An artificial intelligence-based low-altitude unmanned aerial vehicle intelligent inspection method and system

Abstract

The application discloses a low-altitude unmanned aerial vehicle intelligent inspection method and system based on artificial intelligence, relates to the unmanned aerial vehicle flight control technical field, and comprises the following steps: acquiring spatial observation information of an object to be inspected, layering and assigning the spatial observation information to obtain an observation zone and an anchor point chain; discriminating the heading of the observation zone and the anchor point chain and assigning the observation zone and the anchor point chain to segments to obtain a straight-bend segment inspection control chain, performing segment attribution matching on the straight-bend segment inspection control chain to obtain a current inspection segment, performing constraint fusion on the current inspection segment to obtain a current inspection control state; performing control primitive matching on the current inspection control state to obtain a flight control primitive, performing control quantity calculation on the unmanned aerial vehicle according to the flight control primitive to obtain an initial flight control intention; and the application realizes the cooperative closed loop of safety constraint, control intention and actual flight by performing minimum constraint correction on the control intention through flight state prediction and safety certificate value.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) flight control technology, and in particular to an intelligent inspection method and system for low-altitude UAVs based on artificial intelligence. Background Technology

[0002] Low-altitude unmanned aerial vehicle (UAV) inspection technology is an important application direction for the integrated development of intelligent inspection, low-altitude economy, and artificial intelligence flight control. It has been widely used in inspection scenarios such as power lines, photovoltaic power stations, bridges and tunnels, industrial parks, building facades, and public infrastructure. With the improvement of UAVs' perception capabilities, edge computing capabilities, visual recognition capabilities, and autonomous control capabilities, low-altitude inspection is gradually developing from manual remote control and fixed-route flight to autonomous, refined, and intelligent inspection. This places higher demands on the observation stability, flight safety, target approach capability, and continuous mission execution capability of UAVs in complex spaces.

[0003] Existing methods still have room for improvement. First, inspection control relies heavily on unified routes or fixed waypoint logic, resulting in insufficient target alignment stability and fragmented control targeting during the observation process. Second, safety control is mostly limited to obstacle avoidance or amplitude limiting, making it difficult to coordinate closed loops in complex low-altitude environments. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides an artificial intelligence-based intelligent inspection method for low-altitude unmanned aerial vehicles (UAVs) to solve the problem of difficulty in achieving collaborative closed-loop operation.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides an intelligent inspection method for low-altitude unmanned aerial vehicles based on artificial intelligence, comprising:

[0008] Acquire spatial observation information of the object to be inspected, perform hierarchical reduction on the spatial observation information, and obtain the observation zone and anchor chain;

[0009] The observation zone and anchor point chain are subjected to heading discrimination and segment reduction to obtain the straight and curved section inspection control chain. The straight and curved section inspection control chain is subjected to segment assignment matching to obtain the current inspection segment. The current inspection segment is subjected to constraint fusion to obtain the current inspection control state.

[0010] The current inspection control state is matched with control primitives to obtain flight control primitives. Based on the flight control primitives, the control quantities of the UAV are calculated to obtain the initial flight control intent.

[0011] The initial flight control intention is predicted to obtain the flight state prediction result. The safety certificate value is calculated based on the flight state prediction result. The initial flight control intention is then modified with minimum constraints based on the safety certificate value to obtain the safe flight control package.

[0012] The actual flight status is obtained, the response deviation between the safe flight control package and the actual flight status is monitored, and the link is reconstructed for the current inspection segment based on the response deviation and the actual flight status to obtain the updated straight and curved section inspection control chain.

[0013] As a preferred embodiment of the artificial intelligence-based intelligent inspection method for low-altitude unmanned aerial vehicles (UAVs) described in this invention, the specific steps of performing hierarchical reduction of spatial observation information are as follows:

[0014] The spatial observation information includes the location, altitude, shape boundary, target observation surface, low-altitude altitude limit boundary, obstacle boundary, no-entry boundary, UAV take-off and landing point, and UAV flight capability parameters of the object to be inspected;

[0015] Based on the altitude of the object to be inspected and the target observation surface, the object to be inspected is divided into low-altitude observation layers. Recommended flight altitude, minimum safe altitude, maximum permissible altitude, recommended lateral observation distance, target observation window, permissible heading range, permissible speed range, and avoidance exit direction are configured for each low-altitude observation layer to obtain the observation zone.

[0016] Sampling is performed along the outer boundary of each object to be inspected to obtain observation anchor points, and these observation anchor points are connected sequentially to form an anchor point chain.

[0017] As a preferred embodiment of the artificial intelligence-based intelligent inspection method for low-altitude unmanned aerial vehicles (UAVs) described in this invention, the specific steps of performing heading determination and segment reduction are as follows:

[0018] Calculate the heading angles of adjacent observation anchor points in the anchor chain to obtain the anchor chain heading angle sequence, and use the anchor chain heading angle sequence to calculate the heading change between adjacent anchor chain heading angles;

[0019] When the change in heading is less than the preset steering threshold, the corresponding anchor point connection segment is marked as a straight stable observation segment, and straight segment target alignment constraint, straight segment speed constraint, straight segment heading constraint and straight segment lateral safety constraint are configured for the straight stable observation segment to obtain the straight segment control constraint;

[0020] When the change in heading is not less than the preset steering threshold, the corresponding anchor point connection segment is marked as the curve constraint observation segment, and curve constraint observation segment is configured with curve segment target alignment constraint, curve segment speed constraint, curve segment heading constraint, curve segment lateral safety constraint and curve segment deceleration constraint to obtain the curve segment control constraint.

[0021] According to the connection sequence of the anchor point chain, the straight section stability observation segment, the curve constraint observation segment, the straight section control constraint and the curve section control constraint are associated to obtain the straight-curved section inspection control chain.

[0022] As a preferred embodiment of the artificial intelligence-based intelligent inspection method for low-altitude unmanned aerial vehicles (UAVs) described in this invention, the specific steps of constraining and fusing the current inspection segment are as follows:

[0023] The current position, altitude, speed, heading, attitude, lateral distance, remaining return margin, and current inspection progress of the drone are obtained as the current flight status of the drone;

[0024] The observation segment corresponding to the current inspection progress is taken as the current inspection segment;

[0025] The recommended flight altitude, minimum safe altitude, maximum permissible altitude, recommended lateral observation distance, target observation window, permissible heading range, permissible speed range, and avoidance exit direction corresponding to the current inspection segment are used as the basic constraints for the current segment.

[0026] When the current inspection segment is a straight and stable observation segment, the basic constraints of the current segment are fused with the corresponding straight segment control constraints to obtain the current inspection control state;

[0027] When the current inspection segment is a curve constraint observation segment, the basic constraints of the current segment are fused with the corresponding curve segment control constraints to obtain the current inspection control state.

[0028] As a preferred embodiment of the artificial intelligence-based intelligent inspection method for low-altitude unmanned aerial vehicles (UAVs) described in this invention, the specific steps of performing control primitive matching are as follows:

[0029] The current flight status of the UAV is compared with the current inspection control status to obtain the altitude deviation, speed deviation, heading deviation, lateral distance deviation, target alignment deviation, and exit trigger results.

[0030] The altitude deviation, speed deviation, heading deviation, lateral distance deviation, target alignment deviation, exit trigger result, current inspection segment, heading change, current inspection progress, and historical response deviation are feature-encoded to obtain the control feature vector.

[0031] The control feature vector is input into the pre-trained control primitive prediction model for classification, the matching probability of each primitive is obtained, and the primitive with the highest matching probability is selected as the flight control primitive.

[0032] As a preferred embodiment of the artificial intelligence-based intelligent inspection method for low-altitude unmanned aerial vehicles (UAVs) described in this invention, the specific steps for calculating the control variables of the UAV are as follows:

[0033] The altitude constraint, speed constraint, heading constraint, lateral safety constraint, target alignment constraint, avoidance and exit constraint, and the current flight status of the UAV are used as the basis for calculating the control variables.

[0034] When the flight control primitive is a straight-line stability tracking primitive, the speed adjustment, heading correction, and lateral distance adjustment are calculated based on the control quantity calculation and used as the straight-line stability control quantities.

[0035] When the flight control primitive is a cornering deceleration and steering primitive, the calculated speed adjustment, heading correction, attitude compensation, and lateral distance adjustment are used as cornering constraint control quantities.

[0036] When the flight control primitive is a local safety observation adjustment primitive, the corresponding altitude adjustment, speed adjustment, heading correction, lateral distance adjustment, attitude compensation, and hovering time are calculated as local observation control quantities.

[0037] When the flight control primitive is the exit return-to-home primitive, the return-to-home direction, return-to-home speed, and return-to-home altitude are calculated as the return-to-home control variables.

[0038] Configure the control execution sequence for the straight stability control, curve constraint control, local observation control, and return control to obtain the initial flight control intent.

[0039] As a preferred embodiment of the artificial intelligence-based intelligent inspection method for low-altitude unmanned aerial vehicles (UAVs) according to the present invention, the specific steps for calculating the safety certificate value are as follows:

[0040] The flight state for the next control cycle is calculated using the initial flight control intent and the current flight state of the UAV, and the flight state prediction result is obtained.

[0041] The flight status prediction results and the current inspection control status are used to calculate the altitude safety margin, space boundary safety margin, motion safety margin, and curve flight safety margin as safety certificate values.

[0042] As a preferred embodiment of the artificial intelligence-based intelligent inspection method for low-altitude unmanned aerial vehicles (UAVs) described in this invention, the specific steps of performing minimum constraint correction are as follows:

[0043] The safety certificate value is compared with the preset height safety margin threshold, spatial boundary safety margin threshold, motion safety margin threshold, and curve flight safety margin threshold to obtain the safety execution judgment result.

