Binocular vision-based quantitative identification method and system for disturbance area of power transmission line project

Abstract

The invention discloses a power transmission line engineering disturbance area quantitative identification method and system based on binocular vision, and the method comprises the steps: carrying binocular imaging equipment through an unmanned plane, collecting a power transmission line construction site image, obtaining an original binocular image, and carrying out the preprocessing of the original binocular image, thereby obtaining a three-dimensional correction binocular image; performing feature extraction and enhancement on the three-dimensional correction binocular image by adopting a feature extraction model to obtain a depth enhancement feature map; matching left-eye and right-eye feature points of the depth enhanced feature map, establishing a binocular pixel point corresponding relation, calculating parallax data between the left-eye and right-eye feature points, and obtaining depth data based on the parallax data; and identifying a potential engineering disturbance region based on a depth threshold segmentation algorithm and the depth data, optimizing to obtain an engineering disturbance region, segmenting the engineering disturbance region to extract a boundary contour, and calculating the area of the engineering disturbance region. According to the method, the engineering disturbance area is identified by using the depth threshold segmentation algorithm, the identification precision is improved, the method can adapt to complex environmental conditions, and the stability of an identification result is ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of power transmission line disturbance zone identification, and in particular to a method and system for quantitatively identifying power transmission line engineering disturbance zones based on binocular vision. Background Art

[0002] During the construction and maintenance of power transmission lines, accurate identification and quantitative analysis of engineering disturbance areas are important links to ensure construction safety and environmental protection. Traditional methods for identifying engineering disturbance areas mainly rely on ground measurements, which are not only time-consuming and labor-intensive, but also difficult to ensure accuracy and efficiency under complex terrain conditions. They are easily affected by human factors, resulting in subjectivity and inconsistency in the identification results.

[0003] In the existing technology, although the image processing technology based on monocular vision has improved the recognition efficiency to a certain extent, it is difficult to accurately perform quantitative analysis of three-dimensional space due to the lack of depth information. In addition, the monocular vision system is easily affected by lighting changes and occlusion in complex environments, resulting in unstable recognition results. Summary of the invention

[0004] In view of this, the present invention proposes a method and system for quantitatively identifying disturbance areas of power transmission line projects based on binocular vision, which solves the problem that the prior art lacks depth information and is easily affected by lighting changes and occlusions in complex environments, resulting in unstable recognition results.

[0005] The technical solution of the present invention is implemented as follows: In the first aspect, the present invention provides a method for quantitatively identifying disturbance areas of power transmission line engineering based on binocular vision, comprising the following steps:

[0006] S1, using a binocular imaging device mounted on an unmanned aerial vehicle to collect images of a transmission line construction site based on the binocular imaging device to obtain an original binocular image;

[0007] S2, preprocessing the original binocular image to obtain a stereo corrected binocular image;

[0008] S3, using a feature extraction model to perform feature extraction and feature enhancement on the stereo-rectified binocular image to obtain a depth enhancement feature map of the stereo-rectified binocular image;

[0009] S4, matching the left eye feature points and the right eye feature points of the depth enhancement feature map, establishing a binocular pixel correspondence relationship of the binocular imaging device, calculating the disparity data between the left eye feature points and the right eye feature points, and acquiring depth data based on the disparity data;

[0010] S5. Based on the depth threshold segmentation algorithm and the depth data, a potential engineering disturbance area is identified, the potential engineering disturbance area is optimized to obtain an engineering disturbance area, the engineering disturbance area is segmented, and the boundary contour of the engineering disturbance area is extracted, and the area of ​​the engineering disturbance area is calculated.

[0011] On the basis of the above technical solution, preferably, step S2 includes:

[0012] S21, performing image denoising on the original binocular image by a convolutional neural network to obtain a denoised binocular image, and enhancing the denoised binocular image by an adaptive histogram equalization method to obtain a denoised enhanced binocular image;

[0013] S22. Based on the intrinsic parameter matrix and the extrinsic parameter matrix of the binocular imaging device, the epipolar equation of the denoised and enhanced binocular image is calculated, and the denoised and enhanced binocular image is reprojected by the epipolar equation so that the same-name points of the left and right images of the denoised and enhanced binocular image are located on the same horizontal line to obtain a reprojected binocular image, and the reprojected binocular image is resampled by a bilinear interpolation method to obtain a stereo corrected binocular image.

[0014] On the basis of the above technical solution, preferably, step S3 includes:

[0015] S31, constructing an original feature extraction model based on a convolutional neural network, obtaining historical image data of a transmission line construction site, performing transfer learning on the original feature extraction model through pre-trained model parameters and the historical image data, obtaining a feature extraction model, and using the feature extraction model to perform multi-scale feature extraction on the stereo-rectified binocular image to obtain depth feature maps of the left eye and the right eye respectively;

[0016] S32. Use the attention mechanism to process the depth feature map to obtain an attention depth feature map, and perform adaptive feature normalization on the attention depth feature map to obtain a depth enhancement feature map.

