Unmanned aerial vehicle slope inspection autonomous navigation path planning system fused with neural network

Abstract

The invention relates to the technical field of unmanned aerial vehicle path planning, and discloses an unmanned aerial vehicle slope inspection autonomous navigation path planning system fused with a neural network, and the system comprises a sensing module which obtains a continuous optical pixel offset matrix and a three-axis disturbance state vector of a controlled object; the mapping module converts the optical pixel offset matrix into a hidden space feature to calculate an error variance parameter, and generates a dynamic feedback damping gain factor; and the control module determines an adaptive weight according to the three-axis disturbance state vector so as to reset the track update frequency and the repulsion field step length, and injects a dynamic feedback damping gain factor into a cost function to recalculate a cost value and generate an autonomous navigation instruction. Time domain dislocation of data correction and control decision is eliminated, feed-forward compensation of high-frequency motion distortion is realized, and navigation instruction hysteresis divergence is inhibited.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of unmanned aerial vehicle path planning, and particularly relates to an unmanned aerial vehicle slope inspection autonomous navigation path planning system fusing a neural network. BACKGROUND

[0002] The current autonomous navigation mode commonly used at present relies on an on-board visual sensor to obtain spatial coordinates and hand over the spatial coordinates to a controller to generate an inspection flight path. However, when the unmanned aerial vehicle is close to a complex slope surface for operation, the canyon shear wind and the slope updraft will exert high-frequency random disturbance on the machine body, resulting in that the collected displacement data contains motion distortion. In order to maintain the accuracy of the flight path, the conventional technology usually adopts an asynchronous decoupling operation architecture of perception layer data correction and control layer path decision, and relies on the front-end calculation module to filter out image noise, and then delivers the smoothed coordinate data to the path planning operator.

[0003] This improved path relying on improving the front-end filtering algorithm to pursue accurate restoration of coordinates deviates from the inherent physical time lag law of the flight control system. When the perception link outputs the corrected coordinates, the machine body attitude has been secondarily deviated due to the continuous airflow impact. For example, a Chinese invention patent with the authorization announcement number CN120353242B discloses a high slope inspection method based on an unmanned aerial vehicle and an unmanned aerial vehicle. An initial flight path is generated by reading a design drawing, and the position of the navigation point is corrected statically or quasi-statically based on the slope flatness and the laser ranging result. It can be seen that the existing technology mainly has the following deficiencies: 1. There is a space-time misalignment between the perception feedback flow and the action response chain, so that the path planner continuously receives out-of-date environmental true values, and the system loses the ability to predict and compensate when facing sudden airflow disturbance; 2. The controller forcibly tracks the filtered and denoised lag coordinates, which excites navigation instruction oscillation under the action of shear wind, and induces dangerous drift of the inspection flight path to the slope surface; 3. The system lacks an interactive loop that converts airflow disturbance energy into control feedback damping, and can only respond to high-frequency disturbance with a fixed search step, making it difficult to balance the trajectory accuracy of near-field inspection and the anti-collision safety of the machine body.

[0004] Therefore, how to break the operation architecture of mutual separation of perception filtering and control decision, directly translate the environmental distortion residual into the adaptive damping gain of the path planning operator, and block the flight path instability chain caused by instruction lag from the physical feedback level, becomes a technical problem to be solved by the present application. SUMMARY

[0005] To solve the problems proposed in the background art, the technical solution of the present application is as follows: an unmanned aerial vehicle slope inspection autonomous navigation path planning system fusing a neural network, comprising: a first spatial physical quantity perception module; a second disturbance parameter mapping module connected with the first spatial physical quantity perception module; and a third trajectory decision control module connected with the second disturbance parameter mapping module and the first spatial physical quantity perception module; The first spatial physical quantity perception module is configured to obtain a continuous optical pixel shift matrix and a three-axis disturbance state vector of the controlled object in a physical space. The second disturbance parameter mapping module is configured to receive the continuous optical pixel shift matrix, convert the continuous optical pixel shift matrix into a hidden space tensor feature by using a preset dimension reduction mapping network, calculate the dispersion of the hidden space tensor feature along the time dimension to output an error variance parameter, and generate a dynamic feedback damping gain factor according to the error variance parameter. The third trajectory decision control module is configured to receive the dynamic feedback damping gain factor and the three-axis disturbance state vector, extract a frequency energy integral value of the three-axis disturbance state vector as a motion energy level distribution representative value, divide the motion energy level distribution representative value by a preset limit bearing energy level value to output an adaptive weight, multiply a basic update frequency of a preset local path planning algorithm by the adaptive weight to obtain a reset track update frequency, multiply a basic step length of the local path planning algorithm by the adaptive weight to obtain a reset repulsive force field step length, multiply the dynamic feedback damping gain factor by a current generation value of the local path planning algorithm to output a recalculated generation value, and generate an autonomous navigation flight instruction of the controlled object according to the reset track update frequency, the reset repulsive force field step length, and the recalculated generation value.

[0006] Preferably, the second disturbance parameter mapping module internally includes a time sequence construction submodule and a variance estimation submodule. The time sequence construction submodule is configured to receive the continuous optical pixel shift matrix and splice the continuous optical pixel shift matrix into a sliding time window tensor of a fixed length. The variance estimation submodule is configured to receive the sliding time window tensor, input the sliding time window tensor into a long short-term memory network structure to generate a time sequence feature tensor of a current time period, calculate a discrete distribution matrix of the time sequence feature tensor, and extract a main diagonal element value of the discrete distribution matrix as the error variance parameter.