[0044] When the safety execution judgment result is unsafe, the corresponding initial flight control intention is adjusted to not exceed the current inspection control state, and a safe flight control package is obtained.

[0045] As a preferred embodiment of the artificial intelligence-based intelligent inspection method for low-altitude unmanned aerial vehicles (UAVs) described in this invention, the specific steps of reconstructing the link of the current inspection segment are as follows:

[0046] The actual position of a drone that first exceeds the preset allowable response deviation range and whose actual flight state exceeds the current inspection control state is considered an abnormal position.

[0047] Using the abnormal position as a reference, the reentry altitude, reentry speed, reentry heading, reentry lateral distance, and reentry hovering time are calculated using the current inspection control state to form a local reentry control segment. After inserting the segment into the current inspection segment, the updated straight and curved section inspection control chain is obtained.

[0048] Secondly, the present invention provides an intelligent inspection system for low-altitude unmanned aerial vehicles based on artificial intelligence, comprising:

[0049] The configuration module obtains the spatial observation information of the object to be inspected, performs hierarchical reduction on the spatial observation information, and obtains the observation zone and anchor point chain.

[0050] The fusion module performs heading discrimination and segment reduction on the observation zone and anchor point chain to obtain the straight and curved section inspection control chain. It then performs segment assignment matching on the straight and curved section inspection control chain to obtain the current inspection segment. Finally, it performs constraint fusion on the current inspection segment to obtain the current inspection control state.

[0051] The calculation module performs control primitive matching on the current inspection control state to obtain flight control primitives. Based on the flight control primitives, it performs control quantity calculation on the UAV to obtain the initial flight control intent.

[0052] The correction module predicts the flight state of the initial flight control intention, obtains the flight state prediction result, calculates the safety certificate value based on the flight state prediction result, and performs minimum constraint correction on the initial flight control intention based on the safety certificate value to obtain the safe flight control package.

[0053] The reconstruction module acquires the actual flight status, monitors the response deviation between the safe flight control package and the actual flight status, and reconstructs the link of the current inspection segment based on the response deviation and the actual flight status to obtain the updated straight and curved section inspection control chain.

[0054] The beneficial effects of this invention are as follows: By constructing observation zones and anchor point chains through spatial observation information, the target observation surface, flight altitude, lateral distance, heading range, speed range, and avoidance direction can be refined to specific inspection segments. This avoids the problem that a unified route or fixed waypoint is difficult to adapt to different observation levels and differences in straight and curved sections, thus improving the continuity of low-altitude inspections and the targeting of flight control. In addition, by using flight status prediction and safety certificate values ​​to make minimum constraint corrections to control intentions, safety constraints, control intentions, and actual flight responses can form a continuous verification and feedback update relationship before and after control execution. This avoids the problem that safety control remains at the level of local obstacle avoidance or control quantity limitation, improves the closed-loop safety, anomaly recovery capability, and continuous execution capability of inspection tasks in complex low-altitude environments, and realizes a collaborative closed loop of safety constraints, control intentions, and actual flight. Attached Figure Description

[0055] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0056] Figure 1 This is a flowchart of an AI-based intelligent inspection method for low-altitude drones.

[0057] Figure 2 This is a schematic diagram of an AI-based low-altitude drone intelligent inspection system.

[0058] Figure 3 A schematic diagram for generating initial flight control intent.

[0059] Figure 4 A schematic diagram for generating a safe flight control package. Detailed Implementation

[0060] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0061] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0062] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0063] Reference Figures 1-4 This is one embodiment of the present invention, which provides an intelligent inspection method for low-altitude unmanned aerial vehicles based on artificial intelligence, including the following steps:

[0064] S1: Obtain spatial observation information of the object to be inspected, perform hierarchical reduction on the spatial observation information, and obtain the observation zone and anchor chain.

[0065] S1.1: Collect the location, altitude, external boundary, target observation surface, low-altitude altitude limit boundary, obstacle boundary, no-entry boundary, UAV take-off and landing point, and UAV flight capability parameters of the object to be inspected as spatial observation information.

[0066] The target observation surface refers to the surface area selected from the outer boundary of the object to be inspected, which needs to be aligned and observed by the UAV imaging equipment. The target observation surface has an observation surface boundary, which is used to limit the actual surface range of the target observation surface. The target observation window is the projection range of the observation surface boundary in the UAV imaging plane.

[0067] The flight capability parameters of the UAV include minimum stable flight altitude, maximum flightable altitude, minimum controllable altitude interval, maximum inspection speed, minimum stable inspection speed, maximum climb rate, maximum yaw rate, maximum attitude angle, minimum stable observation distance, maximum stable observation distance, minimum stable observation time, safety altitude margin, safety lateral margin, imaging field of view parameters, and maximum hovering time.

[0068] It should also be noted that the drone's flight capability parameters are read from the flight limit parameters recorded by the drone flight control system, the drone product manual, and the observation stability parameters obtained from the pre-mission test calibration.

[0069] Collect the height of the lower edge, the height of the upper edge, and the height of the center of the observation surface of the target.

[0070] The minimum stable flight altitude is used as the starting altitude of the low-altitude observation layer, and the minimum value between the low-altitude altitude limit boundary and the maximum flyable altitude is used as the ending altitude of the low-altitude observation layer, thus obtaining the low-altitude inspection altitude range.

[0071] The low-altitude inspection altitude range is divided from bottom to top according to an integer multiple of the minimum controllable altitude interval. In this embodiment, the low-altitude inspection altitude range is divided from bottom to top according to the minimum controllable altitude interval to obtain low-altitude observation layers. The starting altitude of each low-altitude observation layer is taken as the lower limit altitude of the corresponding low-altitude observation layer, and the ending altitude of each low-altitude observation layer is taken as the upper limit altitude of the corresponding low-altitude observation layer.

[0072] When the ending altitude of the low-altitude observation layer exceeds the ending altitude of the low-altitude inspection altitude range, the ending altitude of the low-altitude inspection altitude range shall be taken as the upper limit altitude of the corresponding low-altitude observation layer.

[0073] The center height of each target observation surface is compared with the lower and upper limits of each low-altitude observation layer, and the low-altitude observation layer to which the center height of the observation surface falls is determined as the low-altitude observation layer to which the target observation surface belongs.

[0074] When the lower and upper edges of the target observation surface cross two or more low-altitude observation layers, the target observation surface is divided according to the lower and upper limits of the crossed low-altitude observation layers and assigned to the corresponding low-altitude observation layers respectively.

[0075] The minimum height of the lower edge of each target observation surface in each low-altitude observation layer is taken as the minimum observation height, the maximum height of the upper edge of each target observation surface is taken as the maximum observation height, and the average of the minimum and maximum observation heights is taken as the observation center height.

[0076] The highest boundary height of the obstacle adjacent to the current low-altitude observation layer is read from the obstacle boundary, and the sum of the highest boundary height of the obstacle and the safe height margin is taken as the safe crossing height of the obstacle.

[0077] It should also be noted that when there are dynamic obstacles in the inspection environment, the highest height of the dynamic obstacles is collected in real time as the highest boundary height of the obstacles.

[0078] The maximum value among the lower limit height of the floor, the minimum stable flight altitude of the drone, and the safe crossing height of the obstacle is taken as the minimum safe height.

[0079] The minimum value among the upper limit of the layer, the low-altitude limit boundary, and the maximum flightable altitude is taken as the maximum allowable altitude.

[0080] When the altitude of the observation center is between the minimum safe altitude and the maximum permissible altitude, the altitude of the observation center will be used as the recommended flight altitude. When the altitude of the observation center does not exceed the minimum safe altitude, the minimum safe altitude will be used as the recommended flight altitude. When the altitude of the observation center is not lower than the maximum permissible altitude, the maximum permissible altitude will be used as the recommended flight altitude.

[0081] It should also be noted that when the minimum safe altitude is greater than the maximum permissible altitude, the corresponding low-altitude observation layer will be marked as an unusable low-altitude observation layer, and the corresponding observation zone will no longer be generated.

[0082] Read the horizontal and vertical field of view angles from the camera's imaging field of view parameters, collect the horizontal and vertical spans of the target observation surface, calculate the first lateral distance based on the horizontal span and horizontal field of view angle, and calculate the second lateral distance based on the vertical span and vertical field of view angle. The expressions are as follows:

[0083] ;

[0084] ;

[0085] in, The first lateral distance, The second lateral distance, The horizontal span of the target observation surface, The vertical span of the target observation surface. For horizontal field of view, This is the vertical field of view angle.

[0086] The maximum value between the first lateral distance and the second lateral distance is used as the candidate lateral observation distance. When the candidate lateral observation distance is between the minimum stable observation distance and the maximum stable observation distance, the candidate lateral observation distance is used as the recommended lateral observation distance. When the candidate lateral observation distance is not greater than the minimum stable observation distance, the minimum stable observation distance is used as the recommended lateral observation distance. When the candidate lateral observation distance exceeds the maximum stable observation distance, the maximum stable observation distance is used as the recommended lateral observation distance. The target observation surface is divided into target observation sub-regions according to the recommended lateral observation distance, so that the boundary of the observation surface corresponding to each target observation sub-region falls into the imaging field of view.

[0087] The observation surface boundary of each target observation sub-region is mapped onto the UAV imaging plane to obtain the target observation window corresponding to each target observation sub-region.

[0088] The direction from the recommended lateral observation distance of the UAV to the target observation surface is taken as the reference heading. Centered on the reference heading, the product of the maximum yaw rate and the control cycle is taken as the maximum heading adjustment angle in a single cycle. The minimum value between the maximum heading adjustment angle in a single cycle and half of the horizontal field of view is taken as the basic allowable deviation range. When an obstacle boundary or no-entry boundary appears within the field of view of the basic allowable deviation range, the basic allowable deviation range on the corresponding side is reduced until no obstacle boundary or no-entry boundary appears, thus obtaining the allowable heading range.

