Unmanned aerial vehicle intelligent RTK inertial navigation system and method based on visual processing

Abstract

The application discloses a UAV intelligent RTK inertial navigation system and method based on visual processing, and relates to the technical field of UAV positioning. The UAV intelligent RTK inertial navigation method based on visual processing acquires satellite RTK signal data, attitude time sequence data and motion environment image time sequence data of a set UAV; based on the satellite RTK signal data, analyzes initial three-dimensional position coordinates and signal confidence characteristic values; based on the attitude time sequence data of the set UAV, analyzes attitude comprehensive stability characteristic values; based on a pre-trained visual environment perception model and in combination with the motion environment image time sequence data, analyzes environment positioning interference characteristic values. The application corrects and processes the initial three-dimensional position coordinates through the above characteristic values, obtains positioning three-dimensional position coordinates of the set UAV, realizes adaptive collaborative positioning of multi-source data, further maintains the stability of the positioning result, and effectively improves the positioning accuracy of the UAV in a complex environment.

Description

Technical Field

[0001] This invention relates to the field of UAV positioning technology, specifically to a vision-based intelligent RTK inertial navigation system and method for UAVs. Background Technology

[0002] With the widespread application of drones in surveying, inspection, emergency rescue and other fields, their flight positioning accuracy and attitude stability have become key factors affecting the quality of mission execution. In existing technologies, drones usually obtain high-precision position coordinates through the Global Positioning System (GPS) and combine them with inertial navigation systems to achieve short-term continuous navigation. However, in complex environments, such as densely built-up areas, forest areas, and canyons, satellite signals are easily interfered with by obstruction, reflection and multipath effects, resulting in a significant decrease in RTK positioning accuracy. In addition, inertial navigation systems have a cumulative drift problem during long-term operation, which will further amplify the position error if there is no external correction.

[0003] Meanwhile, with the development of computer vision and artificial intelligence technologies, visual information-based assisted positioning methods have gradually attracted attention. However, existing methods are mostly limited to independent visual odometry or image matching, lacking deep integration with satellite positioning signals and inertial navigation data.

[0004] The limitations of existing technologies include at least the following problems: Existing technologies are difficult to achieve adaptive correction of UAV positioning in complex environments. As a result, UAVs cannot make timely dynamic corrections based on changes in various data sources in complex environments, which can easily lead to positioning deviations and attitude drift. For example, when the RTK signal strength fluctuates greatly, existing technologies fail to combine the trend of inertial navigation attitude changes with the stability of visual images for confidence correction, resulting in the positioning results being greatly affected by signal interference. In addition, UAVs experience positioning response delays when their attitude changes drastically, and it is difficult to extract the evolution law of environmental interference from image time series. Especially in scenarios with light reflection or many dynamic obstacles, positioning errors will be aggravated, making it difficult to achieve high-precision three-dimensional positioning. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a vision-based intelligent RTK inertial navigation system and method for unmanned aerial vehicles (UAVs), which solves the problem that existing technologies struggle to integrate multi-source data for dynamic correction, resulting in insufficient positioning accuracy for UAVs in complex environments.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a vision-based intelligent RTK inertial navigation method for unmanned aerial vehicles (UAVs), comprising the following steps: acquiring satellite RTK signal data, attitude timing data, and motion environment image timing data of the UAV; analyzing the initial three-dimensional position coordinates and signal confidence feature values ​​of the UAV based on the satellite RTK signal data; analyzing the comprehensive attitude stability feature values ​​of the UAV based on the attitude timing data; analyzing the environmental positioning interference feature values ​​of the UAV based on a pre-trained visual environment perception model and in conjunction with the motion environment image timing data; and correcting the initial three-dimensional position coordinates based on the signal confidence feature values, attitude stability feature values, and environmental positioning interference feature values ​​of the UAV to obtain the positioning three-dimensional position coordinates of the UAV.

[0007] Furthermore, the satellite RTK signal data includes RTK signal data from several satellites, and each RTK signal data specifically includes the signal strength value and pseudo-random code at each signal time point. The specific steps for analyzing and setting the initial three-dimensional position coordinates of the UAV are as follows: Based on the RTK signal data of each satellite of the UAV, extract the pseudorange observation value and carrier phase observation value at the corresponding signal time point; Based on the pseudorange observation value and carrier phase observation value of each satellite of the UAV at each signal time point, analyze and set the initial three-dimensional position coordinates of the UAV.

[0008] Furthermore, the specific steps for analyzing the signal confidence characteristics of the designated UAV are as follows: read the signal strength value of each signal at each time point of each satellite of the designated UAV, and analyze the signal positioning interference characteristic set of the corresponding satellite, including energy deflection characteristic value, energy entropy change characteristic value, and main peak energy shift characteristic value; based on the signal positioning interference characteristic set of each satellite of the designated UAV, analyze the signal confidence characteristics of the designated UAV.

[0009] Furthermore, the attitude time-series data includes the attitude angle set, three-axis angular velocity values, and three-axis acceleration values ​​at each attitude time point. The specific steps for analyzing the comprehensive attitude stability feature value of the UAV are as follows: Based on the attitude time-series data of the UAV, analyze the attitude stability response feature set of the UAV, including attitude disturbance entropy feature value, attitude coordination fluctuation feature value, and attitude response synchronization feature value; perform fusion processing on the attitude stability response feature set of the UAV to obtain the comprehensive attitude stability feature value of the UAV.

[0010] Furthermore, the specific steps for analyzing the attitude stability response feature set of the UAV are as follows: perform time-series processing on the attitude angle set at each attitude time point of the UAV to obtain the attitude disturbance entropy feature value and attitude coordination fluctuation feature value of the UAV; and perform correlation processing on the three-axis angular velocity value and three-axis acceleration value at each attitude time point of the UAV to obtain the attitude response synchronization feature value of the UAV.

[0011] Furthermore, the motion environment image temporal data includes several frames of motion environment image data, and each frame of motion environment image data specifically includes the pixel value, two-dimensional coordinates, and corresponding depth information value of each pixel in the motion environment image. The visual environment perception model includes an input layer, a feature decomposition layer, an LSTM layer, and an output layer.

[0012] Further, the specific steps for analyzing the environmental positioning interference feature values ​​of the UAV are as follows: input the motion environment image time series data of the UAV into the pre-trained visual environment perception model, analyze the environmental interference mapping feature set of the UAV, including propagation obstruction evolution feature values, scattering distribution evolution feature values, and environmental interference cooperative feature values; based on the environmental interference mapping feature set of the UAV, analyze the environmental positioning interference feature values ​​of the UAV.

