Unmanned aerial vehicle-based artificial intelligence inspection system and method

Abstract

The invention discloses an artificial intelligence inspection system and method based on an unmanned aerial vehicle, and particularly relates to the technical field of artificial intelligence of the unmanned aerial vehicle. The method comprises the following steps: acquiring magnetic field interference parameters and terrain complex parameters in pipeline inspection paths in real time, extracting magnetic field intensity fluctuation characteristics and obstacle density abnormal change characteristics, calculating navigation and positioning system accuracy weight assignment of each path in combination with machine learning methods such as polynomial regression, and generating a comprehensive inspection accuracy index; the comprehensive accuracy index is compared with a gradient standard threshold value, so that the accuracy division of the inspection path is realized: for incomplete accuracy inspection, the accuracy abnormal degree of a future navigation system is predicted, and if necessary, the flight height is adjusted and an obstacle dense area is avoided; the problems of false detection and missing detection caused by magnetic field interference and complex terrain in the prior art are effectively solved, the efficiency, accuracy and reliability of the unmanned aerial vehicle inspection task are remarkably improved, and major disasters caused by the fact that leakage is not detected in time are avoided.

Description

Technical Field

[0001] The present invention relates to the technical field of UAV artificial intelligence, and in particular to an UAV artificial intelligence-based inspection system and method. Background Art

[0002] UAV artificial intelligence inspection is an advanced method that combines artificial intelligence (AI) technology with UAVs to automatically inspect specific areas, equipment, or facilities. By carrying high-resolution cameras, infrared sensors, or other detection devices, UAVs can fly autonomously in complex environments and collect data. Artificial intelligence technology plays a key role in data analysis, processing real-time collected information such as image recognition, pattern detection, and anomaly analysis to quickly identify faults or potential hazards. This technology is widely used in infrastructure monitoring (such as power grids, bridges, and railways), agriculture, energy, and security fields. For example, in the oil and gas industry, UAV artificial intelligence inspection can be used to detect pipeline leaks. The AI identifies temperature anomalies through thermal imaging technology and quickly locates the leak points. These technologies significantly improve inspection efficiency, reduce labor costs, and the need for dangerous operations.

[0003] The existing technologies have the following deficiencies: The UAV navigation system may be mispositioned due to magnetic field interference and complex terrain, misrecording the inspection results of healthy pipelines as leak points while missing actual leaking pipe sections. If this error is not detected in time during inspection data analysis, the real leak points may continue to deteriorate until large-scale disasters occur. In addition, if the navigation system has frequent errors, the inspection task may need to be interrupted or replanned frequently, resulting in a significant reduction in task efficiency. Summary of the Invention

[0004] The purpose of the present invention is to provide an UAV artificial intelligence-based inspection system and method to solve the deficiencies in the background art.

[0005] To achieve the above purpose, the present invention provides the following technical solution: An UAV artificial intelligence-based inspection method, comprising the following steps: S1: Real-time obtain magnetic field interference parameters and terrain complexity parameters in different pipeline inspection paths through UAV on-board sensors, where the magnetic field interference parameters include environmental magnetic field intensity data, and the terrain complexity parameters include obstacle density data; S2: Preprocess the obtained environmental magnetic field intensity data and obstacle density data, and extract the magnetic field intensity fluctuation characteristics in the environmental magnetic field intensity data and the obstacle density abnormal change characteristics in the obstacle density data; S3: Determine the accuracy weight assignment of the UAV navigation and positioning system in each pipeline inspection path according to the extracted magnetic field intensity fluctuation characteristics and abnormal change characteristics of obstacle density, and perform weighted average summation calculation to obtain the comprehensive inspection accuracy index; S4: Compare the comprehensive inspection accuracy index in the pipeline inspection path with the gradient standard threshold, and classify the accuracy of the UAV pipeline inspection path according to the comparison result. The classification results include accurate inspection, incomplete accurate inspection, and inaccurate inspection; S5: For accurate inspection, continue the inspection according to the established path without adjusting the flight strategy; for inaccurate inspection, immediately interrupt the UAV inspection task, recalculate the optimal inspection path, and give priority to covering the abnormal pipeline section; S6: For incomplete accurate inspection, predict the degree of accuracy abnormality of the UAV navigation and positioning system within the subsequent fixed time period. If the degree of accuracy abnormality is high, re-plan the UAV inspection path, increase the flight altitude of the UAV, and avoid dense obstacle areas.

[0006] Preferably, in S2, after analyzing the magnetic field intensity fluctuation characteristics in the extracted environmental magnetic field intensity data, a magnetic field intensity continuous fluctuation index is generated. The method for obtaining the magnetic field intensity continuous fluctuation index is as follows: Record the output of the magnetometer sensor to obtain the time series data B(s) of the magnetic field intensity, where s is the time window. Divide the time series B(s) into short-time windows w(n), and the length of each window is N samples; apply the Fourier transform to the signal in each time window to calculate the frequency components: ; where, is the complex spectrum value at time s and frequency f, B(n) is the signal sample within the window, w(n) is the window function value, is the frequency weight in complex form; take the modulus value of the complex spectrum to obtain the amplitude spectrum , and the expression is: ; for each time window s, calculate the spectral energy density , and the expression is: ; where, and are the lowest and highest frequencies of interest respectively, and E(s) represents the spectral energy density of time window s; normalize E(s): ; where, and are the minimum and maximum values of the spectral energy in the entire sequence respectively, is the normalized spectral energy density; perform smoothing processing on the normalized spectral energy density to generate the magnetic field intensity continuous fluctuation index HSK, and the expression is: ; where Z is the smoothing factor, and HSK is the continuous fluctuation index of the magnetic field strength, is the continuous fluctuation index of the magnetic field strength at the moment.

[0007] Preferably, in S2, after analyzing the abnormal change characteristics of the obstacle density in the extracted obstacle density data, an abnormal change index of the obstacle density is generated. The method for obtaining the abnormal change index of the obstacle density is as follows: Collect obstacle point cloud data through a lidar to form a point set ; Each point includes its position coordinates , and use the Euclidean distance to calculate the distance between obstacle points and : ; For each point , find its k-nearest neighbor point set , where k is the number of nearest neighbors specified by the user; the k-nearest neighbor distance is the distance from point to its k-th nearest neighbor point: ; Define to its neighborhood point the reachable distance , and the expression is: ; The local reachable density of point is the reciprocal mean of the reachable distances from all points in its neighborhood to , and the expression is: ; In the formula, is the number of neighborhood points; the local outlier factor of point is the ratio of the average local reachable density of the points in its neighborhood to its own local reachable density: ; Normalize the local outlier factor values of all points to generate an abnormal change index of the obstacle density, and the expression is: ; In the formula, GSH is the abnormal change index of the obstacle density, and are respectively the minimum and maximum local outlier factor values in the dataset.

