A main grid and distribution network tree barrier hidden danger identification method and device based on wire area constraint

Abstract

The application discloses a kind of method and device for identifying main distribution network tree barrier hidden danger based on wire area constraint, method includes: according to the image of inspection, obtain tree barrier area set and wire corridor area;Obtain the correlation characteristics between tree barrier area and wire corridor area;Correlation characteristics include overlap rate, minimum distance and the minimum offset between the center point of tree barrier area and the center line of wire corridor area;According to the score of each tree barrier area obtained according to correlation characteristics, the tree barrier area with score greater than or equal to preset value is recorded as first area;According to correlation characteristics, each first area is reviewed by combining multimodal model, and the first area with review probability greater than or equal to preset value is output.

Description

Technical Field

[0001] This invention relates to the field of distribution network hazard identification, and in particular to a method and device for identifying tree obstruction hazards in main distribution networks based on conductor area constraints. Background Technology

[0002] Tree obstruction hazard identification in main and distribution networks is usually achieved by acquiring visible light images of the main and distribution network lines using inspection drones. The images are then cleaned, filtered, and labeled to construct a tree obstruction hazard training dataset. Based on the training dataset, a target detection model or image recognition model is trained to achieve automatic detection and identification of tree obstruction hazard targets in the inspection images.

[0003] Traditional methods typically rely on single-target detection models to identify trees, power lines, and suspected occlusion areas in images, and then assess potential tree obstruction hazards based on the location, category, and confidence level of the detection boxes. While some solutions can improve detection efficiency to some extent, they still primarily depend on local visual features of the target and lack the ability to jointly analyze the extent of main and distribution network conductor areas, the relative spatial relationship between trees and conductors, and complex scene context information. Furthermore, in actual inspection scenarios, main and distribution network conductors, low-voltage communication lines, tree canopies, background buildings, and other complex interfering targets often appear simultaneously in the same image. Therefore, traditional methods generally suffer from the following drawbacks: (1) Tree obstacles in the main distribution network are easily confused with tree obstacles near low-voltage communication lines, resulting in a high false detection rate. Different types of lines have similar local features in two-dimensional images. When trees are close to low-voltage communication lines, traditional methods are prone to misjudging them as tree obstacles in the main distribution network.

[0004] (2) Lack of effective understanding of the spatial relationship between trees and main and distribution network lines can easily lead to misjudgment in complex perspective scenes. When nearby trees and distant main and distribution network lines overlap in the image, traditional methods often fail to accurately determine their true spatial relationship and may easily misidentify targets that only intersect on the image plane as real tree obstacles.

[0005] (3) Traditional methods do not make sufficient use of conductor area constraints and lack a fine-grained screening mechanism for the main distribution network target area. Traditional methods usually identify the entire image directly and lack the ability to use the main distribution network conductor area to screen and constrain candidate tree obstacles. This results in a large number of trees unrelated to the main distribution network interfering in the subsequent identification process, affecting the accuracy and reliability of the final result.

[0006] (4) Traditional methods mainly rely on single visual features, have limited scene understanding capabilities, and are difficult to balance recognition accuracy and practicality. In complex inspection environments, judging solely based on detection boxes and category results is difficult to comprehensively utilize global scene information, local target information, and the relative positional relationship between trees and wires, resulting in insufficient stability of recognition results.

[0007] Therefore, a new technical solution is urgently needed to address the technical problem of how to improve the accuracy of identifying tree obstruction hazards in the main and distribution networks. Summary of the Invention

[0008] This invention provides a method and apparatus for identifying tree obstruction hazards in main and distribution networks based on conductor region constraints, in order to solve the technical problem of how to improve the accuracy of tree obstruction hazard identification in main and distribution networks.

[0009] To achieve the above objectives, the present invention provides a method for identifying potential tree obstructions in main and distribution networks based on conductor region constraints, comprising: The tree barrier region set and the guide corridor region are obtained from the inspection images; the correlation features between the tree barrier region and the guide corridor region are obtained; the correlation features include the overlap rate, the minimum distance, and the minimum offset between the center point of the tree barrier region and the center line of the guide corridor region; the score of each tree barrier region is obtained according to the correlation features, and the tree barrier region with a score greater than or equal to the preset value is recorded as the first region; the first region is reviewed according to the correlation features and the multimodal model, and the first region with a review probability greater than or equal to the preset value is output.

