An unmanned aerial vehicle intelligent inspection method and system based on artificial intelligence

Abstract

The application relates to the technical field of intelligent identification, and particularly discloses a method and system for intelligent inspection of unmanned aerial vehicles based on artificial intelligence, which determines the target flight height information of an unmanned aerial vehicle by uniformly dividing a farmland area to be inspected to obtain a three-dimensional inspection grid corresponding to a canopy height, performs time sequence analysis on the target flight height information at continuous time points to generate a height credibility parameter, obtains corresponding farmland image data based on the target flight height information, analyzes an imaging field of view coverage range parameter according to a camera internal parameter model parameter, calculates the image spatial resolution corresponding to the current three-dimensional inspection grid, forms a weighted imaging scale description parameter, divides a scale sensitivity level based on a pest identification strategy built based on the first image data, feeds back a height stability adjustment instruction to the unmanned aerial vehicle based on the scale sensitivity level, and constrains the unmanned aerial vehicle to perform secondary supplementary inspection. The application improves the spatial precision and consistency of farmland inspection images, and enhances the reliability of pest identification.

Description

Technical Field

[0001] This invention relates to the field of intelligent recognition technology, and more specifically, to an intelligent inspection method and system for unmanned aerial vehicles (UAVs) based on artificial intelligence. Background Technology

[0002] With the continuous expansion of agricultural production scale and the ongoing improvement of agricultural modernization, traditional agricultural production methods relying on manual experience and manual operations have gradually revealed significant shortcomings in terms of operational efficiency, accuracy, and risk response capabilities. This is particularly evident in applications such as large-scale farmland inspection and pest monitoring, where manual methods struggle to achieve comprehensive and continuous coverage of the farmland environment, failing to meet the practical demands of modern agriculture for high-efficiency, low-cost, and intelligent operations. Agricultural drones, due to their advantages of high maneuverability, high field of view, wide coverage, and the ability to carry various sensors, have gradually become an important technological tool in the field of smart agriculture. However, in complex farmland scenarios, traditional path planning methods are prone to frequent path changes, affecting the stability of drone flight and the quality of operations. During pest inspections in the same farmland, the canopy height varies significantly among different crops or at different growth stages. Path planning methods often suffer from incomplete coverage in complex farmland environments, leading to inconsistent pest image scales, suboptimal imaging angles, and even missed pest areas, severely impacting the accuracy and reliability of pest identification results.

[0003] Therefore, it is necessary to provide an intelligent inspection method and system for drones based on artificial intelligence to solve the above-mentioned technical problems. In order to solve the above problems, a technical solution is provided. Summary of the Invention

[0004] To overcome the aforementioned shortcomings of existing technologies, this invention provides an AI-based intelligent drone inspection method and system to address the problems that arise in traditional drone path planning during complex farmland inspections due to the expansion of farmland and the increasing differences in crop canopy layers. These problems result in incomplete coverage of pest-affected areas and decreased identification accuracy.

[0005] To achieve the above objectives, the present invention provides the following technical solution: An AI-based intelligent inspection method for drones includes the following steps: By evenly dividing the farmland area under inspection into several farmland inspection blocks, a three-dimensional inspection grid is obtained by modeling the farmland inspection blocks, the canopy height corresponding to the three-dimensional inspection grid is obtained, and the target flight altitude information of the UAV is determined based on the preset relative height between the UAV and the three-dimensional inspection grid. The target flight altitude information at continuous time is analyzed in a time series to generate altitude confidence parameters. Based on the target flight altitude information, corresponding farmland image data is obtained. The imaging field of view coverage parameters are analyzed according to the camera intrinsic parameter model parameters. The image spatial resolution corresponding to the current three-dimensional inspection grid is calculated. The altitude confidence parameter is introduced into the imaging scale estimation to form a weighted imaging scale description parameter. Based on the imaging scale description parameters, adaptive scale normalization processing is performed on farmland image data to obtain first image data. Based on the first image data, a pest perception and identification strategy is built to classify scale sensitivity levels. Based on the scale sensitivity level, the drone is fed back with height stability adjustment commands to constrain it to perform a second patrol.

[0006] As a further aspect of the present invention, time-series analysis is performed on the target flight altitude information at consecutive time intervals to generate altitude reliability parameters, specifically as follows: Time series analysis is performed on the altitude time series of UAVs to calculate altitude change characteristic parameters; the altitude change characteristic parameters include altitude change amplitude, altitude change rate, and altitude fluctuation frequency; A height reliability assessment model is constructed based on height variation feature parameters, and the height reliability parameters corresponding to the three-dimensional inspection grid are output.

[0007] As a further aspect of the present invention, a height reliability assessment model is constructed based on height variation feature parameters, and the height reliability parameters corresponding to the three-dimensional inspection grid are output. The specific steps are as follows: Extracting UAV altitude time series ,in, For the target flight altitude information at time i, For the testing cycle, For the j-th 3D inspection grid; Based on the UAV altitude time series, altitude change characteristic parameters including altitude change amplitude, altitude change rate, and altitude fluctuation frequency were calculated. After normalizing the height variation characteristic parameters, a height reliability assessment model is constructed based on the height variation characteristic parameters, and the height reliability parameters are output.

[0008] As a further aspect of the present invention, corresponding farmland image data is obtained based on target flight altitude information. The imaging field of view coverage parameters are analyzed according to camera intrinsic parameter model parameters. The image spatial resolution corresponding to the current farmland image is calculated. The altitude confidence parameter is introduced into the imaging scale estimation to form a weighted imaging scale description parameter. The specific steps are as follows: By controlling the drone to perform inspection flight above the corresponding three-dimensional inspection grid according to the target flight altitude information, the drone uses the airborne camera to collect farmland image data of the corresponding three-dimensional inspection grid and records the target flight altitude information corresponding to the image acquisition time. Obtain the intrinsic parameter model parameters of the UAV's onboard camera. The camera's intrinsic parameter model parameters include focal length, camera pixel size, and pixel size parameters. Based on the target flight altitude information and camera intrinsic parameter model parameters, dynamically calculate the imaging field of view coverage parameters of the current UAV at the target flight altitude; Based on the camera intrinsic parameter model parameters, analyze the imaging field of view coverage parameters and calculate the image spatial resolution corresponding to the current 3D inspection grid. By incorporating a high confidence parameter into the imaging scale estimation process, the image spatial resolution and imaging field of view coverage parameters are weighted to form a weighted imaging scale description parameter. The weighted imaging scale description parameters are associated and stored with the corresponding 3D inspection grid.

[0009] As a further aspect of the present invention, the imaging field of view coverage parameters are analyzed based on the camera intrinsic parameter model parameters, and the image spatial resolution corresponding to the current three-dimensional inspection grid is calculated. The specific steps are as follows: The camera field of view is calculated based on the camera intrinsic parameter model parameters, and the imaging field of view coverage parameters are determined based on the camera field of view. The spatial resolution of the image corresponding to the current 3D inspection grid is calculated based on the imaging field of view coverage parameter.