[0007] Preferably, the step of generating the dynamic feedback damping gain factor according to the error variance parameter by the second disturbance parameter mapping module includes processing the error variance parameter based on a preset exponential decay mathematical relationship. The exponential decay mathematical relationship is as follows: wherein D is the dynamic feedback damping gain factor, V is the error variance parameter, and k is a preset environmental response reference coefficient.

[0008] Preferably, the step of extracting the frequency energy integral value of the three-axis disturbance state vector as the representative value of the motion energy level distribution by the third trajectory decision control module comprises the following steps in sequence: step 401, extracting the high-frequency vibration component of the three-axis disturbance state vector in the three-dimensional space coordinate system; step 402, performing fast Fourier transform on the high-frequency vibration component to generate a frequency energy spectrum; and step 403, calculating the integral area value of the frequency energy spectrum in a frequency band higher than a preset vibration frequency threshold, and setting the integral area value as the frequency energy integral value.

[0009] Preferably, in response to the value of the adaptive weight being greater than a preset safety threshold parameter, the third trajectory decision control module executes a frequency suppression logic; the frequency suppression logic comprises: resetting the second update frequency of the flight path to the product of the base update frequency and 0.5, and resetting the second step length of the repulsive force field to the product of the base step length and 0.33.

[0010] Preferably, the third trajectory decision control module is internally configured with a disturbance evolution tracking sub-module; the disturbance evolution tracking sub-module records a time sequence set of the adaptive weight in a plurality of consecutive sampling periods, and calculates a time derivative change rate of the time sequence set in a fixed time sliding window; in response to the time derivative change rate being greater than the value 0 for 3 consecutive sampling periods, the third trajectory decision control module outputs a hovering blocking signal and generates a wait-in-place hovering instruction for the controlled object to maintain the current three-dimensional space coordinates, instead of the autonomous navigation flight instruction.

[0011] Preferably, the first spatial physical quantity perception module comprises an optical feature capture component and a spatial inertial measurement component; the optical feature capture component and the spatial inertial measurement component are connected through a hardware crystal oscillator clock synchronization triggering module; the optical feature capture component locks exposure to generate a digital image frame containing a continuous optical pixel offset matrix when receiving a synchronization triggering signal; the spatial inertial measurement component synchronously records three-axis acceleration values and three-axis angular velocity values when receiving the synchronization triggering signal, and splices the three-axis acceleration values and the three-axis angular velocity values into a three-axis disturbance state vector.

[0012] Preferably, the local path planning algorithm is a artificial potential field algorithm, and the current generation value is calculated by the gravitational potential field function and the repulsive potential field function; when outputting the recalculated generation value, the third trajectory decision control module only multiplies the dynamic feedback damping gain factor with the output value of the gravitational potential field function, and keeps the output weight value of the repulsive potential field function constant.

[0013] Preferably, the third trajectory decision control module further comprises a space limiting boundary protection submodule; the space limiting boundary protection submodule acquires a three-dimensional straight line distance numerical value between the controlled object and the physical surface of the slope; in response to the three-dimensional straight line distance numerical value being less than 2 m, the space limiting boundary protection submodule outputs a covering instruction, and sets the output weight numerical value of the repulsive potential field function to 5 times of the original numerical value according to the covering instruction.

[0014] Preferably, the environment response reference coefficient is set to 0.05 to 0.15; the acquisition period of the digital image frame is set to 20 ms to 50 ms; and the upper limit of the output frequency of the autonomous navigation flight instruction is set to 10 Hz.

[0015] Compared with the prior art, the unmanned aerial vehicle slope inspection autonomous navigation path planning system fusing a neural network has the following advantages: in the unmanned aerial vehicle slope inspection autonomous navigation path planning, a physical cooperation mechanism of a perception link and a control link is reconstructed, a disturbance feature representing a continuous image pixel offset is converted into a dynamic feedback damping gain factor in a control logic, and the dynamic feedback damping gain factor is injected into a cost function update link of a path planning algorithm, the architecture design deeply fuses an environmental airflow disturbance parameter and a bottom layer track search logic, eliminates time domain misalignment between data correction and control decision in a traditional asynchronous decoupling mode, the path planning module obtains a prediction ability for a complex airflow impact by the gain mechanism, and realizes feedforward compensation for high-frequency motion distortion at a control feedback level; and a navigation instruction is inhibited from hysteresis divergence through dynamic reconstruction of a path search operator, a processor extracts a motion energy level distribution of a body disturbance state vector to generate an adaptive weight, and maps the adaptive weight to a path search step operator in real time, in a working condition where an environmental disturbance intensity increases, the system actively reduces a track update frequency and shrinks a repulsive field step according to the energy level distribution, and the state adjustment closed loop makes the navigation controller no longer mechanically track discrete coordinate fluctuations containing motion distortion, but increases an anti-disturbance damping component for track smoothness, and blocks track drift induced by an out-of-date true value from a physical action level. BRIEF DESCRIPTION OF DRAWINGS

[0016] Fig. 1 is a logic flow diagram of the autonomous navigation instruction generation of the feature mapping and adaptive damping of the application; Fig. 2 is a logic architecture diagram of the unmanned aerial vehicle autonomous navigation system fusing feedforward damping compensation of the application. DETAILED DESCRIPTION

[0017] The technical solutions in the embodiments of the application will be clearly described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only a part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by a person of ordinary skill in the art belong to the protection scope of the application.