[0013] Further, the specific steps for analyzing the environmental interference mapping feature set of the designated UAV are as follows: In the input layer of the visual environment perception model, the motion environment image data of each frame of the designated UAV is received and preprocessed; in the feature decomposition layer of the visual environment perception model, feature extraction processing is performed on the preprocessed motion environment image data of each frame of the designated UAV to obtain the environmental interference temporal feature vector in the corresponding frame motion environment image; in the LSTM layer of the visual environment perception model, temporal analysis is performed on the environmental interference temporal feature vector in each frame motion environment image of the designated UAV to obtain the environmental evolution feature vector of the designated UAV; in the output layer of the visual environment perception model, based on the environmental evolution feature vector of the designated UAV, the environmental interference mapping feature set of the designated UAV is output.

[0014] Furthermore, the specific steps to obtain the three-dimensional position coordinates of the set UAV are as follows: Based on the signal confidence feature value, attitude integrated stability feature value, and environmental positioning interference feature value of the set UAV, analyze the positioning signal feature value of the set UAV; Based on the positioning signal feature value of the set UAV, and combined with the initial three-dimensional position coordinates, analyze the three-dimensional position coordinates of the set UAV.

[0015] A vision-based intelligent RTK inertial navigation system for unmanned aerial vehicles (UAVs) includes: a data acquisition module for acquiring satellite RTK signal data, attitude timing data, and motion environment image timing data of the UAV; an initial positioning module for analyzing the initial three-dimensional position coordinates and signal confidence feature values ​​of the UAV based on the satellite RTK signal data; an attitude analysis module for analyzing the comprehensive attitude stability feature values ​​of the UAV based on the attitude timing data; an environmental positioning interference analysis module for analyzing the environmental positioning interference feature values ​​of the UAV based on a pre-trained visual environment perception model and combined with motion environment image timing data; and a comprehensive positioning analysis module for correcting the initial three-dimensional position coordinates based on the signal confidence feature values, attitude stability feature values, and environmental positioning interference feature values ​​of the UAV to obtain the final three-dimensional position coordinates of the UAV.

[0016] The present invention has the following beneficial effects:

[0017] This vision-based intelligent RTK inertial navigation method for UAVs extracts initial three-dimensional position coordinates and signal confidence feature values ​​through deep analysis of satellite RTK signal data. Based on this, it performs deep fusion processing of the extracted feature values ​​and signal confidence feature values ​​with attitude time series data and motion environment image time series data to generate positioning confidence feature values. This allows for precise correction of the initial three-dimensional position coordinates, thereby achieving adaptive collaborative positioning based on multi-source data. This maintains the stability of positioning results even when there are signal fluctuations or environmental changes, and significantly reduces positioning drift, thus effectively improving the positioning accuracy of UAVs in complex environments.

[0018] (2) The vision-based UAV intelligent RTK inertial navigation method analyzes the attitude time series data and introduces the attitude stability response feature set to generate attitude comprehensive stability feature value, thereby enabling real-time identification of the smoothness and coordination of attitude changes. In turn, it can promptly perceive the attitude change trend when the UAV is rapidly turning, climbing or experiencing disturbances in attitude fluctuation, suppress the transmission effect of cumulative inertial navigation error, and effectively improve the attitude control accuracy of the UAV in high-speed maneuvering or complex airflow environment. At the same time, the attitude comprehensive stability feature value can assist in correcting the initial three-dimensional position coordinates, thereby enabling the UAV to maintain accurate three-dimensional positioning capability under disturbance conditions.

[0019] (3) The UAV intelligent RTK inertial navigation method based on vision processing constructs a visual environment perception model, performs multi-layer feature decomposition and temporal evolution analysis on the time series data of motion environment images, and realizes environmental interference identification and mapping. The model extracts the environmental interference temporal feature vector of each frame image in the feature decomposition layer, realizes the temporal sequence correlation modeling of image features in the LSTM layer, and then outputs the environmental interference mapping feature set, which can reflect the influence of changes in illumination, reflection and dynamic obstacles in the visual scene on the positioning process, thereby generating environmental positioning interference feature values, and dynamically correcting the initial three-dimensional position coordinates based on this, so that the UAV has the ability to adapt to the environment under visual perception, and significantly improves the anti-interference performance of the UAV in complex visual environment positioning.

[0020] (4) The vision-processing-based UAV intelligent RTK inertial navigation system realizes hierarchical analysis and inter-layer collaboration of multi-source data through modular structure design. The data acquisition module ensures that RTK signals, attitude data and motion environment image data participate in positioning analysis within a unified period. The initial positioning module, attitude analysis module and environmental positioning interference analysis module independently extract features such as signal confidence, attitude stability and environmental interference. The comprehensive positioning analysis module realizes accurate correction of the initial three-dimensional position coordinates based on the fusion analysis of the above features, thereby significantly improving the computational efficiency of UAV positioning and thus improving its mission execution reliability under complex working conditions.

[0021] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0022] Figure 1 This is a flowchart of the intelligent RTK inertial navigation method for UAVs based on vision processing according to the present invention.

[0023] Figure 2 This is a schematic diagram illustrating the setting of satellite sequence signal positioning interference feature set data for the UAV in the vision-processing-based intelligent RTK inertial navigation method for UAVs according to the present invention.

[0024] Figure 3 This is a flowchart illustrating the specific steps involved in analyzing and setting the environmental interference mapping feature set of a UAV in the vision-processing-based intelligent RTK inertial navigation method of the present invention.

[0025] Figure 4 This is a block diagram of the vision-processing-based intelligent RTK inertial navigation system for unmanned aerial vehicles (UAVs) according to the present invention. Detailed Implementation

[0026] Please see Figure 1This invention provides a technical solution: a vision-based intelligent RTK inertial navigation method for unmanned aerial vehicles (UAVs), comprising the following steps: within a set period (the period being the least common multiple of the sampling frequencies of RTK signals, attitude timing data, and motion environment image timing data), acquiring satellite RTK signal data, attitude timing data, and motion environment image timing data of a designated UAV; based on the satellite RTK signal data of the designated UAV, analyzing the initial three-dimensional position coordinates and signal confidence feature values ​​of the designated UAV; based on the attitude timing data of the designated UAV, analyzing the attitude comprehensive stability feature values ​​of the designated UAV; based on a pre-trained visual environment perception model and combined with the motion environment image timing data, analyzing the environmental positioning interference feature values ​​of the designated UAV; and correcting the initial three-dimensional position coordinates based on the signal confidence feature values, attitude comprehensive stability feature values, and environmental positioning interference feature values ​​of the designated UAV to obtain the positioning three-dimensional position coordinates of the designated UAV.