[0010] Preferably, the tree obstruction area set and the guide corridor area obtained from the inspection images include: Visible light images of the main and distribution network areas are obtained through inspections to obtain inspection images; A candidate tree obstacle set is obtained by combining the inspection images with a pre-trained target detection model; a conductor region mask is obtained by combining the inspection images with a pre-trained segmentation model. The rectangular regions of each candidate tree barrier are obtained from the candidate tree barrier set to obtain the tree barrier region set; the regions with pixel values ​​of 1 are extracted from the conductor region mask to obtain the conductor corridor region.

[0011] Preferably, the association features between the tree barrier region and the guide corridor region include: Obtain the overlap rate between the tree barrier area and the guide corridor area: ; ; in, The overlap rate, , A larger value indicates a higher degree of correlation between the tree barrier area and the conductor corridor area; This area is a tree-blocked area; This is the area for the conductor corridor; and The coordinates of the center of the candidate bounding box for the candidate tree barrier; and These represent the width and height of the candidate bounding box for the candidate tree barrier, respectively. Obtain the minimum distance between the tree barrier area and the guide corridor area: ; in, For the minimum distance, The smaller the value, the closer the tree barrier area is to the conductor corridor area; p Tree barrier area Any pixel in the array; q For the conductor corridor area Any pixel in the array; for p and q The Euclidean distance between them.

[0012] Preferably, obtaining the association features between the tree barrier region and the guide corridor region further includes: Obtain the minimum offset between the center point of the tree barrier region and the center line of the guide corridor region: Define the set of centerline pixels for the conductor corridor region: ; in, This is the set of centerline pixels in the conductor corridor region. K The number of pixels along the center line; The index of the center line pixel; The first on the center line 1 pixel; The minimum offset between the center point of the tree barrier area and the center line of the guide corridor area is: ; in, This is the minimum offset between the center point of the tree barrier area and the center line of the guide corridor area. This is the center point of the tree-blocked area.

[0013] Preferably, the scores for each tree barrier region obtained based on the association features include: The score for the tree barrier region is obtained by weighted summation based on overlap rate, minimum distance, and minimum offset. ; ; in, Scoring of tree-barrier areas; , and All are weighting coefficients; To prevent extremely small positive numbers with a denominator of zero.

[0014] Preferably, the verification of each first region based on correlation features and a multimodal model includes: Obtain the orientational relationship between each first region and the conductor corridor region; the orientational relationship includes the first region being located within a preset range to the left, right or near the center line of the conductor corridor region. Text prompts for each first region are constructed based on directional relationships and associated features; the guide region mask and the region highlight image of the first region are superimposed on the inspection image with preset weights to obtain the guide image of each first region; each first region is cropped separately to obtain the region cropped image of each first region. The verification input samples are constructed based on the inspection images, cropped images of each first region, guidance images, and text prompts. The verification is performed based on the verification input samples and the pre-selected multimodal model to obtain the verification probability of the existence of real tree obstacles in each first region, and the first region with a verification probability greater than or equal to the preset threshold is output.

[0015] Preferably, the verification is performed based on the input samples and a pre-selected multimodal model to obtain the verification probability of the existence of real tree barriers in each first region, and the first regions with verification probabilities greater than or equal to a preset threshold are output as follows: The kernel probability of each first region output by the multimodal model is expressed as follows: ; in, For the verification probability, A larger value indicates a higher probability of a real tree obstacle hazard. It is a multimodal model; Input samples for verification; Based on the definition of review probability, the multimodal review judgment result : ; in, This is a preset threshold for multimodal verification; when When =1, a genuine tree obstacle hazard is identified. When the value is 0, it is determined that there is no real tree obstacle hazard.

[0016] The present invention also provides a main and distribution network tree obstacle hazard identification device based on conductor area constraints, the device being used to implement the method of the present invention.

[0017] The present invention has the following beneficial effects: This invention presents a method for identifying tree obstructions in main and distribution networks based on conductor region constraints. By introducing a conductor region constraint mechanism and quantifying spatial correlation features such as overlap rate, minimum distance, and center offset between candidate tree obstructions and conductor regions, this method can further filter out interfering tree targets unrelated to the main and distribution networks, improving the relevance of the identification results. By introducing a multimodal verification mechanism and using the original image, local images of candidate targets, conductor region guidance images, and spatial relationship text prompts as verification inputs, the model's ability to understand complex scene contexts and target spatial relationships is enhanced, thereby reducing false detections caused by overlapping foreground and background and similar features. By constructing a cascaded judgment or weighted fusion mechanism for candidate detection confidence, conductor region comprehensive correlation score, and multimodal verification probability, the accuracy and reliability of identifying tree obstructions in main and distribution networks can be significantly improved.