[0018] An autonomous navigation path planning system for unmanned aerial vehicle (UAV) slope inspection integrating neural networks includes: First spatial physical quantity sensing module; The second disturbance parameter mapping module is connected to the first spatial physical quantity sensing module; And the third trajectory decision control module, which is connected to the second disturbance parameter mapping module and the first spatial physical quantity sensing module; The first spatial physical quantity sensing module is configured to acquire the continuous optical pixel offset matrix and the three-axis perturbation state vector of the controlled object in the physical space. The second perturbation parameter mapping module is configured to receive the continuous optical pixel offset matrix, convert the continuous optical pixel offset matrix into latent space tensor features using a preset dimension reduction mapping network, calculate the dispersion of the latent space tensor features along the time dimension to output the error variance parameter, and generate a dynamic feedback damping gain factor based on the error variance parameter. The third trajectory decision control module is configured to receive a dynamic feedback damping gain factor and a three-axis disturbance state vector, extract the frequency domain energy integral value of the three-axis disturbance state vector as a representative value of the motion energy level distribution, divide the representative value of the motion energy level distribution by a preset limit tolerance energy level value to output adaptive weights; multiply the preset base update frequency of the local path planning algorithm by the adaptive weights to obtain the reset track update frequency, and multiply the base step size of the local path planning algorithm by the adaptive weights to obtain the reset repulsion field step size; multiply the dynamic feedback damping gain factor by the current generation value of the local path planning algorithm to output the recalculated generation value, and generate autonomous navigation flight commands for the controlled object based on the reset track update frequency, the reset repulsion field step size, and the recalculated generation value.

[0019] Preferably, the second perturbation parameter mapping module includes a time series construction submodule and a variance estimation submodule. The time series construction submodule is configured to receive a continuous optical pixel offset matrix and concatenate the continuous optical pixel offset matrix into a fixed-length sliding time window tensor. The variance estimation submodule is configured to receive the sliding time window tensor, input the sliding time window tensor into a long short-term memory network structure to generate a temporal feature tensor for the current time period, calculate the discrete distribution matrix of the temporal feature tensor, and extract the values ​​of the main diagonal elements of the discrete distribution matrix as the error variance parameter.

[0020] Preferably, the step of the second disturbance parameter mapping module generating a dynamic feedback damping gain factor based on the error variance parameter includes: processing the error variance parameter based on a preset exponential decay mathematical relationship; the exponential decay mathematical relationship is: Where D is the dynamic feedback damping gain factor, V is the error variance parameter, and k is the preset environmental response baseline coefficient.

[0021] Preferably, the step of extracting the frequency energy integral value of the three-axis disturbance state vector as the representative value of the motion energy level distribution by the third trajectory decision control module comprises the following steps in sequence: step 401, extracting the high-frequency vibration component of the three-axis disturbance state vector in the three-dimensional space coordinate system; step 402, performing fast Fourier transform on the high-frequency vibration component to generate a frequency energy spectrum; and step 403, calculating the integral area value of the frequency energy spectrum in a frequency band higher than a preset vibration frequency threshold, and setting the integral area value as the frequency energy integral value.

[0022] Preferably, in response to the value of the adaptive weight being greater than a preset safety threshold parameter, the third trajectory decision control module executes a frequency suppression logic; the frequency suppression logic comprises: resetting the second update frequency of the flight path to the product of the base update frequency and 0.5, and resetting the second step length of the repulsive force field to the product of the base step length and 0.33.

[0023] Preferably, the third trajectory decision control module is internally configured with a disturbance evolution tracking sub-module; the disturbance evolution tracking sub-module records a time sequence set of the adaptive weight in a plurality of consecutive sampling periods, and calculates a time derivative change rate of the time sequence set in a fixed time sliding window; in response to the time derivative change rate being greater than the value 0 for 3 consecutive sampling periods, the third trajectory decision control module outputs a hovering blocking signal and generates a wait-in-place hovering instruction for the controlled object to maintain the current three-dimensional space coordinates, instead of the autonomous navigation flight instruction.

[0024] Preferably, the first spatial physical quantity perception module comprises an optical feature capture component and a spatial inertial measurement component; the optical feature capture component and the spatial inertial measurement component are connected through a hardware crystal oscillator clock synchronization triggering module; the optical feature capture component locks exposure to generate a digital image frame containing a continuous optical pixel offset matrix when receiving a synchronization triggering signal; the spatial inertial measurement component synchronously records three-axis acceleration values and three-axis angular velocity values when receiving the synchronization triggering signal, and splices the three-axis acceleration values and the three-axis angular velocity values into a three-axis disturbance state vector.

[0025] Preferably, the local path planning algorithm is a artificial potential field algorithm, and the current generation value is calculated by the gravitational potential field function and the repulsive potential field function; when outputting the recalculated generation value, the third trajectory decision control module only multiplies the dynamic feedback damping gain factor with the output value of the gravitational potential field function, and keeps the output weight value of the repulsive potential field function constant.

[0026] Preferably, the third trajectory decision control module further comprises a space limiting boundary protection submodule; the space limiting boundary protection submodule acquires a three-dimensional straight-line distance numerical value between the controlled object and the physical surface of the slope; in response to the three-dimensional straight-line distance numerical value being less than 2 m, the space limiting boundary protection submodule outputs a covering instruction, and sets the output weight numerical value of the repulsive potential field function to 5 times the original numerical value according to the covering instruction.