[0017] On the basis of the above technical solution, preferably, step S3 further includes:

[0018] The calculation formula for multi-scale feature extraction is:

[0019]

[0020] Among them, F is the deep feature map, L is the number of network layers of the feature extraction model, and W (l) is the weight of the lth layer network of the feature extraction model, b (l) is the bias parameter of the lth layer network of the feature extraction model, I (l)is the representation of the stereo-rectified binocular image in the l-th layer network of the feature extraction model, * is the convolution operator, ReLU(·) is the ReLU activation function, is the weight of the residual connection of the l-th layer network of the feature extraction model, is the bias of the residual connection of the lth layer network of the feature extraction model, I (l-1) It is the representation of the stereo-rectified binocular image in the l-1th layer network of the feature extraction model;

[0021] The calculation formula of the depth enhancement feature map is:

[0022] F'=α·f Attention (F)+β·f AdaNorm (F)+(1-α-β)·F;

[0023] f Attention (F) = softmax(W a tanh(W h ·F+W x ·X)+b a )·F;

[0024]

[0025] Among them, F' is the depth enhancement feature map, α and β are the weight of the attention mechanism and the weight of adaptive normalization, respectively, and f Attention (F) is the result of the attention mechanism processing of the deep feature map F, f AdaNorm (F) is the adaptive normalization result of the deep feature map, softmax(·) is the normalized exponential function, tanh(·) is the hyperbolic tangent activation function, W a , W h and W x They are the attention weight matrix, feature map mapping weight matrix and context mapping weight matrix, respectively, a is the attention mechanism bias parameter, X is the context information input vector, γ is the learnable scaling parameter, δ is the learnable offset parameter, τ is the anti-zero constant, μ is the mean of the feature map, σ 2 is the variance of the feature map.

[0026] Based on the above technical solution, preferably, step S4 includes:

[0027] S41, selecting the best matching pair through a feature matching algorithm, and obtaining the left eye feature points and the right eye feature points corresponding to the best matching pair;

[0028] S42: Calculate disparity data between the left eye feature point and the right eye feature point based on a triangulation method, and acquire depth data based on the disparity data.

[0029] On the basis of the above technical solution, preferably, step S4 further includes:

[0030] The calculation formula of the feature matching algorithm is:

[0031]

[0032] Among them, M(x L ) is the left eye feature point x L The best matching feature point, D L and D R are the feature descriptors of the left and right eye feature points, respectively, R is the right eye feature point, Solve the function for minimizing the independent variable.

[0033] Based on the above technical solution, preferably, step S5 includes:

[0034] S51, using a depth threshold segmentation algorithm, marking the area in the depth data that exceeds a preset threshold as a potential engineering disturbance area, and optimizing the potential engineering disturbance area in combination with morphological operations to obtain an optimized engineering disturbance area;

[0035] S52, segmenting the optimized engineering disturbance area through a contour extraction algorithm based on image segmentation, and extracting the boundary contour of the engineering disturbance area;

[0036] S53, calculating the area of ​​the optimized engineering disturbance region by an integration method.

[0037] In a second aspect, the present invention further provides a system for quantitatively identifying disturbance areas of power transmission line engineering based on binocular vision, the system comprising:

[0038] An image acquisition module is used to acquire images of a transmission line construction site by using a binocular imaging device carried by a drone, based on the binocular imaging device, to obtain an original binocular image;

[0039] An image processing module, used for preprocessing the original binocular image to obtain a stereo-corrected binocular image;

[0040] A feature extraction module, used to extract and enhance the features of the stereo-rectified binocular image using a feature extraction model to obtain a depth enhancement feature map of the stereo-rectified binocular image;

[0041] A depth data module is used to match the left eye feature points and the right eye feature points of the depth enhancement feature map, establish a binocular pixel point correspondence relationship of the binocular imaging device, calculate the disparity data between the left eye feature points and the right eye feature points, and obtain depth data based on the disparity data;

[0042] A quantitative identification module is used to identify potential engineering disturbance areas based on a depth threshold segmentation algorithm and the depth data, optimize the potential engineering disturbance areas to obtain engineering disturbance areas, segment the engineering disturbance areas, extract the boundary contours of the engineering disturbance areas, and calculate the area of ​​the engineering disturbance areas.

[0043] In a third aspect, the present invention further provides an electronic device, comprising: at least one processor, at least one memory, a communication interface and a bus;

[0044] Among them, the processor, memory, and communication interface communicate with each other through the bus, the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to implement steps such as a method for quantitatively identifying disturbance areas of power transmission line projects based on binocular vision.