[0008] Preferably, in S3, according to the extracted magnetic field strength fluctuation characteristics and abnormal change characteristics of the obstacle density, determine the accuracy weight assignment of the UAV navigation and positioning system in each pipeline inspection path, and perform weighted average summation calculation to obtain a comprehensive inspection accuracy index, specifically: Convert the continuous fluctuation index of magnetic field intensity and the abnormal change index of obstacle density into a comprehensive feature vector, and use the comprehensive feature vector as the input of the machine learning model. The machine learning model takes the prediction of the accuracy weight assignment label of the UAV navigation and positioning system in each pipeline inspection path as the prediction target, and takes minimizing the sum of the prediction errors of the accuracy weight assignment labels of the UAV navigation and positioning system in all pipeline inspection paths as the training target. Train the machine learning model until the sum of the prediction errors reaches convergence and then stop the model training. Determine the accuracy weight assignment of the UAV navigation and positioning system in each pipeline inspection path according to the model output result. Among them, the machine learning model is a polynomial regression model, and calculate the weighted average sum of the accuracy weight assignments of the UAV navigation and positioning system in each pipeline inspection path to obtain the comprehensive inspection accuracy index.

[0009] Preferably, in S4, compare the comprehensive inspection accuracy index in the pipeline inspection path with the gradient standard threshold, and divide the accuracy of the UAV pipeline inspection path according to the comparison result. The division results include accurate inspection, incomplete accurate inspection, and inaccurate inspection. Compare the obtained comprehensive inspection accuracy index in the pipeline inspection path with the gradient standard threshold. The gradient standard threshold includes a first standard threshold and a second standard threshold, and the first standard threshold is less than the second standard threshold. Compare the comprehensive inspection accuracy index in the pipeline inspection path with the first standard threshold and the second standard threshold respectively. If the comprehensive inspection accuracy index in the pipeline inspection path is greater than the second standard threshold, it indicates that the accuracy of the UAV navigation and positioning system in the pipeline inspection path is high. At this time, generate a high-accuracy inspection signal and divide the accuracy of the UAV pipeline inspection path into accurate inspection. If the comprehensive inspection accuracy index in the pipeline inspection path is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, it indicates that the accuracy of the UAV navigation and positioning system in the pipeline inspection path is medium. At this time, generate a medium-accuracy inspection signal and divide the accuracy of the UAV pipeline inspection path into incomplete accurate inspection. If the comprehensive inspection accuracy index in the pipeline inspection path is less than the first standard threshold, it indicates that the accuracy of the UAV navigation and positioning system in the pipeline inspection path is low. At this time, generate a low-accuracy inspection signal and divide the accuracy of the UAV pipeline inspection path into inaccurate inspection.

[0010] Preferably, in S6, for incomplete accurate inspection, predict the degree of abnormality of the accuracy of the UAV navigation and positioning system in the subsequent fixed time period. If the degree of abnormality of the accuracy is high, re-plan the inspection path of the UAV, increase the flight altitude of the UAV, and avoid areas with dense obstacles. Specifically: For the incomplete accuracy inspection, that is, the comprehensive inspection accuracy index in the generated pipeline inspection path is greater than or equal to the first standard threshold and less than or equal to the second standard threshold. In the next T time period, the abnormal degree of the accuracy of the navigation and positioning system is represented by the comprehensive prediction index as follows: ; where: is the predicted value of the magnetic field intensity fluctuation index within the future time t, GSH(t) is the predicted value of the abnormal change index of the obstacle density within the future time t, CI(t) is the comprehensive inspection accuracy index at the current time t, and α, β, γ are weight coefficients, representing the influence ratios of magnetic field fluctuation, obstacle density, and current comprehensive accuracy on the predicted value, satisfying α + β + γ = 1; use the time series prediction model to perform short-term prediction on HSK(t) and GSH(t): , ; Define the threshold Pthreshold for predicting the abnormal degree of accuracy: If , it indicates a high abnormal degree of accuracy and the path needs to be re-planned; if , continue to inspect according to the existing path.

[0011] Preferably, the path re-planning goal is to avoid the obstacle-dense area with the shortest path and optimize the navigation accuracy. The objective function of the path planning problem , and the expression is: ; In the formula, is the length of path segment i, is the predicted abnormal degree of path segment i, k is the penalty coefficient, used to increase the path weight in the high-abnormal degree area, avoid high-risk areas, and W is the total number of path segments; The flight altitude adjustment rule is: Define the adjusted flight altitude , and the expression is: ; In the formula, is the current flight altitude, and ΔH is the altitude increment coefficient; Mark the high obstacle density area and set it as an impassable area during path planning. The area density weight formula: ; In the formula, is the area weight, GSH(j) is the abnormal change index of the obstacle density within the area. During path re-planning, exclude the areas where the area weight is greater than the area weight threshold preset according to historical data.

[0012] The present invention also provides an unmanned aerial vehicle-based artificial intelligence inspection system, including a data acquisition module, a feature extraction module, a navigation and positioning accuracy evaluation module, an accuracy division module, a path planning and task control module, and a dynamic prediction and strategy optimization module; Data acquisition module: It obtains the magnetic field interference parameters and terrain complexity parameters in different pipeline inspection paths in real time through the onboard sensors of the unmanned aerial vehicle (UAV). The magnetic field interference parameters include ambient magnetic field intensity data, and the terrain complexity parameters include obstacle density data; Feature extraction module: It preprocesses the obtained ambient magnetic field intensity data and obstacle density data, and extracts the magnetic field intensity fluctuation features in the ambient magnetic field intensity data and the abnormal change features of the obstacle density in the obstacle density data; Navigation and positioning accuracy evaluation module: According to the extracted magnetic field intensity fluctuation features and abnormal change features of the obstacle density, it determines the accuracy weight assignment of the UAV navigation and positioning system in each pipeline inspection path, and performs weighted average summation calculation to obtain the comprehensive inspection accuracy index; Accuracy classification module: It compares the comprehensive inspection accuracy index in the pipeline inspection path with the gradient standard threshold, and classifies the accuracy of the UAV pipeline inspection path according to the comparison result. The classification results include accurate inspection, incomplete accurate inspection, and inaccurate inspection; Path planning and task control module: For accurate inspection, it continues the inspection according to the established path without adjusting the flight strategy; for inaccurate inspection, it immediately interrupts the inspection task of the UAV, recalculates the optimal inspection path, and preferentially covers the abnormal pipeline section; Dynamic prediction and strategy optimization module: For incomplete accurate inspection, it predicts the abnormal degree of the accuracy of the UAV navigation and positioning system within a subsequent fixed time period. If the abnormal degree of accuracy is high, it replans the inspection path of the UAV, increases the flight altitude of the UAV, and avoids the dense obstacle area.

