Unmanned aerial vehicle low-altitude inspection method based on visual large model

Abstract

The invention relates to the technical field of unmanned aerial vehicle inspection, and particularly discloses an unmanned aerial vehicle low-altitude inspection method based on a visual large model, and the method comprises the following steps: S1, immediately collecting an inspection image after an unmanned aerial vehicle receives a flight interruption instruction, scanning base stations in a preset calibration range, screening and positioning the base stations, and obtaining the position information of the base stations; s2, the inspection image is segmented according to pixels, the significance index of each pixel is calculated, high-significance positioning points are screened, the pixel coordinates of the high-significance positioning points are recorded, and a positioning coordinate set is formed; s3, defining a safety area boundary of unmanned aerial vehicle flight continuation calibration in combination with a preset deviation distance threshold value and the base station position data; s4, the unmanned aerial vehicle flies in the safe area according to a preset path, images are captured in real time, frames to be calibrated are screened, and a current positioning coordinate set is extracted; through comparison with an original set, a continuous flight point position is accurately determined, and accurate continuous inspection of a breakpoint is realized; the problem of re-flight positioning after flight interruption in a complex environment is solved, and the inspection continuity is improved.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) inspection technology, specifically to a low-altitude UAV inspection method based on a large visual model. Background Technology

[0002] In today's era of rapid technological advancement, drone technology has been widely applied in numerous fields, with inspection work being one of its important applications. With its advantages of efficiency, flexibility, and ability to reach complex areas, drones have gradually become a powerful tool for inspection tasks, playing a vital role in areas such as power line inspection, oil pipeline inspection, and forest resource patrol.

[0003] However, in actual inspection work, existing technologies face some pressing problems. One such problem is the frequent occurrence of interrupted flight resumed operation when using drones for inspections. These interruptions can stem from various factors, such as signal transmission interruptions, insufficient battery power requiring temporary battery replacement, or sudden severe weather causing flight pauses.

[0004] When a drone resumes flight after a break in its designated area, a series of problems can easily arise, with inaccurate positioning being particularly prominent. During a break in flight, the initial set of positioning coordinates may change, and the restarted positioning system requires time to recalibrate and acquire accurate position information. During this process, the drone may experience positional deviations, leading to inaccurate inspection data and deviations from the planned inspection route. This not only affects the efficiency of inspection work but may also result in missed important inspection points due to positioning errors, failing to promptly detect potential safety hazards and posing risks to the normal operation and maintenance of related facilities. Therefore, how to solve the problem of inaccurate positioning during drone inspections when resuming flight after a break in its designated area has become a key technical challenge that needs to be addressed and overcome. Summary of the Invention

[0005] The purpose of this invention is to provide a low-altitude inspection method for unmanned aerial vehicles (UAVs) based on a large visual model, and to solve the following technical problems.

[0006] The objective of this invention can be achieved through the following technical solutions: The method for low-altitude inspection of UAVs based on large visual models includes the following steps: Step S1: The operator sends a flight termination signal to the drone based on the operating device. After receiving the flight termination signal, the drone acquires the inspection image, presets the calibration range of the drone, acquires all drone base stations within the calibration range, filters out several positioning base stations, and acquires the base station information of each positioning base station. Step S2: Divide the inspection image into several pixels, obtain the salience index of each pixel, filter out the positioning points according to the salience index, and obtain the pixel coordinates of the positioning points in the inspection image to obtain the positioning coordinate set. Step S3: Set a deviation distance threshold, and obtain the calibration area of ​​the UAV based on the deviation distance threshold and the information of each base station; Step S4: When the UAV resumes flight, the UAV is placed in the calibration area and flies along the preset calibration path. Real-time images on the calibration path are acquired in real time. Images to be calibrated are selected from the real-time images, and the set of positioning coordinates of the images to be calibrated is obtained and recorded as the current positioning coordinate set. The current positioning coordinate set and the positioning coordinate set are compared to determine the UAV's resuming flight point.