[0027] The specific steps to obtain the three-dimensional position coordinates of the designated UAV are as follows: Based on the signal confidence feature value, attitude integrated stability feature value, and environmental positioning interference feature value of the designated UAV, the positioning signal feature value of the designated UAV is analyzed, and the specific formula is as follows: ;in, To set the location information characteristic values ​​of the UAV, To set the signal confidence characteristic value of the drone, The confidence adjustment coefficients for the signals stored in the database. To set the attitude comprehensive stability eigenvalues ​​of the UAV, These are the attitude stability adjustment coefficients stored in the database. To set the environmental positioning interference characteristic values ​​for the UAV, The environmental location interference adjustment coefficient is stored in the database. , These are the interaction adjustment coefficients stored in the database.

[0028] It needs to be explained that the signal confidence adjustment coefficients stored in the database Attitude stability adjustment coefficient Environmental positioning interference adjustment coefficient The acquisition steps are as follows: Obtain signal confidence feature values, attitude integrated stability feature values, and environmental positioning interference feature values ​​for several historical periods. Extract the mean values ​​of signal confidence features, attitude integrated stability features, and environmental positioning interference features, and sum them to obtain a confidence sum. Ratio the mean values ​​of signal confidence features, attitude integrated stability features, and environmental positioning interference features to the confidence sum, and use the corresponding results as the signal confidence adjustment coefficient. Attitude stability adjustment coefficient Environmental positioning interference adjustment coefficient .

[0029] Interaction adjustment coefficients stored in the database The acquisition steps are as follows: Normalize the signal confidence feature value, attitude integration stability feature value, and environmental positioning interference feature value for each historical period. Extract the Pearson correlation coefficient between the signal confidence feature value and the attitude compensation response feature value to obtain their coordination strength. Simultaneously, extract the negative correlation coefficient (absolute value) between the environmental positioning interference feature value and the aforementioned coordination strength, and perform a ratio operation: coordination strength / (1 + negative correlation coefficient). Use this result as the interaction adjustment coefficient. .

[0030] Based on the positioning information feature value of the UAV and combined with the initial three-dimensional position coordinates, the positioning three-dimensional position coordinates of the UAV are analyzed. Specifically, the positioning information feature value of the UAV is normalized and mapped to the range of 0-1, and compared with the preset positioning information feature threshold. If the positioning information feature value of the UAV is higher than the preset positioning information feature threshold, the initial three-dimensional position coordinates of the UAV are marked as the positioning three-dimensional position coordinates.

[0031] If the positioning information feature value of the UAV is lower than or equal to the preset positioning information feature threshold, then the roll angle, pitch angle, and yaw angle values ​​of the UAV at each attitude time point are read, and the difference between each attitude angle at adjacent attitude time points is processed in turn, and the moving average value is taken to obtain the trend values ​​of roll angle change, pitch angle change, and yaw angle change.

[0032] The system reads the three-dimensional position coordinates of the UAV at each signal time point, performs difference processing on the three-dimensional position coordinates of adjacent signal time points, and performs moving average processing to obtain the changes in the initial three-dimensional position coordinates on the X, Y, and Z axes. It then analyzes whether the signs of the roll angle, pitch angle, and yaw angle change trends are the same as the changes on the X, Y, and Z axes.

[0033] If the roll angle change trend value has the same sign as the change in the X-axis, it indicates that the change trend of the UAV in the lateral attitude change direction is consistent with the initial three-dimensional position coordinates. It is determined that the RTK signal response in this direction is synchronized, and reverse correction needs to be performed to offset the excess drift deviation. That is, based on the initial three-dimensional position coordinates, the coordinate components are reduced proportionally in the lateral X-axis direction. The X-axis coordinate value in the initial three-dimensional position coordinates is: (1 - fixed position signal characteristic value) × (absolute value of the difference between the X-axis coordinate value in the three-dimensional position coordinates of the last signal time point and the previous signal time point in this cycle).

[0034] If the roll angle change trend value has the opposite sign to the change on the X-axis, it indicates that the lateral attitude change direction of the UAV is opposite to the change direction of the initial three-dimensional position coordinates. It is determined that there is a lag or drift in the RTK signal in this direction, and in-direction compensation correction needs to be performed to offset the delay error. That is, the X-axis coordinate value in the initial three-dimensional position coordinates + (1 - fixed position signal feature value) × (absolute value of the difference between the X-axis coordinate value in the three-dimensional position coordinates of the last signal time point and the previous signal time point in this cycle).

[0035] Similarly, if the pitch angle change trend value has the same sign as the change on the Y-axis, then reverse correction is performed; otherwise, same-direction compensation correction is performed. If the yaw angle change trend value has the same sign as the change on the Z-axis, then reverse correction is performed; otherwise, same-direction compensation correction is performed. This is how the three-dimensional position coordinates of the UAV are set.

[0036] Specifically, the satellite RTK signal data includes RTK signal data from several satellites (acquired through the RTK receiver equipped on the UAV), and each RTK signal data specifically includes the signal strength value and pseudo-random code at each signal time point. The specific steps for analyzing and setting the initial three-dimensional position coordinates of the UAV are as follows:

[0037] Based on the RTK signal data of each satellite of the designated UAV, the pseudorange observations and carrier phase observations at the corresponding signal time points are extracted. Specifically, the signal data of each satellite of the designated UAV is frequency-converted, and the pseudo-random code at each signal time point after frequency conversion is compared with the reference pseudo-random code sequence of the corresponding satellite stored in the data (including reference pseudo-random codes for several signal time points, which are standard code sequences generated and stored locally by the RTK receiver equipped with the designated UAV based on the pseudo-random code identifier of the satellite stored in the database) for time offset correlation analysis. That is, the reference pseudo-random code is adjusted step by step on the time axis with a fixed step size. The time offset of the sequence is calculated, and the correlation value between the pseudo-random code and the reference pseudo-random code is analyzed at each time offset (this correlation value can be extracted using a cross-correlation matching algorithm). When the correlation value reaches its maximum, the corresponding time offset is taken as the propagation delay of the satellite signal. The speed of light value stored in the database is read and multiplied with the propagation delay to extract the pseudorange observation value of the corresponding signal time point. At the same time, the carrier phase error of each satellite signal data after frequency conversion is estimated and corrected through a phase-locked loop (PLL) to track the change of satellite carrier phase in the continuous time series in real time, so as to extract the carrier phase observation value corresponding to each signal time point.