[0027] Preferably, the environment response reference coefficient is set to 0.05 to 0.15; the acquisition period of the digital image frame is set to 20 ms to 50 ms; and the upper limit of the output frequency of the autonomous navigation flight instruction is set to 10 Hz.

[0028] Embodiment 1: When the controlled object performs close-range inspection of a canyon slope with a height difference greater than 300 m, the canyon shear wind and the slope updraft alternately act on the fuselage, and the optical feature capture assembly in the first space physical quantity perception module generates a digital image frame after locking exposure, and the digital image frame is mixed with nonlinear motion distortion caused by high-frequency aerodynamic disturbance. Under the conventional architecture of asynchronous decoupling of the perception link and the control link, the action of the front-end calculation module to filter out image noise inevitably causes the spatiotemporal misalignment of the perception feedback flow and the action response chain, so that the third trajectory decision control module continuously receives the environment true value with physical time lag, and further causes the navigation instruction to oscillate and induces the controlled object to produce displacement deviation towards the slope surface.

[0029] For the aforementioned airflow disturbance working condition, the bottom-layer data correction logic is reconstructed into a feedforward control gain mechanism directly acting on the path planning operator. In the specific operation process, the first space physical quantity perception module acquires a continuous optical pixel offset matrix and a three-axis disturbance state vector, and synchronously delivers them to the second disturbance parameter mapping module and the third trajectory decision control module. The second disturbance parameter mapping module converts the continuous optical pixel offset matrix into a hidden space tensor feature by using a pre-set dimension reduction mapping network, calculates the dispersion of the hidden space tensor feature along the time dimension to output an error variance parameter V, and generates a dynamic feedback damping gain factor D based on an exponential decay mathematical relationship , where k is a pre-set environment response reference coefficient. The third trajectory decision control module extracts a high-frequency vibration component of the three-axis disturbance state vector in a three-dimensional space coordinate system, performs a fast Fourier transform on the high-frequency vibration component to generate a frequency energy spectrum, calculates an integral area numerical value of the frequency energy spectrum in a frequency band higher than a pre-set vibration frequency threshold as a frequency energy integral value, extracts the frequency energy integral value as a motion energy level distribution representative value, and divides the motion energy level distribution representative value by a pre-set limit bearing energy level value to output an adaptive weight.

[0030] With the rise of airflow disturbance intensity, the error variance parameter V increases, and the dynamic feedback damping gain factor D calculated according to the above exponential decay mathematical relationship decreases. The third trajectory decision control module multiplies the dynamic feedback damping gain factor D by the output value of the gravitational potential field function, while keeping the output weight value of the repulsive potential field function constant, and recalculates the value of the output. At the same time, the third trajectory decision control module multiplies the basic update frequency of the local path planning algorithm by the adaptive weight to obtain the reset path update frequency, and multiplies the basic step length of the local path planning algorithm by the adaptive weight to obtain the reset repulsive field step length. The third trajectory decision control module generates autonomous navigation flight instructions for the controlled object based on the reset path update frequency, the reset repulsive field step length, and the recalculated value. The system relies on the state flow mechanism across the module to translate the environmental distortion residual into an adaptive damping gain of the path planning operator, actively reduces the gravitational field tracking frequency of the target waypoint, and shrinks the repulsive field detection step length, thereby breaking the chain of outdated true value-induced trajectory drift from the physical action level and maintaining the global stability of the near-field inspection trajectory.

[0031] Example 2: A hardware-in-the-loop simulation test platform is constructed to test the control stability of the dynamic feedback damping gain mechanism under the working condition of controlled object trajectory drift induced by canyon shear wind. A six-degree-of-freedom unmanned aerial vehicle flight dynamics model and a computational fluid dynamics wind field model are integrated. The wind field model solves the Navier-Stokes equation and outputs a continuous three-dimensional space airflow disturbance field. To reproduce the mechanical vibration of the optical sensor in the physical space, a Gaussian white noise with a signal-to-noise ratio of 15 dB is injected into the input link of the optical pixel offset matrix, while a periodic power frequency interference with a frequency of 50 Hz is superimposed. The basic update frequency of the local path planning algorithm is set according to the environmental disturbance frequency bandwidth and the onboard processor computing load. Fixed high-frequency path updates cause instruction accumulation and actuator overload. The system sets the basic update frequency reference value to 50 Hz, and the third trajectory decision control module uses the adaptive weight to multiply the basic update frequency, so that the path update frequency decreases when the disturbance energy exceeds the preset vibration frequency threshold.

[0032] The simulation test platform is started and the shear wind speed is set to 12.0 m / s, the three-dimensional coordinate tracking root mean square error and the minimum distance from the slope surface of the controlled object are measured synchronously, the first control group is peeled off the dynamic feedback damping gain factor D, the adaptive weight is retained, the second control group is peeled off the adaptive weight, the dynamic feedback damping gain factor D is retained, the test group is equipped with the complete dimension reduction mapping network and the parameter adjustment mechanism, the environmental response reference coefficient k is set to 1.0, at the moment of strong wind shear access, the first control group measures that the root mean square error rises to 2.45 m, the minimum distance from the slope surface reduces to 0.85 m, the second control group measures that the root mean square error is 1.98 m, the navigation instruction response time delay is 145.2 ms, the test group measures that the root mean square error converges to 0.62 m, the minimum distance from the slope surface maintains 2.34 m, and the instruction response time delay is reduced to 42.1 ms, the original flight trajectory of the first control group without processing presents high-frequency irregular sawtooth oscillation, the trajectory of the test group presents a smooth damping following curve, and the test data show that the dynamic feedback damping gain factor D suppresses the gravitational potential field tracking tendency, the adaptive weight synchronously shrinks the repulsive force field detection step, and the two parameters construct a feedforward inhibition closed loop at the physical action level, and block the instruction oscillation induced by the environmental true value lag.