[0089] The maximum inspection speed is set as the upper limit of the allowable speed range, and the minimum stable inspection speed is set as the lower limit of the allowable speed range, thus forming the allowable speed range.

[0090] The direction from the observation center point of the target observation surface to the observation location is used as the candidate avoidance and exit direction.

[0091] When a candidate avoidance exit direction crosses an obstacle boundary or a restricted area boundary, the candidate avoidance exit direction is used as the center, and candidate deflection directions are searched step by step to the left and right sides according to a fixed angle step size (5° in this embodiment). The maximum deflection angle is limited to the product of the maximum yaw rate and the preset number of exit control cycles.

[0092] For each candidate deflection direction, determine whether the exit trajectory formed when the UAV exits along the candidate deflection direction intersects with the obstacle boundary, restricted access boundary, or low-altitude height limit boundary, and whether the minimum distance between the exit trajectory and the obstacle boundary or restricted access boundary is not less than the safe lateral margin.

[0093] When there are candidate deflection directions that meet the conditions, the candidate deflection direction with the smallest deflection angle is selected as the avoidance exit direction. When there are no candidate deflection directions that meet the conditions, the corresponding low-altitude observation layer is marked as unsafe to exit the low-altitude observation layer, and no corresponding observation zone is generated.

[0094] When the candidate avoidance exit direction does not cross the obstacle boundary or the restricted area boundary, the candidate avoidance exit direction shall be used as the avoidance exit direction.

[0095] The recommended flight altitude, minimum safe altitude, maximum permissible altitude, recommended lateral observation distance, target observation window, permissible heading range, permissible speed range, and avoidance exit direction are bound to the corresponding low-altitude observation layer to obtain the observation zone.

[0096] S1.2: Calculate the single-sided coverage width on the target observation surface using the recommended lateral observation distance and horizontal field of view. The expression is:

[0097] ;

[0098] in, The coverage width on one side of the target observation surface. To recommend the lateral observation distance, For horizontal field of view, This represents the tangent function.

[0099] Using the single-sided coverage width as the sampling interval, boundary sampling is performed along the observation surface boundary of each target observation sub-region in the observation band, and coverage sampling is performed inside the target observation sub-region at intervals no greater than twice the single-sided coverage width. The boundary sampling points and coverage sampling points are used together as observation anchor points.

[0100] Adjacent observation anchor points are connected sequentially, and connections that cross obstacle boundaries, restricted access boundaries, or low-altitude height limits are eliminated to obtain an anchor point chain.

[0101] It should also be noted that when the connection relationship is removed, resulting in the anchor chain breaking, the observation anchor points before and after the break are used as endpoints to search for bypass connections within an area that does not cross obstacle boundaries, restricted access boundaries, or low-altitude height limits and meets the safety lateral margin and safety height margin.

[0102] When a bypass connection exists, insert the bypass connection into the anchor chain. When no bypass connection exists, mark the corresponding broken section as an uninspectable section, and use the remaining connectable observation anchor connections as the anchor chain.

[0103] S2: Perform heading discrimination and segment reduction on the observation zone and anchor point chain to obtain the straight and curved section inspection control chain. Perform segment assignment matching on the straight and curved section inspection control chain to obtain the current inspection segment. Perform constraint fusion on the current inspection segment to obtain the current inspection control state.

[0104] S2.1: Take the direction from the current observation anchor point to the next observation anchor point as the heading direction of the anchor point connection segment, and arrange the heading directions of each anchor point connection segment according to their order in the anchor point chain to obtain the anchor point chain heading angle sequence.

[0105] The circumference difference is calculated for adjacent heading angles in the heading angle sequence of the anchor chain, and the absolute value of the difference calculation result is taken to obtain the absolute value of the difference between adjacent heading angles. The absolute value of the smallest difference between adjacent heading angles within the angle period is taken as the heading change between adjacent anchor chain heading angles.

[0106] Half of the horizontal field of view is taken as the maximum single-sided heading allowance, the single-sided heading deviation corresponding to the allowable heading range is taken as the heading allowance of the current observation zone, and the minimum value among the maximum heading adjustment angle in a single cycle, the maximum single-sided heading allowance, and the heading allowance of the current observation zone is taken as the preset turning threshold.

[0107] The heading changes between adjacent anchor chain heading angles are continuously accumulated to obtain the continuous cumulative heading changes. When the heading change is less than a preset turning threshold and the continuous cumulative heading changes are less than a preset cumulative turning threshold, the corresponding anchor connection segment is marked as a straight-line stable observation segment. When the heading change is not less than a preset turning threshold or the continuous cumulative heading changes are not less than a preset cumulative turning threshold, the corresponding anchor connection segment is marked as a curve constraint observation segment.

[0108] The preset cumulative turning threshold is the product of the number of consecutively accumulated segments and the preset turning threshold, half of the horizontal field of view, and the minimum value of the single-sided heading deviation corresponding to the current observation zone.

[0109] For a straight-line stable observation segment, the target observation window is used as the target alignment constraint, the allowable speed range is used as the speed constraint, the allowable heading range is used as the heading constraint, and the recommended lateral observation distance and the safe lateral margin are used together as the lateral safety constraint, thus obtaining the straight-line control constraint.

[0110] For the curve constraint observation section, the target observation window is used as the target alignment constraint for the curve section, the allowable heading range and the maximum yaw rate are used together as the heading constraint for the curve section, and the recommended lateral observation distance and the safe lateral margin are used together as the lateral safety constraint for the curve section.

[0111] The reduced upper speed limit is calculated using the upper speed limit, lower speed limit, change in heading, and preset steering threshold. The expression is:

[0112] ;

[0113] in, To reduce the upper limit of speed, This indicates taking the maximum value. As the lower limit of speed, This is the upper limit of speed. For the change in heading, This is the preset steering threshold.

[0114] The reduced upper speed limit is used as the upper speed limit for the curve section, and the lower speed limit is used as the lower speed limit for the curve section. The average of the reduced upper and lower speed limits is used as the target speed for the curve section. The lower speed limit, target speed, and upper speed limit for the curve section are used as speed constraints for the curve section.

[0115] The tangent of the maximum attitude angle is calculated as the horizontal acceleration / deceleration capability ratio, and the product of the horizontal acceleration / deceleration capability ratio and the gravitational acceleration is taken as the maximum usable horizontal deceleration.

[0116] The deceleration distance is calculated using the drone's current speed when entering the curve constraint observation section, the reduced speed limit, and the maximum available horizontal deceleration. The expression is:

[0117] ;

[0118] in, To reduce deceleration distance, For the maximum available horizontal deceleration, This represents the current speed of the UAV when it enters the curve-constrained observation section.

[0119] The deceleration distance and the upper limit of speed in the curve section are used as deceleration constraints for the curve section. The deceleration constraints, target alignment constraints, speed constraints, heading constraints, and lateral safety constraints of the curve section are combined into control constraints for the curve section.

[0120] According to the connection sequence of the anchor point chain, the straight-line stability observation segment, the curve constraint observation segment, the straight-line control constraint and the curve control constraint are associated segment by segment. For each anchor point connection segment, the corresponding recommended flight altitude, target observation window, allowable heading range, allowable speed range, recommended lateral observation distance and avoidance exit direction are retained to obtain the straight-line and curve segment inspection control chain.

[0121] S2.2: Collect the drone's current position, altitude, speed, heading, attitude, lateral distance, remaining return margin, and current inspection progress as the drone's current flight status.

[0122] Collect the starting observation anchor point, ending observation anchor point, and segment type label for each anchor point connection segment, calculate the spatial distance between the starting observation anchor point and the ending observation anchor point, and obtain the segment length of each anchor point connection segment.

[0123] According to the connection order of the anchor chain, the segment length of each anchor connection segment is accumulated to obtain the cumulative progress.

[0124] Project the drone's current position onto the anchor chain to obtain the projection point of the drone's current position on the anchor chain. Take the cumulative length from the starting point of the anchor chain to the projection point as the current spatial progress. When the cumulative progress is not less than the current spatial progress for the first time, take the corresponding anchor connection segment as the current inspection segment.

[0125] Extract recommended flight altitude, minimum safe altitude, maximum permissible altitude, recommended lateral observation distance, target observation window, permissible heading range, permissible speed range, and avoidance exit direction from the observation zone corresponding to the current inspection segment, and use them as the basic constraints for the current segment.

[0126] When the current inspection segment is a straight stable observation segment, the recommended flight altitude, minimum safe altitude, and maximum allowable altitude in the basic constraints of the current segment are merged into the straight segment height requirement, and the target observation window in the basic constraints of the current segment is merged with the target alignment constraint of the straight segment to obtain the target alignment requirement of the straight segment.

[0127] The speed requirement for the straight segment is obtained by intersecting the allowable speed range in the basic constraints of the current segment with the speed constraint of the straight segment. The heading requirement for the straight segment is obtained by intersecting the allowable heading range in the basic constraints of the current segment with the heading constraint of the straight segment.

[0128] The recommended lateral observation distance in the current segment's basic constraints is taken as the target lateral observation distance, the lateral safety constraint of the straight segment is taken as the minimum lateral safety condition, and the target lateral observation distance and the minimum lateral safety condition are merged into the lateral distance requirement of the straight segment. The avoidance exit direction in the current segment's basic constraints is taken as the avoidance exit requirement of the straight segment.

[0129] The requirements for straight segment height, speed, heading, lateral distance, target alignment, and avoidance / exit are merged into a single straight segment fusion constraint. This straight segment fusion constraint is then correlated with the current flight state of the UAV to obtain the current inspection control state.