[0007] As a further aspect of the present invention: the process of setting the calibration range of the UAV includes: All drone base stations are acquired, and the interval between any two adjacent drone base stations is obtained. The average value of the interval is recorded as the average interval. The calibration range of the drone is obtained in real time with the location of the drone as the center and the average interval as the radius.

[0008] As a further aspect of the present invention: the screening process for positioning base stations includes: Set a threshold n for the number of directions, and obtain the central angle 2 based on the threshold. π / n, according to the central angle, the calibration area is divided into n sub-regions; the drone base station closest to the drone in each sub-region is obtained and recorded as the drone's positioning base station.

[0009] As a further aspect of the present invention: the process of obtaining the saliency index of the pixel includes: Get the pixel values ​​of each pixel {RGB1, RGB2, ..., RGB}. m}, where RGB m This represents the pixel value of the m-th pixel, where m is the total number of pixels; and it retrieves the elevation of each pixel's location {Height1, Height2, ..., Height3}. m ...Height m}, where Height m Let m represent the altitude of the m-th pixel; then we obtain the saliency index of the x-th pixel. Where K is a preset weighting coefficient and K > 1, RGB x This represents the pixel value of the x-th pixel, where x ∈ [1, m] and i is a positive integer, RGB. i Height represents the pixel value of the i-th pixel, where i ∈ [1, m] and i is a positive integer. iHeight represents the altitude of the i-th pixel. x This represents the altitude of the x-th pixel.

[0010] As a further aspect of the present invention: the process of setting the significant threshold includes: Obtain the saliency indices {Si1, Si2, ..., Si} of all pixels. num}, where Si num Let denot num, where num represents the total number of significance indices, then the significance threshold is obtained. Si t Let Z represent the t-th significance index, where t∈[1, num] and t is a positive integer, and Z is a preset multiple threshold, where Z is an integer.

[0011] As a further aspect of the present invention: the process of setting the deviation distance threshold includes: Several sample drones and several drone base stations were selected as sample base stations. Each sample drone and each sample base station were randomly paired to obtain several combinations. Under different weather conditions, the distance between the sample drone and the sample base station within each combination was recorded as the acquisition distance, and the distance between the sample drone and the sample base station within each combination was recorded as the actual distance. Based on the acquisition distance and the actual distance, the error |DD´| was obtained, where D is the actual distance and D´ is the acquisition distance. From the errors of each combination, an error set was obtained, and the standard deviation of all errors within the error set was obtained. σ Then, based on the aforementioned standard deviation, the maximum positioning error is obtained. ,in The maximum geometric magnification factor is defined as the maximum positioning error, which is denoted as the deviation distance threshold.

[0012] As a further aspect of the present invention: the process of obtaining the calibration area of ​​the UAV includes: Based on the base station information, the distance between the UAV and each positioning base station before the flight was terminated is obtained, and the unique identifier of the positioning base station is obtained; based on the unique identifier, the position coordinates of each positioning base station are determined and recorded as positioning base station coordinates; a circular area is obtained with the positioning base station coordinates as the center and the distance between the UAV and the positioning base station coordinates before the flight was terminated as the radius; the circular areas of each positioning base station coordinate are obtained, and the areas where the circular areas intersect are obtained, and the center point of the area is recorded as the estimated termination point of the UAV; Obtain the line connecting the estimated flight break point and the coordinates of the positioning base station. Using the estimated flight break point as the midpoint, extract a line of length [length missing]. The line segments are obtained, and finally n line segments are obtained; all the endpoints of the n line segments are obtained, and the endpoints are connected in sequence to obtain the calibration area.

[0013] As a further aspect of the present invention: the process of setting the calibration path includes: Obtain a virtual rectangle of the calibration area, which is the smallest rectangle that completely encloses the calibration area; set a path interval, obtain the length of the virtual rectangle, obtain several parallel lines of the specified length within the virtual rectangle according to the path interval, and trim the portions of each parallel line outside the calibration area through the boundary of the calibration area to obtain the calibration path.