[0033] In order to establish the working boundary of the environmental response reference coefficient k, the parameter gradient control group is configured, the environmental response reference coefficient k is respectively set to 0.05, 1.0 and 5.5, the trajectory deviation extreme value of each group is measured in the continuous airflow impact sequence, when the environmental response reference coefficient k is set to the lower limit 0.05, the error variance parameter V is difficult to drive the exponential decay mathematical relationship to generate sufficient damping, and the measured trajectory deviation extreme value reaches 1.85 m, when the environmental response reference coefficient k is set to 1.0, the trajectory deviation extreme value decreases and maintains at 0.58 m, when the environmental response reference coefficient k is set to the upper limit 5.5, the dynamic feedback damping gain factor D output by the exponential decay mathematical relationship presents nonlinear accelerated decline, the system enters the over-damping stagnation state, the controlled object loses the tracking maneuverability to the target waypoint, the trajectory deviation extreme value climbs to 3.75 m, and the data curve presents a deterioration inflection point, the test data show that the environmental response reference coefficient k is in the range of 0.8 to 1.2, the anti-interference stability and the tracking sensitivity reach a technical balance, the continuous optical pixel deviation matrix is translated into the dynamic feedback damping gain factor D through the dimension reduction mapping network, the adaptive weight based on the three-axis disturbance state vector is calculated, the numerical distribution of the bottom path cost function is changed, the nonlinear motion distortion caused by high-frequency aerodynamic disturbance is directly converted into the feedforward damping parameter, the technical problems between the coordinate tracking accuracy and the global flight anti-collision safety are solved, and the global stability of the inspection trajectory of the controlled object in the wind field environment is maintained.

[0034] Embodiment 3: The embodiment combines Figs. 1-2 , the unmanned aerial vehicle slope inspection autonomous navigation path planning system of the fusion neural network is described, likeFig. 1 As shown, the first spatial physical quantity perception module obtains physical space controlled object data, and outputs a continuous optical pixel offset matrix and a three-axis disturbance state vector downstream respectively; wherein the continuous optical pixel offset matrix enters the second disturbance parameter mapping module, processes features using a preset dimension reduction mapping network, sequentially generates an implicit space tensor feature, i.e., is obtained by converting a preset dimension reduction mapping network, extracts an error variance parameter, i.e., obtains a dispersion along a time dimension, and finally outputs a dynamic feedback damping gain factor, i.e., is generated according to the error variance parameter; at the same time, the three-axis disturbance state vector enters the third trajectory decision control module, performs instruction decision according to the disturbance parameter, obtains a motion energy level distribution representative value, i.e., extracts a frequency energy integral value, and then obtains an adaptive weight, i.e., divides a preset limit bearing energy level value, this adaptive weight is used to calculate a reset repulsive field step, i.e., the algorithm basic step is multiplied by the adaptive weight, and a reset trajectory update frequency, i.e., the algorithm basic update frequency is multiplied by the adaptive weight, in addition, the dynamic feedback damping gain factor generated in the foregoing is introduced to obtain a re-calculation generation value, i.e., multiplied by the current generation value of the local path planning algorithm; finally, the above-mentioned branch parameters are converged to generate autonomous navigation flight instructions, i.e., are generated according to the update frequency, the repulsive field step and the re-calculation generation value, and finally through the circular element node of the arrow pointing to the execution action and the feedback.

[0035] As shown in Fig. 2 The state monitoring results generated by the continuous operation of the system are objectively presented in a two-dimensional distribution system. The horizontal axis represents time and is calibrated in seconds, with a value range of 0 to 60. The vertical axis represents displacement and is calibrated in meters, with a value range of -3.0 to 3.0. The legend clearly distinguishes between two sets of test data sequences, including the first control group and the test group. In this coordinate system, the displacement curve of the first control group presents a large amplitude oscillation pattern that alternately crosses the positive and negative ranges of the vertical axis as it extends along the time axis. The displacement curve of the test group closely follows the zero point region of the horizontal axis, showing a reduced fluctuation amplitude and a convergent state that tends to be flat overall.