[0130] When the current inspection segment is a curve constraint observation segment, the recommended flight altitude, minimum safe altitude, and maximum allowable altitude in the current segment's basic constraints are combined into the curve segment height requirement. The allowable speed range in the current segment's basic constraints is intersected with the curve segment speed constraints to obtain the curve segment speed requirement. The deceleration distance and curve segment speed limit in the curve segment deceleration constraints are used as the curve segment deceleration requirement.

[0131] The intersection of the allowable heading range in the current segment's basic constraints and the heading constraints of the curve segment is used to obtain the heading requirements of the curve segment. The recommended lateral observation distance in the current segment's basic constraints is taken as the target lateral observation distance, and the lateral safety constraints of the curve segment are taken as the lateral safety conditions of the curve. The target lateral observation distance and the lateral safety conditions of the curve are then combined into the lateral distance requirements of the curve segment.

[0132] The target observation window in the current segment's basic constraints is merged with the target alignment constraints in the curve segment to obtain the target alignment requirements for the curve segment. The avoidance and exit direction in the current segment's basic constraints is then used as the avoidance and exit requirements for the curve segment.

[0133] The curve segment height requirement, curve segment speed requirement, curve segment heading requirement, curve segment lateral distance requirement, curve segment target alignment requirement, curve segment deceleration requirement, and curve segment avoidance and exit requirement are used as curve segment fusion constraints. The curve segment fusion constraints are then correlated with the current flight state of the UAV to obtain the current inspection control state.

[0134] S3: Match the control primitives to the current inspection control state to obtain the flight control primitives. Based on the flight control primitives, calculate the control quantities for the UAV to obtain the initial flight control intent.

[0135] S3.1: Compare the drone's current altitude with the corresponding altitude requirements: When the drone's current altitude is lower than the minimum safe altitude, the altitude deviation result is determined as altitude below the safe range; when the drone's current altitude is higher than the maximum permissible altitude, the altitude deviation result is determined as altitude above the permissible range; when the drone's current altitude is between the minimum safe altitude and the maximum permissible altitude, and the absolute value of the difference between the drone's current altitude and the recommended flight altitude is greater than the minimum controllable altitude interval, the altitude deviation result is determined as altitude deviating from the recommended flight altitude; when the drone's current altitude is between the minimum safe altitude and the maximum permissible altitude, and the absolute value of the difference between the drone's current altitude and the recommended flight altitude is not greater than the minimum controllable altitude interval, the altitude deviation result is determined as altitude meeting the observation requirements.

[0136] The drone's current speed is compared with the corresponding speed requirement: when the drone's current speed is lower than the lower limit of the speed requirement, the speed deviation result is determined to be that the speed is lower than the stable inspection requirement; when the drone's current speed is higher than the upper limit of the speed requirement, the speed deviation result is determined to be that the speed exceeds the upper limit of the current inspection segment; when the drone's current speed is between the lower limit and the upper limit of the speed requirement, the speed deviation result is determined to be that the speed meets the requirements of the current inspection segment.

[0137] It should also be noted that when the current inspection segment is a curve constraint observation segment, and the current speed of the UAV is higher than the upper limit of the curve segment speed requirement, the speed deviation result is determined to be that the speed does not meet the curve deceleration requirement.

[0138] The current heading of the UAV is compared with the corresponding heading requirements: when the current heading of the UAV falls outside the allowable heading range in the heading requirements, the heading deviation result is determined to be that the heading deviates from the allowable heading range; when the current heading of the UAV falls within the allowable heading range in the heading requirements, and the current inspection segment is a straight-line stable observation segment, the heading deviation result is determined to be that the heading meets the straight-line stable observation requirements; when the current heading of the UAV falls within the allowable heading range in the heading requirements, and the current inspection segment is a curve constraint observation segment, and the current heading change of the UAV does not exceed the heading constraint of the curve segment, the heading deviation result is determined to be that the heading meets the curve constraint observation requirements; when the current inspection segment is a curve constraint observation segment, and the current heading change of the UAV exceeds the heading constraint of the curve segment, the heading deviation result is determined to be that the heading change of the curve is too fast.

[0139] It should also be noted that the current change in the drone's heading is determined by the orbital difference between the drone's current heading and the heading of the previous control cycle.

[0140] The current lateral distance of the drone is compared with the corresponding lateral distance requirement: when the current lateral distance of the drone is less than the minimum lateral safety condition in the lateral distance requirement, the lateral distance deviation result is determined to be too close; when the current lateral distance of the drone is greater than the target lateral observation distance in the lateral distance requirement, and the difference between the current lateral distance of the drone and the target lateral observation distance is greater than the safe lateral margin, the lateral distance deviation result is determined to be too far; when the current lateral distance of the drone is not less than the minimum lateral safety condition, and the difference between the current lateral distance of the drone and the target lateral observation distance is not greater than the safe lateral margin, the lateral distance deviation result is determined to be that the lateral distance meets the requirements of the current inspection segment.

[0141] The target observation surface is compared with the corresponding target alignment requirements: when the projection center of the target observation surface in the UAV imaging plane falls outside the target observation window, the target alignment deviation result is determined to be that the target deviates from the target observation window; when the projection boundary of the target observation surface in the UAV imaging plane does not fall into the target observation window, the target alignment deviation result is determined to be that the target coverage is insufficient; when the continuous observation time of the target observation surface within the target observation window is less than the minimum stable observation time, the target alignment deviation result is determined to be that the observation time is insufficient; when the projection center of the target observation surface in the UAV imaging plane falls into the target observation window, the projection boundary falls into the target observation window, and the continuous observation time is not less than the minimum stable observation time, the target alignment deviation result is determined to be that the target alignment meets the requirements.

[0142] The remaining return-to-home margin of the drone is compared with the obstacle avoidance exit requirement: when the remaining return-to-home margin of the drone is negative, the exit trigger result is determined to be return-to-home exit trigger; when the remaining return-to-home margin of the drone is non-negative, and the spatial distance between the drone's current position and the obstacle boundary is less than the safe lateral margin, the exit trigger result is determined to be obstacle avoidance exit trigger; when the remaining return-to-home margin of the drone is non-negative, and the distance between the drone's current position and the no-entry boundary is less than the safe lateral margin, the exit trigger result is determined to be obstacle avoidance exit trigger; when the remaining return-to-home margin of the drone is non-negative, and the difference between the drone's current altitude and the low-altitude limit boundary is less than the safe altitude margin, the exit trigger result is determined to be obstacle avoidance exit trigger; when the remaining return-to-home margin of the drone is non-negative, and the lateral distance deviation result is that the lateral distance is too close, the exit trigger result is determined to be obstacle avoidance direction trigger.

[0143] When the drone's current remaining return-to-home margin is non-negative, the spatial distance between the drone's current position and the obstacle boundary is not less than the safe lateral margin, the distance between the drone's current position and the no-entry boundary is not less than the safe lateral margin, the difference between the drone's current altitude and the low-altitude limit boundary is not less than the safe altitude margin, and the lateral distance deviation result is not too close to the lateral distance, the exit trigger result is determined to be no exit trigger required.

[0144] S3.2: The control primitive prediction model consists of an input encoding layer, a feature concatenation layer, a first fully connected layer, a second fully connected layer, and a probability output layer connected in sequence.

[0145] Historical low-altitude UAV inspection flight records were collected and segmented according to the control cycle to obtain training samples.

[0146] Each training sample includes training input features and historical flight control primitives.

[0147] The training input features include historical altitude deviation results, historical speed deviation results, historical heading deviation results, historical lateral distance deviation results, historical target alignment deviation results, historical exit trigger results, historical heading change, and historical current inspection progress.

[0148] Historical flight control primitives include straight-line stability tracking primitives, curve deceleration and steering primitives, local safety observation and adjustment primitives, and exit and return-to-base primitives.

[0149] The input coding layer performs one-hot encoding on the historical altitude deviation results, historical speed deviation results, historical heading deviation results, historical lateral distance deviation results, historical target alignment deviation results, and historical exit trigger results to obtain historical altitude deviation codes, historical speed deviation codes, historical heading deviation codes, historical lateral distance deviation codes, historical target alignment deviation codes, historical exit trigger codes, and historical segment type codes.

[0150] One-hot encoding is performed on historical flight control primitives to obtain label vectors.

[0151] The historical course change and the historical current inspection progress are respectively min-max normalized to obtain the normalized historical course change and the normalized historical current inspection progress.

[0152] The historical altitude deviation code, historical speed deviation code, historical heading deviation code, historical lateral distance deviation code, historical target alignment deviation code, historical exit trigger code, historical segment type code, normalized historical heading change, and normalized historical current inspection progress are concatenated into a training feature vector.

[0153] The first fully connected layer performs a linear transformation on the training feature vector to obtain the first training vector, expressed as:

[0154] ;

[0155] in, As the first training vector, This is the weight matrix of the first fully connected layer. To train feature vectors, This is the bias term for the first fully connected layer.

[0156] The first training vector is substituted into the ReLU function for calculation to obtain the first hidden training vector.

[0157] The second fully connected layer performs a linear transformation on the first hidden training vector to obtain the second training vector, expressed as:

[0158] ;

[0159] in, This is the second training vector. This is the weight matrix of the second fully connected layer. This is the first hidden training vector. This is the bias term for the second fully connected layer.

[0160] The second training vector is substituted into the ReLU function for calculation to obtain the second hidden training vector.

[0161] The probabilistic output layer performs a linear mapping on the second hidden training vector to obtain the candidate flight control primitive output score vector, expressed as:

[0162] ;

[0163] in, Output a score vector for the candidate flight control primitives. The weight matrix of the probability output layer. This is the second hidden training vector. This is the bias term for the probability output layer.

[0164] Softmax normalization is applied to each element in the output score vector of the candidate flight control primitives to obtain the training matching probability of each primitive.

