A method, device and equipment for multi-mode navigation and positioning of a UAV

Abstract

The present invention proposes a multi-mode navigation and positioning method, device and equipment for unmanned aerial vehicle (UAV). The method determines the target positioning accuracy index, obtains GPS, inertial and visual navigation mode positioning data and evaluates and generates a multi-mode navigation state set. Based on the state set, the performance of each navigation mode competes, an inter-mode negotiation mechanism is established, and a mode priority sequence is determined. A navigation mode credibility matrix is ​​established in combination with the priority sequence, and an adaptive weight allocation sequence is established. The weight change trend and mode switching resource overhead are analyzed, and the navigation mode is dynamically screened through a quantitative evaluation of the switching cost to determine the dominant and auxiliary navigation modes. The frequency characteristics of the dominant and auxiliary modes are analyzed, and a frequency domain fusion strategy is determined based on the frequency coherence coefficient to generate a fused positioning trajectory. A high-precision positioning result is output through accuracy evaluation to realize intelligent multi-mode navigation and positioning.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of unmanned aerial vehicle navigation, in particular to a multi-mode navigation positioning method, device and equipment for unmanned aerial vehicle. BACKGROUND

[0002] With the rapid development of unmanned aerial vehicle technology and the continuous expansion of application fields, the requirements for the navigation positioning accuracy and reliability of unmanned aerial vehicles are increasingly improved. The multi-mode navigation positioning technology for unmanned aerial vehicles can provide continuous and stable position information in complex environments by integrating GPS navigation, inertial navigation, visual navigation and other positioning modes, and has become a key technical support for realizing autonomous flight of unmanned aerial vehicles.

[0003] However, the existing multi-mode navigation positioning method has many technical limitations: each navigation mode often works independently, lacks effective performance evaluation and comparison mechanism, and it is difficult to select the optimal mode according to the actual situation; most of the multi-mode data fusion adopts preset fixed parameters, which cannot be dynamically adjusted according to environmental conditions and mode performance; the mode switching decision is simple and crude, without considering the time delay and computational overhead of the switching process; the data processing method is single, mainly concentrated in some conventional dimensions, and the complementary characteristics of different modes cannot be fully tapped; the overall system lacks intelligence and adaptive ability, and the performance sharply decreases under complex conditions such as signal blocking and environmental interference. These problems seriously restrict the navigation performance of unmanned aerial vehicles in complex scenes such as urban canyons, mountainous terrain and marine environment. SUMMARY

[0004] The present application provides a multi-mode navigation positioning method, device and equipment for unmanned aerial vehicles, aiming to realize high-precision intelligent navigation positioning and autonomous task execution of unmanned aerial vehicles in complex environments, integrate key technologies such as navigation mode performance competition, credibility matrix establishment, switching cost evaluation and frequency domain fusion processing, comprehensively control the multi-mode navigation process, realize adaptive selection of navigation mode, optimization of switching decision, complementary advantages in frequency domain, autonomous optimization of positioning accuracy, and form an intelligent multi-mode navigation positioning system with learning ability, decision-making ability, autonomous adaptation and dynamic optimization.

[0005] The present application provides a multi-mode navigation positioning method, device and equipment for unmanned aerial vehicles, aiming to realize high-precision intelligent navigation positioning and autonomous task execution of unmanned aerial vehicles in complex environments, integrate key technologies such as navigation mode performance competition, credibility matrix establishment, switching cost evaluation and frequency domain fusion processing, comprehensively control the multi-mode navigation process, realize adaptive selection of navigation mode, optimization of switching decision, complementary advantages in frequency domain, autonomous optimization of positioning accuracy, and form an intelligent multi-mode navigation positioning system with learning ability, decision-making ability, autonomous adaptation and dynamic optimization.

[0006] Receiving a precision requirement request of a navigation task of an unmanned aerial vehicle, and determining a target positioning precision index according to the precision requirement request;

[0007] Obtaining positioning data of each navigation mode, the navigation modes including GPS navigation mode, inertial navigation mode and visual navigation mode, and generating a multi-mode navigation state set by performing real-time effectiveness evaluation on the positioning data of each navigation mode;

[0008] Compete each navigation mode performance based on the multi-mode navigation state set, generate each navigation mode performance competition result, including: analyze the signal strength and satellite visibility parameters of the GPS navigation mode, generate the GPS mode performance evaluation index; analyze the drift error and stability parameters of the inertial navigation mode, generate the inertial navigation mode performance evaluation index; analyze the feature matching degree and light adaptability parameters of the visual navigation mode, generate the visual navigation mode performance evaluation index; generate each navigation mode performance competition result based on the GPS mode performance evaluation index, the inertial navigation mode performance evaluation index and the visual navigation mode performance evaluation index, establish an inter-mode negotiation mechanism according to each navigation mode performance competition result, and determine a navigation mode priority sequence through the inter-mode negotiation mechanism;

[0009] Establish a navigation mode credibility matrix based on the navigation mode priority sequence and the multi-mode navigation state set, and generate an adaptive weight distribution sequence according to the navigation mode credibility matrix;

[0010] Based on the adaptive weight distribution sequence, analyze the weight change trend of each navigation mode, obtain the system resource overhead required for mode switching according to the weight change trend, quantitatively evaluate the system resource overhead to generate a switching cost index, and perform navigation mode dynamic screening based on the switching cost index to determine the dominant navigation mode and the auxiliary navigation mode at the current time;

[0011] Perform frequency characteristic analysis on the dominant navigation mode and the auxiliary navigation mode to generate a mode feature spectrum, determine a frequency coherence coefficient based on the mode feature spectrum, determine a frequency domain fusion strategy based on the frequency coherence coefficient, and generate a fusion positioning trajectory based on the frequency domain fusion strategy;

[0012] Obtain the precision prediction evaluation result of the fusion positioning trajectory, perform trajectory optimization adjustment according to the precision prediction evaluation result and the target positioning precision index, output a high-precision positioning result, and complete the multi-mode navigation positioning of the unmanned aerial vehicle.

[0013] The second aspect of the present application proposes an unmanned aerial vehicle multi-mode navigation positioning device, comprising:

[0014] The demand receiving module is configured to receive a precision demand request of an unmanned aerial vehicle navigation task, and determine a target positioning precision index according to the precision demand request.

[0015] The data acquisition module is configured to acquire positioning data of each navigation mode, and perform real-time effectiveness evaluation on the positioning data of each navigation mode to generate a multi-mode navigation state set.

[0016] The performance competition negotiation module is configured to perform performance competition of each navigation mode based on the set of multi-mode navigation states, generate performance competition results of each navigation mode, and include: analyzing signal strength and satellite visibility parameters of the GPS navigation mode to generate a GPS mode performance evaluation index; analyzing drift error and stability parameters of the inertial navigation mode to generate an inertial navigation mode performance evaluation index; analyzing feature matching degree and light adaptability parameters of the visual navigation mode to generate a visual navigation mode performance evaluation index; generating performance competition results of each navigation mode based on the GPS mode performance evaluation index, the inertial navigation mode performance evaluation index and the visual navigation mode performance evaluation index; establishing an inter-mode negotiation mechanism according to the performance competition results of each navigation mode; and determining a navigation mode priority sequence through the inter-mode negotiation mechanism.

[0017] The weight distribution module is configured to establish a navigation mode credibility matrix based on the navigation mode priority sequence and the set of multi-mode navigation states, and generate an adaptive weight distribution sequence according to the navigation mode credibility matrix.

[0018] The decision screening module is configured to analyze weight variation trends of each navigation mode based on the adaptive weight distribution sequence, obtain system resource overhead required for mode switching according to the weight variation trends, quantitatively evaluate the system resource overhead to generate a switching cost index, and determine a dominant navigation mode and an auxiliary navigation mode at the current time based on the switching cost index.

[0019] The data fusion module is configured to perform frequency characteristic analysis on the dominant navigation mode and the auxiliary navigation mode to generate a mode feature spectrum, determine a frequency coherence coefficient based on the mode feature spectrum, determine a frequency domain fusion strategy based on the frequency coherence coefficient, and generate a fused positioning trajectory based on the frequency domain fusion strategy.

[0020] The optimization output module is configured to obtain an accuracy prediction evaluation result of the fused positioning trajectory, perform trajectory optimization adjustment according to the accuracy prediction evaluation result and the target positioning accuracy index, output a high-precision positioning result, and complete multi-mode navigation positioning of the unmanned aerial vehicle.

[0021] The third aspect of the present application provides a computer device, including a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the unmanned aerial vehicle multi-mode navigation positioning method disclosed in the first aspect when executing the program.

[0022] The beneficial effects of the present application are embodied in the following points: 1. By establishing a multi-mode navigation state management mechanism based on real-time effectiveness evaluation and performance competition analysis, combined with mode negotiation and priority sorting technology, the advantages of each navigation mode are deeply mined and the synergistic relationship is accurately identified, which improves the accuracy, synergistic efficiency and environmental adaptability of navigation mode selection, so that the multi-mode navigation can quickly respond to environmental changes and maintain the optimal configuration state. 2. Through the credibility matrix construction and adaptive weight distribution technology, a complete weight management framework is established, including basic weight determination, dynamic correction factor generation and weight sequence optimization, combined with weight change trend analysis and switching cost evaluation mechanism, intelligent decision of mode switching is realized, the navigation fusion is transformed from the traditional fixed weight mode to the intelligent adaptive mode, and the fusion accuracy, data utilization efficiency and system resource utilization rate are improved. 3. Through frequency characteristic analysis and frequency domain fusion strategy, the advantages of different navigation modes in frequency domain are complementary, and a differentiated fusion scheme is designed according to the frequency coherence coefficient, which solves the key technical problem that the traditional navigation method only fuses in time domain and ignores the frequency characteristics, forms a multi-mode navigation positioning scheme with full-band optimization capability, and improves the positioning accuracy and trajectory smoothness of the unmanned aerial vehicle in dynamic complex environment.

[0023] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present application. BRIEF DESCRIPTION OF DRAWINGS

[0024] The drawings herein show specific examples of the technical solutions described in the present application, and constitute part of the specification together with the specific embodiments, for explaining the technical solutions, principles and effects of the present application.

[0025] Unless specifically stated or defined otherwise, the same reference signs in different drawings represent the same or similar technical features, and different reference signs may also be used to represent the same or similar technical features.

[0026] Figure 1 is a flowchart of a multi-mode navigation positioning method of an unmanned aerial vehicle according to the present application.

[0027] Figure 2 is a structural block diagram of a multi-mode navigation positioning device of an unmanned aerial vehicle according to the present application.

[0028] Figure 3 is a structural diagram of a computer device according to the present application. DETAILED DESCRIPTION

[0029] In the following description, specific details such as specific system structures and techniques are provided for purposes of illustration rather than limitation to facilitate a thorough understanding of the embodiments of the present application. However, it will be apparent to those skilled in the art that the present application may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid obscuring the description of the present application with unnecessary detail.

[0030] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, integers, steps, operations, elements and / or components, but does not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or collections thereof.

[0031] It will also be understood that the term "and / or" used in this specification and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.

[0032] As used in this specification and the appended claims, the term "if" can be interpreted as "when" or "upon" or "in response to determining" or "in response to detecting," depending on the context. Similarly, the phrase "if it is determined" or "if [described condition or event] is detected" can be interpreted as meaning "upon determination" or "in response to determining" or "upon detection of [described condition or event]" or "in response to detecting [described condition or event]," depending on the context.

[0033] In addition, in the description of the present application specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0034] References to "one embodiment" or "some embodiments" in this specification mean that a particular feature, structure, or characteristic described in conjunction with that embodiment is included in one or more embodiments of the present application. Thus, phrases such as "in one embodiment," "in some embodiments," "in other embodiments," and "in other embodiments" appearing in various places in this specification do not necessarily refer to the same embodiment, but rather mean "one or more but not all embodiments," unless otherwise specifically emphasized. The terms "including," "comprising," "having," and variations thereof all mean "including but not limited to," unless otherwise specifically emphasized.

