Method and system for identifying marker pile based on unmanned aerial vehicle photographing and medium

Abstract

The application discloses a method and system for identifying a marker pile based on unmanned aerial vehicle pile photographing and a medium, relates to the technical field of oil and gas pipeline maintenance, and can provide supplementary and strengthened feature information based on a visible light image in which a marker pile is partially blocked or blurred by using the difference in the penetration or reflection characteristics of different spectral bands (for example, the near-infrared band has a certain penetration ability for plant leaves and can highlight the difference between specific pigments (for example, marker pile paint) and the surrounding natural environment) to generate a feature-enhanced image through pixel-level fusion, thereby solving the problem of partial blocking and making the pile body contour and identification content clearer and more identifiable. Meanwhile, the blocked area can be directly identified through the visible light image, so that maintenance personnel can easily formulate a maintenance plan.

Description

Technical Field

[0001] This invention relates to the field of oil and gas pipeline maintenance technology, specifically to a method, system, and medium for identifying marker posts based on drone-based pile photography. Background Technology

[0002] Oil and gas pipeline marker posts (including mileage markers, corner markers, and warning signs) are a core component of the pipeline ground marking system. They clearly indicate the direction, location, mileage information, and safety warnings of underground pipelines, playing an irreplaceable role in daily pipeline inspections, third-party construction monitoring, emergency repairs, and maintenance and renovations. Therefore, regularly inspecting the marker posts to confirm their condition, such as tilting, collapse, damage, illegible lettering, loss, or obstruction by vegetation, is a fundamental yet demanding task for pipeline operating companies.

[0003] Currently, the inspection and maintenance of pipeline marker posts mainly follow two modes:

[0004] 1. Traditional manual inspection methods rely heavily on inspectors walking or driving along pipeline routes to conduct on-site inspections and record data. This method is inefficient and costly. Pipelines typically stretch for hundreds or even thousands of kilometers, often traversing complex terrains such as wilderness, mountains, and farmland, requiring a significant investment of manpower, resources, and time for manual inspections. Furthermore, the quality of inspections is difficult to guarantee, and inspection records are mostly in paper or simple spreadsheet format, making it difficult to create structured, visualized digital assets and hindering big data analysis and historical status tracking.

[0005] 2. Automated / Semi-Automated Inspection Modes Based on Traditional Computer Vision: To overcome the shortcomings of manual inspection, the industry has begun to explore using inspection vehicles or drones equipped with cameras for data collection and applying image recognition technology for automatic identification and positioning of marker posts. However, existing recognition models generally suffer from inaccurate results and poor robustness in practical applications, making it difficult to meet the requirements of industrial applications. Changes in vegetation cover and surface damage or shading on marker posts need to be identified. Traditional target detection models (such as early versions based on YOLO and SSD) lack the ability to extract features from small, complex targets, resulting in limited generalization performance. Most existing models can only perform basic identification of the presence or absence of marker posts, lacking effective detection and evaluation methods for finer details such as slight tilt angles, surface cracks, and lettering clarity. Manual secondary interpretation of the identified images is still required, resulting in a low degree of automation.

[0006] In conclusion, the current inspection and maintenance of oil and gas pipeline marker posts, whether relying solely on manual labor or employing existing automated identification technologies, faces significant technical bottlenecks. The core issue lies in the lack of an innovative technological approach to improve the accuracy and intelligence of marker post identification. Summary of the Invention

[0007] The technical problem this invention aims to solve is that the inspection and maintenance of oil and gas pipeline marker posts relies heavily on manual labor, and existing recognition models suffer from inaccurate recognition results. The purpose of this invention is to provide a marker post recognition method, system, and medium based on drone-based image capture of marker posts. By fusing visible light images with multispectral images for reinforcement learning, this invention specifically addresses the pain points of existing technologies, such as large occlusion interference, poor imaging conditions, and insufficient robustness of recognition models. It achieves highly accurate and robust automatic image recognition of oil and gas pipeline marker posts in complex field environments.

[0008] This invention is achieved through the following technical solution:

[0009] This solution provides a method for identifying marker posts based on drone-based image capture, including:

[0010] The image data is collected synchronously by the UAV along the target route, and the image data is preprocessed. The image data includes visible light images and multispectral images.