[0045] In a fourth aspect, the present invention further provides a computer-readable storage medium storing computer instructions, wherein the computer instructions enable a computer to implement steps of a method for quantitatively identifying disturbance areas of power transmission line projects based on binocular vision.

[0046] Compared with the prior art, the method and system for quantitatively identifying disturbance areas of power transmission line engineering based on binocular vision of the present invention have the following beneficial effects:

[0047] (1) The stereo images are acquired by the binocular imaging device carried by the UAV, and the convolutional neural network is used for image preprocessing and feature extraction. The deep threshold segmentation algorithm is used to automatically identify the engineering disturbance area, which improves the recognition accuracy and automation level, can adapt to complex environmental conditions, and ensure the stability of the recognition results;

[0048] (2) Through the feature extraction model based on convolutional neural network, combined with transfer learning and multi-scale feature extraction methods, the deep feature map of the transmission line construction site is extracted and enhanced, which improves the accuracy of feature extraction. Combined with the attention mechanism and adaptive feature normalization, the expression ability of the feature map is further enhanced, making it possible to more accurately identify and analyze the engineering disturbance area in complex environments;

[0049] (3) The best matching pair is selected through the feature matching algorithm, the correspondence between the left eye feature points and the right eye feature points is established, and the disparity data is calculated based on the triangulation method to obtain accurate depth data, which improves the accuracy of feature matching and the reliability of depth calculation, and ensures the quantitative analysis of potential engineering disturbance areas. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. 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 creative work.

[0051] Figure 1 The present invention is a flow chart of a method for quantitatively identifying disturbance areas in power transmission line projects based on binocular vision.

[0052] Figure 2 This is a structural diagram of a system for quantitatively identifying disturbance areas in power transmission line projects based on binocular vision according to the present invention. DETAILED DESCRIPTION

[0053] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0054] See also Figure 1 The present invention provides a method for quantitatively identifying disturbance areas of power transmission line engineering based on binocular vision, comprising the following steps:

[0055] S1, using a binocular imaging device mounted on an unmanned aerial vehicle to collect images of a transmission line construction site based on the binocular imaging device to obtain an original binocular image;

[0056] S2, preprocessing the original binocular image to obtain a stereo corrected binocular image;

[0057] S3, using a feature extraction model to perform feature extraction and feature enhancement on the stereo-rectified binocular image to obtain a depth enhancement feature map of the stereo-rectified binocular image;

[0058] S4, matching the left eye feature points and the right eye feature points of the depth enhancement feature map, establishing a binocular pixel correspondence relationship of the binocular imaging device, calculating the disparity data between the left eye feature points and the right eye feature points, and acquiring depth data based on the disparity data;

[0059] S5. Based on the depth threshold segmentation algorithm and the depth data, a potential engineering disturbance area is identified, the potential engineering disturbance area is optimized to obtain an engineering disturbance area, the engineering disturbance area is segmented, and the boundary contour of the engineering disturbance area is extracted, and the area of ​​the engineering disturbance area is calculated.

[0060] Specifically, this embodiment acquires stereo images through a binocular imaging device carried by a drone, combines a convolutional neural network for image preprocessing and feature extraction, and uses a deep threshold segmentation algorithm to automatically identify engineering disturbance areas, thereby improving recognition accuracy and automation, being able to adapt to complex environmental conditions, and ensuring the stability of recognition results.

[0061] Step S1 includes:

[0062] S11. Set the flight parameters of the drone, including flight altitude, flight speed and route planning;

[0063] According to the terrain characteristics and area size of the transmission line construction site, the drone's flight altitude is set to 50-200 meters, the flight speed is 5-15 meters per second, and a grid-shaped route is planned to ensure full coverage of the entire construction site;

[0064] S12, controlling the UAV to fly over the transmission line construction site according to the set flight parameters, and at the same time collecting original binocular images through the binocular imaging device;

[0065] The binocular imaging device is controlled to collect images at fixed time intervals or fixed spatial intervals. The collection interval is dynamically adjusted according to the flight speed and altitude of the UAV to ensure that there is a 30%-60% overlap rate between adjacent images, which is convenient for subsequent image stitching and three-dimensional reconstruction.

[0066] Specifically, this embodiment ensures that the drone can achieve full coverage of the entire transmission line construction site by setting an appropriate flight altitude and planning a grid-shaped route. This all-round coverage can effectively capture every corner of the construction area and avoid the loss of important information due to incomplete coverage.

