Intelligent shooting patrol method of unmanned aerial vehicle

Abstract

The invention provides an unmanned aerial vehicle intelligent shooting patrol method comprising the following steps: 1, determining a to-be-patrolled area, and presetting unmanned aerial vehicle patrol parameters; the method includes: presetting an unmanned aerial vehicle flight height and a lens initial focal length according to a detection target type; step 2, according to preset unmanned aerial vehicle inspection parameters, carrying out aerial beat sub-region division; 3, generating a sub-region aerial photography path according to the divided aerial photography sub-regions; 4, the unmanned aerial vehicle carries out wide-angle shooting and target rough detection according to the sub-region aerial shooting path; 5, performing long-focus fine shooting according to the results of wide-angle shooting and target rough detection to obtain a long-focus photo; and 6, carrying out high-precision classification and defect diagnosis on the obtained long-focus photo, generating a detection report, and outputting a target distribution diagram with coordinates and a defect statistical table. According to the invention, after wide-angle coarse detection, long-focus shooting is only triggered for non-repeated and high-confidence targets, so that the number of invalid zooming times is reduced; and the diagnosis accuracy of long-focus photos is improved through fusion model classification.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned aerial vehicle inspection, and in particular to a method for intelligent photography and inspection by unmanned aerial vehicles. Background Art

[0002] In the field of drone inspections (such as forest pest monitoring or urban infrastructure inspections), existing technologies generally use the following two methods or a combination of the two:

[0003] Wide-angle, full-coverage photography: Drones perform grid-based aerial photography at a fixed altitude and focal length, but this results in a large amount of redundant data (such as duplicate targets in overlapping areas of adjacent photos), increases the back-end processing burden, and makes it difficult to accurately locate small targets (such as diseased trees or cracks in manhole covers).

[0004] Manual zoom: This relies on the pilot to observe the scene in real time and manually switch the lens. This is inefficient and prone to missed inspections, especially during long-distance inspections, where consistency is difficult to ensure.

[0005] A drawback of existing inspection methods is that objects in overlapping areas of adjacent photos are identified multiple times. Traditional methods rely solely on simple IOU (Intersection-over-Union) deduplication, ignoring differences in actual geographic coordinates (such as coordinate errors caused by pixel offset when shooting at high angles). After a lightweight model performs a rough inspection in wide-angle photos, it still requires a telephoto lens to capture all suspected objects, resulting in a large number of false positives or duplicates. Summary of the Invention

[0006] In view of the shortcomings of the existing technology, the present invention provides a drone intelligent shooting and inspection method, which solves the problems mentioned in the background technology.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0008] A drone intelligent photography and inspection method includes the following steps:

[0009] Step 1: Determine the area to be inspected and preset the drone inspection parameters, including: presetting the drone's flight altitude and lens initial focal length based on the type of inspection target;

[0010] Step 2: Divide the aerial photography sub-areas according to the preset drone inspection parameters;

[0011] Step 3: Generate sub-area aerial photography paths based on the divided aerial photography sub-areas;

[0012] Step 4: The drone performs wide-angle photography and rough target inspection based on the sub-area aerial photography path;

[0013] Step 5: Perform telephoto fine shooting based on the results of wide-angle shooting and rough target inspection to obtain a telephoto photo;

[0014] Step 6: Perform high-precision classification and defect diagnosis on the acquired telephoto photos, generate an inspection report, and output a target distribution map with coordinates and a defect statistics table.

[0015] Furthermore, determining the area to be inspected includes determining the inspection area on an electronic map; the inspection area includes forest areas and / or urban roads.

[0016] Furthermore, the drone flight altitude and lens initial focal length are preset according to the detection target type, including: when the detection target type is a forest area, the drone flight altitude is preset to 50 meters and the lens initial focal length is preset to wide-angle mode;

[0017] When the detection target is a city road, the preset drone flight altitude is 30 meters, and the initial focal length of the lens is wide-angle mode.

[0018] Furthermore, the aerial photography sub-area division according to the preset drone inspection parameters includes:

[0019] 2.1. Determine the field of view of the drone's wide-angle lens based on the lens's initial focal length;

[0020] 2.2. Calculate the coverage of a single photo based on the lens’ field of view and the drone’s flight altitude, and divide the inspection area into multiple sub-areas.