[0011] Pixel-level fusion of preprocessed visible light and multispectral images yields a feature-enhanced image;

[0012] The enhanced image is input into the constructed recognition model to obtain the signpost recognition result and the occlusion recognition result.

[0013] A further optimized solution involves preprocessing the image data, including the following methods:

[0014] Based on the SIFT algorithm, feature point detection and feature description are performed on each image to obtain a feature point set and a description subset;

[0015] Feature point matching is performed based on the feature point set and descriptive subset of each image; and the transformation matrix is ​​estimated based on the feature point matching process, with the shape of the marker stake as a constraint.

[0016] Based on the transformation matrix, the registration coordinates of each pixel in the multispectral image are calculated to obtain the registered multispectral image.

[0017] A further optimized solution involves estimating the transformation matrix based on the feature point matching process, using the shape of the marker post as a constraint; this includes the following method:

[0018] When estimating the transformation matrix, the following equation serves as the objective function for optimization:

[0019] ;

[0020] S(T) = ;

[0021] Where Min represents taking the minimum value, and M represents the set of feature point matches, containing elements (p i q j ), where p i =(x i y i ) represents a point in a multispectral image, q j = (x j y j ) represents a point in the visible light image; S(T) represents the marker stake shape constraint function; v1, v2 and v3, v4 represent the midpoints of two pairs of opposite sides of the r-th rectangular template; R represents all rectangular templates; T represents the transformation function corresponding to the transformation matrix; c represents the regularization parameter; ||*|| represents calculating the Euclidean distance between two points.

[0022] A further optimization involves pixel-level fusion of the preprocessed visible light image and the multispectral image; including the following methods:

[0023] Extracting the luminance component L from a visible light image v (x, y);

[0024] If the target region is assumed to have N spectral bands, then the typical reflectance vector of the first target region in the N spectral bands is R = [R1, R2, ..., R]. N The typical reflectance vector of the background region in N spectral bands is R. b =[ R b1 R b2 , ..., R bN ];

[0025] The spectral discrimination between the target region and the background region in each spectral band is calculated, and the spectral discrimination is normalized to obtain the spectral band weights.

[0026] A spectral feature enhancement factor is constructed based on the mean and standard deviation of the background region in multiple spectral bands. A Bruvian transform model for spectral feature enhancement is constructed by combining the brightness component and the weight of the spectral bands. The fused feature-enhanced image is obtained based on the Bruvian transform model.

[0027] A further optimization scheme involves calculating the spectral discrimination between the target region and the background region in each spectral band, and then normalizing the spectral discrimination to obtain the spectral band weights; including the following methods:

[0028] The spectral discrimination D between the target region and the background region in spectral band k is calculated using the following formula. k :

[0029] D k =|R k - R bk | / max(R k R bk );

[0030] Where max(#, *) means taking the larger value between # and *;

[0031] The spectral resolution D for spectral band k is based on the following formula. k Normalization process yields the weight w of spectral band k. k :

[0032] w k =D k / ( D1+ D2+…+ D k +…+ D N ).

[0033] A further optimization scheme involves constructing a spectral feature enhancement factor based on the mean and standard deviation of the background region across multiple spectral bands; including the following methods:

[0034] Obtain the mean value g of the background region in spectral band k. k and standard deviation β k ;

[0035] The spectral feature enhancement factor E of spectral band k is calculated based on the following formula. k (x, y):

[0036] E k (x, y) = 1 + a|I dk -g k | / β k ;

[0037] Among them, I dk This represents the value of the k-th band in the registered multispectral image at position (x, y).

[0038] A further optimized solution is that the Bruvian transform model includes:

[0039] The value F of the fused k-th band image at position (x,y) k (x, y) is:

[0040] ;

[0041] Among them, w l The weights of spectral bands l are represented by l = 1, 2, 3, ..., N.

[0042] A further optimized solution involves inputting the feature-enhanced image into the pre-constructed recognition model to obtain the marker post recognition result or occlusion recognition result; including the following method:

[0043] Using the second region as the target region and the first region as the background region, the recognition model outputs the recognition result for the first region.

[0044] Using the third region as the target region and the second region as the background region, the recognition model outputs the recognition result for the third region.

[0045] In the image data, the second region overlaps the first region, and the third region overlaps the second region.