[0013] In the above technical solution, the technical effects and advantages provided by the present invention are as follows: 1. Through real-time data acquisition, feature extraction and comprehensive analysis based on the onboard sensors of the UAV, the present invention innovatively combines the magnetic field intensity fluctuation index and the abnormal change index of the obstacle density, proposes a weight evaluation method for the accuracy of the navigation and positioning system, and accurately calculates the comprehensive inspection accuracy index. By comparing with the gradient standard threshold, it can efficiently classify the accuracy level of the inspection path, and adopt corresponding inspection strategies for different situations to achieve dynamic adjustment of the inspection task. Especially in the scenario of incomplete accurate inspection, by using the time series prediction model to evaluate the abnormal degree of the future navigation system in advance, combined with path replanning and flight altitude optimization, it effectively avoids the waste of inspection resources and potential risks caused by misdetection or missed detection.

[0014] 2. The present invention significantly improves the reliability, accuracy, and adaptability of the UAV pipeline inspection task. In a complex inspection environment, it can quickly identify and avoid high-risk areas, prioritize covering important pipeline segments, and ensure the comprehensiveness and real-time nature of the inspection. Through path replanning and flight altitude adjustment methods, the risk of navigation errors caused by dense obstacles or magnetic field interference is minimized, while the efficiency of the inspection task and the quality of data collection are improved. The present invention effectively solves the problems of disaster hidden dangers and low efficiency caused by positioning errors in the prior art, and provides strong technical support for UAV intelligent inspection in complex scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] To more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings.

[0016] Figure 1 It is a flowchart of the method of the present invention.

[0017] Figure 2 It is a system module diagram of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present invention.

[0019] Example 1. Please refer to Figure 1 and Figure 2 As shown, a UAV artificial intelligence-based inspection method in this embodiment includes the following steps: S1: Real-time obtain the magnetic field interference parameters and terrain complexity parameters in different pipeline inspection paths through the UAV on-board sensor. The magnetic field interference parameters include ambient magnetic field intensity data, and the terrain complexity parameters include obstacle density data; S2: Preprocess the obtained ambient magnetic field intensity data and obstacle density data, and extract the magnetic field intensity fluctuation characteristics in the ambient magnetic field intensity data and the abnormal change characteristics of the obstacle density in the obstacle density data; S3: Based on the extracted magnetic field intensity fluctuation characteristics and abnormal changes in obstacle density characteristics, determine the accuracy weight assignment of the UAV navigation and positioning system in each pipeline inspection path, and perform weighted average summation calculation to obtain the comprehensive inspection accuracy index; S4: Compare the comprehensive inspection accuracy index in the pipeline inspection path with the gradient standard threshold, and classify the accuracy of the UAV pipeline inspection path according to the comparison result. The classification results include accurate inspection, incomplete accurate inspection, and inaccurate inspection; S5: For accurate inspection, continue the inspection according to the established path without adjusting the flight strategy; for inaccurate inspection, immediately interrupt the UAV inspection task, recalculate the optimal inspection path, and give priority to covering the abnormal pipeline section; S6: For incomplete accurate inspection, predict the degree of accuracy abnormality of the UAV navigation and positioning system within the subsequent fixed time period. If the degree of accuracy abnormality is high, re-plan the UAV inspection path, increase the flight altitude of the UAV, and avoid areas with dense obstacles.

[0020] In S1, the magnetic field interference parameters and terrain complexity parameters in different pipeline inspection paths are obtained in real time through the UAV-borne sensors. The magnetic field interference parameters include ambient magnetic field intensity data, and the terrain complexity parameters include obstacle density data. Specifically: Use a high-sensitivity three-axis magnetometer to detect the ambient magnetic field intensity and its changes in real time. It is integrated in the UAV's navigation system to compensate for the influence of magnetic field anomalies on navigation. The magnetometer measures the ambient magnetic field intensity (unit: gauss or tesla) on the UAV flight path. Record the real-time magnetic field data and analyze whether there are significant magnetic field interference sources (such as high-voltage wires, ferromagnetic materials).

[0021] During the UAV flight, sample the ambient magnetic field intensity data at a fixed time per second and mark its spatial coordinates (GPS position). Extract the fluctuation characteristics of the magnetic field intensity data, such as instantaneous intensity changes, interference frequencies, and spatial distribution patterns. Mark the abnormal areas where the magnetic field intensity is higher than the background noise. Associate the high magnetic field intensity areas with the inspection path to provide a basis for subsequent navigation path optimization.

[0022] Scan the terrain features through a high-resolution three-dimensional lidar and construct an obstacle distribution model. Detect vegetation, water bodies, and other physical obstacles in the terrain, and supplement the area features that cannot be detected by the lidar. The lidar obtains the three-dimensional distribution data of obstacles through the pulse laser reflection time and intensity. The multi-spectral camera distinguishes terrain features, such as buildings, vegetation, etc., by analyzing spectral reflection differences.

[0023] The LiDAR continuously scans along the flight path of the drone, generating real-time point cloud data and recording the height, density and position of obstacles. The multispectral camera captures the spectral image of the terrain surface and extracts the object category (such as trees, rocks). The number and density of obstacles within a unit path are calculated, and the characteristics of the obstacle-dense area are extracted. The high-density obstacle areas in the path are marked and classified into feasibility levels (low, medium, and high).

[0024] The point cloud data and spectral images are combined to construct a three-dimensional terrain model to evaluate the terrain undulation, obstacle distribution and inspection path passability.

[0025] The magnetic field interference parameters and terrain complexity parameters are integrated into the data stream of the navigation system to form a multi-dimensional parameter set for path evaluation. In the path analysis module, the flight area is evaluated in sections and navigation accuracy weights are assigned to each section of the path. Sensor noise is removed through filtering algorithms (such as Kalman filtering) to ensure data accuracy. Machine learning models are used to classify and analyze parameter features to predict the extent to which interference areas may affect navigation.