[0010] As a further aspect of the present invention, a high confidence parameter is introduced into the imaging scale estimation process. The image spatial resolution and imaging field of view coverage parameters are weighted to form a weighted imaging scale description parameter. The specific steps are as follows: By incorporating a high confidence parameter into the calculation of image spatial resolution, a weighted spatial resolution is obtained; The imaging field of view coverage parameter is weighted by a high confidence parameter, and the scale description model is constructed by combining the weighted image spatial resolution and imaging field of view coverage parameter to output the imaging scale description parameter.

[0011] As a further aspect of the present invention, adaptive scale normalization processing is performed on farmland image data according to imaging scale description parameters to obtain first image data. Based on the first image data, a pest perception and identification strategy is built to classify scale sensitivity levels. The specific steps are as follows: Obtain farmland image data and corresponding imaging scale description parameters. Based on the imaging scale description parameters, perform adaptive scale normalization processing on the farmland image data so that the farmland image data acquired under different imaging scale conditions can be expressed in a unified scale space to obtain the first image data. The image data of pest-free farmland obtained in advance by the three-dimensional inspection grid is extracted as the second image data. The image difference of the three-dimensional inspection grid is analyzed by comparing the first image data and the second image data, and the image difference is summarized to evaluate the pest perception sensitivity coefficient of the farmland inspection block. Based on the pest perception sensitivity coefficient, pest targets are divided into different scale sensitivity levels; the scale sensitivity levels include normal sensitivity level and abnormal sensitivity level.

[0012] As a further aspect of the present invention, the pre-acquired pest-free farmland image data of the three-dimensional inspection grid is extracted as the second image data. The image difference of the three-dimensional inspection grid is analyzed by comparing the first image data and the second image data, and the image difference is summarized to evaluate the pest perception sensitivity coefficient of the farmland inspection block. The specific steps are as follows: The pixel data of the first image data is extracted as the first pixel data, and the pixel data of the second image data is extracted as the second pixel data. The pixel difference is calculated based on the first pixel data and the second pixel data, and the pixel difference is summarized to obtain the image difference of the three-dimensional inspection grid. The pest perception sensitivity coefficient is obtained by summarizing the image differences of the three-dimensional inspection grid within the farmland inspection block.

[0013] As a further aspect of the present invention, based on the pest perception sensitivity coefficient, pest targets are divided into different scale sensitivity levels, and the specific steps are as follows: Based on the comparison between the pest perception sensitivity coefficient and the preset sensitivity threshold, if the pest perception sensitivity coefficient is greater than or equal to the preset sensitivity threshold, the corresponding farmland inspection block is at an abnormal sensitivity level; if the pest perception sensitivity coefficient is less than the preset sensitivity threshold, the corresponding farmland inspection block is at a normal sensitivity level.

[0014] An AI-based unmanned aerial vehicle (UAV) intelligent inspection system includes a farmland spatial modeling and analysis module, an imaging scale weighted estimation module, a pest perception module, and a supplementary inspection control feedback module. The farmland spatial modeling and analysis module is used to divide the farmland area under inspection into several farmland inspection blocks, model a three-dimensional inspection grid based on the farmland inspection blocks, obtain the canopy height corresponding to the three-dimensional inspection grid, determine the target flight altitude information of the UAV based on the preset relative height between the UAV and the three-dimensional inspection grid, and perform time-series analysis on the target flight altitude information at continuous time to generate altitude confidence parameters. The imaging scale weighted estimation module is used to acquire corresponding farmland image data based on target flight altitude information, analyze the imaging field of view coverage parameters according to camera intrinsic model parameters, calculate the image spatial resolution corresponding to the current three-dimensional inspection grid, and introduce the altitude confidence parameter into the imaging scale estimation to form a weighted imaging scale description parameter. The pest perception module is used to perform adaptive scale normalization processing on farmland image data according to imaging scale description parameters to obtain first image data, and to build a pest perception and recognition strategy based on the first image data to classify scale sensitivity levels. The supplementary patrol control feedback module is used to feed back altitude stability adjustment commands to the UAV based on the scale sensitivity level, thereby constraining the UAV to perform a second supplementary patrol.

[0015] The present invention discloses an artificial intelligence-based intelligent drone inspection method and system, highlighting its technical effects and advantages. The invention divides the farmland area to be inspected into several inspection blocks and establishes a three-dimensional inspection grid model, simultaneously acquiring the canopy height information corresponding to each grid. Based on a preset relative height between the drone and the three-dimensional grid, the target flight altitude of the drone is determined, and a time-series analysis of the target flight altitude at consecutive moments is performed to generate a height reliability parameter to evaluate the reliability of the flight altitude. Based on this, corresponding farmland image data is acquired through the target flight altitude, and the coverage of the imaging field of view is analyzed in conjunction with camera intrinsic parameter model parameters to calculate the image spatial resolution corresponding to the current grid. Simultaneously, the height reliability parameter is introduced into imaging scale estimation to form a weighted imaging scale description parameter, making image data processing more accurate. Subsequently, adaptive scale normalization processing is performed on the farmland image data according to this imaging scale description parameter to obtain the first image data. Based on this data, a pest perception and recognition strategy is established, classifying pest targets into different scale sensitivity levels. Based on the scale sensitivity level, the system feeds back altitude stability adjustment commands to the UAV, thereby constraining and optimizing the UAV's flight altitude and performing necessary secondary patrols to ensure the integrity of the inspection coverage and the quality of image acquisition.

[0016] This invention, through comprehensive analysis of flight altitude, image resolution, and pest scale sensitivity, not only improves the spatial accuracy and consistency of farmland inspection images but also enhances the reliability of pest identification. Simultaneously, the feedback mechanism dynamically adjusts the drone's flight altitude, enabling precise inspections and secondary follow-up inspections, thereby effectively improving farmland monitoring efficiency, reducing omissions, and enhancing the scientific and refined level of agricultural production management. Attached Figure Description

[0017] Figure 1 A flowchart illustrating an artificial intelligence-based intelligent inspection method for drones, provided as an embodiment of the present invention; Figure 2 This is a system block diagram of an artificial intelligence-based unmanned aerial vehicle (UAV) intelligent inspection system provided in an embodiment of the present invention. Detailed Implementation

[0018] The technical solutions of this invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described technical solutions are only a part of this invention, and not all of it. All other technical solutions obtained by those skilled in the art based on the technical solutions of this invention without inventive effort are within the scope of protection of this invention.