[0036] In the embodiment 4, the controlled object is in a slope inspection wind field environment containing nonlinear motion distortion, the dimension reduction mapping network in the second disturbance parameter mapping module contains a spatial convolution layer and a long short-term memory network structure, the second disturbance parameter mapping module inputs the acquired continuous optical pixel offset matrix into the spatial convolution layer, the spatial convolution layer is composed of 3 layers of depth, each layer is configured with 16 two-dimensional convolution operators with a size of 3 by 3 and a step of 1, and after each layer of convolution, a maximum pooling layer with a step of 2 is connected, so as to compress the original image into a spatial dimension feature map with a dimension of 16 by 16; a long short-term memory network structure containing 64 hidden units is accessed, the sequence correlation is extracted through time step iteration, the output spatial dimension feature map is calculated, and the spatial dimension feature maps of a preset number of sampling periods are spliced along the time axis, the specific sampling strategy is as follows: a first-in first-out circular buffer with a depth of 30 sampling periods is set, when each frame of feature map is enqueued, the oldest 1 frame of data in the buffer is discarded, the overlap rate of adjacent two sliding windows in the buffer is kept at 50%, and sufficient timing smoothness is ensured within a collection period of 20 ms to 50 ms, so as to construct a sliding time window tensor, the second disturbance parameter mapping module inputs the sliding time window tensor into the long short-term memory network structure, generates a timing feature tensor of the current time period, performs element subtraction on adjacent 2 feature vectors in the timing feature tensor and takes an absolute value, and adds the obtained absolute value to a square matrix with a dimension of 25 by 25, so that the main diagonal elements of the square matrix record the feature fluctuation of adjacent time steps, 25 accumulated values on the main diagonal line are extracted, the arithmetic mean value thereof is calculated and defined as an error variance parameter, so as to reflect the time lag uncertainty of the image features in the sliding window, and the discrete distribution matrix of the timing feature tensor is calculated, and the main diagonal element values of the discrete distribution matrix are extracted as error variance parameters.

[0037] For the limit bearing energy level value called by the third trajectory decision control module, based on the hardware-in-the-loop platform, the off-line calibration procedure is executed, the controlled object carrying the standard inspection load is fixed on the wind tunnel test platform, the wind tunnel generator outputs a shear wind speed sequence with increasing flow rate, the on-board sensing component synchronously collects the three-axis disturbance state vector of the controlled object, the third trajectory decision control module extracts the high-frequency vibration component of the three-axis disturbance state vector in the three-dimensional space coordinate system and performs the fast Fourier transform, and the integral area value of the frequency domain energy spectrum in the frequency band higher than the preset vibration frequency threshold is calculated and output as the real-time frequency domain energy integral value. The calibration steps of the limit bearing energy level value are as follows: in the preparation stage at the beginning of the power-on of the controlled object, the wind tunnel or the simulated environment is controlled to increase the shear wind speed by a step increment of 0.5 m / s, the body yaw angle fluctuation is monitored in real time, when the single deviation amplitude of the body yaw angle first reaches 12.0 deg or the physical saturation threshold of the driving motor PWM duty cycle reaches 85%, the real-time frequency domain energy integral value calculated by the current processor is intercepted as the calibration reference, 8 groups of tests are repeatedly executed and the arithmetic mean value is stored in the bottom register, which is set as the limit bearing energy level value; the real-time frequency domain energy integral value is divided by the limit bearing energy level value in the register to obtain a dimensionless ratio value in the range of 0.2 to 1.0, and the dimensionless ratio value is directly assigned to the adaptive weight for subsequent linear scaling control of the flight path update frequency. During the rising process of the gust flow rate, when the body yaw angle of the controlled object breaks through the preset safety angle threshold for a single time, or the output power of the bottom driving motor reaches the rated full load state, the system locks the current time stamp, extracts the real-time frequency domain energy integral value corresponding to the time stamp and sets it as the limit bearing energy level value. Based on the tensor calculation process of the continuous optical pixel offset matrix to the error variance parameter and the limit bearing energy level value physical calibration procedure based on the wind tunnel test platform, the third trajectory decision control module generates a dynamic feedback damping gain factor and an adaptive weight according to the extracted error variance parameter and the calibrated limit bearing energy level value, respectively. The controlled object receives the reset flight path update frequency and the reset repulsive field step determined according to the adaptive weight, and controls the physical actuator to act according to the autonomous navigation instruction containing the above parameters in the slope inspection working condition.

[0038] Embodiment 5: When determining the preset vibration frequency threshold, the offline calibration procedure is started for the inherent mechanical resonance characteristics of the fuselage, the controlled object is rigidly connected to the variable-frequency vibration test bench, the test bench outputs a continuous logarithmic sweep excitation signal, the on-board first spatial physical quantity sensing module synchronously collects the three-axis vibration response sequence of the controlled object under the sweep excitation, the front-end computing module calculates the amplitude-frequency characteristic curve of the three-axis vibration response sequence, extracts the initial frequency node of the amplitude response that exceeds the steady-state baseline and shows nonlinear amplification as the lower boundary of the structure resonance, and the system control unit writes the frequency value corresponding to the lower boundary of the structure resonance into the bottom register, and locks it as the preset vibration frequency threshold. The third trajectory decision control module separates the inherent mechanical vibration of the fuselage from the external wind field disturbance based on the preset vibration frequency threshold, and outputs the frequency energy integral value corresponding to the external airflow impact energy level.

[0039] When the controlled object is in the take-off preparation working condition of starting the rotor blade, the system starts the on-site pre-calibration procedure for the environmental response reference coefficient k based on the current physical mounting state, the controlled object remains in the idle running state on the ground and locks the basic position coordinates, the second disturbance parameter mapping module collects the initial optical pixel offset matrix in the static windless environment, and generates the baseline error variance parameter using the dimension reduction mapping network. The system control unit calculates the logarithmic ratio of the preset target initial damping gain boundary value and the baseline error variance parameter to output the reference coefficient, the system control unit assigns the reference coefficient as the environmental response reference coefficient k, and the system control unit filters out the background static noise introduced by the blade idle vibration based on the assigned environmental response reference coefficient k, locks the trigger zero point of the dynamic feedback damping gain factor, and maintains the control stability of the feedforward damping parameter at the initial stage of slope inspection.