[0026] According to the acquired parameters, the inspection route of the drone is dynamically adjusted to avoid magnetic field interference areas and areas with dense obstacles. High-interference or high-complexity areas are marked to provide data support for subsequent mission planning. The parameter data acquired in real time is stored in the cloud for inspection report generation and system model training optimization.

[0027] S2: preprocessing the acquired environmental magnetic field strength data and obstacle density data, and extracting magnetic field strength fluctuation characteristics in the environmental magnetic field strength data and abnormal obstacle density change characteristics in the obstacle density data.

[0028] Filter the raw data to remove random noise during the sampling process. Common methods include: Low-pass filter: remove high-frequency noise, such as small disturbances that change rapidly. Combine historical and current state prediction values ​​to smooth the curves of magnetic field and obstacle data. Identify and process extreme values ​​that may be caused by sensor errors, and use median filtering or statistical methods (such as 3 times the standard deviation method) to eliminate outliers.

[0029] Calibrate the sensor readings to ensure accurate measurement results: Magnetic field data calibration: Compensate for zero drift or tilt error of the magnetometer, and calibrate the background magnetic field value based on the initial static position of the drone. Correct the time deviation and spatial alignment of the lidar and multispectral data through multiple measurements and comparisons. Normalize the ambient magnetic field intensity data and obstacle density data. Normalization helps eliminate dimensional differences and facilitates subsequent algorithm analysis.

[0030] Extract the characteristics of magnetic field intensity fluctuations, including: Fluctuation analysis: Calculate the first-order derivative: Analyze the rate of change of magnetic field intensity over time or space, and capture the violent fluctuation points: ΔB=B(t+1)−Bt; where Bt is the magnetic field intensity value at the tth moment. A larger ΔB indicates violent fluctuations.

[0031] Fourier Transform (FFT): Converts magnetic field strength data from the time domain to the frequency domain, identifies the dominant frequency characteristics and amplitude characteristics of the fluctuations, and highlights possible interference source patterns.

[0032] Fluctuation amplitude statistics: Calculate the mean, variance, extreme value range and other indicators of magnetic field strength to quantify the fluctuation intensity.

[0033] The features of abnormal changes in obstacle density are extracted as follows: Local density change analysis: Use sliding window technology to calculate the obstacle density in different spatial segments and identify abnormal changes in density: Dwindow = number of obstacles / window area; perform change rate analysis on the density value of each sliding window.

[0034] Cluster analysis: Use K-means or DBSCAN algorithm to cluster the obstacle distribution and locate the dense obstacle areas; abnormal changes are usually manifested as the density value of a certain area deviating significantly from the mean of the adjacent areas.

[0035] Terrain complexity index: A statistical index that comprehensively calculates the height, density, and distribution of obstacles, such as the standard deviation of the obstacle spacing distribution, to assess terrain complexity.

[0036] Set the thresholds for magnetic field fluctuation characteristics and obstacle density change characteristics, and mark the areas exceeding the thresholds as "abnormal" points. Compare the currently extracted features with historical inspection data to identify possible abnormal change trends.

[0037] Superimpose magnetic field fluctuations and obstacle density features on the inspection route map, and use colors to represent the fluctuation amplitude and the intensity of abnormal density changes. Generate a three-dimensional terrain map from LiDAR data to intuitively display the distribution characteristics and dense areas of obstacles. Output magnetic field fluctuation features (such as main frequency, fluctuation amplitude) and obstacle density change features (such as abnormal density distribution) for subsequent analysis. Mark high magnetic field fluctuation areas and obstacle-dense areas as high-risk areas to provide support for inspection route optimization and task planning.

[0038] The magnetic field intensity continuous fluctuation index is generated by analyzing the magnetic field intensity fluctuation characteristics in the extracted environmental magnetic field intensity data. The method for obtaining the magnetic field intensity continuous fluctuation index is as follows: Record the output of the magnetometer sensor to obtain the time series data B(s) of the magnetic field strength, where s is the time window, and the unit is usually Tesla (T) or Gauss (G). Ensure that the sampling frequency fs is high enough to satisfy the sampling theorem (usually at least 2 times the main frequency of the signal). For example, if the main frequency of the magnetic field fluctuation is 10 Hz, then fs ≥ 20 Hz. Divide the time series B(s) into multiple short-time windows w(n), and the length of each window is N samples; the window can use a window function (such as Hanning window, Hamming window) to reduce the spectral leakage phenomenon, and the window function is: Apply the Fourier transform to the signal within each time window to calculate the frequency components: where, is the complex spectral value at time s and frequency f, B(n) is the signal sample within the window, w(n) is the window function value, is the frequency weight in complex form. Take the modulus value of the complex spectrum to obtain the amplitude spectrum , and the expression is: For each time window s, calculate the spectral energy density , and the expression is: where, and are the lowest and highest frequencies of interest respectively (for example, for low-frequency fluctuations, fmin = 0.1 Hz and fmax = 10 Hz can be set). E(s) represents the spectral energy density of the time window s.

[0039] To make the fluctuation index comparable to the magnetic field strength changes in different time windows or paths, normalize E(s): where, and are the minimum and maximum values of the spectral energy in the entire sequence respectively, is the normalized spectral energy density. Smooth the normalized spectral energy density to eliminate short-time fluctuation interference and generate the continuous magnetic field strength fluctuation index HSK, and the expression is: where Z is the smoothing factor, and its value range is 0 < Z < 1 (usually 0.8 to 0.95). HSK is the continuous magnetic field strength fluctuation index, is the continuous magnetic field strength fluctuation index at moment.

[0040] When the continuous fluctuation index of the magnetic field intensity is large, it indicates that the magnetic field environment on the inspection path fluctuates violently, and there may be strong magnetic field interference sources (such as high-voltage transmission lines, substations or large industrial equipment). This will have a significant impact on the navigation and positioning systems of the UAV. For example, the data of the magnetometer may be interfered, resulting in deviation of the navigation direction or drift of the flight trajectory. In addition, severe magnetic field interference may affect the coordinated operation of GPS and IMU (Inertial Measurement Unit), leading to a decrease in positioning accuracy or even temporary failure. In this case, the navigation and positioning accuracy of the inspection path is significantly reduced, and backup navigation strategies (such as RTK-GPS or visual positioning) must be adopted to ensure the reliability of the inspection task.