[0019] like Figure 1 The diagram shown is a flowchart of an artificial intelligence-based intelligent inspection method for drones provided in an embodiment of the present invention. Figure 1 The execution entity of the method shown can be a software and / or hardware device. The execution entity of this application can include, but is not limited to, at least one of the following: user equipment, network equipment, etc. User equipment can include, but is not limited to, computers, smartphones, personal digital assistants (PDAs), and the aforementioned electronic devices. Network equipment can include, but is not limited to, a single network server, a server group consisting of multiple network servers, or a cloud based on cloud computing consisting of a large number of computers or network servers. Cloud computing is a type of distributed computing, consisting of a super virtual computer composed of a group of loosely coupled computers. This embodiment does not limit this. Steps S1 to S4 are detailed as follows: Step S1: By evenly dividing the farmland area to be inspected into several farmland inspection blocks, a three-dimensional inspection grid is obtained by modeling the farmland inspection blocks, the canopy height corresponding to the three-dimensional inspection grid is obtained, and the target flight altitude information of the UAV is determined based on the preset relative height between the UAV and the three-dimensional inspection grid. The target flight altitude information at continuous time is analyzed in a time series to generate altitude confidence parameters. Step S2: Based on the target flight altitude information, obtain the corresponding farmland image data, analyze the imaging field of view coverage parameters according to the camera intrinsic parameter model parameters, calculate the image spatial resolution corresponding to the current three-dimensional inspection grid, and introduce the altitude confidence parameter into the imaging scale estimation to form a weighted imaging scale description parameter. Step S3: Based on the imaging scale description parameters, perform adaptive scale normalization processing on the farmland image data to obtain the first image data, and build a pest perception and identification strategy based on the first image data to classify the scale sensitivity level. Step S4: Based on the scale sensitivity level, a height stability adjustment command is fed back to the UAV to constrain the UAV to perform a second patrol.

[0020] Preferably, the process involves uniformly dividing the farmland area to be inspected into several farmland inspection blocks, creating a three-dimensional inspection grid based on the farmland inspection blocks, obtaining the canopy height corresponding to the three-dimensional inspection grid, determining the drone's target flight altitude information based on the preset relative altitude between the drone and the three-dimensional inspection grid, and performing time-series analysis on the target flight altitude information at continuous times to generate altitude reliability parameters. The specific steps are as follows: Obtain the boundary information of the farmland area to be inspected, and divide the farmland area into several farmland inspection blocks evenly according to the preset spatial resolution; each farmland inspection block corresponds to a sub-region of the farmland area, and is used as the basic unit for UAV inspection and data collection. For each farmland inspection block, a corresponding three-dimensional inspection grid model is constructed based on terrain data and crop distribution information; the three-dimensional inspection grid is used to describe the surface height and the spatial distribution of crop canopy within the inspection block. For each grid cell in the 3D inspection grid, obtain the corresponding crop canopy height information; the canopy height is used to characterize the actual height of the top of the farmland crops relative to the ground. Based on the preset relative height between the UAV and the 3D inspection grid, the canopy height is superimposed with the preset relative height to determine the target flight altitude information of the UAV in the corresponding 3D inspection grid. During the drone inspection flight, the target flight altitude information of the drone flying above each inspection block is continuously acquired according to the preset sampling period to form a drone altitude time series. Time series analysis is performed on the altitude time series of UAVs to calculate altitude change characteristic parameters; the altitude change characteristic parameters include altitude change amplitude, altitude change rate, and altitude fluctuation frequency; A height reliability assessment model is constructed based on height variation characteristic parameters, and the height reliability parameters corresponding to the three-dimensional inspection grid are output. The height reliability parameters are used to reflect the height stability of the UAV during the inspection of the inspection block.

[0021] In one embodiment of this invention, applied to a large-scale farmland planted with contiguous rice and corn, the boundary contour data of the farmland area to be inspected is first obtained through a farmland geographic information system. Then, considering the ground resolution requirements of the camera mounted on the UAV, the farmland area is uniformly divided into multiple farmland inspection blocks according to a preset spatial resolution of 5m × 5m. Each inspection block corresponds to an independent sub-region within the farmland, serving as the smallest spatial unit for UAV flight path planning, inspection scheduling, and pest image acquisition. This method avoids the problem of neglecting local terrain and crop differences caused by traditional uniform modeling of the entire farmland, making subsequent inspections and modeling more targeted.

[0022] For each farmland inspection block, a corresponding 3D inspection grid model is constructed by combining existing farmland topographic elevation data with crop type and planting density information. This 3D inspection grid uses the ground surface as a reference and spatially discretizes the interior of the inspection block, accurately describing the ground undulations at different locations and the spatial distribution differences of crop canopies. For example, within the same inspection block, the canopy height in the corn area is significantly higher than that in the rice area. The 3D inspection grid model can effectively distinguish these differences, providing fundamental data support for subsequent UAV altitude control.

[0023] After the 3D inspection grid is constructed, the corresponding crop canopy height information is extracted for each grid unit. This canopy height can be obtained through lidar point cloud, historical mapping data, or multi-view image inversion, and is used to characterize the true height of the crop top relative to the ground. By obtaining the canopy height in a refined manner, the problem of UAVs flying too low or too high in areas with uneven crop growth or undulating terrain can be avoided when flying at a fixed altitude, thus ensuring inspection safety and imaging consistency.

[0024] Based on this, according to the pre-set safe relative height between the UAV and the 3D inspection grid, the acquired canopy height is superimposed with the preset relative height to calculate the target flight altitude information of the UAV on the corresponding 3D inspection grid. For example, when the canopy height of a certain grid cell is 2.1m and the preset relative height is 3m, the target flight altitude of the UAV at that location is set to 5.1m. In this way, the UAV can dynamically adapt to different crop growth heights, achieving precise flight control that closely follows changes in crop canopy height.

[0025] During the inspection flight of the UAV, the target flight altitude information of the UAV above each inspection block is continuously recorded according to a preset sampling period (such as 0.5s or 1s), forming the corresponding UAV altitude time series. This altitude time series not only reflects the flight status of the UAV within a single inspection block, but also characterizes the altitude changes caused by airflow disturbances, attitude adjustments, or control errors during the flight across inspection blocks.

[0026] Furthermore, time-series analysis was performed on the drone's altitude time series to extract altitude change characteristic parameters such as altitude change amplitude, altitude change rate, and altitude fluctuation frequency. Among them, altitude change amplitude reflects the maximum altitude deviation of the drone within the inspection block, altitude change rate characterizes the drasticness of the drone's altitude adjustment, and altitude fluctuation frequency measures the stability level of the drone's repeated ascents and descents within a short period of time. These characteristic parameters can characterize the actual performance of drone altitude control from multiple dimensions.

[0027] Finally, a height reliability assessment model is constructed based on the extracted height change feature parameters. This model outputs a corresponding height reliability parameter for each 3D inspection grid, comprehensively reflecting the height stability of the UAV during the inspection of that grid block. A higher height reliability parameter corresponds to a smaller height change amplitude, a smoother rate of change, and a lower fluctuation frequency; conversely, a lower height reliability parameter corresponds to a larger change amplitude and a lower fluctuation frequency. Introducing the height reliability parameter provides crucial information for subsequent imaging scale weighting, pest identification reliability assessment, and secondary inspection decisions.