[0021] Furthermore, the calculation formula for calculating the coverage range of a single photo is: coverage range = horizontal coverage length × vertical coverage length; where horizontal coverage length = 2 × H × tan(HFOV / 2), vertical coverage length = 2 × H × tan(VFOV / 2); where H is the flight altitude of the drone, HFOV is the horizontal field of view angle, and VFOV is the vertical field of view angle.

[0022] Furthermore, the logic of the sub-area division includes dividing the patrol area into grids according to the coverage area, and adjacent sub-areas need to maintain a 10% to 20% overlap.

[0023] Furthermore, the aerial photography path of the sub-area is generated according to the divided aerial photography sub-area, including taking the center point of the aerial photography sub-area as the shooting point, and the path of the drone traversing the center point of the sub-area in an "S" shape or "n" shape in sequence is the aerial photography path.

[0024] Furthermore, the UAV performs wide-angle photography and rough target inspection according to the sub-area aerial photography path; including:

[0025] 4.1. The drone flies to the center of each sub-area according to the sub-area aerial photography path and takes wide-angle photos.

[0026] 4.2. Use a lightweight target detection algorithm to identify suspected targets and exclude duplicate targets in the obtained photos.

[0027] Furthermore, the exclusion of duplicate targets includes: converting the pixel coordinates of the detected targets into UTM geographic coordinates according to the GPS position, altitude and camera parameters of the drone; clustering the coordinates using the DBSCAN algorithm, retaining the targets with the highest confidence within the same cluster, and eliminating the rest.

[0028] The present invention provides a method for intelligent drone photography and inspection. Compared with the existing technology, it has the following advantages:

[0029] The present invention adopts wide-angle and telephoto collaborative detection. After a wide-angle rough inspection, telephoto shooting is triggered only for non-repetitive and high-confidence targets, reducing the number of invalid zoom times; the telephoto photos are classified through a fusion model and combined with spectral data to improve the diagnostic accuracy.

[0030] The present invention converts pixel coordinates into real geographic coordinates through drone positioning, altitude and field of view, and combines clustering to solve the error problem of traditional IOU deduplication during high-altitude oblique shooting; only the highest confidence target is retained in the same cluster to reduce the amount of redundant data. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0032] Figure 1 Shown is a schematic flow chart of the method of the present invention;

[0033] Figure 2 A schematic diagram showing the division of the area to be inspected into sub-areas in an embodiment of the present invention is shown;

[0034] Figure 3 A schematic diagram of the aerial photography path in the implementation case of the present invention is shown;

[0035] Figure 4 A schematic diagram comparing wide-angle shooting and telephoto shooting in an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0036] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention are clearly and completely described. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0037] Example

[0038] To solve the technical problems in the background technology, the following drone intelligent photography and inspection method is provided:

[0039] Combine Figure 1 As shown, the present invention provides a drone intelligent shooting and inspection method, comprising the following steps:

[0040] Step 1: Determine the area to be inspected and preset the drone inspection parameters, including: presetting the drone's flight altitude and lens initial focal length based on the type of inspection target; specifically:

[0041] Determining the area to be inspected is to determine the inspection area on an electronic map (such as a GIS system or drone control software). For example, the inspection area includes but is not limited to a forest area or an urban road.

[0042] The drone's flight altitude and initial lens focal length are preset based on the target type. Exemplarily, these targets include, but are not limited to, forest areas, exemplary pine wilt-infected trees with features such as withered crowns and discolored needles, and urban roads, exemplary city manhole covers with defects such as damage, displacement, and flooding. When the target type is pine wilt-infected trees, the drone's flight altitude is preset to 50 meters; when the target type is city manhole covers, the drone's flight altitude is preset to 30 meters. The initial lens focal length is exemplarily set to wide-angle mode, such as 24mm.

[0043] Step 2: Divide the aerial photography sub-areas according to the preset drone inspection parameters; specifically including:

[0044] 2.1. Determine the field of view of the drone's wide-angle lens based on the lens's initial focal length;

[0045] 2.2. Calculate the coverage of a single photo based on the lens’ field of view and the drone’s flight altitude, and divide the inspection area into multiple sub-areas.

[0046] The method in 2.1 for calculating the field of view (FOV) of a drone wide-angle lens based on the initial focal length of the lens is as follows: To calculate the field of view of a drone wide-angle lens, the following parameters are required:

[0047] Lens initial focal length (focal length): such as 24mm; camera sensor size (such as APS-C, 1-inch, full-frame, etc.); sensor crop factor (Crop Factor, required only when the focal length is "full-frame equivalent").