[0018] The main and distribution network tree obstacle hazard identification device based on conductor area constraints of the present invention is used in the method of the present invention and has the same beneficial effects as the method of the present invention.

[0019] In addition to the objectives, features, and advantages described above, the present invention has other objectives, features, and advantages. The invention will now be described in further detail with reference to the accompanying drawings. Attached Figure Description

[0020] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a schematic diagram of the method flow of a preferred embodiment of the present invention.

[0021] Figure 2 This is a schematic diagram of the output of the target detection model according to a preferred embodiment of the present invention.

[0022] Figure 3 This is a schematic diagram of the polygonal region output by the segmentation model of a preferred embodiment of the present invention.

[0023] Figure 4 This is a schematic diagram of the scoring and screening of tree barrier areas and guide wire areas according to a preferred embodiment of the present invention. Detailed Implementation

[0024] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings, but the present invention can be implemented in many different ways as defined and covered by the claims.

[0025] See Figure 1 In a preferred embodiment of the present invention, a method for identifying potential tree obstructions in a main distribution network based on conductor region constraints is provided, comprising: Q1. Based on the inspection images, obtain the tree obstruction area set and the guide corridor area. Q1 specifically includes: The visible light images of the main distribution network area are obtained through inspection to obtain inspection images; a set of candidate tree obstacles is obtained by combining the inspection images with a pre-trained target detection model; a conductor region mask is obtained by combining the inspection images with a pre-trained segmentation model; rectangular regions of each candidate tree obstacle are obtained from the candidate tree obstacle set to obtain a tree obstacle region set; regions with a pixel value of 1 are extracted from the conductor region mask to obtain the conductor corridor region.

[0026] In a preferred embodiment of the present invention, the inspection image I is defined as: ; in, Image height, The width is 3, and 3 represents the RGB color channels of the image. Represents the real number field.

[0027] In a preferred embodiment of the present invention, for an input image I, the pre-trained target detection model outputs a set of candidate tree obstacle targets. : ; in, The number of candidate tree obstacle targets, the first candidate targets Represented as: ; in, and The coordinates of the center of the candidate target box. and These are the width and height of the candidate bounding box, respectively. The detection confidence of the candidate target.

[0028] In a preferred embodiment of the present invention, the inspection images are cleaned, filtered, and labeled to train a target detection model. Labeling can be performed using target detection labeling methods, selecting suspected tree-prone areas and assigning them corresponding category labels. The target detection model is trained to identify candidate target areas that may constitute potential tree-prone hazards from the input inspection images. The target detection model can be implemented using an target detection network, preferably a YOLO series model.

[0029] During the training phase of the object detection model, parameter optimization is achieved by minimizing the detection loss function. Detection loss function It can be represented as: ; in, This represents the classification loss for candidate target categories. This represents the regression loss at the candidate target location. This indicates the existence loss of the candidate target. and These are the weighting coefficients.

[0030] Through the above training method, the target detection model can output one or more candidate tree obstacle target boxes and corresponding confidence information for the image to be identified, providing candidate inputs for subsequent guide zone constraint screening.

[0031] In a preferred embodiment of the present invention, the output diagram of the target detection model is shown below. Figure 2 .

[0032] In a preferred embodiment of the present invention, a mask for the conductor region is provided. Represented as: ; in, Represents a pre-trained segmentation model; wire region mask. Pixels located within the envelope area of ​​the main and distribution network conductor corridor have a value of 1, while pixels located outside the envelope area have a value of 0.

[0033] In a preferred embodiment of the present invention, the inspection images are labeled with main and distribution network conductor regions for use in training the segmentation model. Conductor region labeling can employ instance segmentation labeling or semantic segmentation labeling methods to accurately describe the area range of the main and distribution network conductors in the image.