[0046] This solution also provides a marker post recognition system based on drone-based image capture of marker posts, used to implement the aforementioned marker post recognition method based on drone-based image capture of marker posts; the system includes:

[0047] The acquisition module is used to simultaneously acquire image data of different spectral ranges along the target route using a UAV;

[0048] A preprocessing module is used to preprocess the image data; the image data includes visible light images and multispectral images;

[0049] The fusion module is used to perform pixel-level fusion of preprocessed visible light images and multispectral images to obtain feature-enhanced images;

[0050] The recognition module is used to input the feature-enhanced image into the pre-built recognition model to obtain the recognition result.

[0051] This solution also provides a computer-readable medium storing a computer program, which, when executed by a processor, enables the identification of marker posts based on drone-based image capture, as described above.

[0052] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0053] 1. The present invention provides a method, system, and medium for identifying marker posts based on drone-based image capture. By acquiring and fusing visible light images and multispectral images, and utilizing the differences in the penetrability or reflection characteristics of different spectral bands (such as the near-infrared band having a certain penetrating ability to vegetation leaves and being able to highlight the difference between specific pigments (such as marker post paint) and the surrounding natural environment), the feature-enhanced image generated through pixel-level fusion can provide supplementary and enhanced feature information even when the marker post is partially obscured or blurred in the visible light image, thereby solving the problem of partial obscuration and making the outline of the post and the marking content clearer and more distinguishable. At the same time, the obscured area can be directly identified through the visible light image, which is convenient for maintenance personnel to formulate maintenance plans.

[0054] 2. The present invention provides a method, system, and medium for identifying marker posts based on drone-based image capture of marker posts. The feature-enhanced image (which contains more robust and complementary features) fused from multi-source information (visible light image and multispectral image) is input into the recognition model for identification. This enables the recognition model to learn and infer based on higher quality and more information-rich data, fundamentally improving the model's ability to recognize dirty, faded, low-contrast characters and partially occluded characters, and significantly reducing the false recognition rate and missed recognition rate.

[0055] 3. The method, system, and medium for identifying marker posts based on drone-based image capture provided by this invention form a complete automated processing flow from data acquisition (synchronous acquisition by drone), preprocessing, fusion to identification and error correction. This not only greatly reduces manual intervention and improves inspection efficiency, but more importantly, through multi-level technology fusion (multispectral sensing, image fusion, and recognition models), the system possesses the intelligence to cope with complex field environments, providing a more reliable and efficient technical tool for the safety management of oil and gas pipelines.

[0056] 4. The present invention provides a method, system and medium for identifying marker posts based on drone-based image capture of marker posts; based on feature-enhanced images fused from multiple sources (visible light images and multispectral images), different regions can be used as the target region and background region for identification, and the target region to be identified can be output. Users can customize the identification, and the identification method can be applied to a wider range. Attached Figure Description

[0057] To more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be considered as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort. In the drawings:

[0058] Figure 1 This is a schematic diagram of the process for identifying marker posts based on drone-based image capture of the posts.

[0059] Figure 2 This is a schematic diagram of a marker post recognition system based on drone-based image capture of marker posts. Detailed Implementation

[0060] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments and accompanying drawings. The illustrative embodiments and descriptions of the present invention are only used to explain the present invention and are not intended to limit the present invention.

[0061] During the inspection and maintenance of oil and gas pipeline marker posts, manual labor is heavily relied upon. Existing identification models suffer from inaccurate results. Therefore, this solution provides the following embodiments to address the aforementioned technical problems:

[0062] Example 1

[0063] This embodiment provides a method for identifying marker posts based on drone-based image capture of the posts. Figure 1 As shown, it includes:

[0064] Step 1: Simultaneously collect image data of different spectral ranges along the target route using a drone, and preprocess the image data; the image data includes visible light images and multispectral images;

[0065] In step one, the image data preprocessing includes the following methods:

[0066] S11, based on the SIFT algorithm, feature point detection and feature description are performed on each image to obtain a feature point set and a description subset; feature points include characteristic points such as corner points and edges; assuming the visible light image is I... v (x v ,y v The multispectral image is I. m (x m ,y m );

[0067] Visible light image I detected based on SIFT algorithm v The feature point set Pv and the descriptive subset Zv, as well as the multispectral image I m The feature point set Pm and the description subset Zm;

[0068] S12, feature point matching is performed based on the feature point set and descriptive subset of each image; and the transformation matrix is ​​estimated according to the feature point matching process, using the shape of the marker post as a constraint; the method of estimating the transformation matrix according to the feature point matching process, using the shape of the marker post as a constraint, includes:

[0069] By calculating the Euclidean distance between the feature descriptors of the visible light image and the multispectral image, the closest visible light image descriptor is found for each multispectral image descriptor, and these descriptors form the set M of feature point matching as feature point pairs.