[0048] 2.11 Calculation formula

[0049] The field of view (FOV) is calculated as follows: FOV = 2 × arctan [sensor side length / (2 × actual focal length)];

[0050] Where: FOV: Field of View (horizontal / vertical / diagonal, unit: degrees); Sensor Side Length: horizontal (width), vertical (height), or diagonal size (unit: mm); Actual Focal Length: physical focal length of the lens (unit: mm). Equivalent focal length conversion must be considered.

[0051] Calculation steps

[0052] (1) Determine the sensor size. Different drone cameras have different sensor sizes. Common ones are shown in the following table:

[0053] Sensor Type Horizontal size (mm) Vertical size (mm) Diagonal size (mm) crop factor 1 inch (like DJI Air 3) 13.2 8.8 15.9 2.7 APS-C (such as the Zenmuse X5) 22.2 14.8 26.7 1.5~1.6 Full-frame (such as Zenmuse X9) 36.0 24.0 43.3 1.0

[0054] (2) Determine the actual focal length

[0055] If the lens is marked with "equivalent full-frame focal length" (such as "24mm"), it needs to be converted to the actual focal length: actual focal length = equivalent focal length / crop factor;

[0056] Example:

[0057] 1-inch sensor (crop factor 2.7), equivalent to 24mm → actual focal length = 24 / 2.7 ≈ 8.89mm;

[0058] APS-C sensor (crop factor 1.5), equivalent to 24mm → actual focal length = 24 / 1.5 = 16mm;

[0059] If the lens is marked with the actual focal length (such as 12mm for the MFT system), no conversion is required.

[0060] (3) Calculate the field of view (FOV)

[0061] Select the desired FOV type (horizontal / vertical / diagonal) and substitute into the formula:

[0062] HFOV (horizontal) = 2 × arctan [(sensor horizontal width / (2 × actual focal length)]; VFOV (vertical) = 2 × arctan [(sensor vertical height / (2 × actual focal length)].

[0063] 2.13 Calculation Example

[0064] Case 1: DJI Mavic 3 (1-inch sensor, 24mm-equivalent wide-angle lens)

[0065] Equivalent focal length: 24mm;

[0066] Sensor size: horizontal 13.2mm, vertical 8.8mm;

[0067] Actual focal length: 24 / 2.7 ≈ 8.89mm;

[0068] Horizontal field of view (HFOV): HFOV = 2 × arctan [13.2 / (2 × 8.89)] ≈ 64.6°;

[0069] Vertical field of view (VFOV): VFOV = 2 × arctan [8.8 / (2 × 8.89)] ≈ 46.4°;

[0070] Example 2: APS-C sensor (Zenmuse X5, actual focal length 24mm)

[0071] Actual focal length: 24mm (not equivalent);

[0072] Sensor size: horizontal 22.2mm;

[0073] Horizontal field of view (HFOV): HFOV=2×arctan[22.2 / (2×24)]≈50.5°.

[0074] 2.14. Simplified formula (quick estimate)

[0075] If the sensor size is much smaller than the focal length, the approximate calculation is: FOV ≈ (sensor side length / focal length) × 57.3°;

[0076] Example (APS-C, 24mm focal length):

[0077] HFOV≈(22.2 / 24)×57.3≈53° (close to the exact calculated 50.5°)

[0078] 2.15 Application of UAV Aerial Photography in Planning

[0079] 2.151: Set the initial focal length (e.g., 24mm) → Calculate or use a table to determine the FOV (e.g., 64.6°).

[0080] 2.152: Calculate the coverage of a single photo based on the flight altitude (H):

[0081] Coverage width = 2 × H × tan(HFOV / 2); Example (H = 50m, HFOV = 64.6°):

[0082] Coverage width = 2 × 50 × tan (32.3°) ≈ 63.4.

[0083] 2.2. Calculate the coverage of a single photo based on the lens's field of view and the drone's flight altitude; divide the inspection area into multiple sub-areas.