[0028] Preferably, a height reliability assessment model is constructed based on height variation feature parameters, and the height reliability parameters corresponding to the 3D inspection grid are output. The specific steps are as follows: Extracting UAV altitude time series ,in, For the target flight altitude information at time i, For the testing cycle, For the j-th 3D inspection grid; Based on the UAV altitude time series, altitude change characteristic parameters including altitude change amplitude, altitude change rate, and altitude fluctuation frequency were calculated. After normalizing the height variation characteristic parameters, a height reliability assessment model is constructed based on these parameters, and the height reliability parameters are output. The calculation formula for the height reliability assessment model is as follows: ; ; In the formula: Let j be the height confidence parameter of the j-th 3D inspection grid. It is an exponential function with the natural constant e as the base. The weighting coefficient is the height variation range. For weighting coefficients with high rates of change, Weighting coefficients for high fluctuation frequency It should be noted that the formula for calculating the range of height change is as follows: In the formula: Let j be the height variation range of the j-th 3D inspection grid. This represents the maximum value in the drone's altitude time series. This represents the minimum value in the drone altitude time series.

[0029] Formula for calculating the rate of change of height: In the formula: Let j be the height change rate of the j-th 3D inspection grid. For the target flight altitude information at time i+1, For the target flight altitude information at time i, For time intervals.

[0030] Formula for calculating the frequency of height fluctuations: In the formula: Let j be the frequency of height fluctuation of the j-th 3D inspection grid. This is an indicator function that takes the value 1 when the condition is true and 0 otherwise. For the target flight altitude information at time i+1, For the target flight altitude information at time i, The target flight altitude information at time i-1.

[0031] In one embodiment of the present invention, when a drone inspects a corn planting area corresponding to a three-dimensional inspection grid in actual farmland, the drone performs low-altitude, uniform-speed inspection above the inspection block according to a preset flight strategy. Throughout the entire inspection cycle, the drone continuously collects target flight altitude information above the inspection block at fixed time intervals, thereby forming a corresponding drone altitude time series, which can completely reflect the altitude change process of the drone during the inspection of the inspection block.

[0032] Based on the UAV altitude time series, altitude change characteristic parameters such as altitude change amplitude, altitude change rate, and altitude fluctuation frequency are calculated. Specifically, the altitude change amplitude is obtained by calculating the difference between the maximum and minimum altitude values ​​in the altitude time series, reflecting the maximum possible altitude deviation of the UAV within the inspection block. This parameter increases significantly when the farmland terrain is undulating or airflow disturbances are significant, indicating poor UAV altitude stability.

[0033] The altitude change rate is obtained by statistically calculating the altitude change at adjacent sampling times and is used to characterize the degree of drastic altitude adjustment by the UAV during the inspection process. If the UAV frequently makes rapid ascents and descents to correct its attitude or avoid obstacles, the altitude change rate will increase accordingly; conversely, when the UAV makes relatively smooth altitude fine-tuning, this parameter remains at a low level.

[0034] Altitude fluctuation frequency describes the repeated switching of the drone's altitude change direction. It is determined by whether adjacent altitude change directions reverse, and the number of times the drone switches between "ascending-descending" and "descending-ascending" during the inspection process is counted. When the drone repeatedly adjusts its altitude under conditions of airflow disturbance or insufficient control precision, this fluctuation frequency will increase significantly, reflecting the instability of altitude control.

[0035] After extracting the aforementioned height variation characteristic parameters, each parameter is first normalized to eliminate the influence of different units and numerical ranges on the model calculation. Subsequently, a height reliability assessment model is constructed based on the normalized height variation amplitude, height variation rate, and height fluctuation frequency, and the calculated height reliability is obtained. The height reliability parameter corresponds to each 3D inspection grid. An exponential function is used to non-linearly penalize height instability factors, ensuring that the height reliability parameter decays rapidly when any height change feature increases significantly, thus highlighting the impact of height stability on subsequent imaging quality.

[0036] Using the aforementioned high reliability assessment model, when the UAV's flight altitude changes within a certain inspection block are small, the rate of change is gradual, and the frequency of fluctuations is low, the corresponding high reliability parameter is close to 1. This indicates that the UAV's flight altitude is stable during the inspection of this area, and the farmland images collected have high imaging reliability. Conversely, when the altitude change characteristic parameter increases significantly, the high reliability parameter approaches 0. Based on this, it is determined that the image data within the inspection block may have problems with inconsistent imaging scales or unstable attitudes, providing a basis for subsequent imaging scale weighting, pest identification reliability assessment, and secondary inspection decisions.

[0037] Preferably, based on the target flight altitude information, corresponding farmland image data is acquired; the imaging field of view coverage parameters are analyzed according to the camera intrinsic parameter model parameters; the image spatial resolution corresponding to the current farmland image is calculated; and the altitude confidence parameter is introduced into the imaging scale estimation to form a weighted imaging scale description parameter. The specific steps are as follows: The drone is controlled to perform inspection flight above the corresponding three-dimensional inspection grid according to the target flight altitude information. The onboard camera is used to collect farmland image data of the corresponding three-dimensional inspection grid and the target flight altitude information corresponding to the image acquisition time is recorded. Obtain the intrinsic parameter model parameters of the UAV's onboard camera, including focal length, camera pixel size, and pixel size parameters; based on the camera's intrinsic parameter model parameters, determine the imaging geometry of the camera in the current working state; Based on the target flight altitude information and camera intrinsic parameter model parameters, the imaging field of view coverage parameter of the current UAV at the target flight altitude is dynamically calculated; it should be noted that the imaging field of view coverage parameter is used to characterize the actual ground or canopy area covered by a single farmland image; Based on the camera intrinsic parameter model parameters, analyze the imaging field of view coverage parameters and calculate the image spatial resolution corresponding to the current 3D inspection grid; the image spatial resolution is used to characterize the actual spatial size corresponding to a unit pixel in the image; By incorporating a high confidence parameter into the imaging scale estimation process, the image spatial resolution and imaging field of view coverage parameters are weighted to form a weighted imaging scale description parameter. The weighted imaging scale description parameters are associated and stored with the corresponding 3D inspection grid.

[0038] In one embodiment of the present invention, during actual farmland pest inspection, taking a large-scale farmland with a complex planting structure and significant differences in crop height as an example, the UAV first performs inspection flight above the corresponding three-dimensional inspection grid based on the target flight altitude information determined in the aforementioned steps. During flight, the UAV maintains the target flight altitude in real time through the flight control system, and the onboard camera performs timed or triggered image acquisition above the inspection block, acquiring farmland image data of the corresponding three-dimensional inspection grid. Simultaneously, the target flight altitude information corresponding to the acquisition time of each farmland image is recorded, thereby ensuring a one-to-one correspondence between image data and flight altitude, providing accurate altitude basis for subsequent imaging scale analysis.