[0034] In a preferred embodiment of the present invention, the segmentation model is used to extract a continuous envelope region formed by the same group of main and distribution network conductors along their extension direction from the input image, and output the corresponding region mask. The segmentation model can be implemented using a semantic segmentation model or an instance segmentation model; preferably, an instance segmentation model is used, where a continuous envelope region formed by the same group of main and distribution network conductors is taken as a segmentation instance.

[0035] In a preferred embodiment of the present invention, the segmentation model is trained by minimizing a segmentation loss function, which can be expressed as: ; in, This represents the total loss of the main and distribution network conductor corridor area identification model. This represents the region segmentation loss, used to measure the pixel-level difference between the predicted mask and the real mask; This represents the boundary constraint loss, used to measure the difference between the predicted region boundary and the true polygon boundary; The weighting coefficients for the boundary constraint loss are used to balance the region segmentation accuracy and the boundary positioning accuracy.

[0036] In a preferred embodiment of the present invention, a schematic diagram of the polygonal region output by the segmentation model is shown below. Figure 3 .

[0037] Q2. Obtain the association features between the tree barrier region and the guide corridor region; association features include overlap rate, minimum distance, and minimum offset between the center point of the tree barrier region and the centerline of the guide corridor region. Q2 specifically includes: Obtain the overlap rate between the tree barrier area and the guide corridor area: ; ; in, The overlap rate, , A larger value indicates a higher degree of correlation between the tree barrier area and the conductor corridor area; Tree barrier area, i.e., candidate target The corresponding rectangular area; This is the area for the conductor corridor; and The coordinates of the center of the candidate bounding box for the candidate tree barrier; and These represent the width and height of the candidate bounding box for the candidate tree barrier, respectively. Obtain the minimum distance between the tree barrier area and the guide corridor area: ; in, For the minimum distance, The smaller the value, the closer the tree barrier area is to the conductor corridor area; p Tree barrier area Any pixel in the array; q For the conductor corridor area Any pixel in the array; for p and q The Euclidean distance between them.

[0038] Obtain the minimum offset between the center point of the tree barrier region and the center line of the guide corridor region: Define the set of centerline pixels for the conductor corridor region: ; in, This is the set of centerline pixels in the conductor corridor region. K The number of pixels along the center line; The index of the center line pixel; The first on the center line 1 pixel; The minimum offset between the center point of the tree barrier area and the center line of the guide corridor area is: ; in, This is the minimum offset between the center point of the tree barrier area and the center line of the conductor corridor area. This offset is used to characterize the proximity of the candidate target to the main direction area of ​​the main distribution network conductor. This is the center point of the tree-blocked area.

[0039] Q3. Based on the correlation features, obtain the score of each tree barrier region, and record the tree barrier region with a score greater than or equal to the preset value as the first region.

[0040] In a preferred embodiment of the present invention, the scoring of each tree barrier region based on the associated features includes: The score for the tree barrier region is obtained by weighted summation based on overlap rate, minimum distance, and minimum offset. ; ; in, Scoring of tree-barrier areas; , and All are weighting coefficients; To prevent extremely small positive numbers with a denominator of zero.

[0041] In a preferred embodiment of the present invention, tree-blocked areas with scores lower than a preset value are directly removed.

[0042] By using the above relationship determination steps, background trees with insufficient spatial correlation to the main and distribution network conductor areas, trees near low-voltage communication lines, and other irrelevant interference targets can be effectively eliminated, thereby improving the pertinence and reliability of the subsequent identification process.

[0043] See Figure 4 , Figure 4 The rectangular selection areas are all tree obstacle areas, and the polygonal selection areas are guide wire areas. The score is calculated based on the difference between the rectangular and polygonal selection areas. This allows for the removal or retention of tree-barrier areas, with the retained tree-barrier areas being designated as the first area.

[0044] Q4. Based on the correlation features and a multimodal model, each first region is reviewed, and the first region with a review probability greater than or equal to a preset value is output. Q4 specifically includes: Obtain the orientational relationship between each first region and the conductor corridor region; the orientational relationship includes the first region being located within a preset range to the left, right or near the center line of the conductor corridor region. Text prompts for each first region are constructed based on location relationships and associated features; the guide region mask and the region highlight image of the first region are superimposed onto the inspection image with preset weights to obtain the guide image of each first region; each first region is cropped separately to obtain the region cropped image of each first region; a review input sample is constructed based on the inspection image, the region cropped image of each first region, the guide image, and the text prompts, as follows: ; Local images of candidate targets: ; in, In tree barrier areas The local area expanded according to a preset ratio based on the original structure. For cropping operations; For guiding images: The guiding image is used to highlight the spatial relationship between candidate tree-block targets and the main distribution network conductor area, providing conductor area constraint information for the multimodal verification model. This guides the multimodal verification model to focus on determining whether there is a tree-block hazard association between the candidate targets and the main distribution network conductor, as shown below: ; in, This represents the highlighted region corresponding to the i-th candidate target; To mask the conductor area Convert it into a transition guide image with the same size, number of channels, and pixel range as the original image; To add weighting coefficients, .