[0084] The coverage of a single photo is calculated. The coverage depends on the flight altitude (H) and the field of view (FOV). The formula is as follows:

[0085] Single-sided coverage length = 2 × H × tan(FOV / 2); Example (pine wilt disease tree inspection, H = 50m): Horizontal FOV = 60°, then horizontal coverage length: 2 × 50 × tan(30°) ≈ 57.7 meters; Vertical FOV (assuming 40°): 2 × 50 × tan(20°) ≈ 36.4 meters; Area covered by a single photo: 57.7m × 36.4m ≈ 2100㎡ (rectangular area).

[0086] Manhole cover detection (H=30m): Similarly, the horizontal coverage length is ≈34.6m (when FOV=60°).

[0087] Sub-area division logic and overlap settings include: dividing the patrol area into grids based on coverage, and adjacent sub-areas must maintain a 10% to 20% overlap.

[0088] Implementation case: Figure 2 As shown in the figure, after determining the area to be inspected, when the detection target type is pine wood nematode diseased trees, the drone flight altitude and lens initial focal length are preset; the preset drone flight altitude is 50 meters, the horizontal FOV is 60°, and the vertical FOV is 40°; then the horizontal coverage length is: 2×50×tan(30°)≈57.7 meters; the vertical FOV (assuming it is 40°): 2×50×tan(20°)≈36.4 meters; the coverage area of ​​a single photo is: 57.7m × 36.4m ≈ 2100㎡ (rectangular area).

[0089] If the total length of the area is 100m and a single sheet covers 57.7m, then:

[0090] First image: 0~57.7m;

[0091] Second sheet: 57.7m × (1-20%) = 46.2m starting point, covering to 103.9m (ensuring 11.5m overlap). Divide the inspection area into zones 1, 2, 3, 4, 5, and 6.

[0092] Necessity of overlap: Avoid missing edge targets.

[0093] Step 3: Generate a sub-region aerial photography path based on the divided aerial photography sub-regions; the sub-region aerial photography path needs to meet full coverage without omissions to ensure that all sub-regions are completely photographed to avoid missed inspections (such as diseased trees or manhole cover defects); it must also meet minimum repeated coverage, and the overlap rate of adjacent sub-regions is controlled at 10%~20%, which not only avoids missed inspections but also reduces redundant data; for example, the drone traverses the sub-region center points in an "S" shape, or traverses the sub-region center points in an "n" shape. For example, Figure 3 As shown, the center points of Area 1, Area 2, Area 3, Area 4, Area 5 and Area 6 are A, B, C, D, E and F; the aerial photography path of the UAV is set to A→B→C→D→E→F; it traverses the center points of the sub-areas in an "n" shape.

[0094] Step 4: The drone performs wide-angle photography and rough target inspection based on the sub-area aerial photography path; specifically, it includes:

[0095] 4.1. The drone will fly to the center of each sub-area (e.g., A→B→C→D→E→F) according to the sub-area aerial photography path and capture wide-angle photos. Using GPS / RTK positioning, the drone will fly to the pre-set sub-area center, hover, and adjust its attitude (e.g., ensuring the camera is pointing vertically downward or tilted at a pre-set angle). Hovering stability will be checked (attitude angle <±2°). When lighting conditions are met (e.g., brightness >50 Lux, avoiding backlight), wide-angle photos will be captured, using a pre-set focal length (e.g., 24mm wide-angle) for single or multiple photos (HDR mode is optional).

[0096] The captured photos are then stored and associated with metadata: GPS coordinates (drone location), altitude (e.g., 50 meters), timestamp, and camera parameters (focal length, aperture, ISO). Data is transmitted back in real time. If 4G / 5G or a digital transmission link is supported, the photos can be sent back to a ground station or the cloud for subsequent rough inspection and fine-grained capture of target identification. Otherwise, they are temporarily stored on the onboard SD card for post-processing and identification before the drone takes off again for fine-grained capture.

[0097] 4.2. Use a lightweight target detection algorithm (such as YOLOv5s) to identify suspected targets and exclude duplicate targets in the obtained photos. For example, suspected targets include:

[0098] Pine wilt diseased trees: Mark the bounding box of the browned crown.

[0099] Manhole cover defects: Mark the location of the manhole cover and any abnormal areas on the surface (such as cracks or water accumulation).

[0100] Exclude duplicate targets: If targets are located in overlapping areas, skip subsequent detection.

[0101] Use lightweight target detection models (such as YOLOv5s) and run them on drone-mounted computing units (such as NVIDIA Jetson) or edge servers. Specifically,

[0102] Optimization measures: Quantization (FP16 / INT8) accelerates inference.