[0014] The beneficial effects of this invention are: This invention analyzes the pixel saliency index (combining RGB values ​​and altitude) of inspection images using a large visual model to extract high-discrimination positioning points and generate a set of positioning coordinates. After flight interruption, by comparing the real-time image coordinate set during flight resumption with the original set, sub-meter-level flight resumption point calibration is achieved, solving the positioning failure problem of traditional GPS / RTK in strong electromagnetic interference or complex environments. Based on the base station layout and maximum positioning error model, the calibration area boundary is intelligently delineated to ensure that the UAV is always within a safe range during flight resumption, avoiding the risk of collision due to signal drift. By fusing multi-base station ranging data and visual features, positioning errors are dynamically corrected through the geometrical precision factor (GDOP), significantly improving positioning robustness in adverse weather or complex terrain. It reduces reliance on high-precision GNSS equipment or lidar, and utilizes existing base station networks and visual algorithms to achieve low-cost, high-reliability autonomous flight resumption, suitable for large-scale deployment in industries such as power and oil and gas. This method breaks through the technical bottleneck of traditional UAV inspection during flight interruption and resumption, improves the continuous operation capability in complex scenarios, and provides a standardized solution for automated inspection in the low-altitude economy. Attached Figure Description

[0015] The invention will now be further described with reference to the accompanying drawings.