[0041] When the continuous fluctuation index of the magnetic field intensity is small, it indicates that the magnetic field environment on the inspection path is relatively stable and there are no significant external interference sources. This usually means that the working conditions of the UAV navigation and positioning systems are relatively ideal, the magnetometer can normally sense the earth's magnetic field, and the deviation of the navigation direction and flight trajectory is small. In addition, the coordinated performance of navigation components such as GPS and IMU can also be fully exerted to ensure high-precision path planning and positioning. In this case, the accuracy of the inspection task is relatively high, and the data acquisition results are more reliable, without the need for additional adjustment of the flight path or reliance on redundant navigation means.

[0042] After analyzing the abnormal change characteristics of the obstacle density in the extracted obstacle density data, an abnormal change index of the obstacle density is generated. The method for obtaining the abnormal change index of the obstacle density is as follows: Collect obstacle point cloud data through lidar to form a point set ; Each point contains its position coordinates , and use the Euclidean distance to calculate the distance between obstacle points and : ; The distance can reflect the spatial density characteristics between points.

[0043] For each point , find its set of k-nearest neighbor points , where k is the number of nearest neighbors specified by the user (usually taking 5 to 15, adjusted according to the data scale).

[0044] The k-nearest neighbor distance is the distance from point to its k-th nearest neighbor point: ; Define to its neighborhood point the reachable distance , and the expression is: ; Point 's local reachable density is the reciprocal mean of the reachable distances from all points in its neighborhood to and is expressed as: ; in the formula, is the number of points in the neighborhood of The local outlier factor of point is the ratio of the average local reachability density of the points in its neighborhood to its own local reachability density: ; when > 1, it means that the density of point

[0045] is lower than the average density of its neighborhood and may be an outlier. ; in the formula, GSH is the abnormal change index of obstacle density, and are the minimum and maximum local outlier factor values in the dataset respectively.

[0046] When the abnormal change index of obstacle density is large, it indicates that the distribution of obstacles in the inspection path has significant local anomalies. For example, there may be dense clusters of obstacles, prominent obstacle groups, or suddenly emerging complex terrains in some areas. In this case, the accuracy of the UAV navigation and positioning system may be challenged: the path planning difficulty increases, the obstacle avoidance algorithm needs to adjust the flight trajectory more frequently, which may cause navigation delays or even misjudgments. In addition, high-density obstacle areas may also block GPS signals or affect the effect of visual positioning, further reducing the positioning accuracy, resulting in a decline in inspection efficiency and a weakening of task reliability.

[0047] When the abnormal change index of obstacle density is small, it indicates that the distribution of obstacles in the inspection path is relatively uniform or the density is low, and the environmental complexity is low. In this case, the UAV navigation and positioning system can work smoothly: the path planning can be basically carried out according to the preset without frequent adjustment; the burden on the obstacle avoidance algorithm is small, reducing the risk of navigation delay. At the same time, the uniform terrain conditions are also conducive to the stable operation of the visual navigation system, with high positioning accuracy, thus ensuring the efficiency of the inspection task and the integrity of data collection.

[0048] S3: According to the extracted magnetic field intensity fluctuation characteristics and abnormal change characteristics of obstacle density, determine the accuracy weight assignment of the UAV navigation and positioning system in each pipeline inspection path, and perform weighted average summation calculation to obtain the comprehensive inspection accuracy index.

[0049] Convert the continuous fluctuation index of magnetic field intensity and the abnormal change index of obstacle density into a comprehensive feature vector, and use the comprehensive feature vector as the input of the machine learning model. The machine learning model takes the accuracy weight assignment label of the UAV navigation and positioning system in each pipeline inspection path as the prediction target, and uses minimizing the sum of prediction errors of the accuracy weight assignment labels of the UAV navigation and positioning system in all pipeline inspection paths as the training target to train the machine learning model until the sum of prediction errors reaches convergence and then stops the model training. Determine the accuracy weight assignment of the UAV navigation and positioning system in each pipeline inspection path according to the model output result. Among them, the machine learning model is a polynomial regression model, and the comprehensive inspection accuracy index is obtained by weighted average summation calculation of the accuracy weight assignment of the UAV navigation and positioning system in each pipeline inspection path.

[0050] The method for obtaining the accuracy weight assignment of the UAV navigation and positioning system in each pipeline inspection path is: obtain the corresponding function expression from the comprehensive feature vector training data of the trained machine learning model: ; In the formula, is the output function of the model, HSK is the continuous fluctuation index of magnetic field intensity, GSH is the abnormal change index of obstacle density, is the accuracy weight assignment of the UAV navigation and positioning system in each pipeline inspection path.

[0051] S4: Compare the comprehensive inspection accuracy index in the pipeline inspection path with the gradient standard threshold, and divide the accuracy of the UAV pipeline inspection path according to the comparison result. The division results include accurate inspection, incomplete accurate inspection and inaccurate inspection.

[0052] Compare the obtained comprehensive inspection accuracy index in the pipeline inspection path with the gradient standard threshold. The gradient standard threshold includes a first standard threshold and a second standard threshold, and the first standard threshold is less than the second standard threshold. Compare the comprehensive inspection accuracy index in the pipeline inspection path with the first standard threshold and the second standard threshold respectively; If the comprehensive inspection accuracy index in the pipeline inspection path is greater than the second standard threshold, it indicates that the accuracy of the UAV navigation and positioning system in the pipeline inspection path is high. At this time, generate a high-accuracy inspection signal and divide the accuracy of the UAV pipeline inspection path into accurate inspection; If the comprehensive inspection accuracy index in the pipeline inspection path is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, it indicates that the accuracy of the UAV navigation and positioning system in the pipeline inspection path is medium. At this time, generate a medium-accuracy inspection signal and divide the accuracy of the UAV pipeline inspection path into incomplete accurate inspection; If the comprehensive inspection accuracy index in the pipeline inspection path is less than the first standard threshold, it indicates that the accuracy of the UAV navigation and positioning system in the pipeline inspection path is low. At this time, a low-accuracy inspection signal is generated, and the accuracy of the UAV pipeline inspection path is classified as inaccurate inspection.

[0053] S5: For accurate inspections, continue the inspection according to the established path without adjusting the flight strategy; for inaccurate inspections, immediately interrupt the UAV inspection task, recalculate the optimal inspection path, and prioritize covering abnormal pipeline sections.