[0038] Based on the pseudorange and carrier phase observations of each satellite of the designated UAV at each signal time point, the initial three-dimensional position coordinates of the designated UAV are analyzed. Specifically, based on the pseudorange and carrier phase observations of each satellite of the designated UAV at each signal time point, combined with the spatial three-dimensional coordinates of each satellite stored in the database, a GNSS positioning algorithm is used (such as constructing the observation equations of pseudorange and carrier phase, and solving them using the least squares method, extended Kalman filter, or double-difference RTK solution method, etc.) to extract the three-dimensional position coordinates (including the corresponding coordinate values ​​of the X, Y, and Z axes) of the designated UAV at each signal time point, and a weighted average is performed to extract the initial three-dimensional position coordinates of the designated UAV.

[0039] The specific steps for analyzing the signal confidence characteristics of the UAV are as follows: Read the signal strength value of each satellite of the UAV at each signal time point, and analyze the signal positioning interference characteristic set of the corresponding satellite, including energy deflection characteristic value, energy entropy change characteristic value, and main peak energy shift characteristic value. Specifically, for the signal strength value of each satellite at each signal time point, take every 3 signal time points as a sliding window and one signal time point as the sliding step size, extract the deviation between the signal strength value of the middle signal time point and the average signal strength of the adjacent signal time points (i.e., take the average of the difference), and sum the deviations to obtain the deviation sum value. Ratio the deviation of each sliding window with the deviation sum value, and perform weighted processing based on the results to extract the energy deflection characteristic value of each satellite, which is used to characterize the perturbation of the signal strength. When it is large, it indicates that the signal has abrupt change or asymmetric fluctuation in the current period, and there is a perturbation or reflection phenomenon in the propagation path.

[0040] The signal strength value of each satellite of the UAV is divided into several short-time sub-intervals. Within each short-time sub-interval, the signal strength value is divided into several intensity intervals according to the range of signal strength values. The probability value of the signal strength value in each intensity interval is counted, and the signal strength information entropy of the corresponding short-time sub-interval is extracted. The information entropy difference between adjacent sub-intervals is extracted in turn (absolute value) and averaged to extract the energy entropy change characteristic value of each satellite. This characteristic value is used to characterize the intensity of the change in energy distribution disorder within the current period, that is, to reflect the dynamic fluctuation of signal strength distribution on a short time scale. When the value is large, it indicates that the disorder of signal energy distribution changes drastically within the period, which means that the signal is significantly affected by external disturbances, multipath reflections or blockages.

[0041] Based on the signal strength value of each satellite of the UAV at each signal time point, the maximum signal strength value of each satellite and its corresponding signal time point within the set period are extracted. This signal time point is recorded as the main peak time point. The time difference (absolute value) between this signal time point and the center signal time point of the set period (i.e., the signal time point corresponding to the median of the first and last signal time points of the set period) is then compared with the total duration of the set period (i.e., the total number of signal time points) to extract the main peak energy shift characteristic value of each satellite. This characteristic value is used to characterize the temporal shift of the main energy distribution of the satellite signal in the current period, reflecting the consistency of the signal propagation path and phase. When the value is small, it indicates that the main energy peak of the signal is close to the center of the period, the signal propagation path is stable, and the phase consistency is high.

[0042] Based on the signal positioning interference feature set of each satellite of the designated UAV, the signal confidence feature value of the designated UAV is analyzed. Specifically, based on the energy deflection feature value, energy entropy change feature value, and main peak energy shift feature value of each satellite of the designated UAV, the signal stability feature value of the corresponding satellite is analyzed, and the average value is taken to obtain the signal confidence feature value of the designated UAV.

[0043] The specific formula for calculating the signal stability characteristic value of a certain satellite for a specific UAV is as follows: ;in, To set the signal stability characteristic value of a certain satellite for the drone, To set the energy deflection characteristic value of a certain satellite for the UAV, The energy deflection adjustment coefficients are stored in the database. To define the energy entropy change characteristic value of a certain satellite of the UAV, The energy entropy change adjustment coefficient is stored in the database. To set the peak energy shift characteristic value of a certain satellite for a drone, The main peak energy offset adjustment coefficient is stored in the database. These are the coordination coefficients stored in the database. .

[0044] It should be noted that the energy deflection adjustment coefficients stored in the database Energy entropy change adjustment coefficient Main peak energy shift adjustment coefficient The steps to obtain it are as follows:

[0045] The energy deflection characteristic value, energy entropy change characteristic value, and main peak energy shift characteristic value of each satellite are read. The mean values ​​of the energy deflection characteristic value, energy entropy change characteristic value, and main peak energy shift characteristic value are extracted respectively, and then summed to obtain the signal sum value. The mean values ​​of the energy deflection characteristic value, energy entropy change characteristic value, and main peak energy shift characteristic value are then compared with the signal sum value to obtain the energy deflection adjustment coefficient. Energy entropy change adjustment coefficient Main peak energy shift adjustment coefficient .

[0046] Coordination coefficients stored in the database The acquisition steps are as follows: Read the energy deflection characteristic value, energy entropy change characteristic value, and main peak energy shift characteristic value of each satellite, and use the Pearson correlation coefficient for any two of these characteristic values ​​to analyze their corresponding correlation values ​​(take the absolute value), and take the mean value as the co-regulation coefficient. .

[0047] The following is a specific implementation example for calculating the signal stability characteristic value of a certain satellite of a designated UAV. The available data includes: energy deflection characteristic value, energy entropy change characteristic value, and main peak energy shift characteristic value of four randomly selected satellites of the designated UAV, as detailed in Table 1 and... Figure 2 As shown:

[0048] Table 1. Example of setting up satellite sequence signal positioning interference feature set data for UAVs

[0049] Energy deflection eigenvalue Energy entropy change eigenvalue Main peak energy shift eigenvalue Satellite 1 0.344 0.421 0.213 Satellite 2 0.189 0.327 0.087 Satellite 3 0.246 0.358 0.124 Satellite 4 0.302 0.394 0.149

[0050] Energy deflection adjustment coefficients stored in the database Approximately 0.342;

[0051] Energy entropy change adjustment coefficient Approximately 0.372;

[0052] Main peak energy shift adjustment coefficient Approximately 0.286;

[0053] Coordination coefficients stored in the database Approximately 0.784;

[0054] Substituting the data from Table 1 and the aforementioned coefficients into the specific formula for calculating the signal stability characteristic value of a certain satellite for a given UAV, we obtain:

[0055] The signal stability characteristic value of the first satellite of the UAV is set to exp(-0.784×(0.342×0.344+0.372×0.421+0.286×0.213))≈0.769;

[0056] Set the signal stability characteristic value of the second satellite of the UAV to exp(-0.784×(0.342×0.189+0.372×0.327+0.286×0.087))≈0.847;

[0057] Set the signal stability characteristic value of the third satellite of the UAV to exp(-0.784×(0.342×0.246+0.372×0.358+0.286×0.124))≈0.820;

[0058] Set the signal stability characteristic value of the fourth satellite of the UAV to exp(-0.784×(0.342×0.302+0.372×0.394+0.286×0.149))≈0.795.