[0016] Figure 1 This is a schematic diagram of the steps of the UAV low-altitude inspection method based on a large visual model according to the present invention; Figure 2 This is a flowchart illustrating the UAV low-altitude inspection method based on a large visual model according to the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Please see Figure 1 As shown, this invention is a low-altitude inspection method for UAVs based on a large visual model, comprising the following steps: Step S1: The operator sends a flight termination signal to the drone based on the operating device. After receiving the flight termination signal, the drone acquires the inspection image, presets the calibration range of the drone, acquires all drone base stations within the calibration range, filters out several positioning base stations, and acquires the base station information of each positioning base station. Specifically, step S1 is the initial stage of resuming flight calibration after the drone's flight is interrupted. The operator sends a flight interruption command to the drone through ground operation equipment, such as a remote controller or ground station system. Flight interruption may be triggered by signal interruption, insufficient power, or severe weather. After receiving the command, the drone immediately initiates an emergency response: on the one hand, it collects high-definition images of the current inspection area in real time; on the other hand, it automatically enters the calibration preparation stage, presets the calibration range, and first counts the adjacent intervals of all surrounding drone base stations, calculates the average interval as the reference radius, and then uses the drone's current location as the center to delineate a circular calibration range with this radius to ensure coverage of effective base stations. Subsequently, the drone searches for all base stations within the designated calibration range and selects and locates base stations according to the principle of uniform distribution in multiple directions; a threshold n for the number of directions is set (e.g., n=3 or 4), and the calibration range is divided according to the central angle 2. π The area is divided into n equal fan-shaped sub-regions. Within each sub-region, the base station closest to the drone is selected as the positioning base station to ensure that the base stations are evenly distributed in space and reduce positioning errors. Finally, the drone records key information about these positioning base stations, including the real-time distance between the base station and itself, and the unique geographic identifier of the base station. This information is used for subsequent matching of base station coordinates, laying the data foundation for subsequent calibration area delineation and re-flight point positioning. In a preferred embodiment of the present invention, the process of setting the calibration range of the UAV includes: All drone base stations are acquired, and the interval between each two adjacent drone base stations is obtained. The average value of the interval is recorded as the average interval. The calibration range of the drone is obtained in real time with the location of the drone as the center and the average interval as the radius. In a preferred embodiment of the present invention, the inspection image is an image of the inspection range taken in real time by the UAV during the inspection. In a preferred embodiment of the present invention, the screening process for positioning base stations includes: Set a threshold n for the number of directions, and obtain the central angle 2 based on the threshold. π / n, according to the central angle, the calibration area is divided into n sub-regions; the drone base station closest to the drone in each sub-region is obtained and recorded as the drone's positioning base station; In a preferred embodiment of the present invention, the base station information of the positioning base station includes the distance between the positioning base station and the drone, and the unique identifier of the positioning base station; Step S2: Divide the inspection image into several pixels, obtain the salience index of each pixel, filter out the positioning points according to the salience index, and obtain the pixel coordinates of the positioning points in the inspection image to obtain the positioning coordinate set. Specifically, step S2 is the core step of extracting key positioning features through image analysis. The UAV divides the inspection images collected before the flight suspension into several independent pixel units according to the pixel grid, and calculates the salience index of each pixel one by one. This index comprehensively considers the visual and geographical features of the pixel: on the one hand, it incorporates the RGB color value of the pixel to quantify the visual salience through the color difference with other pixels; on the other hand, it combines the altitude data of the ground location corresponding to the pixel to reflect the terrain features through the altitude difference with the surrounding pixels. During the calculation, the influence of key features is strengthened by preset weight coefficients. Subsequently, the drone statistically analyzes the saliency index of all pixels and sets a saliency threshold. The saliency threshold is calculated by the average of all saliency indices and a preset multiple threshold Z. Pixels with a saliency index exceeding the threshold are marked as high-saliency positioning points. These points usually correspond to obvious ground features, such as building corners, road edges, and special landforms. Finally, a coordinate system is established with any pixel in the inspection image as the origin. The pixel coordinates of the center point of each positioning point in the coordinate system are recorded and integrated to form the original positioning coordinate set, providing a benchmark reference for coordinate comparison during subsequent flight. In a preferred embodiment of the present invention, the process of obtaining the saliency index of the pixel includes: Get the pixel values ​​of each pixel {RGB1, RGB2, ..., RGB}. m}, where RGB m This represents the pixel value of the m-th pixel, where m is the total number of pixels; and it retrieves the elevation of each pixel's location {Height1, Height2, ..., Height3}. m ...Height m}, where Height m Let m represent the altitude of the m-th pixel; then we obtain the saliency index of the x-th pixel. Where K is a preset weighting coefficient and K > 1, RGB x This represents the pixel value of the x-th pixel, where x ∈ [1, m] and i is a positive integer, RGB. i Height represents the pixel value of the i-th pixel, where i ∈ [1, m] and i is a positive integer. i Height represents the altitude of the i-th pixel. x This represents the altitude of the x-th pixel; In a preferred embodiment of the present invention, the process of selecting the positioning points includes: A saliency threshold is set, and pixels whose saliency index exceeds the saliency threshold are recorded as localization points; otherwise, the saliency index of the next pixel is obtained and compared with the saliency threshold. The process of setting the significance threshold includes: Obtain the saliency indices {Si1, Si2, ..., Si} of all pixels. num}, where Si num Let denot num, where num represents the total number of significance indices, then the significance threshold is obtained. Si t Let Z represent the t-th significance index, where t∈[1, num] and t is a positive integer, and Z is a preset multiple threshold, where Z is an integer; In a preferred embodiment of the present invention, the process of obtaining the pixel coordinates includes: In the inspection image, any pixel is selected as the origin, a coordinate system is established based on the origin, and the coordinates of the center point of the positioning point in the coordinate system are obtained and recorded as the pixel coordinates of the positioning point. Step S3: Set a deviation distance threshold, and obtain the calibration area of ​​the UAV based on the deviation distance threshold and the information of each base station; Specifically, step S3 is a crucial step in defining the safety boundary for drone flight continuation calibration. First, a deviation distance threshold is set to quantify the maximum allowable deviation range for positioning. Then, a calibration area is defined. The calibration area is set based on the positioning base station information. The distance between the drone and each positioning base station before the flight termination and the coordinates of the base station are extracted. A circular area is drawn with each base station coordinate as the center and the corresponding distance as the radius. The center point of the intersection of each circular area is the estimated flight termination point. Using the estimated flight termination point as the midpoint, line segments with a length of twice the deviation distance threshold are cut from the line connecting the estimated flight termination point to each base station. The endpoints of all line segments are collected and connected sequentially. The resulting polygonal area is the safe calibration area for drone flight continuation, ensuring that subsequent flight continuation positioning is carried out within a controllable error range. In a preferred embodiment of the present invention, the