[0070] Estimate a geometric transformation T such that Pv = T(Pm) = APm + t, where A = Represents the linear transformation matrix; t= Representing the translation vector, this model maps the multispectral image to the coordinate system of the visible light image based on an affine transformation model.

[0071] ;

[0072] The parameters a, b, c, d, d, f (i.e., the transformation matrix) are estimated using the least squares method. The specific objective function is as follows:

[0073] ;

[0074] S(T) = ;

[0075] Where Min[] represents taking the minimum value, and M represents the set of feature point matching, containing elements (p i q j ), where p i =( x i y i ) represents a point in a multispectral image, q j = (x j y j ) represents a point in the visible light image; S(T) represents the marker stake shape constraint function; v1, v2 and v3, v4 represent the midpoints of two pairs of opposite sides of the r-th rectangular template; R represents all rectangular templates; T represents the transformation function corresponding to the transformation matrix; c represents the regularization parameter; ||*|| represents calculating the Euclidean distance between two points.

[0076] S13, based on the transformation matrix, calculate the registration coordinates for each pixel of the multispectral image to obtain the registered multispectral image. Specifically, the registered multispectral image can be obtained based on the bilinear interpolation method.

[0077] The transformation function T is applied to each band of the multispectral image to obtain the registered multispectral image.

[0078] The preprocessing process mainly involves image registration, which is the process of aligning multiple images so that they correspond to the same points in the same coordinate system. For visible light images and multispectral images, since they may come from different sensors, they need to be registered first to geometrically align the multispectral images with the visible light images, ensuring that the two images correspond to the same physical points in the same coordinate system.

[0079] Step two involves pixel-level fusion of the preprocessed visible light image and multispectral image to obtain a feature-enhanced image; this step specifically includes the following methods:

[0080] S21, Extract the luminance component L from the visible light image. v (x, y); If the visible light image is a color image, convert it to a grayscale image;

[0081] S22, assuming the target area has N spectral bands, then the typical reflectance vector of the first target area in the N spectral bands is R = [R1, R2, ..., R...]. N The typical reflectance vector of the background region in N spectral bands is R. b =[ R b1 R b2 , ..., R bN ];

[0082] S23, calculate the spectral discrimination between the target region and the background region in each spectral band, and normalize the spectral discrimination to obtain the spectral band weights; this step specifically includes the following methods:

[0083] S231, calculate the spectral discrimination D between the target region and the background region in spectral band k according to the following formula. k :

[0084] D k =|R k - R bk | / max(R k R bk );

[0085] Where max(#, *) means taking the larger value between # and *;

[0086] S232, the spectral resolution D of spectral band k is based on the following formula. k Normalization process yields the weight w of spectral band k. k :

[0087] w k =D k / ( D1+ D2+…+ D k +…+ D N ).

[0088] S24, a spectral feature enhancement factor is constructed based on the mean and standard deviation of the background region in multiple spectral bands. A Bruvian transform model for spectral feature enhancement is then constructed by combining the brightness component and spectral band weights. The fused feature-enhanced image is obtained based on the Bruvian transform model. In step S24, the construction of the spectral feature enhancement factor based on the mean and standard deviation of the background region in multiple spectral bands includes the following method:

[0089] Obtain the mean value g of the background region in spectral band k. k and standard deviation β k ;

[0090] The spectral feature enhancement factor E of spectral band k is calculated based on the following formula. k (x, y):

[0091] E k (x, y) = 1 + a|I dk -g k | / β k ;

[0092] Among them, I dk This represents the value of the k-th band in the registered multispectral image at position (x, y).

[0093] The Bruvis transform model includes:

[0094] The value F of the fused k-th band image at position (x,y) k (x, y) is:

[0095] ;

[0096] Among them, w l The weights of spectral bands l are represented by l = 1, 2, 3, ..., N.