[0054] For accurate inspections, it indicates that the navigation and positioning system performs reliably and has high accuracy on the current path. The processing measures include: Continuing the inspection according to the established path: There is no need to adjust the flight strategy, and the task is normally executed along the pre-planned inspection path. Action steps: Maintain the current flight altitude, speed, and navigation strategy of the UAV. Collect data as planned (such as thermal imaging, high-definition images, or gas detection data). Avoid unnecessary path adjustments to improve task efficiency. Regularly monitor the stability of the navigation system during the inspection to ensure that the accuracy remains at a high level. If significant changes occur in the real-time data (such as a sudden increase in magnetic field fluctuations or a sudden increase in the density of obstacles), dynamically evaluate the task status.

[0055] For inaccurate inspections, it indicates that the navigation and positioning system performs poorly and has low accuracy on the current path. The processing measures include: Immediately interrupting the task: Action steps: The UAV automatically returns or flies to a safe standby point to avoid further collecting low-quality data. Mark the inspection status of the current path as "failed" and record information on possible interference sources (such as strong magnetic field areas or obstacle-dense areas). Prevent continued flight from causing false inspections or missed inspections and wasting task resources. Prioritize covering abnormal pipeline sections to ensure that the inspection tasks in high-risk areas are completed as soon as possible. Use the data on the characteristics of magnetic field intensity fluctuations and abnormal changes in obstacle density to mark severely interfered areas and exclude them from the path planning. Adopt an alternative navigation mode (such as RTK-GPS or visual navigation) to improve navigation accuracy. Through path planning algorithms (such as the A* algorithm or Dijkstra algorithm), generate an optimal inspection path that avoids the interference area. The UAV executes the inspection task again according to the new path, focusing on covering the previously interfered areas. Adjust the flight altitude or speed to adapt to the new environmental complexity.

[0056] After completing the task, compare the results of the two inspections to ensure that all key pipeline sections are effectively covered. Feed back the path failure information to the navigation model for optimizing subsequent task planning.

[0057] S6: For incomplete accuracy inspection, predict the degree of accuracy anomaly of the UAV navigation and positioning system within subsequent fixed time periods. If the degree of accuracy anomaly is high, re-plan the UAV inspection path, increase the UAV's flight altitude, and avoid areas with dense obstacles.

[0058] For incomplete accuracy inspection, that is, the comprehensive inspection accuracy index in the generated pipeline inspection path is greater than or equal to the first standard threshold and less than or equal to the second standard threshold. Within the future T time period, the degree of accuracy anomaly of the navigation and positioning system is represented by the comprehensive prediction index as follows: where: is the predicted value of the magnetic field intensity fluctuation index within future time t (based on time series prediction, such as the ARIMA model). GSH(t) is the predicted value of the abnormal change index of obstacle density within future time t (based on the obstacle distribution model). CI(t) is the comprehensive inspection accuracy index at the current time t. α, β, γ are weight coefficients, representing the influence proportions of magnetic field fluctuation, obstacle density, and current comprehensive accuracy on the predicted value, and satisfy α + β + γ = 1.

[0059] Use a time series prediction model (such as LSTM) to perform short-term prediction on HSK(t) and GSH(t): , ; Define the threshold Pthreshold for predicting the degree of accuracy anomaly: If , it indicates that the degree of accuracy anomaly is high and the path needs to be re-planned; if , continue to inspect according to the existing path.

[0060] The goal of path re-planning is to avoid areas with dense obstacles and optimize navigation accuracy with the shortest path. The objective function of the path planning problem: ; In the formula, is the length of path segment i, is the predicted anomaly degree of path segment i, k is the penalty coefficient, used to increase the path weight in high anomaly degree areas, avoid high-risk areas, and W is the total number of path segments. Path planning can use the A* algorithm or the Dijkstra algorithm to achieve optimal path calculation by dynamically adjusting the path weight.

[0061] The flight altitude adjustment rule is: Define the adjusted flight altitude , and the expression is: ; In the formula, is the current flight altitude, and ΔH is the altitude increment coefficient (which can be set according to the distribution characteristics of obstacle density).

[0062] Mark areas with high obstacle density and set them as impassable areas during path planning. The regional density weight formula: ; where, is the regional weight, and GSH(j) is the abnormal change index of obstacle density within the region. During path replanning, regions with regional weights greater than the regional weight threshold preset according to historical data are excluded.

[0063] In this embodiment, the magnetic field interference parameters (including ambient magnetic field intensity data) and terrain complexity parameters (including obstacle density data) in the pipeline inspection path are obtained in real time through the on-board sensors of the UAV, and the data is preprocessed to extract the magnetic field intensity fluctuation characteristics and the abnormal change characteristics of obstacle density. According to the extracted characteristics, the accuracy weight assignment of the UAV navigation and positioning system is calculated, and the comprehensive inspection accuracy index is obtained by weighted average summation. After comparing the comprehensive inspection accuracy index with the gradient standard threshold, the accuracy of the inspection path is divided into three categories: accurate inspection, incomplete accurate inspection, and inaccurate inspection: for accurate inspection, there is no need to adjust the flight strategy; for inaccurate inspection, the task is immediately interrupted and the path is replanned, giving priority to covering the abnormal pipeline section; for incomplete accurate inspection, by predicting the degree of accuracy abnormality of the navigation and positioning system in the subsequent time, if the abnormality degree is high, the path is replanned, increasing the flight altitude and avoiding the obstacle-dense area to ensure the reliability and accuracy of the inspection task.