process of setting the deviation distance threshold includes: Several sample drones and several drone base stations were selected as sample base stations. Each sample drone and each sample base station were randomly paired to obtain several combinations. Under different weather conditions, the distance between the sample drone and the sample base station within each combination was recorded as the acquisition distance, and the distance between the sample drone and the sample base station within each combination was recorded as the actual distance. Based on the acquisition distance and the actual distance, the error |DD´| was obtained, where D is the actual distance and D´ is the acquisition distance. From the errors of each combination, an error set was obtained, and the standard deviation of all errors within the error set was obtained. σ Then, based on the aforementioned standard deviation, the maximum positioning error is obtained. ,in The maximum geometric magnification factor is defined as the maximum positioning error, which is denoted as the deviation distance threshold. Specifically, when locating a drone based on its distance to a nearby drone base station, the positioning results often contain significant errors due to the combined effects of multiple factors. These errors may stem from environmental interference during signal propagation, such as building obstruction, tree reflection, and electromagnetic multipath effects, leading to deviations in ranging data. Simultaneously, the accuracy of the drone base station's geographic coordinates, measurement errors in the drone's own sensors (such as GPS drift and IMU noise), and limitations in signal processing algorithms can further exacerbate positioning inaccuracies. Furthermore, complex terrain or dynamic weather conditions, such as changes in wind speed and temperature gradients, may cause variations in signal propagation speed, resulting in systematic biases in distance calculations based on time or frequency differences. Therefore, based on the maximum error generated, a deviation distance threshold is set to determine the calibration area; and the positioning error depends not only on the ranging error, but also on the spatial layout of the base station. This influence is quantified by the geometric precision factor, and the maximum value of the positioning error = error × maximum geometric precision factor; For the geometrical precision factor (GDOP), the smaller the GDOP value, the more uniform the base station distribution, and the smaller the impact of ranging error on positioning, such as an equilateral triangle or a regular tetrahedron; the larger the GDOP value, the worse the base station layout, and the more significant the error amplification, such as when all base stations are concentrated in the same direction, or when the drone is close to the edge of the base station distribution; therefore, it is necessary to determine the maximum GDOP value under the current deployment through base station layout simulation. For two-dimensional positioning (using several base stations), the relationship between the covariance matrix of the positioning error and GDOP is as follows: ,in Let be the covariance matrix; the covariance matrix describes the dispersion of the positioning results and reflects the statistical characteristics of the error; if the ranging error itself is large, i.e. σThe larger the base station distribution (GDOP), or the worse the base station layout (i.e., the larger the GDOP), the more scattered the positioning results will be, and the greater the error. In a normal distribution, three times the standard deviation corresponds to a probability of 99.7%. This means that if positioning is repeated multiple times, 99.7% of the error will fall within the range of GDOP. Within the range; ultimately obtained ; In a preferred embodiment of the present invention, the process of obtaining the calibration area of ​​the UAV includes: Based on the base station information, the distance between the UAV and each positioning base station before the flight was terminated is obtained, and the unique identifier of the positioning base station is obtained; based on the unique identifier, the position coordinates of each positioning base station are determined and recorded as positioning base station coordinates; a circular area is obtained with the positioning base station coordinates as the center and the distance between the UAV and the positioning base station coordinates before the flight was terminated as the radius; the circular areas of each positioning base station coordinate are obtained, and the areas where the circular areas intersect are obtained, and the center point of the area is recorded as the estimated termination point of the UAV; Obtain the line connecting the estimated flight break point and the coordinates of the positioning base station. Using the estimated flight break point as the midpoint, extract a line of length [length missing]. The line segments are obtained, and finally n line segments are obtained; all the endpoints of the n line segments are obtained, and the endpoints are connected in sequence to obtain the calibration area; Step S4: When the UAV resumes flight, the UAV is placed in the calibration area and flies along the preset calibration path. Real-time images on the calibration path are acquired in real time. Images to be calibrated are selected from the real-time images, and the set of positioning coordinates of the images to be calibrated is obtained and recorded as the current positioning coordinate set. The current positioning coordinate set and the positioning coordinate set are compared to determine the UAV's resuming flight point. Specifically, step S4 is the core stage for achieving accurate resuming of drone flights after a breakpoint. When the drone resumes flight, it first enters the safe calibration area defined in step S3, generates a calibration path according to preset rules, and generates multiple flight paths parallel to the long side within the rectangle by first determining the smallest virtual rectangle that encloses the calibration area, setting the path interval according to the drone sensor field of view parameters, and then using the boundary of the calibration area to cut off the flight path segments outside the area to form an actual flight path that fits the calibration area. The UAV captures high-definition images in real time while flying along the calibration path. Image recognition technology is used to calculate the similarity between the real-time images and the inspection images before the flight termination, and frames with similarity exceeding a threshold are selected for calibration. Step S2 is repeated for these frames to be calibrated: pixels are segmented, saliency index is calculated, high saliency positioning points are selected and their coordinates are recorded to form the current positioning coordinate set. Finally, the current set of positioning coordinates is compared with the original set of positioning coordinates generated in step S2. If the two match (coordinates coincide or deviation is within the allowable range), the real-time position of the UAV corresponding to the frame to be calibrated is determined as the re-flight point to ensure that the UAV accurately resumes inspection from the breakpoint and avoids missed inspections or repeated inspections. In a preferred embodiment of the present invention, the process of setting the calibration path includes: Obtain a virtual rectangle of the calibration area, which is the smallest rectangle that completely encloses the calibration area; set a path interval, obtain the length of the virtual rectangle, obtain several parallel lines of the specified length within the virtual rectangle according to the path interval, and trim the portions of each parallel line outside the calibration area through the boundary of the calibration area to obtain the calibration path; The process of setting the path interval includes: If the side length L of the inspection image is obtained, then the setting range of the path interval is [80%L, 90%]; In a preferred embodiment of the present invention, the screening process of the image to be calibrated includes: Based on image recognition technology, the image similarity between each real-time image and the inspection image is obtained. A similarity threshold is set, and real-time images whose image similarity exceeds the similarity threshold are recorded as images to be calibrated. It should be noted that the process of obtaining the current location coordinate set is the same as the process of obtaining the location coordinate set, and will not be described in detail here; In a preferred embodiment of the present invention, the process for determining the drone's continued flight point includes: Let the current set of positioning coordinates be denoted as J, and the set of positioning coordinates be denoted as K. If J=K, then obtain the position coordinates of the UAV when it acquires the image to be calibrated, and denot it as the UAV's re-flight point. It is worth noting that the Euclidean distance between coordinates can also be calculated to see if it is less than a minimum threshold. If it is less than the minimum threshold, it means that the feature points of the current image to be calibrated are precisely aligned with the feature points of the inspection image before the flight termination in the flight continuation calibration path.