[0165] S3.3: Concatenate the training matching probabilities of each primitive into a training matching probability vector, calculate the cross-entropy loss between the training matching probability vector and the label vector, and calculate the partial derivative of the cross-entropy loss with respect to the parameters of the control primitive prediction model, which is used as the gradient of the parameters of the control primitive prediction model.

[0166] The parameters of the control primitive prediction model are updated using gradient descent based on the gradient of the control primitive prediction model parameters to obtain a pre-trained control primitive prediction model.

[0167] The parameters of the control primitive prediction model include the weight matrix of the first fully connected layer, the bias term of the first fully connected layer, the weight matrix of the second fully connected layer, the bias term of the second fully connected layer, the weight matrix of the probability output layer, and the bias term of the probability output layer.

[0168] The altitude deviation result, speed deviation result, heading deviation result, lateral distance deviation result, target alignment deviation result, exit trigger result, current inspection segment segment type, heading change amount, and current inspection progress are input into the pre-trained control primitive prediction model, and the matching probability of each primitive is calculated according to step S3.2.

[0169] The primitive with the highest matching probability among all primitives is selected as the flight control primitive.

[0170] S3.4: When the flight control primitive is a straight-line stability tracking primitive, the difference between the current speed of the UAV and the target speed required for the straight segment is used as the speed adjustment quantity, the difference between the current heading of the UAV and the corresponding reference heading is used as the heading correction quantity, and the difference between the current lateral distance of the UAV and the target lateral observation distance required for the straight segment is used as the lateral distance adjustment quantity. The speed adjustment quantity, heading correction quantity, and lateral distance adjustment quantity are collectively referred to as straight-line stability control quantities.

[0171] It should also be noted that the target speed of the straight segment in the speed requirement of the straight segment is the average of the lower limit and the upper limit of the speed requirement of the straight segment.

[0172] When the flight control primitive is a cornering deceleration and steering primitive, the difference between the UAV's current speed and the target speed in the cornering segment speed requirement is used as the speed adjustment amount. The average of the upper and lower heading limits in the cornering segment heading requirement is used as the target heading in the cornering segment. The difference between the UAV's current heading and the target heading in the cornering segment is used as the heading correction amount. When the UAV's current attitude exceeds the maximum attitude angle, the difference between the maximum attitude angle and the UAV's current attitude is used as the attitude compensation amount. The difference between the UAV's current lateral distance and the target lateral observation distance in the cornering segment lateral distance requirement is used as the lateral distance adjustment amount. The speed adjustment amount, heading correction amount, attitude compensation amount, and lateral distance adjustment amount are collectively referred to as cornering constraint control amounts.

[0173] When the flight control primitive is a local safety observation adjustment primitive, the difference between the UAV's current altitude and the recommended flight altitude in the altitude requirement is used as the altitude adjustment amount; the difference between the UAV's current speed and the target speed in the speed requirement is used as the speed adjustment amount; the difference between the UAV's current heading and the reference heading corresponding to the heading requirement is used as the heading correction amount; when the lateral distance deviation result is that the lateral distance is too far, the difference between the UAV's current lateral distance and the target lateral observation distance in the lateral distance requirement is used as the lateral distance adjustment amount; when the lateral distance deviation result is that the lateral distance is too close, the difference between the UAV's current lateral distance and the minimum lateral safety condition in the lateral distance requirement is used as the lateral distance adjustment amount; and the difference between the UAV's current attitude and the maximum attitude angle is used as the attitude compensation amount.

[0174] It should also be noted that when the current inspection segment is a curve constraint observation segment, the target speed in the speed requirement is the target speed of the curve segment; when the current inspection segment is a straight stability observation segment, the target speed in the speed requirement is the target speed of the straight segment.

[0175] The initial value of hovering time is set to zero. When the target alignment deviation result is insufficient observation time, the difference between the minimum stable observation time and the continuous observation time is used as the candidate hovering time. The minimum value between the candidate hovering time and the maximum hovering time is used as the hovering time. The altitude adjustment, speed adjustment, heading correction, lateral distance adjustment, attitude compensation, and hovering time are collectively referred to as local observation control variables.

[0176] When the flight control primitive is the exit / return primitive, the exit direction in the exit requirement is taken as the exit heading for the UAV to leave the current inspection segment, the direction between the UAV's current position and the UAV's take-off and landing point is taken as the return direction, the difference between the UAV's current speed and the lower limit of the allowed speed range is taken as the return speed adjustment amount, and the difference between the UAV's current altitude and the recommended flight altitude is taken as the return altitude adjustment amount. The exit heading, return direction, return speed adjustment amount, and return altitude adjustment amount are collectively referred to as the return control quantities.

[0177] When the flight control primitive is a straight-line stability tracking primitive, the straight-line stability control quantities are sorted in the order of heading correction, speed adjustment, and lateral distance adjustment, and then bound to the starting observation anchor point, ending observation anchor point, and control cycle of the current inspection segment to obtain the initial flight control intent.

[0178] When the flight control primitive is a curve deceleration and steering primitive, the curve constraint control quantities are sorted in the order of speed adjustment, heading correction, attitude compensation, and lateral distance adjustment, and then bound to the starting observation anchor point, ending observation anchor point, and control cycle of the current inspection segment to obtain the initial flight control intent.

[0179] When the flight control primitive is a local safety observation adjustment primitive, and the altitude deviation result is that the altitude is below the safe range or above the allowable range, the local observation control quantities are sorted in the order of altitude adjustment, lateral distance adjustment, heading correction, speed adjustment, attitude compensation, and hovering time, and then bound to the starting observation anchor point, ending observation anchor point, and control cycle of the current inspection segment to obtain the initial flight control intent.

[0180] When the flight control primitive is a local safety observation adjustment primitive, and the lateral distance deviation result is that the lateral distance is too close, the local observation control quantities are sorted in the order of lateral distance adjustment, heading correction, speed adjustment, altitude adjustment, attitude compensation, and hovering time, and then bound to the starting observation anchor point, ending observation anchor point, and control cycle of the current inspection segment to obtain the initial flight control intent.

[0181] When the flight control primitive is a local safety observation adjustment primitive, and the target alignment deviation result is that the target deviates from the target observation window, the target coverage is insufficient, or the observation time is insufficient, the local observation control quantities are sorted in the order of heading correction, lateral distance adjustment, altitude adjustment, speed adjustment, attitude compensation, and hovering time, and then bound to the starting observation anchor point, ending observation anchor point, and control cycle of the current inspection segment to obtain the initial flight control intent.

[0182] When the flight control primitive is the exit return primitive, the return control quantities are sorted in the order of exit heading, return direction, return altitude adjustment, and return speed adjustment, and then bound to the starting observation anchor point, ending observation anchor point, and control cycle of the current inspection segment to obtain the initial flight control intent.

[0183] S4: Perform flight state prediction on the initial flight control intention to obtain the flight state prediction result. Calculate the safety certificate value based on the flight state prediction result. Perform minimum constraint correction on the initial flight control intention based on the safety certificate value to obtain the safe flight control package.

[0184] S4.1: The altitude, speed, heading, attitude, and lateral distance of the UAV in its current flight state are superimposed element by element with the altitude adjustment, speed adjustment, heading correction, attitude compensation, and lateral distance adjustment in the initial flight control intention to obtain the predicted altitude, predicted speed, predicted heading, predicted attitude, and predicted lateral distance for the next control cycle, which are used as the flight state prediction results.

[0185] It should also be noted that the altitude adjustment, speed adjustment, heading correction, attitude compensation, and lateral distance adjustment in the initial flight control intent are constrained correction increments.

[0186] Calculate the difference between the predicted height and the minimum safe height in the height requirement to obtain the lower boundary margin of height. Calculate the difference between the maximum allowable height in the height requirement and the predicted height to obtain the upper boundary margin of height. The minimum value between the lower boundary margin and the upper boundary margin of height is taken as the height safety margin.

[0187] The average of the current speed and the predicted speed is used as the average speed of the prediction period, and the intermediate heading between the current heading and the predicted heading is used as the average heading of the prediction period. The horizontal displacement is calculated using the average speed of the prediction period, the average heading of the prediction period, and the control period, and then the horizontal displacement is superimposed on the current position of the UAV to obtain the predicted position of the UAV in the next control period.

[0188] Collect the safe envelope radius of the UAV body, and determine the predicted body envelope with the predicted position as the center and the safe envelope radius of the UAV body as the outline.

[0189] Calculate the minimum spatial distance between the predicted body envelope and the obstacle boundary to obtain the obstacle boundary safety margin; calculate the minimum spatial distance between the predicted body envelope and the restricted boundary to obtain the restricted boundary safety margin; calculate the minimum spatial distance between the predicted body envelope and the outer boundary of the object to be inspected to obtain the target outer boundary safety margin.

[0190] The difference between the predicted lateral distance and the minimum lateral safety condition in the lateral distance requirement is calculated to obtain the lateral safety margin. The difference between the low-altitude height limit boundary and the predicted height is calculated to obtain the low-altitude height limit margin. The minimum value among the obstacle boundary safety margin, the no-entry boundary safety margin, the target shape boundary safety margin, the lateral safety margin, and the low-altitude height limit margin is taken as the spatial boundary safety margin.

[0191] Calculate the difference between the predicted speed and the lower speed limit in the speed requirement to obtain the lower speed margin. Calculate the difference between the upper speed limit in the speed requirement and the predicted speed to obtain the upper speed margin. Calculate the minimum difference between the predicted heading and the boundary of the allowable heading range in the heading requirement to obtain the heading safety margin. Calculate the difference between the maximum attitude angle and the predicted attitude to obtain the attitude safety margin. The minimum value between the lower speed margin and the upper speed margin is taken as the speed safety margin. The speed safety margin, heading safety margin, and attitude safety margin are taken as the motion safety margin.