[0064] Embodiment 2, the artificial intelligence-based UAV inspection system described in this embodiment includes a data acquisition module, a feature extraction module, a navigation and positioning accuracy evaluation module, an accuracy division module, a path planning and task control module, and a dynamic prediction and strategy optimization module; Data acquisition module: The magnetic field interference parameters and terrain complexity parameters in different pipeline inspection paths are obtained in real time through the on-board sensors of the UAV. The magnetic field interference parameters include ambient magnetic field intensity data, and the terrain complexity parameters include obstacle density data; Feature extraction module: The acquired ambient magnetic field intensity data and obstacle density data are preprocessed, and the magnetic field intensity fluctuation characteristics in the ambient magnetic field intensity data and the abnormal change characteristics of obstacle density in the obstacle density data are extracted; Navigation and positioning accuracy evaluation module: According to the extracted magnetic field intensity fluctuation characteristics and abnormal change characteristics of obstacle density, determine the accuracy weight assignment of the UAV navigation and positioning system in each pipeline inspection path, and calculate the comprehensive inspection accuracy index after weighted average summation; Accuracy division module: Compare the comprehensive inspection accuracy index in the pipeline inspection path with the gradient standard threshold, and divide the accuracy of the UAV pipeline inspection path according to the comparison result. The division results include accurate inspection, incomplete accurate inspection, and inaccurate inspection; Path Planning and Task Control Module: For accurate inspections, continue the inspection along the established path without adjusting the flight strategy; for inaccurate inspections, immediately interrupt the inspection task of the UAV, recalculate the optimal inspection path, and prioritize covering abnormal pipeline sections. Dynamic Prediction and Strategy Optimization Module: For inspections with incomplete accuracy, predict the degree of abnormality in the accuracy of the UAV's navigation and positioning system within a subsequent fixed time period. If the degree of accuracy abnormality is high, re-plan the inspection path of the UAV, increase the flight altitude of the UAV, and avoid areas with dense obstacles.

[0065] The above formulas are all dimensionless and take their numerical values for calculation. The formulas are obtained by collecting a large amount of data for software simulation to obtain a formula that is closest to the actual situation. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0066] It should be understood that the term "and / or" in this article is merely a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone. Here, A and B can be singular or plural. In addition, the character " / " in this article generally represents an "or" relationship between the preceding and following associated objects, but it may also represent an "and / or" relationship. The specific meaning can be understood by referring to the context.

[0067] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed in this article can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.

[0068] As described above, this is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art within the technical scope disclosed in this application can easily think of changes or substitutions, which should all be covered within the protection scope of this application.

Claims

1. An artificial intelligence-based inspection method for drones, characterized in that: It includes the following steps: S1: Real-time obtain the magnetic field interference parameters and terrain complexity parameters in different pipeline inspection paths through the on-board sensors of the UAV. The magnetic field interference parameters include environmental magnetic field intensity data, and the terrain complexity parameters include obstacle density data; S2: Preprocess the obtained environmental magnetic field intensity data and obstacle density data, and extract the magnetic field intensity fluctuation characteristics in the environmental magnetic field intensity data and the abnormal change characteristics of the obstacle density in the obstacle density data; S3: According to the extracted magnetic field intensity fluctuation characteristics and abnormal change characteristics of the obstacle density, determine the accuracy weight assignment of the UAV navigation and positioning system in each pipeline inspection path, and perform weighted average summation calculation on it to obtain the comprehensive inspection accuracy index; S4: Compare the comprehensive inspection accuracy index in the pipeline inspection path with the gradient standard threshold, and divide the accuracy of the UAV pipeline inspection path according to the comparison result. The division results include accurate inspection, incomplete accurate inspection, and inaccurate inspection; S5: For accurate inspection, continue the inspection according to the established path without adjusting the flight strategy; for inaccurate inspection, immediately interrupt the inspection task of the UAV, recalculate the optimal inspection path, and give priority to covering the abnormal pipeline section; S6: For incomplete accurate inspection, predict the degree of accuracy abnormality of the UAV navigation and positioning system in the subsequent fixed time period. If the degree of accuracy abnormality is high, re-plan the inspection path of the UAV, increase the flight altitude of the UAV, and avoid the area with dense obstacles.

2. The method for artificial intelligence inspection based on unmanned aerial vehicle according to claim 1, wherein: In S2, after analyzing the magnetic field intensity fluctuation characteristics in the extracted environmental magnetic field intensity data, a magnetic field intensity continuous fluctuation index is generated. The acquisition method of the magnetic field intensity continuous fluctuation index is: Record the output of the magnetometer sensor to obtain the time series data B(s) of the magnetic field intensity, where s is the time window. Divide the time series B(s) into short-time windows w(n), and the length of each window is N samples; Apply the Fourier transform to the signals within each time window to calculate the frequency components: ; where is the complex spectral value at time s and frequency f, B(n) is the signal sample within the window, w(n) is the window function value, is the frequency weight in complex form; Take the modulus of the complex spectrum to obtain the amplitude spectrum , and the expression is: ; For each time window s, calculate the spectral energy density , and the expression is: ; where and are the lowest and highest frequencies of interest respectively, and E(s) represents the spectral energy density of time window s; Normalize E(s): ; where and are the minimum and maximum values of the spectral energy in the entire sequence respectively, is the normalized spectral energy density; Smooth the normalized spectral energy density to generate the continuous magnetic field intensity fluctuation index HSK, and the expression is: ; where Z is the smoothing factor, HSK is the continuous magnetic field intensity fluctuation index, is the continuous magnetic field intensity fluctuation index at time 3. The method for artificial intelligence inspection based on unmanned aerial vehicle according to claim 2, wherein: In S2, after analyzing the abnormal change characteristics of the obstacle density in the extracted obstacle density data, an obstacle density abnormal change index is generated. The acquisition method of the obstacle density abnormal change index is: Collect obstacle point cloud data through lidar to form a point set ; Each point contains its position coordinates , and calculate the distance between obstacle points using Euclidean distance and : ; For each point , find its set of k-nearest neighbor points , where k is the number of nearest neighbors specified by the user; The k-nearest neighbor distance is the distance from the point to its k-th nearest neighbor point: ; Define to the reachable distance of its neighborhood points , and the expression is: ; point The local reachable density is all the points in its neighborhood The reciprocal mean of the reachable distance is expressed as: ; In the formula, yes The number of neighboring points of The local outlier factor is the ratio of the average local reachability density of the points in its neighborhood to its own local reachability density: ; Normalize the local outlier factor values ​​of all points to generate the obstacle density abnormal change index, the expression is: ; In the formula, GSH is the obstacle density abnormal change index, and are the minimum and maximum local outlier factor values ​​in the data set, respectively.

4. The method for artificial intelligence inspection based on unmanned aerial vehicle according to claim 3, wherein: In S3, according to the extracted magnetic field intensity fluctuation characteristics and abnormal change characteristics of the obstacle density, determine the accuracy weight assignment of the UAV navigation and positioning system in each pipeline inspection path, and perform weighted average summation calculation on it to obtain the comprehensive inspection accuracy index. Specifically: Convert the continuous fluctuation index of magnetic field intensity and the abnormal change index of obstacle density into a comprehensive feature vector, and use the comprehensive feature vector as the input of a machine learning model. The machine learning model takes the prediction of the accuracy weight assignment label of the drone navigation and positioning system in each pipeline inspection path as the prediction target, and takes minimizing the sum of the prediction errors of the accuracy weight assignment labels of the drone navigation and positioning system in all pipeline inspection paths as the training target. Train the machine learning model until the sum of the prediction errors converges and then stop the model training. Determine the accuracy weight assignment of the drone navigation and positioning system in each pipeline inspection path according to the model output results. Among them, the machine learning model is a polynomial regression model, and perform a weighted average summation calculation on the accuracy weight assignment of the drone navigation and positioning system in each pipeline inspection path to obtain a comprehensive inspection accuracy index.