[0019] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the present invention should still fall within the scope of the invention.

Claims

1. A method for low-altitude UAV inspection based on a large visual model, characterized in that, Includes the following steps: Step S1: The operator sends a flight termination signal to the drone based on the operating device. After receiving the flight termination signal, the drone acquires the inspection image, presets the calibration range of the drone, acquires all drone base stations within the calibration range, filters out several positioning base stations, and acquires the base station information of each positioning base station. Step S2: Divide the inspection image into several pixels, obtain the salience index of each pixel, filter out the positioning points according to the salience index, and obtain the pixel coordinates of the positioning points in the inspection image to obtain the positioning coordinate set. Step S3: Set a deviation distance threshold, and obtain the calibration area of ​​the UAV based on the deviation distance threshold and the information of each base station; Step S4: When the UAV resumes flight, the UAV is placed in the calibration area and flies along the preset calibration path. Real-time images on the calibration path are acquired in real time. Images to be calibrated are selected from the real-time images, and the set of positioning coordinates of the images to be calibrated is obtained and recorded as the current positioning coordinate set. The current positioning coordinate set and the positioning coordinate set are compared to determine the UAV's resuming flight point.

2. The UAV low-altitude inspection method based on a large visual model according to claim 1, characterized in that, In step S1, the process of setting the calibration range of the UAV includes: All drone base stations are acquired, and the interval between any two adjacent drone base stations is obtained. The average value of the interval is recorded as the average interval. The calibration range of the drone is obtained in real time with the location of the drone as the center and the average interval as the radius.