[0059] In this implementation scheme, the initial three-dimensional coordinates are calculated by performing deep analysis on the RTK signal. Signal confidence analysis is introduced for the RTK signal, taking into account the fluctuation pattern of signal strength over time, changes in energy distribution, and the time-domain shift of the main peak. This allows for the sensitive capture of subtle changes in signal stability. Based on this, energy deflection adjustment coefficient, energy entropy change adjustment coefficient, main peak energy shift adjustment coefficient, and collaborative adjustment coefficient are set to enable the characteristic results of each satellite to have adaptive balancing capabilities. This avoids abnormal signals from having an excessive impact on the overall positioning accuracy, enhances the positioning accuracy of the UAV in complex scenarios, and enables the UAV to maintain high confidence in navigation stability during dynamic flight and rapid attitude changes.

[0060] Specifically, the attitude time-series data includes the attitude angle set, three-axis angular velocity values, and three-axis acceleration values ​​at each attitude time point. The specific steps for analyzing the comprehensive attitude stability characteristic value of the UAV are as follows: Based on the attitude time-series data of the UAV, analyze the attitude stability response characteristic set of the UAV, including attitude disturbance entropy characteristic value, attitude coordination fluctuation characteristic value, and attitude response synchronization characteristic value; perform fusion processing on the attitude stability response characteristic set of the UAV to obtain the comprehensive attitude stability characteristic value of the UAV, that is, perform weighted processing on the attitude disturbance entropy characteristic value, attitude coordination fluctuation characteristic value, and attitude response synchronization characteristic value of the UAV, and perform reciprocal processing on the result, that is, the reciprocal of the weighted processing result + 1, to obtain the comprehensive attitude stability characteristic value of the UAV, which is used to characterize the dynamic stability of the UAV's attitude within the set period. The larger the value, the more stable the UAV's attitude and the smaller the compensation requirement.

[0061] The attitude angle set includes roll angle, pitch angle, and yaw angle values, all of which can be obtained by an IMU installed on the UAV body.

[0062] The three-axis angular velocity values ​​are set to the angular velocities of the drone along the X, Y, and Z axes, which are acquired in real time via gyroscopes.

[0063] The three-axis acceleration values ​​are set as the linear accelerations of the UAV in the X, Y, and Z axes, which are acquired in real time by accelerometers.

[0064] The specific steps for analyzing the attitude stability response feature set of the UAV are as follows: Time-series processing is performed on the attitude angle set at each attitude time point of the UAV to obtain the attitude perturbation entropy feature value and attitude coordination fluctuation feature value of the UAV. Specifically:

[0065] For each attitude time point, the roll, pitch, and yaw angle values ​​are differentially processed (absolute values ​​are taken) to obtain several sets of roll, pitch, and yaw angle changes at adjacent attitude time points. Sample entropy processing is then applied to each set of roll, pitch, and yaw angle changes, using a window length (e.g., 3 sets of adjacent attitude time points). This results in several data segments. For any given data segment, it is compared one by one with the other data segments in the same sequence, and the maximum absolute difference between the corresponding data points is calculated. The data is compared with a preset threshold. If the maximum absolute difference is less than the preset threshold, it is determined to be a similar data segment pair. The number of similar data segment pairs is counted to obtain the similarity probability. The window length is increased by 1, and the above process is repeated to obtain the expanded similarity probability. The sample entropy value is then analyzed to obtain the roll angle change entropy value, pitch angle change entropy value, and yaw angle change entropy value. A weighted average is then performed to obtain the attitude disturbance entropy feature value, which is used to characterize the complexity of the attitude change of the UAV within a set period. The larger the value, the more disordered the attitude change and the worse the attitude stability.

[0066] Based on the roll angle, pitch angle, and yaw angle values ​​at each attitude time point, the variance of each attitude angle and the covariance between any two attitude angles are extracted, and a covariance matrix is ​​formed. The trace value (the three elements on the main diagonal of the covariance matrix, i.e., the algebraic sum of the variance values ​​of each attitude angle, which can be obtained by sequentially accumulating the elements on the diagonal of the matrix) and the determinant value of the covariance matrix are extracted and weighted to extract attitude coordination fluctuation feature values, which are used to characterize the degree of fluctuation of the overall attitude coordination and stability of the UAV within a set period.

[0067] The three-axis angular velocity and three-axis linear acceleration values ​​of the UAV at each attitude time point are correlated to obtain the attitude response synchronization characteristic value of the UAV. Specifically, the coaxial angular velocity and linear acceleration at each attitude time point are cross-correlated. That is, the angular velocity and linear acceleration of that axis are arranged point by point in chronological order. Then, the time position relationship between the two sets of data is changed sequentially, so that the angular velocity data as a whole is shifted forward or backward by a certain time interval. Under each shift position, the angular velocity data and linear acceleration data are multiplied point by point at the same time point and summed. This allows for comparison of the two values ​​under different time offsets. The similarity of the trend of the data set is compared; the similarity results under all time offsets are compared, and the data set with the highest similarity is found. The time interval corresponding to this set is the similarity time difference of that axis. The similarity time difference of each coaxial axis is obtained, and the synchronization value of the corresponding coaxial axis is extracted, i.e., 1 / (1+similar time difference). The weighted average is then processed to extract the attitude response synchronization feature value, which is used to characterize the synchronization coordination and timing matching degree between the attitude change and dynamic response of the UAV within a set period. The larger the attitude response synchronization feature value, the more synchronized the changes in angular velocity and linear acceleration are, and the higher the attitude and dynamic coordination is.

[0068] In this implementation scheme, by performing time-series analysis on attitude time-series data, the attitude evolution trend of the UAV during flight can be revealed. For example, by introducing characteristics such as attitude disturbance entropy, coordination fluctuations, and response synchronization, a comprehensive assessment of flight stability can be conducted. When the UAV is in a sharp turn, climbing, or subjected to airflow disturbances, the disorder and structural deviations of attitude changes can be identified in a timely manner, and the cumulative error of the inertial navigation system can be corrected immediately. At the same time, the combined analysis of covariance matrix and sample entropy can not only determine the overall coordination trend of attitude, but also detect abnormal fluctuations on a single axis, avoiding overall deviations caused by local instability. Furthermore, by performing cross-correlation analysis of angular velocity and linear acceleration, the time lag relationship between dynamic response and attitude change can be identified, thereby providing a lead time for attitude stability judgment. This enables the UAV to maintain stable flight even in environments such as high-speed maneuvers or wind disturbances, and provides a more reliable data basis for positioning correction.