[0035] The technical solutions of the embodiments of this application are introduced below.

[0036] like Figure 1As shown, the embodiment of the present application provides a multi-mode navigation positioning method for unmanned aerial vehicles, including the following steps S110-S170:

[0037] Step S110, receiving a precision requirement request of an unmanned aerial vehicle navigation task, and determining a target positioning precision index according to the precision requirement request.

[0038] Specifically, the precision requirement request of the unmanned aerial vehicle navigation task is received through a multi-level interface architecture of a task management platform, a standardized precision requirement description protocol is established, and multi-dimensional requirement expressions such as position precision, speed precision, attitude precision, and time synchronization precision are supported. The precision requirement request adopts a hierarchical classification description framework, including task type identification, precision level requirement, environmental constraint condition, and performance index range, and other key information. The task type identification covers different application scenarios such as reconnaissance surveillance, cargo transportation, search and rescue, agricultural plant protection, power inspection, and ocean surveying, and each task type corresponds to different baseline precision requirements. The precision level is divided into four levels: ultra-high precision level (position precision better than 0.5 meters), high precision level (position precision 0.5-2 meters), standard precision level (position precision 2-10 meters), and rough precision level (position precision 10-50 meters), and the speed precision requirements correspond to error ranges of 0.05 m / s, 0.1 m / s, 0.5 m / s, and 1.0 m / s, respectively. The environmental constraint conditions include flight height range, meteorological condition limitation, electromagnetic interference intensity, geographical environment complexity, shelter distribution density, and signal availability parameters. In the express delivery task in urban environment, the precision requirement request specifies the position precision as 1.5 meters, the speed precision as 0.15 m / s, the attitude precision as 0.8 degrees, and labels the high building shelter, the multipath interference intensity as medium level, and the GPS signal availability time proportion as 80%. Through the standardized protocol interface and the multi-dimensional requirement description framework, the precision requirement request of the unmanned aerial vehicle navigation task is comprehensively received.

[0039] The target positioning accuracy index is determined according to the received accuracy requirement request. The target positioning accuracy index determination process adopts a multi-criteria decision method, and comprehensively considers multiple constraint conditions such as task requirements, hardware capabilities and environmental restrictions, to generate the optimal accuracy target that can be achieved. The accuracy index system includes five dimensions of absolute positioning accuracy, relative positioning accuracy, dynamic tracking accuracy, long-term stable accuracy and real-time response accuracy. Each dimension sets a clear numerical target and a confidence interval. The absolute positioning accuracy represents the deviation range of the position of the unmanned aerial vehicle relative to the true geographical coordinates, which is quantified by statistical indicators such as the circular error probable (CEP) and the spherical error probable (SEP). CEP50% and CEP95% represent the position error range under 50% and 95% confidence, respectively. The relative positioning accuracy describes the consistency and continuity of the position change of the unmanned aerial vehicle in a continuous time period, which is evaluated by the position drift rate and the trajectory smoothness index. The drift rate is controlled to be not more than a set threshold per kilometer of flight distance. The dynamic tracking accuracy evaluates the positioning stability of the unmanned aerial vehicle during maneuvering flight, including parameters such as turn radius error, climb rate error, speed tracking error and angular velocity tracking error. Through multi-dimensional accuracy index quantification and achievability analysis, the target positioning accuracy index that meets the task requirements is determined.

[0040] In step S120, positioning data of each navigation mode is acquired, each navigation mode including a GPS navigation mode, an inertial navigation mode and a visual navigation mode, and real-time validity evaluation is performed on the positioning data of each navigation mode to generate a multi-mode navigation state set.

[0041] Specifically, the positioning data of each navigation mode is acquired by using a parallel data acquisition architecture, and real-time positioning information of the GPS navigation mode, the inertial navigation mode and the visual navigation mode is simultaneously acquired through independent data acquisition channels. The GPS navigation mode data acquisition relies on a high-precision GNSS receiver, which is configured with L1 / L2 / L5 three-frequency receiving capability and supports multiple constellation signal reception such as GPS, GLONASS, BDS and Galileo, with positioning accuracy reaching sub-meter level. The GPS data acquisition frequency is set to 20 Hz, and the three-dimensional position coordinates, velocity vector, time stamp and satellite geometric distribution factor (GDOP) of the unmanned aerial vehicle are continuously recorded. At the same time, signal quality parameters such as signal strength, carrier-to-noise ratio, pseudo-range residual error and multipath error are monitored. The inertial navigation mode data acquisition is realized through a high-precision inertial measurement unit (IMU), which is composed of a fiber-optic gyroscope and a quartz flexible accelerometer. The angular velocity measurement accuracy reaches 0.01° / h, and the acceleration measurement accuracy reaches 10 -5g. The inertial data acquisition frequency is set to 100 Hz, and the three-axis angular velocity, three-axis acceleration, attitude angle and angular velocity integrated position information of the UAV are obtained in real time. The optimal estimation value of position, velocity and attitude is calculated in real time by Kalman filtering algorithm. The visual navigation mode data acquisition is based on the combination architecture of binocular stereo vision camera and monocular vision odometer. The baseline distance of the binocular camera is 12 cm, the resolution is 1920x1080, the frame rate is 30 fps, the monocular camera resolution is 2048x1536, and it has automatic exposure and white balance adjustment function. The ORB feature extraction algorithm is used for visual data processing, 500-1000 feature points are extracted from each image, the relative displacement and rotation angle are calculated through stereo matching and optical flow tracking, and the visual odometer trajectory is generated. Through the parallel data acquisition architecture and multi-frequency multi-mode receiving technology, the positioning data of GPS navigation mode, inertial navigation mode and visual navigation mode are fully acquired.

[0042] The positioning data of each navigation mode is evaluated in real time, and a multi-mode navigation state set is generated through multiple verification mechanisms such as data quality analysis, accuracy evaluation and reliability test. Real-time effectiveness evaluation adopts a multi-index fusion evaluation system, including data integrity, time consistency, spatial continuity, precision stability and anomaly detection. Data integrity evaluation is quantified by data packet loss rate, data update frequency and data format checking, etc. The GPS mode requires a data loss rate of less than 1%, the inertial mode requires a data update frequency of more than 95Hz, and the visual mode requires an image quality score of more than 0.8. Time consistency evaluation analyzes the timestamp alignment of each mode data, calculates the time synchronization error and delay jitter, and requires that the GPS and inertial data time synchronization error be less than 10 milliseconds and the visual data processing delay be controlled within 50 milliseconds. Spatial continuity evaluation verifies the rationality of positioning data through methods such as trajectory smoothness, velocity mutation detection and position jump identification, and sets the position jump threshold to 1.5 times the maximum possible speed and the velocity mutation threshold to 2 times the maximum acceleration. Precision stability evaluation quantifies the stability performance of each mode through standard deviation calculation, precision change rate analysis and confidence interval estimation in a sliding window. The abnormal detection adopts a 3σ criterion based on statistical analysis |x - μ|>3σ and a hybrid method based on machine learning, where S_statistical = (x -μ) / σ, S_ML is the output of the machine learning model, and when S>T_threshold, it is determined to be abnormal, identifying abnormal patterns such as outliers, drifts, oscillations and failures, and establishing an abnormal detection threshold matrix. In the marine inspection task, the GPS mode has an effectiveness score of 0.95 in open sea environment, the inertial mode has an effectiveness score of 0.88 under the influence of ship vibration, and the visual mode has an effectiveness score reduced to 0.72 due to sea surface reflection. Through real-time effectiveness evaluation and multiple verification mechanisms, a multi-mode navigation state set reflecting the current working state and performance of each navigation mode is generated.

[0043] Step S130, based on the multi-mode navigation state set, the performance of each navigation mode is competed, the performance competition results of each navigation mode are generated, and a mode-to-mode negotiation mechanism is established according to the performance competition results of each navigation mode. The navigation mode priority sequence is determined through the mode-to-mode negotiation mechanism.

[0044] Specifically, the performance competition of each navigation mode analyzes the comprehensive performance of the GPS navigation mode, the inertial navigation mode and the visual navigation mode under the current environmental conditions, establishes a dynamic competition evaluation and negotiation decision mechanism, and finally determines the priority order of each navigation mode.

[0045] In some embodiments, the performance competition of each navigation mode is performed based on the multi-mode navigation state set to generate the performance competition results of each navigation mode, including: analyzing the signal strength and satellite visibility parameters of the GPS navigation mode to generate a GPS mode performance evaluation index; analyzing the drift error and stability parameters of the inertial navigation mode to generate an inertial navigation mode performance evaluation index; analyzing the feature matching degree and illumination adaptability parameters of the visual navigation mode to generate a visual navigation mode performance evaluation index; generating the performance competition results of each navigation mode based on the GPS mode performance evaluation index, the inertial navigation mode performance evaluation index and the visual navigation mode performance evaluation index.

[0046] First, the signal strength and satellite visibility parameters of the GPS navigation mode in the multi-mode navigation state set are analyzed to generate GPS mode performance evaluation indicators. The GPS mode performance evaluation focuses on two core elements: signal quality and geometric configuration. The current performance level of the GPS mode is quantified by real-time monitoring of key parameters such as carrier-to-noise ratio (C / N0), signal strength indicator (RSSI), satellite elevation angle distribution and geometric precision dilution factor (GDOP). Signal strength analysis uses multi-frequency signal power measurement. The L1 band carrier-to-noise ratio threshold is set to 35dB-Hz, the L2 band carrier-to-noise ratio threshold is set to 32dB-Hz, and the L5 band carrier-to-noise ratio threshold is set to 38dB-Hz. When the signal strength of any frequency band is lower than the threshold, a signal quality warning is triggered. Satellite visibility parameters include the total number of visible satellites, the number of effectively tracked satellites, satellite elevation angle distribution and azimuth angle coverage. It is required that the number of visible satellites is not less than 6, at least 4 of the effectively tracked satellites have an elevation angle higher than 15 degrees, and the azimuth angle coverage reaches more than 270 degrees. The geometric configuration quality is evaluated by the precision dilution factor, where , , , , each parameter from The diagonal elements of the matrix are obtained, where A is the observation matrix. A GDOP less than 6 is considered a good configuration, less than 3 is considered an excellent configuration, and greater than 10 is considered a poor configuration, requiring the GPS mode weight to be reduced. Multipath effect assessment is achieved through pseudorange residual analysis and carrier phase observation quality inspection. When the multipath error exceeds 2 meters, the GPS performance evaluation index is correspondingly degraded. In express delivery tasks in urban environments, the GPS signal is affected by obstruction by high-rise buildings. The number of visible satellites is reduced from 12 to 7, the GDOP value increases from 2.1 to 5.8, the carrier-to-noise ratio generally decreases by 3-5dB, and the GPS mode performance evaluation index decreases from 0.92 to 0.73. Through signal strength monitoring and geometric configuration analysis, a GPS mode performance evaluation index that comprehensively reflects the real-time performance level of the GPS navigation mode is generated.

[0047] Further, the drift error and stability parameters of the inertial navigation mode in the multi-mode navigation state set are analyzed to generate an inertial navigation mode performance evaluation index. The inertial navigation mode performance evaluation mainly investigates key parameters such as gyroscope drift, accelerometer bias, random walk noise, and temperature stability, and quantifies the precision decay law of the inertial mode through error propagation analysis and statistical property evaluation. The gyroscope drift error is divided into two parts: deterministic drift and random drift. The deterministic drift is determined by the zero bias stability test, and the angle random walk coefficient is required to be less than 0.1° / √h, and the bias stability is better than 1° / h. The accelerometer bias stability is required to be better than 50μg, and the velocity random walk coefficient is less than 0.01m / s / √h. The temperature sensitivity of the inertial device is calibrated by the temperature coefficient, and the temperature coefficient of the gyroscope is required to be less than 0.01° / h / ℃, and the temperature coefficient of the accelerometer is required to be less than 10μg / ℃. The decay law of the inertial navigation precision with time is described by the Schuler oscillation model, and the position error growth rate is about 1 nautical mile / hour without external correction. The short-term stability evaluation adopts the Allan variance analysis method, and the stability index under different integration times is calculated, and the position drift is controlled within 0.5 meters within 1 minute, and the position drift is controlled within 15 meters within 10 minutes. The influence of environmental vibration on the inertial measurement precision is verified by the vibration test platform, and the precision decay is required to be not more than 20% under the condition of vibration acceleration 0.5g in the frequency range of 1-100Hz. Through drift error analysis and stability parameter evaluation, the inertial navigation mode performance evaluation index quantifying the current precision level and reliability degree of the inertial navigation mode is generated.