[0045] For text prompts: ; in, This is a template generation function that converts spatial relationship features into text prompts. This indicates the azimuth relationship of the candidate target relative to the main distribution network conductor area.

[0046] The multimodal kernel input constructed in the above manner enables the multimodal model to simultaneously utilize global scene information in the original image, local detail information of candidate targets, prompt information of the main and distribution network conductor regions, and spatial relationship information between the target and the conductor regions, thereby improving the understanding and identification ability of real tree obstacle hazards.

[0047] The verification process is performed based on the input samples and a pre-selected multimodal model. The verification probability of the presence of real tree barriers in each first region is obtained, and the first regions with verification probabilities greater than or equal to a preset threshold are output, including: The kernel probability of each first region output by the multimodal model is expressed as follows: ; in, For the verification probability, A larger value indicates a higher probability of a real tree obstacle hazard. It is a multimodal model; Input samples for verification; Based on the definition of review probability, the multimodal review judgment result : ; in, This is a preset threshold for multimodal verification; when When =1, a genuine tree obstacle hazard is identified. When the value is 0, it is determined that there is no real tree obstacle hazard.

[0048] Multimodal model output The verification confidence level pi of each candidate target is calculated, and the verification confidence level pi is compared with the decision threshold. The comparison results are used to output the corresponding review and judgment results. The review and assessment results include both tree-obstruction hazards and non-tree-obstruction hazards.

[0049] This further eliminates targets that merely overlap with conductors on a two-dimensional image plane but are not actual tree obstructions in the main and distribution networks, as well as similar interference targets related to low-voltage communication lines.

[0050] This invention presents a method for identifying tree obstructions in main and distribution networks based on conductor region constraints. By introducing a conductor region constraint mechanism and quantifying spatial correlation features such as overlap rate, minimum distance, and center offset between candidate tree obstructions and conductor regions, this method can further filter out interfering tree targets unrelated to the main and distribution networks, improving the relevance of the identification results. By introducing a multimodal verification mechanism and using the original image, local images of candidate targets, conductor region guidance images, and spatial relationship text prompts as verification inputs, the model's ability to understand complex scene contexts and target spatial relationships is enhanced, thereby reducing false detections caused by overlapping foreground and background and similar features. By constructing a cascaded judgment or weighted fusion mechanism of candidate detection confidence, conductor region comprehensive correlation score, and multimodal verification probability, the accuracy and reliability of identifying tree obstructions in main and distribution networks can be significantly improved.

[0051] In a preferred embodiment of the present invention, a main distribution network tree obstacle hazard identification device based on conductor region constraints is also provided, the device being used to implement the method of the present invention.

[0052] The main and distribution network tree obstacle hazard identification device based on conductor area constraints of the present invention is used in the method of the present invention and has the same beneficial effects as the method of the present invention.

[0053] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. 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 potential tree obstructions in main and distribution networks based on conductor region constraints, characterized in that, include: The inspection images yield a set of tree-blocked areas and a guide corridor area; the correlation features between the tree-blocked areas and the guide corridor area are obtained; the correlation features include overlap rate, minimum distance, and minimum offset between the center point of the tree-blocked area and the center line of the guide corridor area; the correlation features are used to obtain a score for each tree-blocked area, and tree-blocked areas with scores greater than or equal to a preset value are recorded as the first area; the correlation features are combined with a multimodal model to verify each first area, and the first area with a verification probability greater than or equal to a preset value is output.