5. The method for artificial intelligence inspection based on drones according to claim 4, characterized in that: In S4, compare the comprehensive inspection accuracy index in the pipeline inspection path with the gradient standard threshold, and classify the accuracy of the drone pipeline inspection path according to the comparison result. The classification results include accurate inspection, incomplete accurate inspection, and inaccurate inspection; Compare the obtained comprehensive inspection accuracy index in the pipeline inspection path with the gradient standard threshold. The gradient standard threshold includes a first standard threshold and a second standard threshold, and the first standard threshold is less than the second standard threshold. Compare the comprehensive inspection accuracy index in the pipeline inspection path with the first standard threshold and the second standard threshold respectively; If the comprehensive inspection accuracy index in the pipeline inspection path is greater than the second standard threshold, it indicates that the accuracy of the drone navigation and positioning system in the pipeline inspection path is high. At this time, generate a high-accuracy inspection signal and classify the accuracy of the drone pipeline inspection path as accurate inspection; If the comprehensive inspection accuracy index in the pipeline inspection path is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, it indicates that the accuracy of the drone navigation and positioning system in the pipeline inspection path is medium. At this time, generate a medium-accuracy inspection signal and classify the accuracy of the drone pipeline inspection path as incomplete accurate inspection; If the comprehensive inspection accuracy index in the pipeline inspection path is less than the first standard threshold, it indicates that the accuracy of the drone navigation and positioning system in the pipeline inspection path is low. At this time, generate a low-accuracy inspection signal and classify the accuracy of the drone pipeline inspection path as inaccurate inspection.

6. The method for artificial intelligence inspection based on unmanned aerial vehicle according to claim 5, wherein: In S6, for incomplete accurate inspection, predict the degree of abnormality of the accuracy of the drone navigation and positioning system in the subsequent fixed time period. If the degree of abnormality of the accuracy is high, re-plan the inspection path of the drone, increase the flight altitude of the drone, and avoid areas with dense obstacles. Specifically: For the incomplete accuracy inspection, that is, the comprehensive inspection accuracy index in the generated pipeline inspection path is greater than or equal to the first standard threshold and less than or equal to the second standard threshold. In the next T time period, the abnormal degree of the accuracy of the navigation and positioning system is determined by the comprehensive prediction index as follows: where: is the predicted value of the magnetic field intensity fluctuation index in the future time t, GSH(t) is the predicted value of the abnormal change index of the obstacle density in the future time t, CI(t) is the comprehensive inspection accuracy index at the current time t, and α, β, γ are weight coefficients, indicating the influence ratios of magnetic field fluctuation, obstacle density, and current comprehensive accuracy on the predicted value, satisfying α + β + γ = 1. The time series prediction model is used to perform short-term prediction on HSK(t) and GSH(t): , ; Define the threshold Pthreshold for predicting the abnormal degree of accuracy: If , it indicates a high abnormal degree of accuracy and the path needs to be re-planned; if , continue to inspect according to the existing path.

7. A method for artificial intelligence inspection based on an unmanned aerial vehicle according to claim 6, characterized in that: The goal of path replanning is to avoid densely populated obstacle areas with the shortest path and optimize navigation accuracy. The objective function of the path planning problem , and the expression is: ; In the formula, is the length of path segment i, is the predicted abnormality degree of path segment i, k is the penalty coefficient, which is used to increase the path weight of high-abnormality areas and avoid high-risk areas, and W is the total number of path segments; The flight altitude adjustment rule is: Define the adjusted flight altitude , and the expression is: ; In the formula, is the current flight altitude, and ΔH is the altitude increment coefficient; Mark the high obstacle density area and set it as an impassable area during path planning. The area density weight formula is: ; In the formula, is the area weight, and GSH(j) is the abnormal change index of the obstacle density in the area. During path replanning, exclude the areas where the area weight is greater than the area weight threshold preset according to historical data.

8. An unmanned aerial vehicle (UAV)-based artificial intelligence inspection system for implementing an UAV-based artificial intelligence inspection method according to any one of claims 1 to 7, characterized in that: It includes a data acquisition module, a feature extraction module, a navigation and positioning accuracy evaluation module, an accuracy classification module, a path planning and task control module, and a dynamic prediction and strategy optimization module; Data acquisition module: It obtains the magnetic field interference parameters and terrain complexity parameters in different pipeline inspection paths in real time through the on-board sensors of the drone. The magnetic field interference parameters include environmental magnetic field intensity data, and the terrain complexity parameters include obstacle density data; Feature extraction module: It preprocesses the obtained environmental magnetic field intensity data and obstacle density data, and extracts the magnetic field intensity fluctuation features in the environmental magnetic field intensity data and the abnormal change features of the obstacle density in the obstacle density data; Navigation and positioning accuracy evaluation module: According to the extracted magnetic field intensity fluctuation features and abnormal change features of the obstacle density, it determines the accuracy weight assignment of the drone navigation and positioning system in each pipeline inspection path, and calculates the weighted average sum to obtain the comprehensive inspection accuracy index; Accuracy classification module: It compares the comprehensive inspection accuracy index in the pipeline inspection path with the gradient standard threshold, and classifies the accuracy of the drone pipeline inspection path according to the comparison result. The classification results include accurate inspection, incomplete accurate inspection, and inaccurate inspection; Path planning and task control module: For accurate inspection, continue the inspection according to the established path without adjusting the flight strategy; for inaccurate inspection, immediately interrupt the inspection task of the drone, recalculate the best inspection path, and give priority to covering the abnormal pipeline section; Dynamic prediction and strategy optimization module: For incomplete accurate inspection, it predicts the degree of accuracy abnormality of the drone navigation and positioning system in the subsequent fixed time period. If the degree of accuracy abnormality is high, it replans the inspection path of the drone, increases the flight altitude of the drone, and avoids the area with dense obstacles.

Images