[0192] When the current inspection segment is a curve constraint observation segment, the difference between the upper limit of the curve segment speed requirement and the predicted speed is calculated to obtain the curve speed margin. The product of the maximum yaw rate and the control period is used as the upper limit of the single-cycle heading change of the curve. The difference between the upper limit of the single-cycle heading change of the curve and the predicted heading change is calculated to obtain the curve heading change margin. The difference between the predicted lateral distance and the curve lateral safety condition in the curve segment lateral distance requirement is calculated to obtain the curve lateral safety margin. The curve speed margin, curve heading change margin, and curve lateral safety margin are used as the curve flight safety margin.

[0193] When the current inspection segment is not a curve constraint observation segment, the curve flight safety margin is not calculated, and the curve flight safety judgment result is marked as inapplicable, but is regarded as safe in the safety execution judgment.

[0194] The height safety margin, space boundary safety margin, motion safety margin, and curve flight safety margin are combined into a single safety certificate value.

[0195] S4.2: The product of the maximum lifting speed and the control cycle is used as the maximum height change in a single cycle, and the maximum value among the safety height margin, the minimum controllable height interval, and the maximum height change in a single cycle is used as the height safety margin threshold.

[0196] The product of the maximum inspection speed and the control cycle is taken as the maximum horizontal displacement in a single cycle, and the maximum value among the safety lateral margin, the maximum horizontal displacement in a single cycle, and the height safety margin threshold is taken as the spatial boundary safety margin threshold.

[0197] It should also be noted that the safety altitude margin is used to compensate for altitude control errors, positioning errors, wind drift disturbances, and obstacle boundary measurement errors; the minimum controllable altitude interval is used to characterize the minimum altitude adjustment resolution that the UAV flight control can perform; and the maximum altitude change in a single cycle is used to cover the maximum vertical displacement that the UAV may produce in the next control cycle.

[0198] Since the height safety margin threshold needs to be no less than the safety height margin, no less than the minimum controllable height interval, and no less than the maximum height change in a single cycle, the maximum value among the safety height margin, the minimum controllable height interval, and the maximum height change in a single cycle is used as the height safety margin threshold.

[0199] The product of the maximum available horizontal deceleration and the control period is used as the speed safety margin threshold, and the product of the maximum yaw rate and the control period is used as the heading safety margin threshold.

[0200] Historical pitch and roll angles of the UAV during low-altitude inspection test flights are collected. The average historical pitch angle is used as the historical pitch reference angle, and the average historical roll angle is used as the historical roll reference angle. The absolute value of the difference between the historical pitch angle and the historical pitch reference angle for each control cycle is calculated to obtain the historical pitch angle fluctuation value. The absolute value of the difference between the historical roll angle and the historical roll reference angle for each control cycle is calculated to obtain the historical roll angle fluctuation value. The maximum value between the historical pitch angle fluctuation value and the historical roll angle fluctuation value is used as the attitude safety margin threshold.

[0201] Speed ​​safety margin threshold, heading safety margin threshold, and attitude safety margin threshold are collectively referred to as motion safety margin threshold.

[0202] The product of the maximum available horizontal deceleration and the control period is used as the cornering speed margin threshold, the product of the maximum yaw rate and the control period is used as the cornering heading change margin threshold, and the safe lateral margin is used as the cornering lateral safety margin threshold.

[0203] The cornering speed margin threshold, the cornering heading change margin threshold, and the cornering lateral safety margin threshold are collectively referred to as the cornering flight safety margin threshold.

[0204] When the height safety margin is not less than the height safety margin threshold, the height safety judgment result is determined to be safe; when the height safety margin is less than the height safety margin threshold, the height safety judgment result is determined to be unsafe.

[0205] When the safety margin of the spatial boundary is not less than the preset safety margin threshold, the spatial boundary is determined to be safe; when the safety margin of the spatial boundary is less than the preset safety margin threshold, the spatial boundary is determined to be unsafe.

[0206] When the speed safety margin is not less than the speed safety margin threshold, the heading safety margin is not less than the heading safety margin threshold, and the attitude safety margin is not less than the attitude safety margin threshold, the motion safety judgment result is determined to be safe.

[0207] When the speed safety margin is less than the speed safety margin threshold, the heading safety margin is less than the heading safety margin threshold, or the attitude safety margin is less than the attitude safety margin threshold, the motion safety judgment result is determined to be unsafe.

[0208] When the current inspection segment is a curve constraint observation segment, and the curve speed margin is not less than the curve speed margin threshold, the curve heading change margin is not less than the curve heading change margin threshold, and the curve lateral safety margin is not less than the curve lateral safety margin threshold, the curve flight safety judgment result is determined to be safe.

[0209] When the current inspection segment is a curve constraint observation segment, and the curve speed margin is less than the curve speed margin threshold, the curve heading change margin is less than the curve heading change margin threshold, or the curve lateral safety margin is less than the curve lateral safety margin threshold, the curve flight safety judgment result is determined to be unsafe.

[0210] When the altitude safety assessment result, the spatial boundary safety assessment result, the motion safety assessment result, and the curve flight safety assessment result are all safe, the safety execution assessment result is determined to be safe, and the initial flight control intention is taken as the safe flight control package.

[0211] When any one of the altitude safety judgment result, spatial boundary safety judgment result, motion safety judgment result, and curve flight safety judgment result is unsafe, the safety execution judgment result is determined to be unsafe, and the safety margin type that does not meet the corresponding safety margin threshold in the safety certificate value is marked to obtain the safety margin type to be corrected.

[0212] S4.3: When the type of safety margin to be corrected includes altitude safety margin, if the predicted altitude is lower than the minimum safe altitude in the altitude requirements, the difference between the minimum safe altitude and the current altitude of the UAV shall be used as the corrected altitude adjustment amount. If the predicted altitude is higher than the maximum allowable altitude in the altitude requirements, the difference between the maximum allowable altitude and the current altitude of the UAV shall be used as the corrected altitude adjustment amount.

[0213] When the type of safety margin to be corrected includes motion safety margin, if the predicted speed is lower than the lower speed limit in the speed requirement, the difference between the lower speed limit and the current speed of the UAV is used as the corrected speed adjustment amount. If the predicted speed is higher than the upper speed limit in the speed requirement, the difference between the upper speed limit and the current speed of the UAV is used as the corrected speed adjustment amount. If the predicted heading exceeds the allowable heading range in the heading requirement, the minimum difference between the allowable heading range and the current heading of the UAV is used as the corrected heading correction amount. If the predicted attitude exceeds the maximum attitude angle, the difference between the maximum attitude angle and the current attitude of the UAV is used as the corrected attitude compensation amount.

[0214] When the type of safety margin to be corrected includes spatial boundary safety margin, and the predicted lateral distance is less than the minimum lateral safety condition in the lateral distance requirement, the difference between the minimum lateral safety condition and the current lateral distance of the UAV is used as the corrected lateral distance adjustment amount.

[0215] When the type of safety margin to be corrected includes curve flight safety margin, and the predicted speed is higher than the upper limit of the curve segment speed requirement, the difference between the upper limit of the curve segment speed and the current speed of the UAV is used as the corrected speed adjustment amount. If the predicted heading change exceeds the upper limit of the single-cycle heading change in the curve segment heading requirement, the difference between the corresponding curve segment target heading and the current heading of the UAV is used as the corrected heading correction amount. If the predicted lateral distance is less than the curve lateral safety condition in the curve segment lateral distance requirement, the difference between the curve lateral safety condition and the current lateral distance of the UAV is used as the corrected lateral distance adjustment amount.

[0216] The corrected altitude adjustment, speed adjustment, heading correction, lateral distance adjustment, and attitude compensation are replaced with the initial flight control intent to obtain the complete flight control intent. The complete flight control intent is then bound to the starting observation anchor point, ending observation anchor point, and control cycle of the current inspection segment to obtain the safe flight control package.

[0217] S5: Obtain the actual flight status, monitor the response deviation between the safe flight control package and the actual flight status, and reconstruct the link of the current inspection segment based on the response deviation and the actual flight status to obtain the updated straight and curved section inspection control chain.

[0218] S5.1: At the end of the control cycle, collect the UAV's position, altitude, speed, heading, attitude, lateral distance, remaining return margin, continuous observation time, and the projection center and projection boundary of the target observation surface in the UAV's imaging plane as the actual flight status.

[0219] The altitude adjustment, speed adjustment, heading correction, attitude compensation, lateral distance adjustment, and hovering time in the safe flight control package are added element by element to the altitude, speed, heading, attitude, lateral distance, and continuous observation time of the UAV before executing the safe flight control package, respectively, to obtain the target control altitude, target control speed, target control heading, target control attitude, target control lateral distance, and target control hovering time corresponding to the safe flight control package.

[0220] It should also be noted that when the safety flight control package does not include hover time, the hover time should be set to zero.

[0221] The difference between the altitude in actual flight and the target control altitude is taken as the altitude response deviation; the difference between the speed in actual flight and the target control speed is taken as the speed response deviation; the difference between the heading in actual flight and the target control heading is taken as the heading response deviation; the difference between the attitude in actual flight and the target control attitude is taken as the attitude response deviation; the difference between the lateral distance in actual flight and the target control lateral distance is taken as the lateral distance response deviation; and the difference between the continuous observation time in actual flight and the target control hovering time is taken as the hovering response deviation.

[0222] Altitude response deviation, speed response deviation, heading response deviation, attitude response deviation, lateral distance response deviation, and hovering response deviation are considered as the response deviations between the safe flight control package and the actual flight conditions.

[0223] S5.2: The difference between the target control height and the minimum safe height and the maximum allowable height shall be used as the lower boundary margin and the upper boundary margin of the target height, respectively. The minimum value among the minimum controllable height interval, the maximum height change in a single period, the lower boundary margin and the upper boundary margin of the target height shall be used as the allowable height response deviation.