[0039] After image data acquisition is completed, the intrinsic parameter model parameters of the UAV's onboard camera are further obtained. These parameters include the camera's focal length, pixel physical size, and image pixel size. Using these intrinsic parameter model parameters, the geometric relationship between optical projection and pixel mapping during camera imaging can be accurately established, thus providing a foundation for the precise calculation of imaging field of view and spatial resolution under different flight altitudes.

[0040] Based on the acquired target flight altitude information and camera intrinsic parameter model parameters, the imaging field of view coverage parameters of the UAV at the current flight altitude are dynamically calculated. Specifically, as the UAV's flight altitude increases, the area of ​​ground or crop canopy covered by a single image increases; conversely, as the flight altitude decreases, the imaging field of view coverage shrinks accordingly. By calculating the imaging field of view coverage parameters in real time, the actual spatial coverage area corresponding to each farmland image can be accurately characterized, avoiding the uncertainty of coverage caused by altitude changes.

[0041] Furthermore, based on the camera's intrinsic parameter model parameters, the imaging field of view coverage is analyzed, and the image spatial resolution corresponding to the current 3D inspection grid is calculated. This image spatial resolution is used to characterize the size of a single pixel in the image within the actual farmland space, such as the length of each pixel in centimeters or millimeters. Through this calculation, the differences in the spatial scale of farmland images under different inspection blocks and different flight altitudes can be clearly identified, providing a basis for subsequent scale consistency processing of pest targets.

[0042] After obtaining the image spatial resolution and imaging field of view coverage parameters, the altitude confidence parameter obtained in the preceding steps is introduced into the imaging scale estimation process. Specifically, the image spatial resolution and imaging field of view coverage parameters are weighted so that the imaging scale parameter has a higher weight in subsequent processing within inspection blocks with high altitude stability; while the influence of the imaging scale parameter is reduced in inspection blocks with large altitude fluctuations and low confidence. Through this weighting method, a weighted imaging scale description parameter that can comprehensively reflect flight altitude stability and imaging geometry characteristics is formed.

[0043] Finally, the weighted imaging scale description parameters are associated and stored with the corresponding 3D inspection grid. By establishing a mapping relationship between the imaging scale description parameters and the spatial inspection grid, not only can direct input be provided for subsequent adaptive scale normalization of farmland images and pest identification, but it can also serve as an important basis for triggering secondary inspections and flight altitude adjustments when the imaging scale reliability of certain inspection blocks is found to be low, thereby improving the accuracy and reliability of the overall farmland inspection and pest monitoring results.

[0044] Preferably, the imaging field of view coverage parameters are analyzed based on the camera intrinsic parameter model parameters, and the image spatial resolution corresponding to the current 3D inspection grid is calculated. The specific steps are as follows: The camera field of view is calculated based on the camera intrinsic parameter model parameters, and the imaging field of view coverage parameters are determined based on the camera field of view. The calculation formula is as follows: In the formula: The horizontal field of view of the camera. The camera's field of view in the vertical direction. This represents the number of pixels in the horizontal direction of the image. The horizontal camera pixel size. This represents the number of pixels in the vertical direction of the image. The vertical camera pixel size. This refers to the focal length parameter; In the formula: The field of view coverage length for imaging farmland in the horizontal direction. The field of view coverage length for farmland imaging in the vertical direction. The target flight altitude for the drone. The horizontal field of view of the camera. This refers to the camera's field of view in the vertical direction; The image spatial resolution corresponding to the current 3D inspection grid is calculated based on the imaging field of view coverage parameter. The calculation formula is as follows: In the formula: This represents the image spatial resolution corresponding to the current 3D inspection grid.

[0045] Specifically, a high confidence parameter is introduced into the imaging scale estimation process, and the image spatial resolution and imaging field of view coverage parameters are weighted to form a weighted imaging scale description parameter. The specific steps are as follows: By incorporating a high confidence level parameter into the image spatial resolution calculation, the weighted spatial resolution is obtained, and the calculation formula is as follows: In the formula: Let j be the weighted spatial resolution of the j-th 3D inspection grid. Let j be the image spatial resolution corresponding to the j-th 3D inspection grid. Let be the height reliability parameter of the j-th 3D inspection grid; The imaging field-of-view coverage parameter is weighted by a high-confidence parameter, and the calculation formula is as follows: , In the formula: Let be the horizontal field of view coverage length of the farmland imaging for the j-th 3D inspection grid. Let be the field of view coverage length of the farmland imaging in the vertical direction of the j-th 3D inspection grid. Let be the horizontal field of view coverage length of the j-th 3D inspection grid in the farmland imaging field of view. Let be the field of view coverage length of the j-th 3D inspection grid in the vertical direction for farmland imaging; By combining the weighted image spatial resolution and imaging field of view coverage parameters, a scale description model is constructed to output imaging scale description parameters. , These are the weighting coefficients for the image spatial resolution. These are the weighting coefficients for the imaging field of view coverage parameter.

[0046] In one embodiment of the present invention, during actual farmland inspection, a drone equipped with a visible light camera with a resolution of 4000×3000 pixels is used as an example. The focal length of the camera is... The pixel size is First, the camera's intrinsic parameter model parameters are read, and the camera's field of view in the horizontal and vertical directions is calculated based on these parameters. By substituting the number of pixels in the horizontal direction, the number of pixels in the vertical direction, and the corresponding pixel size into the field of view calculation formula, the camera's field of view in the horizontal direction and the field of view in the vertical direction are obtained. These parameters characterize the imaging angular range of the camera in the current operating state and are an important basis for subsequent calculations of the ground imaging coverage.

[0047] After obtaining the camera's field of view, the system further combines the UAV's target flight altitude on the corresponding 3D inspection grid to calculate the imaging field of view coverage parameters under the current altitude conditions. Specifically, by substituting the field of view and target flight altitude into the imaging field of view coverage calculation formula, the horizontal and vertical imaging coverage lengths are obtained, respectively. These imaging field of view coverage parameters accurately reflect the actual area size covered by a single farmland image on the ground or crop canopy, providing a quantitative basis for evaluating image coverage integrity and inspection efficiency.

[0048] After determining the imaging field of view coverage, the image spatial resolution corresponding to the current 3D inspection grid is calculated based on the relationship between the coverage parameters and the camera pixel distribution. By calculating the ratio of the coverage length in the horizontal and vertical directions to the corresponding number of pixels, and averaging the resolutions in the two directions, a comprehensive image spatial resolution is obtained. This image spatial resolution characterizes the actual size of a single pixel in the farmland space and is an important indicator of the image's ability to express detail.

[0049] Furthermore, the high confidence parameter obtained in the aforementioned steps is introduced into the imaging scale estimation process to perform confidence-weighted processing on the image spatial resolution and imaging field of view coverage parameters. When the UAV's flight altitude is stable within a certain 3D inspection grid and the corresponding high confidence parameter is high, the weighted spatial resolution and imaging field of view coverage parameters change little; while when the high confidence parameter is low, the corresponding spatial resolution and coverage are amplified by dividing by the high confidence parameter, thereby reducing the reliability weight of the image in that area in subsequent processing. Through this method, the first... The weighted spatial resolution and the weighted imaging field of view coverage length corresponding to each 3D inspection grid.