[0069] Specifically, such as Figure 3 As shown, the motion environment image temporal data includes several frames of motion environment image data, and each frame of motion environment image data specifically includes the pixel value, two-dimensional coordinates and corresponding depth information value of each pixel in the motion environment image. The visual environment perception model includes an input layer, a feature decomposition layer, an LSTM layer and an output layer.

[0070] The specific steps for analyzing the environmental positioning interference feature values ​​of the designated UAV are as follows: Input the time-series image data of the UAV's motion environment into the pre-trained visual environment perception model, analyze the environmental interference mapping feature set of the designated UAV, including propagation obstruction evolution feature values, scattering distribution evolution feature values, and environmental interference cooperative feature values; Based on the environmental interference mapping feature set of the designated UAV, analyze the environmental positioning interference feature values ​​of the designated UAV, specifically by weighting the propagation obstruction evolution feature values, scattering distribution evolution feature values, and environmental interference cooperative feature values ​​of the designated UAV to obtain the environmental positioning interference feature values ​​of the designated UAV.

[0071] It should be explained that in the weighted processing involved in this implementation example, the weight coefficient corresponding to each parameter can be obtained through the proportional self-normalization method, taking the weighted processing of the environmental positioning interference feature value of the set UAV as an example:

[0072] The propagation obstruction evolution characteristic value, scattering distribution evolution characteristic value, and environmental interference cooperative characteristic value for several historical periods are obtained. The mean values ​​of the propagation obstruction evolution characteristic value, scattering distribution evolution characteristic value, and environmental interference cooperative characteristic value are extracted and summed to obtain the interference sum value. The mean values ​​of the propagation obstruction evolution characteristic value, scattering distribution evolution characteristic value, and environmental interference cooperative characteristic value are respectively compared with the interference sum value to obtain the weight coefficients corresponding to the propagation obstruction evolution characteristic value, scattering distribution evolution characteristic value, and environmental interference cooperative characteristic value.

[0073] The specific steps for analyzing and setting the environmental interference mapping feature set of the UAV are as follows: In the input layer of the visual environment perception model, the motion environment image data of each frame of the UAV is received and preprocessed, such as normalizing each frame of the input motion environment image, mapping the pixel values ​​to the [0, 1] interval, resampling the size of each frame of the motion environment image to unify the input resolution, and removing random noise points in the image by using Gaussian filtering or median filtering.

[0074] In the feature decomposition layer of the visual environment perception model, feature extraction is performed on each frame of motion environment image data of the preprocessed UAV to obtain the temporal feature vector of environmental interference in the corresponding frame of motion environment image, specifically as follows:

[0075] For each pixel in each frame of the motion environment image, HSI transformation is performed to extract the luminance component of the corresponding pixel. Convolution operation is then performed to extract the horizontal and vertical luminance gradients of the corresponding pixel. The square root of the sum of the squares of these two gradients is taken as the luminance change rate of the pixel. Simultaneously, several surrounding pixels (e.g., 3×3 or 5×5 windows) are taken as the center of each pixel as a neighborhood window, and their luminance gradient direction vectors are extracted. That is, the horizontal and vertical luminance gradient values ​​of each pixel in the neighborhood window are read, and the arctangent value of these two values ​​is used to calculate the luminance gradient direction angle of the pixel. The average value is then taken to obtain the local texture direction similarity value of the corresponding pixel. This value is weighted with the luminance change rate, and the average value of the result is taken as the light reflection feature (used to characterize the surface reflection intensity of the area where the pixel is located and its potential interference to the satellite signal propagation path).

[0076] The depth information value of each pixel is compared with the depth information value distribution of each pixel in the neighborhood window (absolute value is taken), and the average value is taken to obtain the average depth difference of the pixel. The average depth difference of all pixels with a preset occlusion threshold is extracted (if the depth difference is higher than the preset occlusion threshold, the pixel is determined to be at the edge of an obstacle or in a rapidly undulating area). The average value is taken as the spatial occlusion feature (used to characterize the blocking strength of the obstacle to the direction of signal propagation), and weighted with the light reflection feature to extract the environmental propagation obstruction feature.

[0077] Discrete Fourier Transform (DFT) is performed on the luminance component of each pixel in each frame of the moving environment image to extract the high-frequency energy and total energy in each frame of the moving environment image. The ratio is then processed to obtain the high-frequency energy proportion (characterizing the proportion of high-frequency details in the image, such as the proportion of rough surfaces). Local luminance variance is extracted using a fixed-size window (e.g., 5×5) (i.e., variance processing is performed on the luminance component of each pixel), and the mean is taken. This mean is then weighted with the high-frequency energy proportion to extract the environmental scattering distribution characteristics (used to characterize the non-directional scattering intensity of satellite signals from the environmental surface).

[0078] The environmental propagation obstruction features and environmental scattering distribution features in each frame of motion environment image are then concatenated in chronological order to form an environmental interference temporal feature vector.

[0079] In the LSTM layer of the visual environment perception model, temporal analysis is performed on the environmental interference temporal feature vectors in each frame of the motion environment image of the designated UAV to obtain the environmental evolution feature vector of the designated UAV. Specifically, the LSTM continuously updates the hidden state information of each time step during the processing through its internal gating unit structure (including input gate, forget gate, and output gate). This process can learn the temporal dependence of features between consecutive frames, capture the long-term evolution law of environmental interference, and automatically filter short-term noise and unstable fluctuations. At each time step, the LSTM layer calculates a new hidden state representation based on the hidden state of the previous time step and the current input feature vector, thereby preserving key environmental change features throughout the entire time sequence. Finally, the LSTM layer outputs an environmental dynamic interference feature vector that comprehensively reflects the impact of the external environment on satellite signal propagation, such as:

[0080] For the environmental propagation obstacle features in each frame of motion environment image, a moving average processing is performed. That is, the sliding window length is set to several frames (e.g., 3 or 5 frames), the sliding step size is one frame, the mean value of the environmental propagation obstacle features is extracted in each sliding window, and the variance is processed to extract the propagation obstacle variance. The mean values ​​of the environmental propagation obstacle features in adjacent sliding windows are then processed by difference processing and the mean value is taken. The propagation obstacle variances are then weighted to extract the propagation obstacle evolution features, which are used to characterize the time evolution trend of interference factors such as reflection and blockage in the external environment on the stability of satellite signal propagation path. When the value is larger, it indicates that the propagation obstacle intensity increases over time and the external interference gradually strengthens.