[0048] Meanwhile, the feature matching degree and the light adaptability parameter of the visual navigation mode in the multi-mode navigation state set are analyzed to generate a visual navigation mode performance evaluation index. The visual navigation mode performance evaluation focuses on core elements such as image quality, feature extraction effect, matching accuracy and environmental adaptability, and analyzes the working state and navigation performance of the visual sensor in real time through computer vision algorithms. The feature matching degree evaluation adopts feature description operators such as ORB, SIFT and SURF, and counts the number of successfully extracted feature points, the uniformity of feature point distribution and the distinguishability of feature descriptors in each frame of image. It is required that the number of feature points in each frame of image is not less than 300, the distribution coverage rate of feature points in the image is more than 80%, and the Hamming distance matching success rate of feature descriptors is higher than 85%. The light adaptability parameters include image brightness distribution, contrast level, exposure time stability and white balance accuracy. The image brightness histogram is required to be uniformly distributed to avoid overexposure and underexposure. The contrast evaluation adopts double standards of Weber contrast and Michelson contrast, and the contrast value is required to be kept in the range of 0.3-0.8. Motion blur detection is realized through image gradient analysis and frequency energy distribution evaluation. When the motion blur index is more than 0.6, the visual navigation performance significantly decreases. The depth estimation accuracy is evaluated by a binocular stereo matching algorithm, and it is required that the depth estimation error is less than 0.3 meters at a distance of 10 meters, and the depth estimation error is less than 2 meters at a distance of 50 meters. The visual odometry trajectory accuracy is verified by a closed-loop detection and global optimization algorithm, and the cumulative trajectory error growth rate is required to be controlled within 1%. Through feature matching analysis and light adaptability evaluation, the visual navigation mode performance evaluation index is generated, which comprehensively reflects the image processing ability and navigation accuracy of the visual navigation mode.

[0049] Finally, the performance competition results of each navigation mode are generated based on the GPS mode performance evaluation index, the inertial navigation mode performance evaluation index and the visual navigation mode performance evaluation index. The performance competition results are generated by using a weighted comprehensive scoring method. The performance evaluation indexes of each navigation mode are normalized and weighted to form a competition ranking result which can be directly compared. The competition scoring algorithm considers the differentiated weight configuration of task requirements. The precision index weight is increased for high-precision tasks, the stability index weight is increased for long-time navigation tasks, and the adaptability index weight is increased for complex environment tasks. The GPS mode competition scoring comprehensively considers three main factors, i.e., signal quality, satellite geometry configuration and multipath suppression capability. The comprehensive score is calculated by using a weighted summation method. The inertial mode competition scoring mainly evaluates the drift stability, noise level and temperature compensation effect. The time-weighted average method is used to reflect the long-term stability characteristics. The visual mode competition scoring mainly considers the feature quality, light adaptability and motion robustness. The scoring weight under different light conditions is adjusted by using an environmental adaptability coefficient. The comprehensive competition ranking is determined by using a ranking algorithm to generate a dynamic updated priority ranking result. The competition results also include the cooperative work suggestions and complementary enhancement strategies of each mode. When the GPS mode score is high, it is suggested to be used as the leading mode. When the inertial mode score is outstanding, it is suggested to undertake short-term high-precision navigation tasks. When the visual mode score is excellent, it is suggested to be responsible for relative positioning and environment perception tasks. In the city express delivery scene, the GPS mode score is 0.94, ranking first. The inertial mode score is 0.82, ranking second. The visual mode score is 0.75, ranking third. By using the weighted comprehensive scoring and dynamic ranking algorithm, the performance competition results of each navigation mode under the current environmental conditions are generated.

[0050] In some embodiments, the mode-to-mode negotiation mechanism is established according to the performance competition results of each navigation mode, including: a performance comparison matrix is established based on the performance competition results of each navigation mode; a mode-to-mode complementarity analysis is performed according to the performance comparison matrix to obtain a cooperative enhancement relationship of each navigation mode; a mode-to-mode resource allocation demand analysis is performed based on the cooperative enhancement relationship to obtain a resource configuration strategy; and a mode-to-mode negotiation process is started for the resource configuration strategy to establish the mode-to-mode negotiation mechanism.

[0051] A performance comparison matrix is established based on the performance competition results of each navigation mode. The performance comparison matrix adopts a three-dimensional matrix structure. The GPS navigation mode, the inertial navigation mode and the visual navigation mode are used as row and column indexes. The matrix elements represent the performance difference and advantage relationship between different modes. In the matrix construction process, the score data in the performance competition results of each navigation mode are compared in pairs to calculate the performance difference degree and the complementarity degree between any two modes. The performance difference degree is calculated by using a normalized score difference , S are the performance scores of modes i and j. Positive values ​​indicate that the row mode outperforms the column mode, while negative values ​​indicate that the column mode outperforms the row mode. The absolute value reflects the degree of advantage. The complementarity assessment considers the performance differences between modes under different environmental conditions and task scenarios. Variance analysis and correlation calculations are used to determine the complementary potential between modes. The matrix also includes time dimension information, recording the dynamic performance trends of each mode at different times, supporting time series performance prediction and dynamic adjustment. In urban express delivery tasks, the GPS mode has an advantage of 0.12 over the inertial mode in open areas and 0.19 over the visual mode. In tunnel environments, the inertial mode has an advantage of 0.76 over the GPS mode and 0.05 over the visual mode, reflecting significant differences in environmental adaptability. Through pairwise comparison and differential analysis of performance competition results, a performance comparison matrix reflecting the relative performance relationships among navigation modes was established.

[0052] Inter-mode complementarity analysis is conducted based on the performance comparison matrix to identify the synergistic enhancement relationships among navigation modes. Complementarity analysis utilizes covariance analysis and principal component analysis to identify the complementary characteristics and synergistic potential of each navigation mode across different performance dimensions. The synergistic enhancement relationship determines the synergistic effect and enhancement factor of each mode combination by analyzing the off-diagonal elements and temporal variation patterns in the performance comparison matrix. The synergistic relationship between GPS and inertial modes is primarily reflected in the complementarity of long-term and short-term accuracy. GPS provides a long-term stable absolute positioning reference, while the inertial mode provides short-term, high-frequency relative position updates. The synergistic relationship between GPS and visual modes is reflected in the complementarity of global and local positioning. GPS provides a global coordinate framework, while the visual mode provides local environmental perception and relative positioning capabilities. The synergistic relationship between inertial and visual modes is reflected in the complementarity of continuity and discreteness. The inertial mode provides continuous motion state estimation, while the visual mode provides discrete position correction and environmental constraints. Through complementarity analysis and synergistic effect evaluation, a synergistic enhancement relationship is derived that describes the collaborative potential and enhancement effect of each navigation mode.

[0053] Based on the inter-mode resource allocation demand analysis of the synergistic enhancement relationship, the resource configuration strategy is obtained. The resource allocation demand analysis considers multiple dimensions such as computing resources, storage resources, communication resources and power consumption resources, and determines the optimal resource allocation scheme through the performance improvement data in the synergistic enhancement relationship. The computing resource allocation determines the processor time slice allocation proportion according to the algorithm complexity and real-time requirement of each mode. The GPS mode needs less computing resources mainly for signal processing and positioning calculation, the inertial mode needs medium computing resources for filtering and error compensation, and the vision mode needs a large amount of computing resources for image processing and feature matching. The storage resource allocation considers the data cache demand and historical data saving requirement of each mode, and the vision mode needs the largest storage space for image data and feature map storage. The communication resource allocation determines the network resource allocation according to the data transmission frequency and bandwidth demand of each mode. The GPS mode has small data volume but is sensitive to time delay, the inertial mode has medium data volume and high update frequency, and the vision mode has large data volume but high tolerance to time delay. The power consumption resource allocation considers the energy consumption characteristics and task duration requirement of each mode, and establishes a dynamic power consumption management strategy to optimize the energy consumption allocation under the premise of ensuring performance. For example, in the forest fire monitoring task, the resource configuration strategy allocates 40% of the computing resources to the vision mode for smoke detection, 35% to the inertial mode to maintain stable flight, and 25% to the GPS mode to provide a position reference. Through resource demand analysis and optimization configuration algorithm, the resource configuration strategy balancing performance demand and resource constraint is obtained.

[0054] A mode-to-mode negotiation mechanism is established for the inter-mode negotiation process of resource allocation strategy. The negotiation process adopts a distributed decision architecture, and each navigation mode acts as an independent negotiation subject to compete for resources and make performance commitments based on the resource allocation strategy. The negotiation mechanism includes four stages: resource application, performance evaluation, scheme coordination, and agreement reaching. Each stage has clear negotiation rules and decision criteria. In the resource application stage, each mode proposes resource demand application based on the current performance state and task demand, including minimum resource demand and expected resource allocation. In the performance evaluation stage, each mode promises corresponding performance indicators and quality of service according to the allocated resource amount, establishing a resource-performance mapping relationship. In the scheme coordination stage, resource conflicts and performance contradictions are solved through multiple rounds of negotiation and strategy adjustment, and a Pareto optimal solution is found using game theory. In the agreement reaching stage, the final resource allocation scheme and performance responsibility division are determined, and a dynamic adjustment mechanism is established to respond to environmental changes. The negotiation mechanism also includes a breach detection and punishment mechanism to ensure that each mode fulfills its performance commitments. In the complex terrain environment of mountain search and rescue tasks, the GPS mode actively reduces resource application and promises limited positioning services due to signal obstruction, the inertial mode applies for more computing resources and promises high-precision short-term navigation, and the vision mode applies for additional storage resources and promises terrain matching and obstacle detection. Through three rounds of negotiation, a resource allocation agreement is reached. Based on the distributed negotiation process and dynamic adjustment mechanism, a mode-to-mode negotiation mechanism is established to support the coordination of navigation modes and the optimization of resource allocation.

[0055] The navigation mode priority sequence is determined through an inter-mode negotiation mechanism. The priority determination process mainly considers two core elements: current performance and environmental adaptability. By analyzing the actual performance level, stability characteristics, and response ability to environmental changes exhibited by each mode in the negotiation, a relative advantage evaluation framework is established. The priority evaluation of the GPS mode focuses on its signal quality stability, positioning accuracy level, and satellite geometry configuration. In open environments with good signals, it usually obtains a high priority ranking, while in obstructed environments, the priority decreases accordingly. The priority evaluation of the inertial mode mainly considers its short-term accuracy retention ability, drift error control level, and independent working characteristics. In environments where GPS signals are limited or completely invalid, the priority significantly increases. The priority evaluation of the visual mode focuses on feature extraction quality, light adaptability, and environmental texture richness. In environments with good light and obvious texture features, it obtains a high priority, while in harsh weather or monotonous environments, the priority decreases. The priority sequence generation adopts a weighted ranking method, dynamically ranks according to the comprehensive performance of each mode, and ensures that the priority sequence can accurately reflect the relative advantages of each mode in the current environmental conditions. In the open area of the city delivery task, the GPS mode obtains the first priority due to its stable signal quality and excellent absolute positioning accuracy. The inertial mode obtains the second priority due to its good short-term stability and high-frequency update characteristics. The visual mode obtains the third priority due to its environmental dependence. In the signal-obstructed environment of the tunnel inspection task, the priority sequence is significantly adjusted. The inertial mode jumps to the first priority due to its independence from external signals and stable performance in closed environments. The visual mode rises to the second priority due to its effective identification of tunnel wall texture features and relative positioning ability. The GPS mode drops to the third priority due to severe signal obstruction. Through the intelligent decision-making process of the negotiation mechanism, a navigation mode priority sequence is generated that accurately reflects the current performance state, environmental adaptability, and relative advantage relationship of each navigation mode.