[0050] After weighting the spatial resolution and imaging field of view coverage parameters based on their reliability, a scale description model is constructed by combining the weighted image spatial resolution and imaging field of view coverage parameters, outputting the imaging scale description parameters for the corresponding 3D inspection grid. By setting the image spatial resolution weight coefficient and the imaging field of view coverage weight coefficient, the calculation results of the imaging scale description parameters can be flexibly adjusted according to the emphasis on imaging accuracy or coverage for different inspection tasks. Through the construction of these imaging scale description parameters, a unified and comparable scale representation can be achieved for farmland images under different inspection blocks and different flight stability conditions, providing a reliable basis for subsequent adaptive scale normalization, pest identification sensitivity analysis, and secondary inspection decisions.

[0051] Preferably, based on the imaging scale description parameters, adaptive scale normalization processing is performed on the farmland image data to obtain the first image data. A pest perception and identification strategy is then built based on the first image data to classify scale sensitivity levels. The specific steps are as follows: Obtain farmland image data and corresponding imaging scale description parameters. Based on the imaging scale description parameters, perform adaptive scale normalization processing on the farmland image data so that the farmland image data acquired under different imaging scale conditions can be expressed in a unified scale space to obtain the first image data. The image data of pest-free farmland obtained in advance by the three-dimensional inspection grid is extracted as the second image data. The image difference of the three-dimensional inspection grid is analyzed by comparing the first image data and the second image data, and the image difference is summarized to evaluate the pest perception sensitivity coefficient of the farmland inspection block. Based on the pest perception sensitivity coefficient, pest targets are divided into different scale sensitivity levels; the scale sensitivity levels include normal sensitivity level and abnormal sensitivity level.

[0052] Specifically, the pre-acquired pest-free farmland image data of the 3D inspection grid is extracted as the second image data. The image difference of the 3D inspection grid is analyzed by comparing the first image data and the second image data, and the image difference is summarized to evaluate the pest perception sensitivity coefficient of the farmland inspection block. The specific steps are as follows: The pixel data of the first image data is extracted as the first pixel data, and the pixel data of the second image data is extracted as the second pixel data. The pixel difference is calculated based on the first pixel data and the second pixel data, and the pixel difference is summarized to obtain the image difference of the three-dimensional inspection grid. The pest perception sensitivity coefficient is obtained by summarizing the image differences of the three-dimensional inspection grid within the farmland inspection block.

[0053] Specifically, based on the pest perception sensitivity coefficient, pest targets are divided into different scale sensitivity levels. The specific steps are as follows: Based on the comparison between the pest perception sensitivity coefficient and the preset sensitivity threshold, if the pest perception sensitivity coefficient is greater than or equal to the preset sensitivity threshold, the corresponding farmland inspection block is at an abnormal sensitivity level; if the pest perception sensitivity coefficient is less than the preset sensitivity threshold, the corresponding farmland inspection block is at a normal sensitivity level.

[0054] In one embodiment of the present invention, during a pest inspection of a rice paddy in the jointing stage, the drone flies over different three-dimensional inspection grids. Due to differences in crop canopy height and slight fluctuations in flight altitude, the farmland images collected by each inspection block differ in imaging scale and spatial resolution. First, the data for each farmland image and its corresponding associated imaging scale description parameters are acquired. Based on the imaging scale description parameters, adaptive scale normalization processing is performed on the farmland image data, uniformly mapping the images acquired under different imaging scale conditions to the same scale space for expression. This eliminates the scale inconsistency problem caused by changes in flight altitude or field of view, resulting in scale-uniform first image data.

[0055] After image scale normalization, pre-established pest-free farmland image data is retrieved from the 3D inspection grid as the second image data. This pest-free farmland image data can originate from the initial growth stage of the farmland or images of healthy crops confirmed manually, serving as a benchmark reference for pest identification. By comparing and analyzing the first and second image data, the image differences between the same 3D inspection grid under the current inspection state and the pest-free state are calculated, thereby extracting change features that reflect the occurrence of pests.

[0056] Specifically, pixel information is extracted from the first image data as first pixel data, and pixel information is extracted from the second image data as second pixel data. Pixel-level change information is obtained by calculating the difference between the first and second pixel data at corresponding pixel locations. Subsequently, the pixel differences of all three-dimensional inspection grids within the inspection block are summarized to form an overall image difference description of the farmland inspection block, reflecting the comprehensive degree of change in crop appearance, texture, or color within that area.

[0057] Based on this, the pest perception sensitivity coefficient of the farmland inspection blocks is calculated according to the summarized image differences. The pest perception sensitivity coefficient is used to characterize the degree of response of pest targets to changes in imaging scale and image differences. When the changes in crop appearance caused by pests still show significant differences after scale normalization, the corresponding pest perception sensitivity coefficient is high; otherwise, the coefficient is low. In this way, the degree of pest impact can be effectively decoupled from imaging scale factors, improving the reliability of pest perception results.

[0058] Finally, the calculated pest perception sensitivity coefficient is compared with a preset sensitivity threshold. Based on the comparison results, the pest targets corresponding to the farmland inspection blocks are classified into different scale sensitivity levels. When the pest perception sensitivity coefficient is greater than or equal to the preset sensitivity threshold, the farmland inspection block is determined to be in an abnormal sensitivity level, indicating that the pest features in this area are highly sensitive to changes in imaging scale, and there may be a risk of pest occurrence or spread. When the pest perception sensitivity coefficient is less than the preset sensitivity threshold, the farmland inspection block is determined to be in a normal sensitivity level, indicating that the image changes in this area are within an acceptable range. Through the above-mentioned classification of scale sensitivity levels, a clear basis can be provided for subsequent UAV altitude stability feedback control and secondary patrol decisions, thereby improving the accuracy and reliability of farmland pest inspection.

[0059] Preferably, based on the pest perception sensitivity coefficient, pest targets are divided into different scale sensitivity levels. In this embodiment of the invention, when inspecting a rice-corn mixed-crop field in its mid-to-late growth stage during actual farmland pest inspection, after the UAV completes the inspection flight and acquires the corresponding first image data, it calculates the corresponding pest perception sensitivity coefficient for each inspected field block based on the scale-normalized image and a pre-established pest-free farmland image benchmark. This pest perception sensitivity coefficient comprehensively reflects the degree of appearance change of the current inspected image relative to the healthy crop image under a unified imaging scale, and is used to characterize the sensitivity response level of pest targets to changes in imaging scale.