[0224] The difference between the target control speed and the lower and upper speed limits in the speed requirements are respectively taken as the lower and upper boundary margins of the target speed. The minimum value among the maximum speed change in a single cycle, the lower and upper boundary margins of the target speed is taken as the allowable speed response deviation.

[0225] The minimum difference between the target control heading and the boundary of the allowable heading range in the heading requirement is taken as the target heading boundary margin, and the minimum of the maximum heading adjustment angle in a single cycle and the target heading boundary margin is taken as the heading allowable response deviation.

[0226] The absolute value of the difference between the target control attitude and the maximum attitude angle is taken as the allowable attitude response deviation.

[0227] It should also be noted that when the target control attitude includes the target control pitch angle and the target control roll angle, the absolute value of the difference between the maximum attitude angle and the target control pitch angle is calculated as the pitch angle allowable deviation, and the absolute value of the difference between the maximum attitude angle and the target control roll angle is calculated as the roll angle allowable deviation. The minimum value between the pitch angle allowable deviation and the roll angle allowable deviation is taken as the attitude allowable response deviation.

[0228] The differences between the target control lateral distance and the minimum lateral safety condition, the minimum stable observation distance, and the maximum stable observation distance are respectively taken as the target lateral safety margin, the target lateral lower boundary margin, and the target lateral upper boundary margin. The minimum value among the safety lateral margin, the maximum horizontal displacement in a single cycle, the target lateral safety margin, the target lateral lower boundary margin, and the target lateral upper boundary margin is taken as the allowable lateral distance response deviation.

[0229] The difference between the maximum hovering time and the target hovering time is taken as the upper boundary margin of hovering, the difference between the target control hovering time and the minimum stable observation time is taken as the stable observation margin, and the minimum value between the upper boundary margin of hovering and the stable observation margin is taken as the hovering allowable response deviation.

[0230] The allowable response deviations for altitude, speed, heading, attitude, lateral distance, and hovering are set as preset allowable response deviation ranges.

[0231] S5.3: When the absolute value of the altitude response deviation exceeds the permissible altitude response deviation, the result is determined as altitude response deviation exceeding the limit. When the absolute value of the speed response deviation exceeds the permissible speed response deviation, the result is determined as speed response deviation exceeding the limit. When the absolute value of the heading response deviation exceeds the permissible heading response deviation, the result is determined as heading response deviation exceeding the limit. When the absolute value of the attitude response deviation exceeds the permissible attitude response deviation, the result is determined as attitude response deviation exceeding the limit. When the absolute value of the lateral distance response deviation exceeds the permissible lateral distance response deviation, the result is determined as lateral distance response deviation exceeding the limit. When the absolute value of the hovering response deviation exceeds the permissible hovering response deviation, the result is determined as hovering response deviation exceeding the limit.

[0232] When the absolute value of the altitude response deviation does not exceed the permissible altitude response deviation, the absolute value of the speed response deviation does not exceed the permissible speed response deviation, the absolute value of the heading response deviation does not exceed the permissible heading response deviation, the absolute value of the attitude response deviation does not exceed the permissible attitude response deviation, the absolute value of the lateral distance response deviation does not exceed the permissible lateral distance response deviation, and the absolute value of the hovering response deviation does not exceed the permissible hovering response deviation, the result of exceeding the response deviation limit is determined as "response deviation not exceeding the limit".

[0233] When the actual flight altitude is lower than the minimum safe altitude required by the current inspection and control status, the actual flight altitude is determined to be below the safe range.

[0234] When the actual flight altitude is higher than the maximum allowable altitude required by the current inspection control status, the actual flight altitude is determined to be higher than the allowable range.

[0235] When the speed in the actual flight state is lower than the lower limit of the speed required in the current inspection control state, the actual flight state is judged to have exceeded the result of speed being lower than the stable inspection requirement.

[0236] When the speed in the actual flight state is higher than the speed limit required in the current inspection control state, the result of the actual flight state is determined to be that the speed exceeds the speed limit of the current inspection segment.

[0237] When the heading in the actual flight state falls outside the allowable heading range required by the heading in the current inspection control state, the actual flight state is judged to have deviated from the allowable heading range.

[0238] When the lateral distance in the actual flight state is less than the minimum lateral safety condition required by the lateral distance in the current inspection control state, the actual flight state is judged to have exceeded the result as the lateral distance being too close.

[0239] When the lateral distance in the actual flight state is greater than the target lateral observation distance required by the current inspection control state, and the difference between the lateral distance in the actual flight state and the target lateral observation distance is greater than the safe lateral margin, the actual flight state is judged to have exceeded the result as the lateral distance is too far.

[0240] When the projection center of the target observation surface on the UAV's imaging plane falls outside the target observation window required for target alignment in the current inspection control state, the actual flight state is determined to be outside the target observation window.

[0241] When the projection boundary of the target observation surface on the UAV imaging plane does not fall completely into the target observation window required for target alignment in the current inspection control state, the actual flight state is judged to exceed the result as insufficient target coverage.

[0242] When the continuous observation time of the target observation surface within the target observation window required for target alignment in the current inspection control state is less than the minimum stable observation time, the actual flight state is judged to exceed the result as insufficient observation time.

[0243] S5.4: When the response deviation exceeds the preset allowable response deviation range for the first time, and the actual flight status exceeds any of the following: altitude below the safe range, altitude above the allowable range, speed below the stable inspection requirements, speed exceeding the speed limit of the current inspection segment, heading deviating from the allowable heading range, lateral distance too close, lateral distance too far, target deviating from the target observation window, insufficient target coverage, or insufficient observation time, the position of the UAV in the corresponding control cycle will be regarded as an abnormal position.

[0244] Using the abnormal position as a reference, when the altitude in the actual flight state is lower than the minimum safe altitude in the altitude requirements, the minimum safe altitude is used as the reentry altitude; when the altitude in the actual flight state is higher than the maximum permissible altitude in the altitude requirements, the maximum permissible altitude is used as the reentry altitude; when the altitude in the actual flight state is between the minimum safe altitude and the maximum permissible altitude, the recommended flight altitude is used as the reentry altitude.

[0245] When the current inspection segment is a straight stable observation segment, the target speed in the speed requirement of the straight segment is used as the reentry speed. When the current inspection segment is a curve constraint observation segment, the target speed of the curve segment in the speed requirement of the curve segment is used as the reentry speed. When the speed response deviation exceeds the speed allowable response deviation, the lower limit of the speed requirement is used as the reentry speed.

[0246] When the current inspection segment is a straight-line stable observation segment, the reference heading corresponding to the heading requirement of the straight segment is used as the re-entry heading. When the current inspection segment is a curve constraint observation segment, the average of the upper and lower heading limits in the heading requirement of the curve segment is used as the re-entry heading. When the actual flight state triggers the avoidance exit requirement, the avoidance exit direction is used as the re-entry heading.

[0247] When the lateral distance in the actual flight state is less than the minimum lateral safety condition in the lateral distance requirement, the minimum lateral safety condition is used as the re-entry lateral distance. When the lateral distance in the actual flight state is greater than the target lateral observation distance in the lateral distance requirement, and the difference between the lateral distance in the actual flight state and the target lateral observation distance is greater than the safe lateral margin, the target lateral observation distance is used as the re-entry lateral distance. When the lateral distance in the actual flight state is not less than the minimum lateral safety condition in the current inspection control state, and the absolute value of the difference between the lateral distance in the actual flight state and the target lateral observation distance is not greater than the safe lateral margin, the target lateral observation distance is used as the re-entry lateral distance.

[0248] When the projection center of the target observation surface in the UAV imaging plane falls outside the target observation window, or the projection boundary of the target observation surface in the UAV imaging plane does not fall into the target observation window, the difference between the minimum stable observation time and the continuous observation time in the actual flight state is used as the candidate re-entry hovering time, and the minimum value between the candidate re-entry hovering time and the maximum hovering time is used as the re-entry hovering time. When the target observation surface meets the target alignment requirements, the re-entry hovering time is set to zero.

[0249] By segmenting the abnormal location, reentry altitude, reentry speed, reentry heading, reentry lateral distance, and reentry hovering time, a local reentry control segment is formed.

[0250] Based on the abnormal location, the current inspection segment is divided into executed segments and remaining inspection segments. After the abnormal location, the local re-entry control segment is inserted. The local re-entry control segment, the remaining inspection segment, and the anchor point connection segment after the current inspection segment are re-associated according to the connection order of the anchor point chain to obtain the updated straight and curved segment inspection control chain.

[0251] This embodiment also provides an artificial intelligence-based intelligent inspection system for low-altitude unmanned aerial vehicles, including:

[0252] The configuration module obtains the spatial observation information of the object to be inspected, performs hierarchical reduction on the spatial observation information, and obtains the observation zone and anchor point chain.

[0253] The fusion module performs heading discrimination and segment reduction on the observation zone and anchor point chain to obtain the straight and curved section inspection control chain. It then performs segment assignment matching on the straight and curved section inspection control chain to obtain the current inspection segment. Finally, it performs constraint fusion on the current inspection segment to obtain the current inspection control state.

[0254] The calculation module performs control primitive matching on the current inspection control state to obtain flight control primitives. Based on the flight control primitives, it performs control quantity calculation on the UAV to obtain the initial flight control intent.

[0255] The correction module predicts the flight state of the initial flight control intention, obtains the flight state prediction result, calculates the safety certificate value based on the flight state prediction result, and performs minimum constraint correction on the initial flight control intention based on the safety certificate value to obtain the safe flight control package.

[0256] The reconstruction module acquires the actual flight status, monitors the response deviation between the safe flight control package and the actual flight status, and reconstructs the link of the current inspection segment based on the response deviation and the actual flight status to obtain the updated straight and curved section inspection control chain.