[0081] For the environmental scattering distribution characteristics in each frame of motion environment image, a moving average processing is performed to extract the scattering distribution evolution characteristics, which are used to characterize the temporal change trend of the intensity of multipath scattering of signals by surface roughness and texture structure in the external environment. When the value is larger, it indicates that the surface roughness of the environment is enhanced and the signal scattering effect is aggravated.

[0082] For each frame of motion environment image, correlation analysis (using Pearson correlation coefficient method) is performed on the environmental propagation obstruction features and environmental scattering distribution features to extract environmental interference cooperative features, which are used to characterize the degree of interference of the synchronous change of reflection shielding effect and scattering effect in the external environment on signal propagation; and the propagation obstruction evolution features, scattering distribution evolution features, and environmental interference cooperative features are concatenated into an environmental evolution feature vector.

[0083] In the output layer of the visual environment perception model, based on the environmental evolution feature vector of the UAV, the environmental interference mapping feature set of the UAV is output. Specifically, the propagation obstacle evolution feature, scattering distribution evolution feature, and environmental interference cooperative feature in the environmental evolution feature vector are activated by the Sigmoid function to obtain the propagation obstacle evolution feature value, scattering distribution evolution feature value, and environmental interference cooperative feature value between 0 and 1.

[0084] The pre-training steps for the visual environment perception model are as follows:

[0085] The labeled dataset consists of time-series images of motion environment collected by UAVs in different complex environments, along with their corresponding ground truth labels for environmental interference. The data collection scenarios include densely built-up areas, forest areas, canyons, open areas, and areas with strong reflectivity. The data is jointly labeled by flight test results and ground illumination occlusion observations to ensure that the training data covers a variety of typical interference patterns. Each sample in the labeled dataset includes several consecutive frames of motion environment images. Each frame contains the pixel value, two-dimensional coordinates, and depth information of each pixel, along with its corresponding true interference level label (including propagation obstruction level, scattering distribution level, and cooperative interference intensity).

[0086] Secondly, the labeled dataset is preprocessed, including pixel value normalization, size resampling, brightness standardization and noise removal, and divided into training set, validation set and test set in chronological order, with proportions of, for example, 80%, 10% and 10%, to ensure that the temporal continuity is not disrupted.

[0087] During the model training phase, preprocessed image data for each frame is input into the input layer of the visual environment perception model. Environmental propagation obstruction features and scattering distribution features are extracted through the feature decomposition layer, and temporal feature learning is performed in the LSTM layer. The LSTM optimizes the weights of the gated units through the backpropagation algorithm (BPTT) to capture the temporal dependence and evolution trend of environmental interference features. During training, the Adam optimizer is used to minimize the loss function (such as mean squared error MSE or cross-entropy), and the learning rate and the number of hidden units in the LSTM are continuously adjusted to improve the convergence speed and generalization ability of the model.

[0088] By monitoring the prediction error of the model in real time under different interference scenarios using the validation set, the training process is stopped early and hyperparameters are fine-tuned. Finally, the model's performance in unseen data is verified using the test set, and its stability and robustness under multiple scenarios and lighting conditions are evaluated. After training, the optimal model parameters are saved for use in environmental perception and localization interference feature extraction during actual drone flight, enabling intelligent adaptation and robust recognition of complex visual environments.

[0089] In this implementation scheme, a visual environment perception model is introduced to analyze the temporal data of motion environment images. This enables the UAV to adaptively identify and perceive external interference in complex visual environments. For example, by performing feature decomposition and temporal analysis on multi-frame images, the model can extract actual spatial interference features from details such as light reflection and occlusion structures. For instance, when the UAV passes through forests, canyons, or urban building complexes, the model can identify the impact trends of reflection, occlusion, and scattering on satellite signals, and make positioning corrections accordingly. Furthermore, the introduction of the LSTM layer enables the model to understand the evolution process of environmental interference and construct an environmental interference mapping feature set. This allows for accurate assessment of interference intensity when signal distortion or multipath effects are significant, thereby significantly improving the UAV's environmental adaptability in complex scenarios such as strong reflections and building complexes, making positioning more stable, and ensuring that the flight process is not affected by sudden short-term visual interference.

[0090] Please see Figure 4 This invention provides a technical solution: a vision-processing-based intelligent RTK inertial navigation system for unmanned aerial vehicles (UAVs), comprising: a data acquisition module for acquiring satellite RTK signal data, attitude timing data, and motion environment image timing data of a designated UAV; an initial positioning module for analyzing the initial three-dimensional position coordinates and signal confidence feature values ​​of the designated UAV based on the satellite RTK signal data; an attitude analysis module for analyzing the comprehensive attitude stability feature values ​​of the designated UAV based on the attitude timing data; an environmental positioning interference analysis module for analyzing the environmental positioning interference feature values ​​of the designated UAV based on a pre-trained visual environment perception model and combined with the motion environment image timing data; and a comprehensive positioning analysis module for correcting the initial three-dimensional position coordinates based on the signal confidence feature values, attitude stability feature values, and environmental positioning interference feature values ​​of the designated UAV to obtain the positioning three-dimensional position coordinates of the designated UAV.