[0060] In practical applications, the first step is to preset corresponding sensitivity thresholds for different crop types and growth stages. For example, for rice areas with relatively regular leaf structures and obvious pest characteristics, the sensitivity threshold can be set to a lower level; while for corn areas with high leaf overlap and complex texture variations, the sensitivity threshold can be increased accordingly. This approach avoids misjudgments caused by differences in crop types and improves the adaptability of the scale sensitivity level classification.

[0061] When the pest perception sensitivity coefficient of a certain farmland inspection area is greater than or equal to a preset sensitivity threshold, the inspection area is determined to be at an abnormally sensitive level. This abnormally sensitive level indicates that, after scale normalization, the image changes in this area are still significant, and the pest targets show high sensitivity to changes in imaging scale, suggesting a high probability of pest occurrence or spread. In this case, a more refined analysis process can be triggered, such as guiding a drone to conduct local low-altitude supplementary inspections to acquire higher-resolution images, thereby improving the accuracy of pest identification.

[0062] Conversely, when the pest perception sensitivity coefficient of a certain farmland inspection block is less than a preset sensitivity threshold, the inspection block is determined to be at a normal sensitivity level. This normal sensitivity level indicates that, under uniform imaging scale conditions, the crop image changes in this area are within the normal growth fluctuation range, pest characteristics are not obvious, and the scale changes of the current inspection image have little impact on the pest identification results. In this case, the original inspection strategy can be maintained without additional adjustments to the drone's flight altitude or increasing the inspection frequency, thereby reducing the overall inspection cost.

[0063] By classifying farmland inspection blocks into scale-sensitive levels based on pest perception sensitivity coefficients, it is possible to achieve hierarchical perception and differentiated processing of pest targets in complex farmland environments. On the one hand, limited drone inspection resources can be prioritized for areas with abnormally high sensitivity levels, improving the timeliness and accuracy of pest monitoring; on the other hand, it also avoids unnecessary repeated inspections in areas with normal sensitivity levels, improving the overall efficiency and intelligence level of farmland pest inspection.

[0064] An AI-based unmanned aerial vehicle (UAV) intelligent inspection system includes a farmland spatial modeling and analysis module, an imaging scale weighted estimation module, a pest perception module, and a supplementary inspection control feedback module. The farmland spatial modeling and analysis module is connected to the imaging scale weighted estimation module, the imaging scale weighted estimation module is connected to the pest perception module, and the pest perception module is connected to the supplementary inspection control feedback module. The farmland spatial modeling and analysis module is used to divide the farmland area under inspection into several farmland inspection blocks, model a three-dimensional inspection grid based on the farmland inspection blocks, obtain the canopy height corresponding to the three-dimensional inspection grid, determine the target flight altitude information of the UAV based on the preset relative height between the UAV and the three-dimensional inspection grid, and perform time-series analysis on the target flight altitude information at continuous time to generate altitude confidence parameters. The imaging scale weighted estimation module is used to acquire corresponding farmland image data based on target flight altitude information, analyze the imaging field of view coverage parameters according to camera intrinsic model parameters, calculate the image spatial resolution corresponding to the current three-dimensional inspection grid, and introduce the altitude confidence parameter into the imaging scale estimation to form a weighted imaging scale description parameter. The pest perception module is used to perform adaptive scale normalization processing on farmland image data according to imaging scale description parameters to obtain first image data, and to build a pest perception and recognition strategy based on the first image data to classify scale sensitivity levels. The supplementary patrol control feedback module is used to feed back altitude stability adjustment commands to the UAV based on the scale sensitivity level, thereby constraining the UAV to perform a second supplementary patrol.

[0065] like Figure 2 The diagram shown is a system block diagram of an artificial intelligence-based unmanned aerial vehicle (UAV) intelligent inspection system according to an embodiment of the present invention, which can be used to execute... Figure 1 The steps in the method embodiments shown are implemented in a similar manner and have similar technical effects, and will not be repeated here.

[0066] Through the above embodiments, this invention divides the farmland area to be inspected into several farmland inspection blocks, and establishes a three-dimensional inspection grid model on this basis, while acquiring the canopy height information corresponding to each grid. Based on the preset relative height between the UAV and the three-dimensional grid, the target flight altitude of the UAV is determined, and time-series analysis is performed on the target flight altitude at continuous moments to generate a height reliability parameter to evaluate the reliability of the flight altitude. On this basis, corresponding farmland image data is acquired through the target flight altitude, and the coverage of the imaging field of view is analyzed in combination with the camera intrinsic parameter model parameters, thereby calculating the image spatial resolution corresponding to the current grid. At the same time, the height reliability parameter is introduced into the imaging scale estimation to form a weighted imaging scale description parameter, making the image data processing more accurate. Subsequently, adaptive scale normalization processing is performed on the farmland image data according to the imaging scale description parameter to obtain the first image data, and a pest perception and recognition strategy is established based on this data, classifying pest targets into different scale sensitivity levels. According to the scale sensitivity level, a height stability adjustment command is fed back to the UAV to constrain and optimize the UAV's flight altitude, and necessary secondary inspections are performed to ensure the integrity of the inspection coverage and the quality of image acquisition.

[0067] This invention, through comprehensive analysis of flight altitude, image resolution, and pest scale sensitivity, not only improves the spatial accuracy and consistency of farmland inspection images but also enhances the reliability of pest identification. Simultaneously, the feedback mechanism dynamically adjusts the drone's flight altitude, enabling precise inspections and secondary follow-up inspections, thereby effectively improving farmland monitoring efficiency, reducing omissions, and enhancing the scientific and refined level of agricultural production management.

[0068] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

[0069] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An intelligent inspection method for unmanned aerial vehicles (UAVs) based on artificial intelligence, characterized in that, Includes the following steps: By uniformly dividing the farmland area to be inspected into several farmland inspection blocks, a three-dimensional inspection grid is obtained by modeling the farmland inspection blocks. The canopy height corresponding to the three-dimensional inspection grid is obtained. Based on the preset relative height between the UAV and the three-dimensional inspection grid, the target flight altitude information of the UAV is determined. Time-series analysis is performed on the target flight altitude information at continuous time intervals to generate altitude reliability parameters. Specifically: Time series analysis is performed on the altitude time series of UAVs to calculate altitude change characteristic parameters; the altitude change characteristic parameters include altitude change amplitude, altitude change rate, and altitude fluctuation frequency; A height reliability assessment model is constructed based on height variation feature parameters, and the height reliability parameters corresponding to the 3D inspection grid are output. The specific steps are as follows: Extracting UAV altitude time series ,in, For the target flight altitude information at time i, For the testing cycle, For the j-th 3D inspection grid; Based on the UAV altitude time series, altitude change characteristic parameters including altitude change amplitude, altitude change rate, and altitude fluctuation frequency were calculated. After normalizing the height variation characteristic parameters, a height reliability assessment model is constructed based on the height variation characteristic parameters, and the height reliability parameters are output. Based on the target flight altitude information, corresponding farmland image data is obtained. The imaging field of view coverage parameters are analyzed according to the camera intrinsic parameter model parameters. The image spatial resolution corresponding to the current three-dimensional inspection grid is calculated. The altitude confidence parameter is introduced into the imaging scale estimation to form a weighted imaging scale description parameter. Based on the imaging scale description parameters, adaptive scale normalization processing is performed on farmland image data to obtain first image data. Based on the first image data, a pest perception and identification strategy is built to classify scale sensitivity levels. Based on the scale sensitivity level, the drone is fed back with height stability adjustment commands to constrain it to perform a second patrol.