[0040] To determine the second disturbance parameter mapping module to process the continuous optical pixel offset matrix parameter, the dimension reduction mapping network offline state uses the windless environment to collect the calibration displacement sequence and the preset amplitude disturbed displacement sequence to complete the supervised learning, the spatial feature extraction layer converts the optical pixel offset matrix into an N-dimensional feature vector, the long short-term memory network structure extracts the time correlation of the feature vector within the preset length sliding time window, and the generated time sequence feature tensor is calculated by the main diagonal element of the discrete distribution matrix to obtain the error variance parameter V. In order to make the dynamic feedback damping gain factor D output meet the preset control accuracy requirement, the static noise reference is obtained through the on-site pre-calibration procedure, the initial optical pixel offset matrix is collected under the blade idle state before the controlled object takes off, the corresponding baseline error variance parameter is calculated, the environmental response reference coefficient k is determined according to the logarithmic ratio of the target initial damping gain boundary value and the parameter, the trigger zero point of the dynamic feedback damping gain factor is locked, and the measurement deviation introduced by the inherent vibration of the power system is eliminated.

[0041] In a real slope inspection flight control, the third trajectory decision control module extracts the three-axis disturbance state vector in real time and dynamically calculates the adaptive weight. When the real-time value of the adaptive weight is greater than the preset safety threshold parameter, the third trajectory decision control module triggers the hard constraint intervention of the frequency suppression logic. The third trajectory decision control module stops executing the normal proportional attenuation calculation link, resets the flight path update frequency to the product of the base update frequency and 0.5, and simultaneously resets the repulsive force field step to be enlarged according to the preset multiple. The safety threshold parameter obtained based on the physical saturation feature combined with the condition-triggered frequency reduction multiplication operation enables the controlled object to enter a low-frequency conservative detection mode when it is subjected to extreme airflow impact, blocks the cascading conduction path of high-frequency noise from the environment to the attitude control link, and maintains the stability of the bottom layer power system of the controlled object in a harsh environment.

[0042] In a real slope inspection flight control, the third trajectory decision control module extracts the three-axis disturbance state vector in real time and dynamically calculates the adaptive weight. When the real-time value of the adaptive weight is greater than the preset safety threshold parameter, the third trajectory decision control module triggers the hard constraint intervention of the frequency suppression logic. The third trajectory decision control module stops executing the normal proportional attenuation calculation link, resets the flight path update frequency to the product of the base update frequency and 0.5, and simultaneously resets the repulsive force field step to be enlarged according to the preset multiple. The safety threshold parameter obtained based on the physical saturation feature combined with the condition-triggered frequency reduction multiplication operation enables the controlled object to enter a low-frequency conservative detection mode when it is subjected to extreme airflow impact, blocks the cascading conduction path of high-frequency noise from the environment to the attitude control link, and maintains the stability of the bottom layer power system of the controlled object in a harsh environment.

[0043] When determining the limit bearing energy level value and the safety threshold parameter of the third trajectory decision control module, hardware-in-the-loop simulation combined with physical wind tunnel test is adopted to complete quantitative calibration. In the state of carrying a standard inspection load on the controlled object, the wind tunnel generator outputs an increasing shear wind, the high-frequency vibration component of the three-axis disturbance state vector is extracted in real time and fast Fourier transform is performed. When the controlled object body yaw angle single offset exceeds the preset safety angle threshold or any rotor drive motor output power reaches the rated full load state, the current timestamp corresponding frequency domain energy integral value is intercepted and solidified as the limit bearing energy level value, which is multiplied by 0.8 engineering redundancy coefficient to determine the safety threshold parameter, which is used as the trigger frequency suppression logic hard constraint boundary. When the real-time calculation adaptive weight exceeds its boundary, the system is forced to lower the reset flight path update frequency and shrink the reset repulsive force field step, so that the controlled object maintains physical dynamic balance and inspection flight path smoothness by adjusting the path tracking sensitivity under extreme airflow conditions.

[0044] The embodiments of the present application are described above in combination with the drawings, and the embodiments and features in the embodiments of the present application can be combined with each other without conflict, and the present application is not limited to the above specific embodiments, which are only illustrative and not restrictive. Those skilled in the art can make many forms under the inspiration of the present application without departing from the purpose of the present application and the scope of the present application, which are all within the protection of the present application.

Claims

1. A drone-based autonomous navigation path planning system for slope inspection integrating neural networks, characterized in that, include: First spatial physical quantity sensing module; The second disturbance parameter mapping module is connected to the first spatial physical quantity sensing module; And the third trajectory decision control module, which is connected to the second disturbance parameter mapping module and the first spatial physical quantity sensing module; The first spatial physical quantity sensing module is configured to acquire the continuous optical pixel offset matrix and the three-axis perturbation state vector of the controlled object in the physical space. The second perturbation parameter mapping module is configured to receive the continuous optical pixel offset matrix, convert the continuous optical pixel offset matrix into latent space tensor features using a preset dimension reduction mapping network, calculate the dispersion of the latent space tensor features along the time dimension to output the error variance parameter, and generate a dynamic feedback damping gain factor based on the error variance parameter. The third trajectory decision control module is configured to receive the dynamic feedback damping gain factor and the three-axis disturbance state vector, extract the frequency domain energy integral value of the three-axis disturbance state vector as the representative value of the motion energy level distribution, and divide the representative value of the motion energy level distribution by the preset limit tolerance energy level value to output the adaptive weight. The preset local path planning algorithm's base update frequency is multiplied by the adaptive weights to obtain the reset track update frequency, and the local path planning algorithm's base step size is multiplied by the adaptive weights to obtain the reset repulsion field step size; the dynamic feedback damping gain factor is multiplied by the current generation value of the local path planning algorithm to output the recalculated generation value, and the autonomous navigation flight command of the controlled object is generated based on the reset track update frequency, the reset repulsion field step size, and the recalculated generation value.