2. The method for identifying tree-related hazards in main and distribution networks based on conductor region constraints according to claim 1, characterized in that, Based on the inspection images, the tree obstruction area set and the guide corridor area include: The inspection image is obtained by acquiring visible light images of the main distribution network area through inspection; A candidate tree obstacle set is obtained by combining the inspection images with a pre-trained target detection model; a conductor region mask is obtained by combining the inspection images with a pre-trained segmentation model. The rectangular regions of each candidate tree barrier are obtained from the candidate tree barrier set to obtain the tree barrier region set; the regions with pixel values ​​of 1 are extracted from the conductor region mask to obtain the conductor corridor region.

3. The method for identifying tree obstructions in main and distribution networks based on conductor region constraints according to claim 2, characterized in that, The association features between the tree barrier region and the guide corridor region include: Obtain the overlap rate between the tree barrier area and the guide corridor area: ； ； in, The overlap rate, , A larger value indicates a higher degree of correlation between the tree barrier area and the conductor corridor area; This area is a tree-blocked area; This is the area for the conductor corridor; and The coordinates of the center of the candidate target box of the candidate tree barrier; and These are the width and height of the candidate target bounding box for the candidate tree barrier, respectively.

4. The method for identifying tree obstructions in main and distribution networks based on conductor region constraints according to claim 3, characterized in that, Obtaining association features between tree barrier regions and conductor corridor regions also includes: Obtain the minimum distance between the tree barrier area and the guide corridor area: ； in, For the minimum distance, The smaller the value, the closer the tree barrier area is to the conductor corridor area; p Tree barrier area Any pixel in the array; q For the conductor corridor area Any pixel in the array; for p and q The Euclidean distance between them.

5. The method for identifying tree obstruction hazards in main and distribution networks based on conductor region constraints according to claim 4, characterized in that, Obtaining association features between tree barrier regions and conductor corridor regions also includes: Obtain the minimum offset between the center point of the tree barrier region and the center line of the guide corridor region: Define the set of centerline pixels for the conductor corridor region: ； in, This is the set of centerline pixels in the conductor corridor region. K The number of pixels along the center line; The index of the center line pixel; The first on the center line 1 pixel; The minimum offset between the center point of the tree barrier area and the center line of the guide corridor area is: ； in, This is the minimum offset between the center point of the tree barrier area and the center line of the guide corridor area. This is the center point of the tree-blocked area.

6. The method for identifying tree obstruction hazards in main and distribution networks based on conductor region constraints according to claim 5, characterized in that, The scores for each tree barrier region obtained based on the aforementioned correlation features include: The score for the tree barrier region is obtained by weighted summation based on the overlap rate, the minimum distance, and the minimum offset. ； ； in, Scoring of tree-barrier areas; , and All are weighting coefficients; To prevent extremely small positive numbers with a denominator of zero.

7. The method for identifying tree obstruction hazards in main and distribution networks based on conductor region constraints according to claim 6, characterized in that, The verification of each first region based on the aforementioned correlation features and a multimodal model includes: Obtain the orientational relationship between each first region and the conductor corridor region; the orientational relationship includes the first region being located within a preset range to the left, right or near the center line of the conductor corridor region. Based on the directional relationship and the associated features, text prompts for each first region are constructed; the mask of the conductor region and the region highlight image of the first region are superimposed on the inspection image with a preset weight to obtain the guide image of each first region; each first region is cropped separately to obtain the region cropped image of each first region. Based on the inspection image, the cropped image of each first region, the guide image, and the text prompt, a verification input sample is constructed; the verification is performed based on the verification input sample and a pre-selected multimodal model to obtain the verification probability of the existence of real tree obstacles in each first region, and the first region with a verification probability greater than or equal to a preset threshold is output.

8. The method for identifying tree obstruction hazards in main and distribution networks based on conductor region constraints according to claim 7, characterized in that, The verification is performed based on the input samples and a pre-selected multimodal model to obtain the verification probability of the existence of real tree obstacles in each first region, and the first regions with verification probabilities greater than or equal to a preset threshold are output as follows: The kernel probability of each first region output by the multimodal model is expressed as follows: ； in, For the verification probability, A larger value indicates a higher probability of a real tree obstacle hazard. It is a multimodal model; Input samples for verification; The multimodal review determination result is defined based on the review probability. : ； in, This is a preset threshold for multimodal verification; when When =1, a genuine tree obstacle hazard is identified. When the value is 0, it is determined that there is no real tree obstacle hazard.

9. A device for identifying tree obstruction hazards in main and distribution networks based on conductor region constraints, characterized in that, The apparatus is used to implement the method according to any one of claims 1 to 8.

Images