2. The artificial intelligence-based intelligent inspection method for unmanned aerial vehicles according to claim 1, characterized in that, Based on the target flight altitude information, corresponding farmland image data is acquired. The imaging field of view coverage parameters are analyzed according to the camera intrinsic parameter model parameters. The image spatial resolution corresponding to the current farmland image is calculated. The altitude confidence parameter is introduced into the imaging scale estimation to form a weighted imaging scale description parameter. The specific steps are as follows: By controlling the drone to perform inspection flight above the corresponding three-dimensional inspection grid according to the target flight altitude information, the drone uses the airborne camera to collect farmland image data of the corresponding three-dimensional inspection grid and records the target flight altitude information corresponding to the image acquisition time. Obtain the intrinsic parameter model parameters of the UAV's onboard camera. The camera's intrinsic parameter model parameters include focal length, camera pixel size, and pixel size parameters. Based on the target flight altitude information and camera intrinsic parameter model parameters, dynamically calculate the imaging field of view coverage parameters of the current UAV at the target flight altitude; Based on the camera intrinsic parameter model parameters, analyze the imaging field of view coverage parameters and calculate the image spatial resolution corresponding to the current 3D inspection grid. By incorporating a high confidence parameter into the imaging scale estimation process, the image spatial resolution and imaging field of view coverage parameters are weighted to form a weighted imaging scale description parameter. The weighted imaging scale description parameters are associated and stored with the corresponding 3D inspection grid.

3. The intelligent inspection method for unmanned aerial vehicles based on artificial intelligence according to claim 2, characterized in that, Based on the camera intrinsic parameter model parameters, analyze the imaging field of view coverage parameters, and calculate the image spatial resolution corresponding to the current 3D inspection grid. The specific steps are as follows: The camera field of view is calculated based on the camera intrinsic parameter model parameters, and the imaging field of view coverage parameters are determined based on the camera field of view. The spatial resolution of the image corresponding to the current 3D inspection grid is calculated based on the imaging field of view coverage parameter.

4. The artificial intelligence-based intelligent inspection method for unmanned aerial vehicles according to claim 2, characterized in that, By incorporating a high-confidence parameter into the imaging scale estimation process, and weighting the image spatial resolution and imaging field of view coverage parameters, a weighted imaging scale description parameter is formed. The specific steps are as follows: By incorporating a high confidence parameter into the image spatial resolution calculation, a weighted spatial resolution is obtained; The imaging field of view coverage parameter is weighted by a high confidence parameter, and the scale description model is constructed by combining the weighted image spatial resolution and imaging field of view coverage parameter to output the imaging scale description parameter.

5. The artificial intelligence-based intelligent inspection method for unmanned aerial vehicles according to claim 1, characterized in that, Based on the imaging scale description parameters, adaptive scale normalization is performed on the farmland image data to obtain the first image data. A pest perception and identification strategy is then built based on this first image data to classify scale sensitivity levels. The specific steps are as follows: Obtain farmland image data and corresponding imaging scale description parameters. Based on the imaging scale description parameters, perform adaptive scale normalization processing on the farmland image data so that the farmland image data acquired under different imaging scale conditions can be expressed in a unified scale space to obtain the first image data. The image data of pest-free farmland obtained in advance by the three-dimensional inspection grid is extracted as the second image data. The image difference of the three-dimensional inspection grid is analyzed by comparing the first image data and the second image data, and the image difference is summarized to evaluate the pest perception sensitivity coefficient of the farmland inspection block. Based on the pest perception sensitivity coefficient, pest targets are divided into different scale sensitivity levels; Scale sensitivity levels include normal sensitivity levels and abnormal sensitivity levels.

6. The artificial intelligence-based intelligent inspection method for unmanned aerial vehicles according to claim 5, characterized in that, The pre-acquired pest-free farmland image data of the 3D inspection grid is extracted as the second image data. The image difference of the 3D inspection grid is analyzed by comparing the first and second image data, and the image difference is summarized to evaluate the pest perception sensitivity coefficient of the farmland inspection block. The specific steps are as follows: The pixel data of the first image data is extracted as the first pixel data, and the pixel data of the second image data is extracted as the second pixel data. The pixel difference is calculated based on the first pixel data and the second pixel data, and the pixel difference is summarized to obtain the image difference of the three-dimensional inspection grid. The pest perception sensitivity coefficient is obtained by summarizing the image differences of the three-dimensional inspection grid within the farmland inspection block.

7. The artificial intelligence-based intelligent inspection method for unmanned aerial vehicles according to claim 6, characterized in that, Based on the pest perception sensitivity coefficient, pest targets are classified into different scale sensitivity levels. The specific steps are as follows: Based on the comparison between the pest perception sensitivity coefficient and the preset sensitivity threshold, if the pest perception sensitivity coefficient is greater than or equal to the preset sensitivity threshold, the corresponding farmland inspection block is at an abnormally sensitive level. If the pest perception sensitivity coefficient is less than the preset sensitivity threshold, the corresponding farmland inspection block is at the normal sensitivity level.

8. An AI-based intelligent drone inspection system, applied to the AI-based intelligent drone inspection method as described in any one of claims 1-7, characterized in that, include: The module includes farmland spatial modeling and analysis, imaging scale weighted estimation, pest perception, and supplementary patrol control feedback. The farmland spatial modeling and analysis module is used to divide the farmland area under inspection into several farmland inspection blocks, model a three-dimensional inspection grid based on the farmland inspection blocks, obtain the canopy height corresponding to the three-dimensional inspection grid, determine the target flight altitude information of the UAV based on the preset relative height between the UAV and the three-dimensional inspection grid, and perform time-series analysis on the target flight altitude information at continuous time to generate altitude confidence parameters. The imaging scale weighted estimation module is used to acquire corresponding farmland image data based on target flight altitude information, analyze the imaging field of view coverage parameters according to camera intrinsic model parameters, calculate the image spatial resolution corresponding to the current three-dimensional inspection grid, and introduce the altitude confidence parameter into the imaging scale estimation to form a weighted imaging scale description parameter. The pest perception module is used to perform adaptive scale normalization processing on farmland image data according to imaging scale description parameters to obtain first image data, and to build a pest perception and recognition strategy based on the first image data to classify scale sensitivity levels. The supplementary patrol control feedback module is used to feed back altitude stability adjustment commands to the UAV based on the scale sensitivity level, thereby constraining the UAV to perform a second supplementary patrol.

Images