and generate real-time navigation control instructions; at the same time, it does not rely on real-time sensor data, but only generates basic heading instructions based on the current position of the unmanned boat and the position of obstacles; when the sensor data is abnormal and the credibility of the control instruction decreases, the unmanned boat automatically switches to the basic navigation instruction, and triggers an abnormal warning signal to be sent to the control center; After the control center receives the abnormal warning signal, the path planning module determines whether the unmanned boat can reach the designated location to complete the mission based on the basic heading instruction. If it is determined that it cannot be completed, the return route is planned based on real-time data; The command operation module generates corresponding navigation instructions according to the path planning scheme, splits the instructions into multiple fragments, and sends them separately through different channels. After receiving the command data, the unmanned boat verifies and merges them to execute the return mission.

7. The intelligent sensing navigation system for unmanned boat according to claim 6, characterized in that: The data acquisition module includes a sensor unit and a data processing unit; the sensor unit includes a laser radar, a camera, a millimeter-wave radar, a sonar and a GPS positioning device, which is responsible for collecting data about the surrounding environment of the unmanned boat in real time; the data processing unit fuses multi-sensor data, identifies and eliminates abnormal data, and dynamically adjusts the sensor trust weight.

8. The intelligent sensing navigation system for unmanned boat according to claim 6, characterized in that: The decision control module includes a first control unit and a second control unit; the first control unit builds a dynamic decision model based on environmental data and sensor trust weights, generates intelligent navigation instructions, and realizes autonomous navigation of the unmanned boat; the second control unit serves as a backup control mechanism to generate basic heading instructions according to the current position of the unmanned boat and the position of obstacles.

9. The intelligent sensing navigation system for unmanned boat according to claim 6, characterized in that: The path planning module includes an environmental analysis unit and a path planning unit; the environmental analysis unit receives environmental data, constructs a grid map and marks obstacle distribution, and evaluates navigation feasibility in combination with a historical energy consumption model; the path planning unit integrates target distance, energy consumption and obstacle information to plan a return route.

10. The intelligent sensing navigation system for unmanned boat according to claim 6, characterized in that: The instruction operation module includes an instruction generation unit and an instruction execution unit; the instruction generation unit converts the path planning results into JSON format instructions, splits the fragments and marks the numbers, timestamps and check codes, and sends them through multiple channels; the instruction execution unit receives and verifies the instruction fragments, verifies the data integrity and sequence, merges and generates complete instructions, and drives the unmanned boat to perform navigation tasks.

Images