[0056] In step S140, a navigation mode credibility matrix is established based on the navigation mode priority sequence and the multi-mode navigation state set, and an adaptive weight distribution sequence is generated according to the navigation mode credibility matrix.

[0057] Specifically, the navigation mode credibility matrix is established by fusing the performance information in the navigation mode priority sequence and the multi-mode navigation state set, quantifying the credibility of each navigation mode in the current environmental conditions, and generating a dynamically adjusted weight distribution scheme based on the credibility evaluation results.

[0058] In some embodiments, the establishing the navigation mode credibility matrix based on the navigation mode priority sequence and the multi-mode navigation state set comprises: determining a basic credibility weight of each navigation mode according to the navigation mode priority sequence; analyzing a real-time performance deviation of each navigation mode based on the multi-mode navigation state set; generating a dynamic credibility correction factor according to the real-time performance deviation; and establishing the navigation mode credibility matrix based on the basic credibility weight and the dynamic credibility correction factor.

[0059] The basic credibility weight of each navigation mode is determined according to the navigation mode priority sequence. The basic credibility weight is set based on the ranking result in the navigation mode priority sequence and the relative performance difference, and a hierarchical weight allocation method is used to establish the baseline credibility level of each mode. The weight allocation uses normalization processing to ensure that the sum of the weights of all modes is equal to 1, while maintaining the relative advantage relationship between the modes. The navigation mode ranked first obtains the highest basic credibility weight, which is usually set to be in the range of 0.5-0.7, reflecting its dominant position in the current environment. The mode ranked second obtains a medium credibility weight, which is set to be in the range of 0.2-0.4, serving as an important auxiliary support. The mode ranked third obtains a lower credibility weight, which is set to be in the range of 0.1-0.3, mainly assuming the functions of supplement and verification. The weight setting also considers the degree of performance gap between modes. When the performance difference between modes is large, the weight difference is increased, and when the performance is similar, a more balanced weight allocation is used. The basic credibility weight serves as a stable reference benchmark and remains relatively stable over a long period of time, avoiding frequent weight shocks that affect navigation stability. In the open area of the city express delivery task, the GPS mode is ranked first and obtains a basic credibility weight of 0.65, the inertial mode is ranked second and obtains a weight of 0.25, and the visual mode is ranked third and obtains a weight of 0.10. Through hierarchical weight allocation and normalization processing, the basic credibility weight reflecting the baseline credibility level of each navigation mode is determined.

[0060] The real-time performance deviation of each navigation mode is analyzed based on a multi-mode navigation state set. By comparing the current performance of each navigation mode with its historical average performance level, the fluctuation degree and deviation range of mode performance are quantified. The performance deviation calculation adopts a sliding window statistical method to calculate the performance mean, variance and change trend of each navigation mode within a time window, establishing a performance baseline and deviation threshold. The real-time performance deviation of the GPS mode mainly reflects in the positioning accuracy change, signal quality fluctuation and satellite geometry configuration change, etc. The performance deviation degree is evaluated by monitoring the GDOP value change, carrier-to-noise ratio fluctuation and multipath error increase and decrease. The real-time performance deviation of the inertial mode focuses on the drift error growth, temperature influence and vibration interference, etc. The deviation level is quantified by Allan variance analysis and error propagation calculation. The real-time performance deviation of the vision mode mainly reflects in the feature extraction quality change, light adaptability fluctuation and motion blur influence, etc. The deviation degree is evaluated by the feature point number change, matching success rate fluctuation and image quality score change. The performance deviation analysis also considers the influence degree of environmental changes on the performance of each mode, establishes an environment-performance correlation model to predict the performance change trend. In the forest fire monitoring task, the smoke concentration increase leads to the performance deviation of the vision mode decreasing from the baseline level of 0.88 to 0.54, with a deviation amplitude of 38.6%; the GPS mode is less affected by the smoke, with a performance deviation decreasing from 0.92 to 0.87, with a deviation amplitude of 5.4%; the inertial mode is affected by the hot air flow, with a performance deviation decreasing from 0.82 to 0.75, with a deviation amplitude of 8.5%. Through real-time performance monitoring and statistical analysis, the real-time performance deviation of each navigation mode relative to the baseline performance is determined.

[0061] For example, the dynamic confidence correction factor is generated according to the real-time performance deviation, including: analyzing the numerical distribution range and change trend of the real-time performance deviation; determining the performance deviation grade division threshold according to the numerical distribution range; performing grade classification on the real-time performance deviation based on the performance deviation grade division threshold; and generating a dynamic confidence correction factor according to the grade classification result.

[0062] Firstly, the numerical distribution range and variation trend of real-time performance deviation are analyzed, and the distribution characteristics and time sequence variation law of performance deviation of each navigation mode are determined by statistical analysis method. The numerical distribution range analysis adopts quantile statistics and extreme value analysis method to calculate the minimum value, maximum value, median and quartile of performance deviation, and establishes the statistical description of deviation distribution. The variation trend analysis identifies the linear trend, periodic variation and random fluctuation components of performance deviation by regression analysis and time series analysis method, and predicts the development direction and variation rate of deviation. The GPS mode deviation distribution presents normal distribution characteristics with small variance, and the deviation range is usually within ± 15%, and the variation trend is relatively stable; the inertial mode deviation distribution shows a gradually increasing trend, the deviation is small in the short term but accumulates with time, and the variation trend presents exponential growth characteristics; the visual mode deviation distribution is greatly affected by environmental factors, and has a wide distribution range and obvious environmental correlation, and the variation trend is closely related to light conditions and scene complexity.

[0063] Then, the performance deviation level division threshold is determined according to the numerical distribution range, and the continuous deviation value is converted into discrete level classification by threshold setting. The level division adopts five-level classification method, including excellent level (deviation less than 5%), good level (deviation 5%-15%), general level (deviation 15%-30%), poor level (deviation 30%-50%) and failure level (deviation greater than 50%), and each level corresponds to different confidence correction coefficient. The threshold setting considers the characteristic differences of each navigation mode, and the GPS mode threshold is relatively strict to reflect its high precision requirement, the inertial mode threshold considers the time accumulation characteristics, and the visual mode threshold adapts to the sensitivity of environmental changes.

[0064] Then, the real-time performance deviation of each navigation mode is classified into levels based on the performance deviation level division threshold, and the real-time performance deviation value obtained by analysis is mapped to the corresponding level category. The level classification process adopts fuzzy classification method to handle boundary conditions, and weighted average method is used to determine the level attribution when the deviation value is close to the threshold boundary. The classification results include the current level, level stability and level variation trend of each navigation mode, forming a dynamically updated level state record. For example, in the marine oil platform inspection task, the real-time deviation of GPS mode is 8.2% and classified as good level, the real-time deviation of inertial mode is 12.5% and classified as good level, and the real-time deviation of visual mode is 35.8% and classified as poor level.

[0065] Finally, the dynamic confidence correction factor is generated according to the grade classification result, and the grade classification is converted into a quantitative correction coefficient through the grade-correction factor mapping relationship. The correction factor adopts a nonlinear mapping function, the excellent level corresponds to the correction factor 1.1-1.2, the good level corresponds to 0.9-1.0, the general level corresponds to 0.7-0.8, the poor level corresponds to 0.4-0.6, and the failure level corresponds to 0.1-0.3. Through the bias analysis, grade classification and factor mapping, the dynamic confidence correction factor reflecting the real-time confidence degree change of each navigation mode is generated.

[0066] The navigation mode confidence matrix is established based on the basic confidence weight and the dynamic confidence correction factor. The confidence matrix is established by using the weighted fusion method, the basic confidence weight is taken as a stable reference component, the dynamic confidence correction factor is taken as a real-time adjustment component, and a comprehensive confidence evaluation result is generated through mathematical operation. The matrix construction adopts a square matrix form of three rows and three columns, and the rows and columns correspond to the GPS navigation mode, the inertial navigation mode and the visual navigation mode respectively. The diagonal elements represent the self-confidence of each mode, and the non-diagonal elements represent the cross-confidence and the synergistic effect between modes. The confidence calculation adopts the fusion formula C(i,j) = w_base(i) × f_dynamic(i) × α_env(i,j), wherein w_base(i) is the basic weight of mode i, f_dynamic(i) is the dynamic correction factor, and α_env(i,j) is the adaptability coefficient in environment j. The diagonal element C(i,i) represents the self-confidence, and the non-diagonal element C(i,j) = ρ(i,j)√(C(i,i)×C(j,j)) represents the cross-confidence, wherein ρ(i,j) is the synergistic coefficient. The environmental adaptability coefficient is adjusted according to the influence degree of the current environmental conditions on each mode. The GPS mode coefficient is higher in the sunny and open environment, the inertial mode coefficient is improved in the tunnel environment, and the visual mode coefficient is enhanced in the environment with rich texture. The confidence matrix also includes a time decay mechanism, and the confidence evaluation that is not updated for a long time will gradually decay, ensuring the timeliness of the confidence evaluation. The matrix update frequency is set to 1 Hz, which is synchronized with the navigation data update frequency, realizing real-time confidence monitoring and evaluation. In the complex terrain conditions of mountain search and rescue tasks, the GPS mode is significantly affected by the terrain shielding, the confidence is significantly reduced, the inertial mode maintains a high confidence in the short term, and the visual mode is affected by the light and terrain texture, the confidence is at a medium level, forming a dynamic confidence distribution reflecting the confidence degree of each mode in the current environment. Through the fusion calculation of the basic weight and the dynamic correction, the navigation mode confidence matrix reflecting the confidence degree and the synergistic relationship of each navigation mode is established.

[0067] In some embodiments, the generating the adaptive weight distribution sequence according to the navigation mode credibility matrix comprises: analyzing credibility distribution characteristics of each mode in the navigation mode credibility matrix; determining a weight distribution reference value according to the credibility distribution characteristics; performing adaptive weight adjustment based on the weight distribution reference value to generate an initial weight distribution sequence; and performing normalization processing on the initial weight distribution sequence to generate the adaptive weight distribution sequence.

[0068] The credibility distribution characteristics of each mode in the navigation mode credibility matrix are analyzed. The credibility distribution characteristic analysis extracts the numerical distribution law and characteristic parameters of each navigation mode in the credibility matrix through statistical analysis methods, including the mean, variance, skewness, and kurtosis of the credibility. The distribution characteristic analysis uses a time window sliding statistical method to calculate the statistical distribution parameters of the credibility of each mode within a set time window, and identifies the concentration trend, dispersion degree, and distribution shape of the credibility. The length of the time window is set according to the response characteristics of the navigation mode, with a 30-second window for the GPS mode, a 60-second window for the inertial mode, and a 15-second window for the visual mode to adapt to its fast environmental change characteristics. The GPS mode credibility distribution characteristics usually present a relatively stable normal distribution mode, with a small mean fluctuation range and a low variance, reflecting its stable performance in open environments. In the urban express delivery task, when the UAV enters a high-rise valley from an open square, the GPS mode credibility drops sharply from 0.92 to 0.35, and the distribution characteristic analysis immediately captures this drastic change trend and updates the distribution parameters. The inertial mode is in sharp contrast to the GPS mode, with its credibility distribution characteristics showing dynamic characteristics over time, with high and stable credibility in the short term and a slow downward trend in the long term, with the distribution skewness gradually shifting to low values. The visual mode credibility distribution characteristics show obvious environmental dependence, with the credibility distribution concentrated in the high value interval in good lighting and textured environments, and the distribution shifting to the low value interval in harsh environmental conditions, with a relatively large variance reflecting its environmental sensitivity. The environmental adaptability index is calculated by analyzing the credibility change amplitude of each mode under different environmental conditions, A_env = 1 - |C_max - C_min| / C_avg, where C_max and C_min are the maximum and minimum credibility values in the environmental change process, and C_avg is the average credibility. This index reflects the adaptability of each mode to environmental changes, with a larger value indicating stronger adaptability. Through statistical analysis and distribution modeling, the credibility distribution characteristics reflecting the credibility change law and distribution mode of each navigation mode are determined.