[0091] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0092] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. An unmanned aerial vehicle intelligent RTK inertial navigation method based on visual processing, characterized in that, The method comprises the following steps: Obtaining satellite RTK signal data, attitude time series data and motion environment image time series data of the set unmanned aerial vehicle; Based on the satellite RTK signal data of the set unmanned aerial vehicle, the initial three-dimensional position coordinates and signal confidence characteristic values of the set unmanned aerial vehicle are analyzed; Based on the attitude time series data of the set unmanned aerial vehicle, the attitude comprehensive stability characteristic values of the set unmanned aerial vehicle are analyzed; Based on the pre-trained visual environment perception model and combined with the motion environment image time series data, the environment positioning interference characteristic values of the set unmanned aerial vehicle are analyzed; Based on the signal confidence characteristic values, the attitude comprehensive stability characteristic values and the environment positioning interference characteristic values of the set unmanned aerial vehicle, the initial three-dimensional position coordinates are corrected to obtain the positioning three-dimensional position coordinates of the set unmanned aerial vehicle, which is specifically: Based on the signal confidence characteristic values, the attitude comprehensive stability characteristic values and the environment positioning interference characteristic values of the set unmanned aerial vehicle, the positioning signal characteristic values of the set unmanned aerial vehicle are analyzed, and the specific formula is as follows: ; wherein, , , , are respectively set position signal confidence characteristic value, signal confidence characteristic value, attitude comprehensive stability characteristic value, environment positioning interference characteristic value of the unmanned aerial vehicle, , , are respectively stored in the database signal confidence adjustment coefficient, attitude stability adjustment coefficient, environment positioning interference adjustment coefficient, , is the interaction adjustment coefficient stored in the database; Based on the positioning signal characteristic values of the set unmanned aerial vehicle and combined with the initial three-dimensional position coordinates, the positioning three-dimensional position coordinates of the set unmanned aerial vehicle are analyzed. 2.The visual processing based intelligent RTK and inertial navigation method for UAVs according to claim 1, wherein, The satellite RTK signal data comprises RTK signal data of a plurality of satellites, and each RTK signal data is specifically signal intensity values and pseudo-random code at each signal time point, and the specific steps of analyzing the initial three-dimensional position coordinates of the set unmanned aerial vehicle are as follows: Based on the RTK signal data of each satellite of the set unmanned aerial vehicle, the pseudo-range observation values and carrier phase observation values at the corresponding signal time points are extracted; Based on the pseudo-range observation values and carrier phase observation values at each signal time point of each satellite of the set unmanned aerial vehicle, the initial three-dimensional position coordinates of the set unmanned aerial vehicle are analyzed. 3.The visual processing based UAV intelligent RTK INS method of claim 2, wherein, The specific steps of analyzing the signal confidence characteristic values of the set unmanned aerial vehicle are as follows: Read the signal intensity values at each signal time point of each satellite of the set unmanned aerial vehicle, analyze the signal positioning interference characteristic set of the corresponding satellite, including energy deflection characteristic values, energy entropy change characteristic values and main peak energy shift characteristic values; Based on the signal positioning interference characteristic set of each satellite of the set unmanned aerial vehicle, the signal confidence characteristic values of the set unmanned aerial vehicle are analyzed. 4.The visual processing based UAV intelligent RTK INS method of claim 1, wherein, The attitude time series data comprises attitude angle set, three-axis angular velocity values and three-axis linear acceleration values at each attitude time point, and the specific steps of analyzing the attitude comprehensive stability characteristic values of the set unmanned aerial vehicle are as follows: Based on the attitude time series data of the set unmanned aerial vehicle, the attitude stability response characteristic set is analyzed, including attitude disturbance entropy characteristic values, attitude coordination fluctuation characteristic values and attitude response synchronization characteristic values; The attitude stability response characteristic set of the set unmanned aerial vehicle is fused to obtain the attitude comprehensive stability characteristic values of the set unmanned aerial vehicle. 5.The visual processing based intelligent RTK and inertial navigation method for UAVs according to claim 4, wherein, The specific steps of analyzing the attitude stability response characteristic set of the set unmanned aerial vehicle are as follows: The attitude angle set at each attitude time point of the set unmanned aerial vehicle is processed in time series to obtain the attitude disturbance entropy characteristic values and the attitude coordination fluctuation characteristic values of the set unmanned aerial vehicle; And the three-axis angular velocity values and the three-axis linear acceleration values at each attitude time point of the set unmanned aerial vehicle are associated to obtain the attitude response synchronization characteristic values of the set unmanned aerial vehicle. 6.The visual processing based UAV intelligent RTK INS method of claim 1, wherein, The motion environment image time series data includes several frames of motion environment image data, and each frame of motion environment image data specifically includes the pixel value, two-dimensional coordinates and corresponding depth information value of each pixel point in the motion environment image. 7.The visual processing based UAV intelligent RTK INS method of claim 6, wherein, The specific steps for analyzing the environment positioning interference characteristic value of the set unmanned aerial vehicle are as follows: The motion environment image time series data of the set unmanned aerial vehicle is input into the pre-trained visual environment perception model, and the environment interference mapping feature set of the set unmanned aerial vehicle is analyzed, including the propagation obstacle evolution characteristic value, the scattering distribution evolution characteristic value and the environment interference coordination characteristic value. Based on the environment interference mapping feature set of the set unmanned aerial vehicle, the environment positioning interference characteristic value of the set unmanned aerial vehicle is analyzed. 8.The visual processing based UAV intelligent RTK INS method of claim 7, wherein, The specific steps for analyzing the environment interference mapping feature set of the set unmanned aerial vehicle are as follows: In the input layer of the visual environment perception model, each frame of motion environment image data of the set unmanned aerial vehicle is received and preprocessed; In the feature decomposition layer of the visual environment perception model, the feature extraction processing is performed on the preprocessed each frame of motion environment image data of the set unmanned aerial vehicle, and the environment interference time series feature vector in the corresponding frame of motion environment image is obtained; In the LSTM layer of the visual environment perception model, the environment interference time series feature vector in each frame of motion environment image of the set unmanned aerial vehicle is analyzed in time series, and the environment evolution feature vector of the set unmanned aerial vehicle is obtained; In the output layer of the visual environment perception model, based on the environment evolution feature vector of the set unmanned aerial vehicle, the environment interference mapping feature set of the set unmanned aerial vehicle is output. 9.The visual processing based UAV intelligent RTK INS method of claim 1, wherein, The specific steps for obtaining the positioning three-dimensional position coordinates of the set unmanned aerial vehicle are as follows: Based on the signal confidence characteristic value, the attitude comprehensive stability characteristic value and the environment positioning interference characteristic value of the set unmanned aerial vehicle, the positioning signal characteristic value of the set unmanned aerial vehicle is analyzed; Based on the positioning signal characteristic value of the set unmanned aerial vehicle, and combined with the initial three-dimensional position coordinates, the positioning three-dimensional position coordinates of the set unmanned aerial vehicle are analyzed.

10. An unmanned aerial vehicle intelligent RTK inertial navigation system based on visual processing, applying the unmanned aerial vehicle intelligent RTK inertial navigation method based on visual processing of any one of claims 1-9, characterized in that, It includes: The data acquisition module is used for acquiring satellite RTK signal data, attitude time series data and motion environment image time series data of the set unmanned aerial vehicle; The initial positioning module is used for analyzing the initial three-dimensional position coordinates and signal confidence characteristic value of the set unmanned aerial vehicle based on the satellite RTK signal data of the set unmanned aerial vehicle; The attitude analysis module is used for analyzing the attitude comprehensive stability characteristic value of the set unmanned aerial vehicle based on the attitude time series data of the set unmanned aerial vehicle; The environment positioning interference analysis module is used for analyzing the environment positioning interference characteristic value of the set unmanned aerial vehicle based on the pre-trained visual environment perception model and combined with the motion environment image time series data; The comprehensive positioning analysis module is used for correcting the initial three-dimensional position coordinates based on the signal confidence characteristic value, the attitude comprehensive stability characteristic value and the environment positioning interference characteristic value of the set unmanned aerial vehicle, and obtaining the positioning three-dimensional position coordinates of the set unmanned aerial vehicle.

Images