2. The UAV slope inspection autonomous navigation path planning system integrating neural networks according to claim 1, characterized in that, The second perturbation parameter mapping module includes a time series construction submodule and a variance estimation submodule. The time series construction submodule is configured to receive continuous optical pixel offset matrices and concatenate the continuous optical pixel offset matrices into a fixed-length sliding time window tensor. The variance estimation submodule is configured to receive a sliding time window tensor, input the sliding time window tensor into a long short-term memory network structure to generate a temporal feature tensor for the current time period, calculate the discrete distribution matrix of the temporal feature tensor, and extract the values ​​of the main diagonal elements of the discrete distribution matrix as the error variance parameter.

3. The UAV slope inspection autonomous navigation path planning system integrating neural networks according to claim 1, characterized in that, The second perturbation parameter mapping module generates a dynamic feedback damping gain factor based on the error variance parameter, including: processing the error variance parameter based on a preset exponential decay mathematical formula; the exponential decay mathematical formula is: Where D is the dynamic feedback damping gain factor, V is the error variance parameter, and k is the preset environmental response baseline coefficient.

4. The UAV slope inspection autonomous navigation path planning system integrating neural networks according to claim 1, characterized in that, The steps of the third trajectory decision control module to extract the frequency domain energy integral value of the three-axis disturbance state vector as the representative value of the motion energy level distribution include: step 401, extracting the high-frequency vibration component of the three-axis disturbance state vector in the three-dimensional spatial coordinate system; step 402, performing a fast Fourier transform on the high-frequency vibration component to generate a frequency domain energy spectrum; step 403, calculating the integral area value of the frequency domain energy spectrum in the frequency band higher than the preset vibration frequency threshold, and setting the integral area value as the frequency domain energy integral value.

5. The UAV slope inspection autonomous navigation path planning system integrating neural networks according to claim 1, characterized in that, In response to the adaptive weight value being greater than the preset safety threshold parameter, the third trajectory decision control module executes frequency suppression logic; the frequency suppression logic includes: setting the reset track update frequency to the product of the base update frequency and 0.5, and setting the reset repulsion field step size to the product of the base step size and 0.

33.

6. The UAV slope inspection autonomous navigation path planning system integrating neural networks according to claim 1, characterized in that, The third trajectory decision control module is internally configured with a disturbance evolution tracking submodule; the disturbance evolution tracking submodule records the time series set of adaptive weights over multiple consecutive sampling periods and calculates the rate of change of the time derivative of the time series set within a fixed time sliding window; In response to the time derivative rate of change being greater than 0 for three consecutive sampling periods, the third trajectory decision control module outputs a hovering blocking signal and generates a hovering waiting command to keep the controlled object in place with its current three-dimensional spatial coordinates, in lieu of the autonomous navigation flight command.

7. The UAV slope inspection autonomous navigation path planning system integrating neural networks according to claim 1, characterized in that, The first spatial physical quantity sensing module includes an optical feature capture component and a spatial inertial measurement component; The optical feature capture component and the space inertial measurement component are connected via a hardware crystal oscillator clock synchronization trigger module; When the optical feature capture component receives the synchronization trigger signal, it locks the exposure to generate a digital image frame containing a continuous optical pixel offset matrix; when the space inertial measurement component receives the synchronization trigger signal, it synchronously records the triaxial acceleration and triaxial angular velocity values, and concatenates the triaxial acceleration and triaxial angular velocity values ​​into a triaxial disturbance state vector.

8. The UAV slope inspection autonomous navigation path planning system integrating neural networks according to claim 1, characterized in that, The local path planning algorithm is an artificial potential field algorithm, and the current generation value is calculated by the gravitational potential field function and the repulsive potential field function. When the third trajectory decision control module outputs the recalculated generation value, it multiplies the dynamic feedback damping gain factor only with the output value of the gravitational potential field function, and keeps the output weight value of the repulsive potential field function constant.

9. The UAV slope inspection autonomous navigation path planning system fused with neural networks according to claim 8, characterized in that, The third trajectory decision control module also includes a spatial limit boundary protection submodule; the spatial limit boundary protection submodule obtains the three-dimensional straight-line distance value between the controlled object and the physical surface of the slope; in response to the three-dimensional straight-line distance value being less than 2m, the spatial limit boundary protection submodule outputs a coverage command, and sets the output weight value of the repulsive potential field function to 5 times the original value according to the coverage command.

10. The UAV slope inspection autonomous navigation path planning system integrating neural networks according to claim 3, characterized in that, The environmental response baseline coefficient is set to 0.05 to 0.15; the digital image frame acquisition period is set to 20ms to 50ms; and the upper limit of the output frequency of autonomous navigation flight commands is set to 10Hz.

Images