[0103] Crop the model input resolution (e.g. 640×640).

[0104] Detection process: Input preprocessing: The photo is scaled to the model input size (such as 640×640).

[0105] Normalized pixel value (0~1).

[0106] Inference output: Obtain the detection box (Bounding Box), category (such as "Disease Tree" and "Defective Manhole Cover") and confidence score.

[0107] Threshold filtering:

[0108] Only keep targets with confidence > 0.5 (tunable parameter).

[0109] Target classification and feature extraction: Detection of pine wilt diseased trees:

[0110] Features: withered and yellow tree crowns, reddish-brown needles (the model needs to be trained to recognize specific colors / textures).

[0111] Urban manhole cover defect detection: Features: damaged edges, reflections from accumulated water, and displacement (abnormal samples need to be marked).

[0112] 4.3. Duplicate target elimination method

[0113] (1) Deduplication based on overlapping areas

[0114] Principle: There is a 10% to 20% overlap between adjacent sub-region photos, and the same target may be detected in multiple photos.

[0115] Deduplication step: Calculate the target geographic coordinates: Based on the drone's GPS position, altitude, and camera FOV, convert the detection box pixel coordinates into actual geographic coordinates (UTM / WGS84).

[0116] Formula: Target longitude = UAV longitude + (pixel horizontal offset / image width) × ground coverage width; (the same applies to latitude).

[0117] Spatial clustering and deduplication: All detected targets are clustered by coordinates (such as the DBSCAN algorithm, radius = 2m, and the minimum number of samples is set to 1).

[0118] Only the targets with the highest confidence are retained within the same cluster.

[0119] (2) Visual feature matching (optional)

[0120] If the target coordinates are close, SIFT / ORB feature matching is used to confirm whether they are the same object.

[0121] 4. Output results

[0122] Rough inspection results table:

[0123] Target ID category Confidence longitude latitude Photo source 001 sick tree 0.82 120.123°E 30.456°N Area A Photo 1 002 Manhole cover defect 0.76 120.125°E 30.454°N Area B Photo 1

[0124] Exception handling:

[0125] Low confidence targets (such as 0.3-0.5) are marked as "pending review" and subsequently manually screened.

[0126] Implementation cases, such as Figure 3 After completing the shooting and detection of area 1 at point A, the system flies to point B to take a wide-angle photo of area 2. The target detection algorithm detects five suspected targets, 4, 5, 6, 7, and 8. Among them, target 4 is in the overlapping area with area 1 and has been identified and can be excluded. That is, the valid suspected targets are 5, 6, 7, and 8.

[0127] Step 5: Based on the results of wide-angle shooting and target rough inspection, perform telephoto fine shooting to obtain a telephoto photo; specifically, it includes suspected target positioning and zooming: based on the pixel coordinates of the suspected target in the wide-angle photo, calculate its actual position in the drone's field of view. The comparison between wide-angle shooting and telephoto shooting is shown in the figure below. Figure 4 shown.

[0128] The drone adjusts the gimbal angle and switches to a telephoto lens (e.g. 200mm) to shoot each suspected target from multiple angles, such as:

[0129] Pine wilt diseased tree: photograph the crown details (close-up of needles).

[0130] Manhole cover defects: multi-focal length shooting (global defects, local cracks).

[0131] Step 6: Perform high-precision classification and defect diagnosis on the obtained telephoto photos, and generate an inspection report outputting a target distribution map with coordinates and a defect statistics table (such as the number of diseased trees and the damage rate of manhole covers).

[0132] Input telephoto images into a large multimodal model (e.g., a ViT+ResNet fusion model): Pine wilt diseased trees: Distinguish diseased trees from healthy dead branches (requires spectral auxiliary data). Manhole cover defects: Classify the defect type (e.g., "edge chipping" or "completely missing").

[0133] Save and filter results: Confirm the target (e.g., diseased trees 1 and 2 in zone 1), record the coordinates and classification results, and exclude misdetected targets (e.g., shadow interference from trees 3 and 4).