[0097] Step 3: Input the enhanced feature image into the constructed recognition model to obtain the recognition result; including the following methods:

[0098] Using the second region as the target region and the first region as the background region, the recognition model outputs the recognition result for the first region.

[0099] Using the third region as the target region and the second region as the background region, the recognition model outputs the recognition result for the third region.

[0100] In the image data, the second region covers the first region, and the third region covers the second region. Specifically, when detecting the status of a marker post, if the marker post needs to be identified, the marker post is used as the second region, and the background of the marker post is used as the first region, i.e., the background region. According to the above steps, the recognition model can output the marker post recognition result. At this time, the visible light image can be combined to determine whether the marker post is occluded. If the identification content on the marker post needs to be detected and identified, the marker post is used as the second region, i.e., the background region, and the identification content is used as the third region, i.e., the target region. According to the above steps, the recognition model can output the identification content recognition result. At this time, it can be determined whether the identification content needs maintenance. Specifically, the model for the marker post and the identification model for the identification content are different models. The marker post model can be a conventional identification model, while the identification model for the identification content can be a text-related identification model such as an OCR model.

[0101] Drones regularly conduct inspections of the entire pipeline according to a pre-defined method of photographing the marker posts. Using the aforementioned identification methods, they automatically identify the condition of marker posts and any obstructions along the pipeline, allowing maintenance personnel to promptly perform maintenance and remove obstructions. After taking photos, the drones upload the collected data to a cloud platform via a collaborative communication protocol. Simultaneously, a manual inspection terminal activates AR navigation, guiding inspection personnel to the corresponding marker post locations according to the marker post photographing plan. Inspection personnel then take supplementary photos of the continuous marker posts and their surrounding environment according to the "one marker post, one standard" photographing angle and field of view specifications, and upload the captured data to the cloud platform. Through regular inspections and data collection, real-time monitoring and data updates of the entire pipeline are ensured.

[0102] Example 2

[0103] This embodiment provides a marker post identification system based on drone-based image capture of marker posts, as shown in the figure, for implementing the marker post identification method based on drone-based image capture of marker posts described in Embodiment 1; the system includes:

[0104] The acquisition module is used to simultaneously acquire image data of different spectral ranges along the target route using a UAV;

[0105] A preprocessing module is used to preprocess the image data; the image data includes visible light images and multispectral images;

[0106] The fusion module is used to perform pixel-level fusion of preprocessed visible light images and multispectral images to obtain feature-enhanced images;

[0107] The recognition module is used to input the feature-enhanced image into the pre-constructed recognition model to obtain the marker post recognition result;

[0108] The error correction module is used to correct the identification results of the identifier stake based on business rules and context.

[0109] Example 3

[0110] This embodiment provides a computer-readable medium storing a computer program. The computer program, when executed by a processor, can implement the marker post identification method based on drone-based marker post photography as described in Embodiment 1; specifically, it performs the following steps:

[0111] Step 1: Simultaneously collect image data of different spectral ranges along the target route using a drone, and preprocess the image data; the image data includes visible light images and multispectral images;

[0112] Step 2: Perform pixel-level fusion of the preprocessed visible light image and multispectral image to obtain a feature-enhanced image;

[0113] Step 3: Input the enhanced feature image into the constructed recognition model to obtain the marker post recognition result.

[0114] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for identifying marker posts based on drone-based continuous post photography, characterized in that, include: The system uses drones to simultaneously collect image data of marker posts in different spectral ranges and preprocesses the image data; the image data includes visible light images and multispectral images. Pixel-level fusion of preprocessed visible light and multispectral images yields a feature-enhanced image; The enhanced image is input into the constructed recognition model to obtain the signpost recognition result and the occlusion recognition result.

2. The method for identifying marker posts based on UAV-based continuous post photography according to claim 1, characterized in that, Preprocessing the image data includes the following methods: Based on the SIFT algorithm, feature point detection and feature description are performed on each image to obtain a feature point set and a description subset; Feature point matching is performed based on the feature point set and descriptive subset of each image; and the transformation matrix is ​​estimated based on the feature point matching process, with the shape of the marker stake as a constraint. Based on the transformation matrix, the registration coordinates of each pixel in the multispectral image are calculated to obtain the registered multispectral image.