[0257] In summary, this invention constructs observation zones and anchor point chains using spatial observation information, enabling the target observation surface, flight altitude, lateral distance, heading range, speed range, and avoidance direction to be refined into specific inspection segments. This avoids the problem of uniform routes or fixed waypoints being difficult to adapt to different observation levels and differences in straight and curved sections, improving the continuity of low-altitude inspections and the targeted nature of flight control. Furthermore, by using flight status prediction and safety certificate values ​​to perform minimum constraint correction on control intentions, safety constraints, control intentions, and actual flight responses can form a continuous verification and feedback update relationship before and after control execution. This avoids the problem of safety control remaining at the level of local obstacle avoidance or control quantity limitation, improving closed-loop safety, anomaly recovery capability, and continuous execution capability of inspection tasks in complex low-altitude environments, and achieving a collaborative closed loop of safety constraints, control intentions, and actual flight.

[0258] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for intelligent inspection of low-altitude unmanned aerial vehicles based on artificial intelligence, characterized in that, include: The spatial observation information of the object to be inspected is obtained, and the spatial observation information is layered and reduced to obtain the observation zone and anchor point chain. The specific steps are as follows: The spatial observation information includes the location, altitude, shape boundary, target observation surface, low-altitude altitude limit boundary, obstacle boundary, no-entry boundary, UAV take-off and landing point, and UAV flight capability parameters of the object to be inspected; Based on the altitude of the object to be inspected and the target observation surface, the object to be inspected is divided into low-altitude observation layers. Recommended flight altitude, minimum safe altitude, maximum permissible altitude, recommended lateral observation distance, target observation window, permissible heading range, permissible speed range, and avoidance exit direction are configured for each low-altitude observation layer to obtain the observation zone. Sampling is performed along the outer boundary of each object to be inspected to obtain observation anchor points, and the observation anchor points are connected sequentially to form an anchor point chain; The observation zone and anchor point chain are subjected to heading discrimination and segment reduction to obtain the straight and curved section inspection control chain. The segment assignment matching of the straight and curved section inspection control chain is performed to obtain the current inspection segment. The current inspection segment is constrained and fused to obtain the current inspection control state. The current inspection control state is matched with control primitives to obtain flight control primitives. Based on the flight control primitives, the control quantities of the UAV are calculated to obtain the initial flight control intent. The initial flight control intention is predicted to obtain the flight state prediction result. The safety certificate value is calculated based on the flight state prediction result. The initial flight control intention is then modified with minimum constraints based on the safety certificate value to obtain the safe flight control package. The actual flight status is obtained, the response deviation between the safe flight control package and the actual flight status is monitored, and the link is reconstructed for the current inspection segment based on the response deviation and the actual flight status to obtain the updated straight and curved section inspection control chain.

2. The intelligent inspection method for low-altitude unmanned aerial vehicles based on artificial intelligence as described in claim 1, characterized in that, The specific steps for determining the course and assigning segments are as follows: Calculate the heading angles of adjacent observation anchor points in the anchor chain to obtain the anchor chain heading angle sequence, and use the anchor chain heading angle sequence to calculate the heading change between adjacent anchor chain heading angles; When the change in heading is less than the preset steering threshold, the corresponding anchor point connection segment is marked as a straight stable observation segment, and straight segment target alignment constraint, straight segment speed constraint, straight segment heading constraint and straight segment lateral safety constraint are configured for the straight stable observation segment to obtain the straight segment control constraint; When the change in heading is not less than the preset steering threshold, the corresponding anchor point connection segment is marked as the curve constraint observation segment, and curve constraint observation segment is configured with curve segment target alignment constraint, curve segment speed constraint, curve segment heading constraint, curve segment lateral safety constraint and curve segment deceleration constraint to obtain the curve segment control constraint. According to the connection sequence of the anchor point chain, the straight section stability observation segment, the curve constraint observation segment, the straight section control constraint and the curve section control constraint are associated to obtain the straight-curved section inspection control chain.

3. The intelligent inspection method for low-altitude unmanned aerial vehicles based on artificial intelligence as described in claim 1, characterized in that, The specific steps for constraining and fusing the current inspection segment are as follows: The current position, altitude, speed, heading, attitude, lateral distance, remaining return margin, and current inspection progress of the drone are obtained as the current flight status of the drone; The observation segment corresponding to the current inspection progress is taken as the current inspection segment; The recommended flight altitude, minimum safe altitude, maximum permissible altitude, recommended lateral observation distance, target observation window, permissible heading range, permissible speed range, and avoidance exit direction corresponding to the current inspection segment are used as the basic constraints for the current segment. When the current inspection segment is a straight and stable observation segment, the basic constraints of the current segment are fused with the corresponding straight segment control constraints to obtain the current inspection control state; When the current inspection segment is a curve constraint observation segment, the basic constraints of the current segment are fused with the corresponding curve segment control constraints to obtain the current inspection control state.

4. The intelligent inspection method for low-altitude unmanned aerial vehicles based on artificial intelligence as described in claim 1, characterized in that, The specific steps for performing control primitive matching are as follows: The current flight status of the UAV is compared with the current inspection control status to obtain the altitude deviation, speed deviation, heading deviation, lateral distance deviation, target alignment deviation, and exit trigger results. The altitude deviation, speed deviation, heading deviation, lateral distance deviation, target alignment deviation, exit trigger result, current inspection segment, heading change, current inspection progress, and historical response deviation are feature-encoded to obtain the control feature vector. The control feature vector is input into the pre-trained control primitive prediction model for classification, the matching probability of each primitive is obtained, and the primitive with the highest matching probability is selected as the flight control primitive.

5. The intelligent inspection method for low-altitude unmanned aerial vehicles based on artificial intelligence as described in claim 1, characterized in that, The specific steps for calculating the control variables of the UAV are as follows: The altitude constraint, speed constraint, heading constraint, lateral safety constraint, target alignment constraint, avoidance and exit constraint, and the current flight status of the UAV are used as the basis for calculating the control variables. When the flight control primitive is a straight-line stability tracking primitive, the speed adjustment, heading correction, and lateral distance adjustment are calculated based on the control quantity calculation and used as the straight-line stability control quantities. When the flight control primitive is a cornering deceleration and steering primitive, the calculated speed adjustment, heading correction, attitude compensation, and lateral distance adjustment are used as cornering constraint control quantities. When the flight control primitive is a local safety observation adjustment primitive, the corresponding altitude adjustment, speed adjustment, heading correction, lateral distance adjustment, attitude compensation, and hovering time are calculated as local observation control quantities. When the flight control primitive is the exit return-to-home primitive, the return-to-home direction, return-to-home speed, and return-to-home altitude are calculated as the return-to-home control variables. Configure the control execution sequence for the straight stability control, curve constraint control, local observation control, and return control to obtain the initial flight control intent.

6. The intelligent inspection method for low-altitude unmanned aerial vehicles based on artificial intelligence as described in claim 1, characterized in that, The specific steps for calculating the security certificate value are as follows: The flight state for the next control cycle is calculated using the initial flight control intent and the current flight state of the UAV, and the flight state prediction result is obtained. The flight status prediction results and the current inspection control status are used to calculate the altitude safety margin, space boundary safety margin, motion safety margin, and curve flight safety margin as safety certificate values.

7. The intelligent inspection method for low-altitude unmanned aerial vehicles based on artificial intelligence as described in claim 1, characterized in that, The specific steps for performing minimum constraint correction are as follows: The safety certificate value is compared with the preset height safety margin threshold, spatial boundary safety margin threshold, motion safety margin threshold, and curve flight safety margin threshold to obtain the safety execution judgment result. When the safety execution judgment result is unsafe, the corresponding initial flight control intention is adjusted to not exceed the current inspection control state, and a safe flight control package is obtained.

8. The intelligent inspection method for low-altitude unmanned aerial vehicles based on artificial intelligence as described in claim 1, characterized in that, The specific steps for reconstructing the link of the current inspection segment are as follows: The actual position of a drone that first exceeds the preset allowable response deviation range and whose actual flight state exceeds the current inspection control state is considered an abnormal position. Using the abnormal position as a reference, the reentry altitude, reentry speed, reentry heading, reentry lateral distance, and reentry hovering time are calculated using the current inspection control state to form a local reentry control segment. After inserting the segment into the current inspection segment, the updated straight and curved section inspection control chain is obtained.

9. An artificial intelligence-based intelligent inspection system for low-altitude unmanned aerial vehicles (UAVs), based on the artificial intelligence-based intelligent inspection method for low-altitude UAVs as described in any one of claims 1 to 8, characterized in that, include: The configuration module obtains the spatial observation information of the object to be inspected, performs hierarchical reduction on the spatial observation information, and obtains the observation zone and anchor point chain. The fusion module performs heading discrimination and segment reduction on the observation zone and anchor point chain to obtain the straight and curved section inspection control chain. It then performs segment assignment matching on the straight and curved section inspection control chain to obtain the current inspection segment. Finally, it performs constraint fusion on the current inspection segment to obtain the current inspection control state. The calculation module performs control primitive matching on the current inspection control state to obtain flight control primitives. Based on the flight control primitives, it performs control quantity calculation on the UAV to obtain the initial flight control intent. The correction module predicts the flight state of the initial flight control intention, obtains the flight state prediction result, calculates the safety certificate value based on the flight state prediction result, and performs minimum constraint correction on the initial flight control intention based on the safety certificate value to obtain the safe flight control package. The reconstruction module acquires the actual flight status, monitors the response deviation between the safe flight control package and the actual flight status, and reconstructs the link of the current inspection segment based on the response deviation and the actual flight status to obtain the updated straight and curved section inspection control chain.

Images