[0069] The weight distribution reference value is determined according to the credibility distribution characteristics. The weight distribution reference value is determined by using a quantization mapping method based on the credibility distribution characteristics, which converts the credibility distribution parameters of each mode into a reference value of weight distribution. The mapping process is implemented by a mathematical function W_base=β1·μ_C+β2·(1 / σ_C)+β3·A_env, where μ_C is the credibility mean, σ_C is the credibility standard deviation, A_env is the environmental adaptability index, β1, β2, β3 are weight coefficients and β1+β2+β3=1, and a corresponding relationship between the distribution characteristic parameters and the weight reference value is established. The credibility mean is used as the main reference factor, which reflects the average performance level of each mode. The higher the mean value, the greater the corresponding reference weight. The mapping relationship uses a linear function to ensure the rationality of weight distribution. The stability index is calculated by the reciprocal of the credibility variance. The smaller the variance, the better the stability, and the higher the corresponding weight reference value. The stability coefficient and the reference weight are positively correlated. Taking the marine oil platform inspection as an example, the GPS mode has a high mean value (0.89) and low variance (0.02) in open sea areas, which makes it obtain a reference weight of 0.55. The high variance (0.15) of the visual mode caused by the influence of sea surface reflection reduces its reference weight to 0.15. The environmental adaptability factor is determined according to the change range of the credibility distribution under different environmental conditions. The adaptability level is quantified by calculating the change rate of the distribution parameters. The mode with a small change rate obtains a higher adaptability score. The reference value calculation also considers the synergistic effect of each mode. When the credibility distribution characteristics of multiple modes show good complementarity, the reference weight values of these modes are correspondingly improved. The weight distribution reference value is normalized to ensure that the sum of the reference values of all modes is equal to 1, maintaining the mathematical consistency of weight distribution. Through the quantization analysis and mapping calculation of the credibility distribution characteristics, the weight distribution reference value of each navigation mode in the weight distribution is determined.

[0070] The adaptive weight adjustment is based on the weight distribution reference value, and an initial weight distribution sequence is generated. The adaptive weight adjustment adopts a dynamic adjustment algorithm, which adaptively corrects the reference weight value according to real-time environmental changes and task requirements, and generates a weight distribution scheme that adapts to the current situation. The adjustment algorithm establishes a mapping relationship between the degree of environmental change and the weight adjustment range, and contains adjustment strategies corresponding to three levels of slight change, moderate change and drastic change. The adjustment algorithm considers the immediate impact of environmental changes on the performance of each mode, and automatically adjusts the weight value of the corresponding mode when the environmental condition changes are detected. The adjustment response time is dynamically set according to the degree of change. Changes in task requirements will also trigger weight adjustment. High-precision tasks increase the weight of high-precision modes, fast-response tasks increase the weight of high-frequency update modes, and long-time tasks increase the weight of stable modes. Taking a mountain search and rescue task as an example, when dense fog suddenly arrives, causing the visibility to drop from 8 kilometers to 200 meters, the adaptive adjustment algorithm identifies that the visual mode credibility has dropped sharply, and adjusts its weight from the reference value of 0.30 to 0.18, while increasing the weight of the inertial mode to 0.45. The entire adjustment process is completed within 1.5 seconds. The adaptive adjustment adopts incremental adjustment, and the adjustment range is limited to ±30% of the reference value. By setting the adjustment rate limit, the instability of the navigation performance caused by the drastic change of the weight is avoided. The weight adjustment also introduces a smoothing filter mechanism to eliminate high-frequency noise and sudden interference in weight adjustment through sliding average and low-pass filtering. The filter time constant is set to 2-5 seconds. Multi-mode coordinated adjustment ensures the coordinated adjustment of the weight of each mode. When the weight of one mode increases, the weight of other modes is adjusted accordingly to maintain the balance of the total weight. Through the adaptive adjustment algorithm and the dynamic balance mechanism, an initial weight distribution sequence reflecting the current environment and task requirements is generated.

[0071] The initial weight distribution sequence is normalized to generate an adaptive weight distribution sequence. The normalization process uses a mathematical standardization method to ensure that the sum of the weight values of all navigation modes is strictly equal to 1, maintaining the mathematical constraints and physical meaning of weight distribution. The normalization algorithm first performs a weight validity test to verify the reasonableness and integrity of the weight values, then calculates the sum of all mode weights in the initial weight distribution sequence, and adjusts the mode weights to standardized values through proportional scaling. The normalization process also includes a weight lower limit protection mechanism to ensure that the weight of any mode does not drop to zero or negative, maintaining a minimum weight threshold of 0.05 to ensure the basic participation of each mode and prevent complete failure of a single mode. In the complex scenario of forest fire monitoring, the initial weight sequence shows that the GPS, inertial, and visual mode weights are 0.28, 0.45, and 0.38, respectively, with a total of 1.11. The normalization process adjusts them to the precise ratio of 0.25, 0.41, and 0.34 through proportional scaling. The weight upper limit constraint prevents excessive concentration of a single mode weight, with a maximum single mode weight threshold of 0.80. The weight dispersion mechanism avoids the risk of excessive dependence on a single navigation mode. The normalized weight distribution sequence includes data integrity verification to ensure the feasibility of the weight distribution scheme. The sequence output uses timestamp markers and state identifiers to support time sequence management and state tracking of weight configuration. Through mathematical normalization and constraint processing, an adaptive weight distribution sequence that meets real-time navigation requirements is generated.

[0072] In step S150, based on the adaptive weight distribution sequence, the weight change trend of each navigation mode is analyzed, the system resource overhead required for mode switching is obtained according to the weight change trend, the switching cost index is generated by quantitatively evaluating the system resource overhead, and the navigation mode is dynamically screened based on the switching cost index to determine the dominant navigation mode and the auxiliary navigation mode at the current time.

[0073] The weight change trend of each navigation mode is analyzed based on the adaptive weight distribution sequence. The weight change trend analysis adopts a time series analysis method to perform feature extraction and pattern recognition on the weight time series data of each navigation mode in the adaptive weight distribution sequence. The trend analysis algorithm first establishes a sliding time window, and the window length is set to an integer multiple of the weight update period. The change trend is identified through statistical analysis of the weight data in the window. The weight change rate is calculated using the difference method. The first-order difference AW(t) = W(t) - W(t-1) reflects the immediate change of the weight, and the second-order difference AW(t) = AW(t) - AW(t-1) reflects the acceleration of the change. The direction and speed of the weight change are determined by the multi-order difference sequence. The trend type identification includes four basic modes: rising trend, falling trend, oscillation trend, and stable trend. The trend direction and intensity are determined by the positive and negative and size of the trend slope k = Σ(t·W(t) - t_avg·W_avg) / Σ(t - t_avg)². The GPS mode weight change trend shows a stable high weight maintenance in an open environment, and a rapid downward trend when entering a shielding environment. The inertial mode weight remains stable or slowly increases in the short term, and shows a downward trend due to error accumulation in the long term. The weight change of the visual mode is highly related to the environmental conditions, and the weight fluctuation caused by light changes shows a periodic oscillation feature. The trend prediction adopts the autoregressive moving average model ARMA(p,q), which predicts the weight change direction in the short term based on the historical weight data, and provides forward-looking information for mode switching decision. In the urban canyon navigation scenario, the rapid downward trend of the GPS weight from the stable high value indicates the need to switch to other navigation modes. The trend analysis algorithm can identify this change mode at the early stage of weight decline. Through time series analysis and trend identification, the weight change trend of each navigation mode is determined.

[0074] In some embodiments, the system resource overhead required for mode switching according to the weight change trend includes: identifying a mode switching trigger point in the weight change trend; determining the source mode and the target mode to be switched according to the mode switching trigger point; analyzing the resource requirements for switching from the source mode to the target mode, including calculating the resource requirements, time delay requirements, and data transmission requirements; and calculating the system resource overhead required for mode switching based on the resource requirements.

[0075] A mode switching trigger point in the weight change trend is identified. The mode switching trigger point identification adopts a multi-criteria decision method to determine the time when mode switching needs to be performed by analyzing weight change rate, weight intersection point and weight mutation point and other characteristics. The weight change rate trigger criterion is set to mark a potential switching trigger point when the weight change rate of any navigation mode |dW / dt| exceeds the threshold value 0.1 / s. The weight intersection trigger criterion monitors the intersection of the weight curves of different navigation modes, and when the weight of the original main mode decreases and the weight of the candidate mode rises to form an intersection, and the weight difference after the intersection exceeds 0.15, it is confirmed as a switching trigger point. The weight mutation trigger criterion identifies the mutation point by detecting the second derivative of the weight, and when |d²W / dt²|>0.5 / s², it is determined that the switching demand is caused by a dramatic change in the environment. The trigger point verification mechanism avoids false triggering caused by transient disturbances through duration test, and requires that the trigger condition be continuously met for more than 0.5 seconds to confirm validity. For example, when entering a city canyon from an open area, the GPS weight decreases from 0.75 to 0.35 in 2 seconds, the decrease rate reaches 0.2 / s, and at the same time the inertial mode weight rises from 0.20 to 0.55, at t=1.2s the weight curves of the two cross, confirming that this time is the mode switching trigger point.

[0076] The source mode and the target mode to be switched are determined according to the mode switching trigger point. The source mode identification determines the highest weight navigation mode as the current dominant source mode by analyzing the weight distribution before the trigger point. The target mode selection is based on the weight development trend after the trigger point, predicting the mode with the highest weight in the future time window as the switching target. The mode pair determination process also needs to consider the feasibility of switching, and direct switching between some modes may have technical limitations, requiring transition through intermediate modes. The switching path planning adopts a state transition graph method, taking GPS, inertial and visual modes as state nodes, and feasible switching paths as directed edges, and determines the optimal switching sequence through the shortest path algorithm. For example, when the visual mode needs to be switched due to sudden changes in light, if the current GPS signal is also weak, the switching path is determined to be a two-step switching scheme of visual→inertial→GPS, rather than directly switching from visual to GPS. Special case handling includes emergency switching strategies when multiple modes fail simultaneously, prioritizing at least one mode to maintain working state.

[0077] The resource requirement for switching from the source mode to the target mode is analyzed, including computing resource requirement, time delay requirement and data transmission requirement. The computing resource requirement analysis includes CPU cycles and memory occupation required by coordinate conversion calculation, filter re-initialization, parameter reconfiguration and other operations in the switching process. GPS to inertial switching needs to perform conversion from ECEF coordinate system to carrier coordinate system, with a computational complexity of O(n2), and is expected to occupy 15% of CPU resources for 0.8 seconds. The time delay requirement includes four components of switching decision delay, data buffering delay, algorithm convergence delay and system stabilization delay, with a total delay of about 1.2 seconds for GPS to inertial and about 2.1 seconds for inertial to vision. The data transmission requirement analyzes the data format difference and transmission bandwidth requirement of each mode, with a GPS data packet size of about 200 bytes / time, an update rate of 20 Hz, a required bandwidth of 4 KB / s; an inertial data packet of about 500 bytes / time, an update rate of 100 Hz, a required bandwidth of 50 KB / s; and vision data containing image information, with a compressed size of about 100 KB / frame, a required bandwidth of 3 MB / s at 30 fps. The resource requirement also includes the overhead of double-mode parallel running in the switching process, which needs to maintain the resource occupation of both modes during the transition period.