3. The method for identifying marker posts based on UAV-based continuous post photography according to claim 2, characterized in that, The transformation matrix is ​​estimated based on the feature point matching process, using the shape of the marker post as a constraint; including... method: When estimating the transformation matrix, the following equation serves as the objective function for optimization: ； S(T)= ; Where Min represents taking the minimum value, and M represents the set of feature point matches, containing elements (p i q j ), where p i =( x i y i ) represents a point in a multispectral image, q j = (x j y j ) represents a point in the visible light image; S(T) represents the marker stake shape constraint function; v1, v2 and v3, v4 represent the midpoints of two pairs of opposite sides of the r-th rectangular template; R represents all rectangular templates; T represents the transformation function corresponding to the transformation matrix; c represents the regularization parameter; ||*|| represents calculating the Euclidean distance between two points.

4. The method for identifying marker posts based on UAV-based continuous post photography according to claim 1, characterized in that, Pixel-level fusion of preprocessed visible light and multispectral images; including methods: Extracting the luminance component L from a visible light image v (x, y); If the target region is assumed to have N spectral bands, then the typical reflectance vector of the first target region in the N spectral bands is R = [R1, R2, ..., R]. N The typical reflectance vector of the background region in N spectral bands is R. b =[ R b1 R b2 , ..., R bN ]; The spectral discrimination between the target region and the background region in each spectral band is calculated, and the spectral discrimination is normalized to obtain the spectral band weights. A spectral feature enhancement factor is constructed based on the mean and standard deviation of the background region in multiple spectral bands. A Bruvian transform model for spectral feature enhancement is constructed by combining the brightness component and the weight of the spectral bands. The fused feature-enhanced image is obtained based on the Bruvian transform model.

5. The method for identifying marker posts based on UAV-based continuous post photography according to claim 4, characterized in that, The spectral discrimination between the target region and the background region in each spectral band is calculated, and the spectral discrimination is normalized to obtain the spectral band weights. Including methods: The spectral discrimination D between the target region and the background region in spectral band k is calculated using the following formula. k : D k =|R k - R bk | / max(R k ,R bk ); Where max(#, *) means taking the larger value between # and *; The spectral resolution D for spectral band k is based on the following formula. k Normalization process yields the weight w of spectral band k. k : w k =D k / ( D1+ D2+…+ D k +…+ D N ) 。 6. The method for identifying marker posts based on UAV-based continuous post photography according to claim 4, characterized in that, The spectral feature enhancement factor is constructed based on the mean and standard deviation of the background region in multiple spectral bands. Including methods: Obtain the mean value g of the background region in spectral band k. k and standard deviation β k ; The spectral feature enhancement factor E of spectral band k is calculated based on the following formula. k (x, y): E k (x,y)=1+a|I dk - g k | / β k ; Among them, I dk This represents the value of the k-th band in the registered multispectral image at position (x, y).

7. The method for identifying marker posts based on UAV-based continuous post photography according to claim 4, characterized in that, The Bruvis transform model includes: The value F of the fused k-th band image at position (x,y) k (x, y) is: ； Among them, w l The weights of spectral bands l are represented by l = 1, 2, 3, ..., N.

8. The method for identifying marker posts based on UAV-based continuous post photography according to claim 4 or 5, characterized in that, The enhanced image is input into the pre-constructed recognition model to obtain the signpost recognition result or the occlusion recognition result. Including methods: Using the second region as the target region and the first region as the background region, the recognition model outputs the recognition result for the first region. Using the third region as the target region and the second region as the background region, the recognition model outputs the recognition result for the third region. In the image data, the second region overlaps the first region, and the third region overlaps the second region.

9. A marker post identification system based on drone-based continuous marker post photography, characterized in that, The system is used to implement the marker post identification method based on drone-based marker post photography as described in any one of claims 1-8; the system includes: The acquisition module is used to simultaneously acquire image data of different spectral ranges along the target route using a UAV; A preprocessing module is used to preprocess the image data; the image data includes visible light images and multispectral images; The fusion module is used to perform pixel-level fusion of preprocessed visible light images and multispectral images to obtain feature-enhanced images; The recognition module is used to input the feature-enhanced image into the pre-built recognition model to obtain the recognition result.

10. A computer-readable medium having a computer program stored thereon, characterized in that, The computer program, when executed by a processor, can implement the marker post identification method based on drone-based pile photography as described in any one of claims 1-8.

Images