3. The UAV low-altitude inspection method based on a large visual model according to claim 1, characterized in that, In step S1, the screening process for location base stations includes: Set a threshold n for the number of directions, and obtain the central angle 2 based on the threshold. π / n, according to the central angle, the calibration area is divided into n sub-regions; the drone base station closest to the drone in each sub-region is obtained and recorded as the drone's positioning base station.

4. The UAV low-altitude inspection method based on a large visual model according to claim 1, characterized in that, In step S2, the process of obtaining the saliency index of the pixel includes: Get the pixel values ​​of each pixel {RGB1, RGB2, ..., RGB}. m }, where RGB m This represents the pixel value of the m-th pixel, where m is the total number of pixels; and it retrieves the elevation of each pixel's location {Height1, Height2, ..., Height3}. m ...Height m }, where Height m Let m represent the altitude of the m-th pixel; then we can obtain the saliency index of the x-th pixel. Where K is a preset weighting coefficient and K > 1, RGB x This represents the pixel value of the x-th pixel, where x ∈ [1, m] and i is a positive integer, RGB. i Height represents the pixel value of the i-th pixel, where i ∈ [1, m] and i is a positive integer. i Height represents the altitude of the i-th pixel. x This represents the altitude of the x-th pixel.

5. The UAV low-altitude inspection method based on a large visual model according to claim 4, characterized in that, In step S2, the process of setting the significance threshold includes: Obtain the saliency indices {Si1, Si2, ..., Si} of all pixels. num }, where Si num Let denot num, where num represents the total number of significance indices, then the significance threshold is obtained. Si t Let Z represent the t-th significance index, where t∈[1, num] and t is a positive integer, and Z is a preset multiple threshold, where Z is an integer.

6. The UAV low-altitude inspection method based on a large visual model according to claim 1, characterized in that, In step S3, the process of setting the deviation distance threshold includes: Several sample drones and several drone base stations were selected as sample base stations. Each sample drone and each sample base station were randomly paired to obtain several combinations. Under different weather conditions, the distance between the sample drone and the sample base station within each combination was recorded as the acquisition distance, and the distance between the sample drone and the sample base station within each combination was recorded as the actual distance. Based on the acquisition distance and the actual distance, the error |DD´| was obtained, where D is the actual distance and D´ is the acquisition distance. From the errors of each combination, an error set was obtained, and the standard deviation of all errors within the error set was obtained. σ Then, based on the standard deviation, the maximum positioning error is obtained. max =3× σ ×GDOP max GDOP max The maximum geometric magnification factor is defined as the maximum positioning error, which is denoted as the deviation distance threshold.

7. The UAV low-altitude inspection method based on a large visual model according to claim 1, characterized in that, In step S3, the process of obtaining the calibration area of ​​the UAV includes: Based on the base station information, the distance between the UAV and each positioning base station before the flight was terminated is obtained, and the unique identifier of the positioning base station is obtained; based on the unique identifier, the position coordinates of each positioning base station are determined and recorded as positioning base station coordinates; a circular area is obtained with the positioning base station coordinates as the center and the distance between the UAV and the positioning base station coordinates before the flight was terminated as the radius; the circular areas of each positioning base station coordinate are obtained, and the areas where the circular areas intersect are obtained, and the center point of the area is recorded as the estimated termination point of the UAV; Obtain the line connecting the estimated flight break point and the coordinates of the positioning base station. Using the estimated flight break point as the midpoint, extract a line with a length of 2error from the line. max The line segments are obtained, and finally n line segments are obtained; all the endpoints of the n line segments are obtained, and the endpoints are connected in sequence to obtain the calibration area.

8. The UAV low-altitude inspection method based on a large visual model according to claim 1, characterized in that, In step S4, the process of setting the calibration path includes: Obtain a virtual rectangle of the calibration area, which is the smallest rectangle that completely encloses the calibration area; set a path interval, obtain the length of the virtual rectangle, obtain several parallel lines of the specified length within the virtual rectangle according to the path interval, and trim the portions of each parallel line outside the calibration area through the boundary of the calibration area to obtain the calibration path.

Images