[0078] The system resource overhead required for mode switching is calculated based on the resource demand. A weighted comprehensive evaluation method is used to unify the computing resource demand, time delay demand and data transmission demand obtained by the foregoing analysis into a comparable overhead metric space. The computing resource overhead is quantified by the formula C compute = (CPU usage × T duration) + (Memory usage × K memory), where C compute is the computing resource overhead, CPU usage is the processor usage rate, T duration is the resource occupation duration, Memory usage is the memory occupation amount, and K memory is the memory cost coefficient. The resource demand of the coordinate conversion calculation, filter reinitialization and other operations obtained by the foregoing analysis is substituted into the formula to obtain the computing resource overhead. The time overhead evaluation is C time = T switch × V accuracy loss, where C time is the time overhead, T switch is the total delay determined in the foregoing, and V accuracy loss is the unit time precision loss value coefficient, reflecting the influence caused by the navigation precision decline during switching. The data transmission overhead is calculated by C data = (Bandwidth × T transfer) / Bandwidth max, where C data is the data transmission overhead, Bandwidth is the actual bandwidth demand obtained by the foregoing analysis, T transfer is the data transmission time, and Bandwidth max is the maximum available bandwidth of the system for normalization processing. The comprehensive system resource overhead is C total = α·C compute + β·C time + γ·C data, where C total is the total system resource overhead, and α, β and γ are weight coefficients, respectively reflecting the relative importance of the computing resource, time delay and data transmission in the total overhead. The weight settings are dynamically adjusted according to the task type and environmental conditions. The system resource overhead required for switching from the source mode to the target mode is finally obtained.

[0079] System resource overhead is quantitatively evaluated to generate a switching cost metric. This quantitative evaluation of switching cost establishes a comprehensive cost function, unifying resource overheads across different dimensions into a comparable cost metric space. The cost function is designed as C_switch = w1·C_compute + w2·C_time + w2·C_accuracy + w4·C_risk, where C_compute represents the compute resource cost, C_time represents the time delay cost, C_accuracy represents the accuracy loss cost, and C_risk represents the switching risk cost. w1, w2, w3, and w4 are weight coefficients for each cost. The compute resource cost is calculated as the weighted sum of the incremental CPU utilization and incremental memory consumption: C_compute = k_cpu·ΔCPU + k_mem·ΔMemory. Normalization is performed to ensure comparability across different resource types. The time delay cost is modeled using a logarithmic function: C_time = log(1 + T_switch / T_ref), where T_ref is the reference time constant, reflecting the diminishing marginal effect of time delay. The accuracy loss cost reflects the degradation in positioning accuracy during the handover process: C_accuracy = ∫(σ_switch(t) - σ_normal(t))dt. The cumulative accuracy loss during the handover period is calculated by integration. The handover risk cost considers the probability of handover failure and the severity of the consequences: C_risk = P_failure·S_consequence, where P_failure is the probability of handover failure and S_consequence is the severity score of the consequences of failure. The cost metric also incorporates environmental adaptability corrections, increasing the handover cost in adverse conditions and reducing it in stable environments. This dynamic cost assessment is achieved through environmental factor adjustment. During an emergency supply delivery mission in heavy rain, severe visual mode performance degradation necessitated a handover. However, an evaluation of the cost of switching to GPS mode revealed that cloud cover reduced GPS signal quality. After comprehensive consideration of the handover cost metric, it was recommended to maintain the inertial mode as the primary mode. Through multi-dimensional quantification and comprehensive evaluation, a handover cost metric reflecting the true cost of handover was generated.

[0080] The navigation mode dynamic screening based on the switching cost index determines the dominant navigation mode and the auxiliary navigation mode at the current time. The dynamic screening adopts a cost-benefit balance decision mechanism, compares the cost of maintaining the current mode with the cost of switching to a new mode, and selects the navigation configuration scheme with the minimum total cost. The decision algorithm establishes a state transition model, takes the navigation mode configuration as the system state and the switching action as the state transition, and solves the optimal state sequence through dynamic programming. The dominant mode selection follows the stability priority principle, and switching is only performed when the performance benefit of the new mode minus the switching cost is still significantly better than the current mode. The judgment condition is Performance_new - C_switch> Performance_current + Δ_threshold, where Δ_threshold is the switching threshold. The auxiliary mode selection considers the complementarity with the dominant mode, and preferentially selects the navigation mode that can compensate for the defects of the dominant mode as the auxiliary. The synergy effect of different mode combinations is quantified through a complementarity matrix. The screening process also introduces mode health monitoring, which continuously evaluates the running state and availability of each navigation mode. Modes with a health degree below the threshold are temporarily excluded from the candidate set. Multi-mode cooperative configuration allows multiple auxiliary modes to be activated simultaneously, but needs to balance performance improvement and resource consumption. The optimal number of auxiliary modes is determined through marginal utility analysis. In the complex scene conversion of urban express delivery, when the UAV enters the building-dense area from the open area, the dynamic screening algorithm evaluates the GPS performance decline, but the cost of switching to pure inertial navigation is too high. The decision result is that the inertial mode is upgraded to the dominant mode, the GPS is downgraded to the auxiliary mode, and the visual mode is activated as the second auxiliary mode, forming a multi-mode cooperative navigation configuration. Through cost balance and dynamic optimization, the dominant navigation mode and auxiliary navigation mode that adapt to the current environment and task requirements are determined.

[0081] In step S160, frequency characteristic analysis is performed on the dominant navigation mode and the auxiliary navigation mode to generate a mode characteristic spectrum, a frequency coherence coefficient is determined based on the mode characteristic spectrum, a frequency domain fusion strategy is determined based on the frequency coherence coefficient, and a fused positioning trajectory is generated based on the frequency domain fusion strategy.

[0082] The frequency characteristics of the main navigation mode and the auxiliary navigation mode are analyzed to generate the mode characteristic spectrum. The frequency characteristics analysis uses the fast Fourier transform (FFT) algorithm to convert the time-domain navigation positioning data to the frequency domain space and extract the spectral characteristics of each navigation mode. Data preprocessing includes detrending, windowing, and zero padding operations. The least squares method is used to fit and eliminate the linear trend of the data. The Hanning window function is used to reduce spectral leakage. Zero padding extends the data length to an integer power of 2 to improve the efficiency of FFT calculation. The spectral calculation uses the segmented average power spectral density estimation method. The long time series is divided into multiple overlapping segments, each with a length of 1024 sampling points and an overlap rate of 50%. The random noise influence is suppressed by multiple segment averaging. The spectral characteristics of the GPS navigation mode show that the low-frequency component is dominant, with the main energy concentrated in the 0-0.1 Hz frequency band, reflecting its steady-state positioning characteristics and slow drift characteristics. The energy in the 0.1-1 Hz frequency band is low, indicating that GPS has limited response capability to rapid motion changes. The inertial navigation mode spectrum presents a wide frequency characteristic. The low-frequency band (0-0.5 Hz) reflects the integral drift and bias error, the medium-frequency band (0.5-5 Hz) contains the carrier motion information, and the high-frequency band (5-50 Hz) is mainly the sensor noise and vibration interference. The spectral characteristics of the visual navigation mode are related to the image processing frame rate, with energy peaks appearing at discrete frequency points. The main peak is located at the image update frequency (e.g., 30 Hz), and the secondary peak appears at the feature tracking update frequency. The spectral amplitude normalization ensures the comparability between different modes. The phase spectrum analysis reveals the time delay characteristics of each mode. GPS has satellite signal propagation delay, the inertial mode has the smallest delay, and the visual mode has a fixed delay due to image processing. In the offshore wind farm inspection task, the dominant inertial mode spectrum is concentrated in the 0.2-2 Hz range, reflecting the periodic disturbance caused by wind and waves. The auxiliary GPS mode provides a stable reference below 0.05 Hz, and the spectral characteristics of the two modes form a good complement. Through FFT transformation and power spectrum analysis, the mode characteristic spectrum reflecting the frequency domain characteristics of each navigation mode is generated.

[0083] The frequency coherence coefficient is determined based on the mode feature spectrum. The frequency coherence analysis quantifies the correlation degree of two modes in the frequency domain by calculating the cross-power spectral density and the self-power spectral density of the dominant mode and the auxiliary mode at each frequency point. The coherence coefficient calculation formula is γ²(f) = |S_xy(f)|² / (S_xx(f)·S_yy(f)), where S_xy(f) is the cross-power spectral density of the dominant mode x and the auxiliary mode y, S_xx(f) and S_yy(f) are the self-power spectral densities respectively, and γ²(f) takes the value range [0, 1]. The coherence coefficient close to 1 indicates that the two modes are highly correlated at this frequency point, and the output signal has a certain amplitude and phase relationship; close to 0 indicates that the two modes are independent of each other, and are suitable for complementary fusion. The frequency resolution is set to 0.01 Hz, and the coherence coefficients of 5000 frequency points in the range of 0-50 Hz are calculated to form a continuous coherence curve. The coherence analysis of GPS and inertial mode shows that the coherence coefficient reaches more than 0.85 in the 0-0.1 Hz ultra-low frequency band, indicating that both can accurately reflect the slow motion of the carrier; in the 0.5-2 Hz medium frequency band, the coherence decreases to below 0.3, and GPS cannot track rapid changes while the inertial mode responds well. The inertial and visual modes show moderate coherence (0.4-0.6) in the 1-10 Hz frequency band, both can perceive the dynamic motion of the carrier but the mechanisms are different, leaving room for fusion optimization. The coherence coefficient also needs to consider the influence of signal-to-noise ratio, and the coherence coefficient calculated under low signal-to-noise ratio needs to be corrected for deviation, and the correction formula is γ_corrected = γ_measured·(1 - 1 / SNR). The time-varying coherence is analyzed by short-time Fourier transform (STFT), with a window length of 256 sampling points and a sliding step of 64 points, to generate a coherence distribution map in the time-frequency domain. In the urban canyon navigation scene, the multipath effect of GPS signal causes abnormal peaks in the coherence of GPS and inertial mode in certain frequency bands, which need to be identified and removed through statistical test. Through cross-spectrum analysis and coherence calculation, the frequency coherence coefficient between the dominant and auxiliary modes is determined.

[0084] In some embodiments, the frequency domain fusion strategy is determined based on the frequency coherence coefficient, including: dividing the frequency bands according to the frequency coherence coefficient, identifying high coherence frequency bands and low coherence frequency bands; determining the fusion mode of each frequency band based on the distribution characteristics of the high coherence frequency bands and the low coherence frequency bands; designing the frequency domain fusion filter parameters according to the fusion mode of each frequency band; and generating the frequency domain fusion strategy based on the frequency domain fusion filter parameters.

[0085] The frequency bands are divided according to the frequency coherence coefficient, and the high coherence frequency band and the low coherence frequency band are identified. The adaptive threshold method is used for frequency band division, and the division threshold is determined by statistical analysis of the distribution characteristics of the coherence coefficient. The high coherence frequency band is defined as the frequency interval with a coherence coefficient γ²(f)>0.7, indicating that the dominant mode and the auxiliary mode have strong correlation in these frequency bands; the low coherence frequency band is defined as the interval with γ²(f)<0.3, indicating that the two modes are relatively independent; the medium coherence frequency band is 0.3 ≤γ²(f) ≤ 0.7, which needs special processing strategy. The frequency band boundary is determined by the zero point of the first derivative of the coherence coefficient, avoiding the discontinuity caused by artificial setting. The continuity constraint ensures the smooth transition of adjacent frequency bands, and the transition region is defined by the 3dB bandwidth criterion. The frequency band division results of GPS-inertial combination show that 0-0.08Hz is the high coherence frequency band, and both modes can accurately reflect the steady-state information; 0.08-0.5Hz is the transition frequency band; 0.5-10Hz is the low coherence frequency band, and the GPS response is insufficient while the inertial mode is dominant; above 10Hz is the noise dominant frequency band. The frequency band division also considers the signal power distribution, and the low power frequency band is not the main fusion frequency band even if the coherence is high. In the smoke environment of forest fire monitoring task, the high coherence frequency band of vision-inertial combination is concentrated in 2-8Hz, which corresponds to the main flight frequency of unmanned aerial vehicle; the low coherence frequency band is below 0.1Hz and above 20Hz, corresponding to slow drift and high frequency noise respectively. Through adaptive threshold analysis, the high coherence frequency band and the low coherence frequency band with different coherence characteristics are identified.