[0134] Comparison of the embodiments: Traditional method (method in the background art): UAV wide-angle shooting combined with manual review; The method of the present invention: According to the above wide-angle and telephoto coordination;

[0135] index Traditional method (wide angle + manual review) This invention (wide-angle-telephoto synergy) False detection rate of single inspection target 25% 8% Telephoto shots 200 times / square kilometer 120 times / square kilometer Data processing time 2 hours / square kilometer 0.5 hours / square kilometer

[0136] The present invention adopts wide-angle and telephoto collaborative detection. After a rough wide-angle inspection, telephoto shooting is triggered only for non-repetitive and high-confidence targets, reducing the number of invalid zoom times; the telephoto photos are classified through a fusion model to improve the diagnostic accuracy.

[0137] The present invention converts pixel coordinates into real geographic coordinates through drone positioning, altitude and field of view, and combines clustering to solve the error problem of traditional IOU deduplication during high-altitude oblique shooting; only the highest confidence target is retained in the same cluster to reduce the amount of redundant data.

[0138] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply the existence of any such actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or device comprising the element.

[0139] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A drone intelligent photography and inspection method, characterized by: The following steps are included Step 1: Determine the area to be inspected and preset the drone inspection parameters, including: presetting the drone's flight altitude and lens initial focal length based on the type of inspection target; Step 2: Divide the aerial photography sub-areas according to the preset drone inspection parameters; Step 3: Generate sub-area aerial photography paths based on the divided aerial photography sub-areas; Step 4: The drone performs wide-angle photography and rough target inspection based on the sub-area aerial photography path; Step 5: Perform telephoto fine shooting based on the results of wide-angle shooting and rough target inspection to obtain a telephoto photo; Step 6: Perform high-precision classification and defect diagnosis on the acquired telephoto photos, generate an inspection report, and output a target distribution map with coordinates and a defect statistics table.

2. The method for intelligent photography and inspection using a drone according to claim 1, characterized in that: Determining the area to be inspected includes determining the inspection area on an electronic map; the inspection area includes a forest area and / or an urban road.

3. The method for intelligent photography and inspection using a drone according to claim 2, characterized in that: The preset UAV flight altitude and lens initial focal length according to the detection target type include: when the detection target type is a forest area, the preset UAV flight altitude is 50 meters and the lens initial focal length is wide-angle mode; When the detection target is a city road, the preset drone flight altitude is 30 meters, and the initial focal length of the lens is wide-angle mode.

4. The method for intelligent photography and inspection using a drone according to claim 1, wherein: The aerial photography sub-area division according to the preset drone inspection parameters includes: 2.

1. Determine the field of view of the drone's wide-angle lens based on the lens's initial focal length; 2.

2. Calculate the coverage of a single photo based on the lens’ field of view and the drone’s flight altitude, and divide the inspection area into multiple sub-areas.

5. The method for intelligent photography and inspection using a drone according to claim 4, characterized in that: The calculation formula for calculating the coverage of a single photo is: Coverage = Horizontal Coverage Length × Vertical Coverage Length; Horizontal Coverage Length = 2 × H × tan(HFOV / 2), Vertical Coverage Length = 2 × H × tan(VFOV / 2); H is the flight altitude of the drone, HFOV is the horizontal field of view angle, and VFOV is the vertical field of view angle.

6. The method for intelligent photography and inspection using a drone according to claim 5, characterized in that: The logic of the sub-area division includes dividing the patrol area into grids according to the coverage area, and adjacent sub-areas need to maintain a 10% to 20% overlap.

7. The method for intelligent photography and inspection using a drone according to claim 6, characterized in that: The sub-area aerial photography path is generated according to the divided aerial photography sub-area, including taking the center point of the aerial photography sub-area as the shooting point, and the path of the drone traversing the sub-area center point in sequence in an "S" shape or "n" shape is the aerial photography path.

8. The method for intelligent photography and inspection using a drone according to claim 1, characterized in that: The drone performs wide-angle photography and rough target inspection according to the sub-area aerial photography path, including: 4.

1. The drone flies to the center of each sub-area according to the sub-area aerial photography path and takes wide-angle photos. 4.

2. Use a lightweight target detection algorithm to identify suspected targets and exclude duplicate targets in the obtained photos.

9. The method for intelligent photography and inspection using a drone according to claim 8, characterized in that: The method of excluding duplicate targets includes: converting the pixel coordinates of the detected targets into UTM geographic coordinates according to the GPS position, altitude and camera parameters of the UAV; clustering the coordinates using the DBSCAN algorithm, retaining the targets with the highest confidence within the same cluster, and eliminating the rest.

Images