[0086] Based on the distribution characteristics of high- and low-coherence frequency bands, the fusion mode for each frequency band is determined. The fusion mode design adopts a differentiated strategy based on the coherence characteristics of the frequency bands. High-coherence frequency bands use weighted average fusion to fully utilize redundant information, while low-coherence frequency bands use complementary fusion to leverage the strengths of each. The fusion mode for high-coherence frequency bands is Y_high(f) = α(f)·X_main(f) + (1-α(f))·X_aux(f), where α(f) = γ²(f) / (γ²(f) + σ²_noise) is the frequency-dependent weighting coefficient, ensuring a more balanced weight distribution with higher coherence. Low-coherence frequency bands use frequency-selective fusion, determining the dominance of each mode based on its signal-to-noise ratio in a specific frequency band. Y_low(f) = X_main(f)·H_main(f) + X_aux(f)·H_aux(f), where H_main and H_aux are complementary transfer functions. A phase compensation mechanism is introduced in medium-coherence frequency bands to eliminate pseudo-coherence effects by estimating and correcting phase differences, thereby improving fusion quality. A gradual fusion mode is used in transition frequency bands, with an S-shaped transition function ensuring smooth switching of fusion strategies and avoiding discontinuities at frequency band boundaries. GPS-inertial fusion uses a fixed weight ratio of 0.6:0.4 in high-coherence low-frequency bands to fully leverage the long-term stability of GPS. In low-coherence medium-frequency bands, the inertial mode weight is increased to 0.9, with GPS providing only auxiliary corrections. Noise bands are suppressed through band-stop filtering. An abnormality handling mechanism automatically adjusts the fusion mode upon detecting sudden changes in coherence, preventing erroneous data from contaminating the fusion results. During emergency rescue missions in heavy rain, GPS signal attenuation degrades the previously high-coherence frequency bands. The fusion mode is adaptively adjusted to inertial-dominated, while increasing the weight in frequency bands where vision is still effective. Through differentiated design and adaptive adjustment, a fusion mode tailored to the characteristics of each frequency band is determined.

[0087] According to the fusion mode of each frequency band, the frequency domain fusion filter parameters are designed. The filter design adopts the frequency response matching method, and the corresponding transfer function is designed according to the requirements of the fusion mode of each frequency band. The low-pass filter is used for the high-coherent frequency band, and is designed as a Butterworth four-order filter, with the cutoff frequency set at the boundary between the high and low coherent frequency bands, the passband ripple less than 0.5 dB, and the stopband attenuation greater than 40 dB. The band-pass filter is used for selective fusion, with the center frequency aligned with the target frequency band, the bandwidth adaptively adjusted according to the coherence distribution, and the Q value adjustable in the range of 2-10. The complementary filter bank design ensures that H_main(f) + H_aux(f) = 1 is always true, avoiding signal distortion, and the linear phase FIR structure is adopted to ensure the time delay consistency of each frequency band. The filter order is determined by the frequency resolution and the transition bandwidth requirement, and the typical configuration is 128-order FIR or 8-order IIR, balancing performance and computational complexity. The filter parameters of GPS-inertial fusion: the low-pass cutoff frequency is 0.1 Hz for extracting GPS steady-state information, the high-pass cutoff frequency is 0.1 Hz for extracting inertial dynamic information, and the gain attenuation is 3 dB at the cross frequency to ensure smooth transition. The adaptive parameter adjustment mechanism dynamically modifies the filter coefficients according to the real-time coherence changes, and the adjustment rate is limited to within 10% per second to prevent filter instability. The fixed-point operation precision is considered in the implementation of the digital filter, and 16-bit fixed-point representation is adopted to ensure real-time performance in embedded systems. Under the complex terrain conditions of mountain search and rescue, the GPS signal is sometimes present and sometimes absent, resulting in a dramatic change in coherence. The adaptive filter adjusts the cutoff frequency and filter order in real time to ensure the continuity and stability of the fusion process. Through transfer function design and parameter optimization, the frequency domain fusion filter parameters that meet the fusion requirements of each frequency band are generated.

[0088] The frequency domain fusion filter parameter is used to generate a frequency domain fusion strategy. The fusion strategy integrates the frequency band division, fusion mode and filter parameter to form a complete frequency domain fusion processing flow. The strategy implementation first performs FFT transformation on the input primary and auxiliary navigation data to obtain the frequency domain representation X_main(f) and X_aux(f). Then, according to the preset frequency band division scheme, the spectrum data is allocated to the corresponding processing channel, and the high-coherent frequency band data is sent to the weighted fusion channel, and the low-coherent frequency band data is sent to the selective fusion channel. Each channel is processed according to the designed fusion mode and filter parameter, the weighted fusion channel executes Y_weighted(f) = W(f)·[α·X_main(f) + (1-α)·X_aux(f)], and the selective fusion channel executes Y_selective(f) = X_main(f)·H_LP(f) + X_aux(f)·H_HP(f). The channel output is superimposed in the frequency domain to form a complete spectrum Y_fused(f) = Y_weighted(f) + Y_selective(f) + Y_transition(f), wherein Y_transition is the smoothing processing result of the transition frequency band. The inverse FFT transformation converts the fused frequency domain data back to the time domain to generate the fused positioning data sequence. Post-processing includes phase continuity check, outlier rejection and time delay compensation to ensure the integrity and real-time performance of the output data. The strategy also contains a performance monitoring mechanism to evaluate the fusion effect in real time and feedback the adjustment of the fusion parameter. In the long-haul task of offshore wind farm inspection, the frequency domain fusion strategy successfully solves the problem of long-term drift of GPS and short-term noise of inertia, and realizes the optimization performance of the full frequency band through frequency selective fusion. Through system integration and process optimization, a complete frequency domain fusion strategy based on frequency domain analysis is generated.

[0089] The fusion positioning trajectory is generated based on a frequency domain fusion strategy. The fusion positioning trajectory generation optimally combines the data of the primary and auxiliary navigation modes in the frequency domain by performing the frequency domain fusion strategy, and outputs high-quality positioning results. The trajectory generation process processes the navigation data stream in real time, and adopts a sliding window mechanism to ensure calculation efficiency and real-time performance. The window length is 1024 points, and the update rate is 100 Hz. In each processing period, the newly entered navigation data and the historical data form a complete window, which is preprocessed and then subjected to FFT transformation, conversion to the frequency domain, and fusion processing. The fusion processing is strictly performed in accordance with the frequency domain fusion strategy. In the high-coherence frequency band, the random errors are suppressed by weighted averaging, in the low-coherence frequency band, the effective information is retained by advantage selection, and in the transition frequency band, the continuity is ensured by gradual processing. The time domain data after inverse transformation is subjected to overlap-add processing to eliminate the blocking effect, the output points are subjected to continuity inspection with the historical trajectory to ensure the smoothness and physical rationality of the trajectory. The trajectory quality evaluation indexes include position accuracy, velocity continuity and acceleration rationality, and the fusion effect is verified by comparison with the reference trajectory. When the fusion result is detected to be abnormal, the abnormal situation processing mechanism automatically switches to a backup fusion strategy or a single mode output to ensure the robustness of the system. The fusion trajectory output includes complete information such as position coordinates, velocity vector, accuracy estimation and confidence, and supports the decision-making needs of downstream applications. Through frequency domain fusion processing and quality control, the fusion positioning trajectory with high precision and high reliability is generated.

[0090] In step S170, the accuracy prediction evaluation result of the fusion positioning trajectory is obtained, the trajectory is optimized and adjusted according to the accuracy prediction evaluation result and the target positioning accuracy index, the high-precision positioning result is output, and the multi-mode navigation positioning of the unmanned aerial vehicle is completed.

[0091] Specifically, the precision prediction evaluation of the fusion positioning trajectory is performed by using a method combining statistical analysis and error propagation theory to calculate key indicators such as absolute precision, relative precision and time stability of the fusion positioning trajectory. The absolute precision evaluation is performed by comparing the position deviation of the fusion trajectory and the reference benchmark to calculate statistical parameters such as root mean square error, circular probability error and maximum error, thereby quantifying the absolute positioning precision level of the trajectory. The relative precision evaluation analyzes the consistency and continuity of the trajectory, and evaluates the relative precision performance of the trajectory by distance error and angle error between adjacent position points. The time stability evaluation analyzes the change trend of the precision with time, identifies the time period and reason of the precision deterioration, and calculates the precision drift rate and stability coefficient. For example, in the urban express delivery task, when the unmanned aerial vehicle flies in the complex building group environment, the precision evaluation algorithm analyzes the fusion trajectory to find that the absolute positioning precision reaches 1.2 meters in the open area and decreases to 2.8 meters in the shielding area, the relative precision remains 0.5 meters in the straight flight segment and increases to 1.5 meters in the turning maneuver segment. The evaluation process uses a sliding window method to calculate the local precision indicators in different time windows to form a time-space distribution atlas of the precision. The error propagation analysis calculates the contribution degree of the errors of each navigation mode to the fusion result to identify the main error source and weak link. Through multi-dimensional precision analysis and error propagation calculation, the precision prediction evaluation result of the fusion positioning trajectory is obtained.

[0092] According to the precision prediction evaluation result and the target positioning precision index, the trajectory is optimized and adjusted, and the overall precision level of the fusion positioning trajectory is improved through an iterative optimization algorithm. The trajectory optimization and adjustment adopts a segmented optimization strategy, and the low-precision trajectory segment identified in the precision evaluation is optimized, and the high-precision trajectory segment is kept. The optimization algorithm first compares the gap between the current trajectory precision and the target precision index, and the trajectory segment whose precision meets the requirement is kept unchanged, and the trajectory segment whose precision is insufficient starts the optimization program. The optimization method includes three strategies of weight redistribution, filter parameter adjustment and outlier elimination, and the appropriate optimization method is selected according to the error source and characteristics. The weight redistribution reduces the weight of the low-precision mode and improves the influence of the high-precision mode by adjusting the contribution proportion of each navigation mode in the fusion. The filter parameter adjustment improves the filter performance and state estimation precision by optimizing the process noise and observation noise parameters of the Kalman filter. The outlier elimination adopts a statistical test-based outlier detection method to identify and eliminate observation data that deviates obviously from the normal range. In the mountain search and rescue task, when the GPS precision decline period caused by terrain obstruction is detected, the optimization algorithm automatically increases the fusion weights of the inertial mode and the visual mode, reduces the GPS weight from 0.6 to 0.3, increases the inertial weight from 0.25 to 0.45, and increases the visual weight from 0.15 to 0.25. After optimization, the trajectory precision of this period is improved from 3.5 meters to 2.1 meters. The optimization and adjustment process adopts an iterative mechanism, and the precision evaluation and parameter adjustment are repeatedly executed until the trajectory precision reaches the target requirement or the maximum iteration number is reached. In the forest fire monitoring task, after 3 times of iterative optimization, the overall precision of the fusion positioning trajectory is improved from the initial 2.8 meters to 1.9 meters, meeting the target precision requirement of 2.5 meters. Through the segmented optimization and iterative adjustment mechanism, the high-precision positioning result meeting the target precision index is output, and the multi-mode navigation and positioning of the unmanned aerial vehicle is completed.

[0093] In order to perform the unmanned aerial vehicle multi-mode navigation and positioning method corresponding to the method embodiment to realize the corresponding functions and technical effects. Referring to Figure 2 , Figure 2 The structure block diagram of the unmanned aerial vehicle multi-mode navigation and positioning device 200 provided by the embodiment of the application is shown. For ease of illustration, only the parts related to the embodiment are shown, and the unmanned aerial vehicle multi-mode navigation and positioning device 200 provided by the embodiment of the application includes:

[0094] The demand receiving module 201 is configured to receive a precision demand request of an unmanned aerial vehicle navigation task, and determine a target positioning precision index according to the precision demand request;

[0095] The data acquisition module 202 is configured to acquire positioning data of each navigation mode, and perform real-time validity evaluation on the positioning data of each navigation mode to generate a multi-mode navigation state set;

[0096] The performance competition negotiation module 203 is configured to perform performance competition of each navigation mode based on the set of multi-mode navigation states, and generate performance competition results of each navigation mode, including: analyzing signal strength and satellite visibility parameters of the GPS navigation mode to generate a GPS mode performance evaluation index; analyzing drift error and stability parameters of the inertial navigation mode to generate an inertial navigation mode performance evaluation index; analyzing feature matching degree and light adaptability parameters of the visual navigation mode to generate a visual navigation mode performance evaluation index; generating performance competition results of each navigation mode based on the GPS mode performance evaluation index, the inertial navigation mode performance evaluation index and the visual navigation mode performance evaluation index; establishing an inter-mode negotiation mechanism according to the performance competition results of each navigation mode; and determining a navigation mode priority sequence through the inter-mode negotiation mechanism.

[0097] The weight distribution module 204 is configured to establish a navigation mode credibility matrix based on the navigation mode priority sequence and the set of multi-mode navigation states, and generate an adaptive weight distribution sequence according to the navigation mode credibility matrix.

[0098] The decision screening module 205 is configured to analyze weight variation trends of each navigation mode based on the adaptive weight distribution sequence, obtain system resource overhead required for mode switching according to the weight variation trends, quantitatively evaluate the system resource overhead to generate a switching cost index, and perform dynamic screening of navigation modes based on the switching cost index to determine a dominant navigation mode and an auxiliary navigation mode at the current time.

[0099] The data fusion module 206 is configured to perform frequency characteristic analysis on the dominant navigation mode and the auxiliary navigation mode to generate a mode feature spectrum, determine a frequency coherence coefficient based on the mode feature spectrum, determine a frequency domain fusion strategy based on the frequency coherence coefficient, and generate a fused positioning trajectory based on the frequency domain fusion strategy.

[0100] The optimization output module 207 is configured to obtain an accuracy prediction evaluation result of the fused positioning trajectory, perform trajectory optimization adjustment according to the accuracy prediction evaluation result and the target positioning accuracy index, output a high-precision positioning result, and complete multi-mode navigation positioning of the unmanned aerial vehicle.

[0101] The unmanned aerial vehicle multi-mode navigation positioning apparatus 200 described above can implement the unmanned aerial vehicle multi-mode navigation positioning method of the method embodiment described above. The optional items in the method embodiment described above are also applicable to this embodiment, and will not be described in detail herein. The remaining content of the present embodiment can be referred to the content of the method embodiment described above, and will not be described in detail herein.

[0102] As Figure 3As shown, the third embodiment of the present application further provides a computer device, comprising a memory 301, a processor 302, and a computer program stored in the memory 301 and capable of running on the processor 302, characterized in that the processor 302 implements the steps of the unmanned aerial vehicle multi-mode navigation positioning method according to the first embodiment of the present application when executing the program.

[0103] The above embodiments are intended to exemplarily reproduce and deduce the technical solutions of the present application, and to completely describe the technical solutions, objects and effects of the present application, so as to make the public more thoroughly and comprehensively understand the disclosed content of the present application, and not to limit the protection scope of the present application.

[0104] The above embodiments are not exhaustive enumeration based on the present application, and there can be a plurality of other embodiments not listed. Any replacement and improvement made without violating the concept of the present application is within the protection scope of the present application.

Claims

1. A multi-mode navigation and positioning method for an unmanned aerial vehicle, characterized in that: include: Receive an accuracy requirement request for a UAV navigation mission, and determine a target positioning accuracy index based on the accuracy requirement request; Acquire positioning data of each navigation mode, wherein each navigation mode includes a GPS navigation mode, an inertial navigation mode, and a visual navigation mode, and perform real-time validity evaluation on the positioning data of each navigation mode to generate a multi-mode navigation state set; Performing performance competition among the navigation modes based on the multi-mode navigation state set to generate performance competition results of the navigation modes, including: analyzing the signal strength and satellite visibility parameters of the GPS navigation mode to generate a GPS mode performance evaluation index; analyzing the drift error and stability parameters of the inertial navigation mode to generate an inertial navigation mode performance evaluation index; analyzing the feature matching degree and illumination adaptability parameters of the visual navigation mode to generate a visual navigation mode performance evaluation index; generating performance competition results of the navigation modes based on the GPS mode performance evaluation index, the inertial navigation mode performance evaluation index, and the visual navigation mode performance evaluation index, establishing an inter-mode negotiation mechanism according to the performance competition results of the navigation modes, and determining a navigation mode priority sequence through the inter-mode negotiation mechanism; establishing a navigation mode credibility matrix based on the navigation mode priority sequence and the multi-mode navigation state set, and generating an adaptive weight allocation sequence according to the navigation mode credibility matrix; Analyzing weight change trends of each navigation mode based on the adaptive weight allocation sequence, obtaining system resource overhead required for mode switching based on the weight change trend, quantitatively evaluating the system resource overhead to generate a switching cost index, and dynamically screening navigation modes based on the switching cost index to determine a dominant navigation mode and an auxiliary navigation mode at the current moment; Performing frequency characteristic analysis on the dominant navigation mode and the auxiliary navigation mode to generate a mode characteristic spectrum, determining a frequency coherence coefficient based on the mode characteristic spectrum, determining a frequency domain fusion strategy based on the frequency coherence coefficient, and generating a fused positioning trajectory based on the frequency domain fusion strategy; Obtain the accuracy prediction evaluation result of the fused positioning trajectory, perform trajectory optimization adjustment based on the accuracy prediction evaluation result and the target positioning accuracy index, output a high-precision positioning result, and complete the multi-mode navigation positioning of the UAV.

2. The multi-mode navigation and positioning method for unmanned aerial vehicles according to claim 1, characterized in that: The establishing of an inter-mode negotiation mechanism according to the performance competition results of the navigation modes includes: Establishing a performance comparison matrix based on the performance competition results of each navigation mode; Performing inter-mode complementarity analysis based on the performance comparison matrix to obtain the synergistic enhancement relationship of each navigation mode; Based on the collaborative enhancement relationship, perform resource allocation demand analysis between modes to obtain resource configuration strategies; An inter-mode negotiation process is initiated for the resource allocation strategy, and an inter-mode negotiation mechanism is established.

3. The multi-mode navigation and positioning method for unmanned aerial vehicles according to claim 1, characterized in that: The establishing of a navigation mode credibility matrix based on the navigation mode priority sequence and the multi-mode navigation state set includes: Determining a basic credibility weight of each navigation mode according to the navigation mode priority sequence; Analyzing the real-time performance deviation of each navigation mode based on the multi-mode navigation state set; generating a dynamic credibility correction factor according to the real-time performance deviation; A navigation mode credibility matrix is ​​established based on the basic credibility weight and the dynamic credibility correction factor.

4. The multi-mode navigation and positioning method for unmanned aerial vehicles according to claim 1, characterized in that: Generating an adaptive weight allocation sequence according to the navigation mode credibility matrix includes: Analyzing the credibility distribution characteristics of each mode in the navigation mode credibility matrix; Determining a weight distribution benchmark value according to the credibility distribution characteristics; Performing adaptive weight adjustment based on the weight allocation reference value to generate an initial weight allocation sequence; Normalizing the initial weight distribution sequence to generate an adaptive weight distribution sequence.

5. The multi-mode navigation and positioning method for unmanned aerial vehicles according to claim 1, characterized in that: The acquiring of system resource overhead required for mode switching according to the weight change trend includes: Identifying a mode switching trigger point in the weight change trend; Determining a source mode and a target mode to be switched according to the mode switching trigger point; Analyzing resource requirements for switching from the source mode to the target mode, including computing resource requirements, time delay requirements, and data transmission requirements; The system resource overhead required for mode switching is calculated based on the resource requirement.

6. The multi-mode navigation and positioning method for unmanned aerial vehicles according to claim 1, characterized in that: The determining of the frequency domain fusion strategy based on the frequency coherence coefficient includes: Dividing the frequency bands according to the frequency coherence coefficients to identify high coherence frequency bands and low coherence frequency bands; Determining a fusion mode for each frequency band based on distribution characteristics of the high coherence frequency band and the low coherence frequency band; Designing frequency domain fusion filter parameters according to the fusion mode of each frequency band; A frequency domain fusion strategy is generated based on the frequency domain fusion filter parameters.

7. The multi-mode navigation and positioning method for unmanned aerial vehicles according to claim 3, characterized in that: Generating a dynamic credibility correction factor according to the real-time performance deviation includes: Analyze the numerical distribution range and change trend of the real-time performance deviation; Determining a performance deviation level classification threshold based on the numerical distribution range; Classifying the real-time performance deviation based on the performance deviation classification threshold; A dynamic credibility correction factor is generated according to the grade classification result.

8. A multi-mode navigation and positioning device for an unmanned aerial vehicle, characterized in that: include: A demand receiving module is used to receive the accuracy requirement request of the UAV navigation task and determine the target positioning accuracy index according to the accuracy requirement request; A data acquisition module is used to obtain positioning data of each navigation mode, including GPS navigation mode, inertial navigation mode and visual navigation mode, and perform real-time validity evaluation on the positioning data of each navigation mode to generate a multi-mode navigation state set; a performance competition negotiation module, configured to perform performance competition among the navigation modes based on the multi-mode navigation state set and generate performance competition results for the navigation modes, including: analyzing the signal strength and satellite visibility parameters of the GPS navigation mode to generate a GPS mode performance evaluation index; analyzing the drift error and stability parameters of the inertial navigation mode to generate an inertial navigation mode performance evaluation index; analyzing the feature matching degree and illumination adaptability parameters of the visual navigation mode to generate a visual navigation mode performance evaluation index; generating performance competition results for the navigation modes based on the GPS mode performance evaluation index, the inertial navigation mode performance evaluation index, and the visual navigation mode performance evaluation index; establishing an inter-mode negotiation mechanism based on the performance competition results of the navigation modes, and determining a navigation mode priority sequence through the inter-mode negotiation mechanism; a weight allocation module, configured to establish a navigation mode credibility matrix based on the navigation mode priority sequence and the multi-mode navigation state set, and generate an adaptive weight allocation sequence according to the navigation mode credibility matrix; a decision screening module, configured to analyze a weight change trend of each navigation mode based on the adaptive weight allocation sequence, obtain a system resource overhead required for mode switching based on the weight change trend, quantitatively evaluate the system resource overhead to generate a switching cost index, and dynamically screen the navigation modes based on the switching cost index to determine a dominant navigation mode and an auxiliary navigation mode at the current moment; a data fusion module, configured to perform frequency characteristic analysis on the dominant navigation mode and the auxiliary navigation mode to generate a mode characteristic spectrum, determine a frequency coherence coefficient based on the mode characteristic spectrum, determine a frequency domain fusion strategy based on the frequency coherence coefficient, and generate a fused positioning trajectory based on the frequency domain fusion strategy; The optimization output module is used to obtain the accuracy prediction evaluation result of the fusion positioning trajectory, perform trajectory optimization adjustment according to the accuracy prediction evaluation result and the target positioning accuracy index, output a high-precision positioning result, and complete the multi-mode navigation positioning of the UAV.

9. A computer device, characterized in that: The method comprises a memory and a processor; the memory is used to store a computer program; the processor is used to execute the computer program and implement the method according to any one of claims 1 to 7 when executing the computer program.

Images