[0067] Adjust the flight altitude and speed of the drone according to the terrain characteristics and size of the construction site, so that the drone can flexibly adapt to the needs of different terrains, improve the effectiveness of data collection, and ensure that high-quality image data can still be obtained under complex or changing site conditions;

[0068] By adjusting the time interval or spatial interval of image acquisition during flight, the overlap rate between adjacent images is maintained at 30%-60%, thereby improving the accuracy and completeness of the reconstructed model.

[0069] Step S2 includes:

[0070] S21, performing image denoising on the original binocular image by a convolutional neural network to obtain a denoised binocular image, and enhancing the denoised binocular image by an adaptive histogram equalization method to obtain a denoised enhanced binocular image;

[0071] S22. Based on the intrinsic parameter matrix and the extrinsic parameter matrix of the binocular imaging device, the epipolar equation of the denoised and enhanced binocular image is calculated, and the denoised and enhanced binocular image is reprojected by the epipolar equation so that the same-name points of the left and right images of the denoised and enhanced binocular image are located on the same horizontal line to obtain a reprojected binocular image, and the reprojected binocular image is resampled by a bilinear interpolation method to obtain a stereo corrected binocular image.

[0072] Specifically, this embodiment performs image denoising through a convolutional neural network to remove noise in the original binocular image, improve image clarity, reduce the interference of noise on feature extraction and depth calculation, and enhance the contrast of the denoised binocular image through an adaptive histogram equalization method.

[0073] By calculating the epipolar equations of the denoised and enhanced binocular image and reprojecting it, we ensure that the same-name points of the left and right images are on the same horizontal line and the geometric consistency of the binocular image is guaranteed. The bilinear interpolation method is used to resample the reprojected binocular image to reduce image distortion and ensure that the structural integrity of the image is maintained at different resolutions.

[0074] Step S3 includes:

[0075] S31, constructing an original feature extraction model based on a convolutional neural network, obtaining historical image data of a transmission line construction site, performing transfer learning on the original feature extraction model through pre-trained model parameters and the historical image data, obtaining a feature extraction model, and using the feature extraction model to perform multi-scale feature extraction on the stereo-rectified binocular image to obtain depth feature maps of the left eye and the right eye respectively;

[0076] S32. Use the attention mechanism to process the depth feature map to obtain an attention depth feature map, and perform adaptive feature normalization on the attention depth feature map to obtain a depth enhancement feature map.

[0077] Specifically, this embodiment constructs an original feature extraction model and uses historical image data for transfer learning, so that the model can better adapt to the specific construction environment and improve the ability to recognize the characteristics of the engineering disturbance area. By adopting a multi-scale feature extraction method, the detailed information and global features of the image are captured at different scales, thereby improving the ability to understand complex scenes.

[0078] By introducing the attention mechanism to process the deep feature map, important features are highlighted, irrelevant information is suppressed, and the expressiveness of the feature map is enhanced. Through adaptive feature normalization, the consistency of the feature map is maintained under different conditions, the comparability of features is improved, and the feature fluctuations caused by lighting changes or other environmental factors are reduced.

[0079] Step S3 also includes:

[0080] The calculation formula for multi-scale feature extraction is:

[0081]

[0082] Among them, F is the deep feature map, L is the number of network layers of the feature extraction model, and W (l) is the weight of the lth layer network of the feature extraction model, b (l) is the bias parameter of the lth layer network of the feature extraction model, I (l) is the representation of the stereo-rectified binocular image in the l-th layer network of the feature extraction model, * is the convolution operator, ReLU(·) is the ReLU activation function, is the weight of the residual connection of the l-th layer network of the feature extraction model, is the bias of the residual connection of the lth layer network of the feature extraction model, I (l-1) It is the representation of the stereo-rectified binocular image in the l-1th layer network of the feature extraction model;

[0083] The calculation formula of the depth enhancement feature map is:

[0084] F'=α·f Attention (F)+β·f AdaNorm (F)+(1-α-β)·F;

[0085] f Attention (F) = softmax(W a tanh(W h ·F+W x ·X)+b a )·F;

[0086]

[0087] Among them, F' is the depth enhancement feature map, α and β are the weight of the attention mechanism and the weight of adaptive normalization, respectively, and f Attention (F) is the result of the attention mechanism processing of the deep feature map F, f AdaNorm (F) is the adaptive normalization result of the deep feature map, softmax(·) is the normalized exponential function, tanh(·) is the hyperbolic tangent activation function, W a , W h and W x They are the attention weight matrix, feature map mapping weight matrix and context mapping weight matrix, respectively, a is the attention mechanism bias parameter, X is the context information input vector, γ is the learnable scaling parameter, δ is the learnable offset parameter, τ is the anti-zero constant, μ is the mean of the feature map, σ 2 is the variance of the feature map.

[0088] Specifically, this embodiment uses a multi-layer structure of a convolutional neural network and utilizes network weights and bias parameters of different layers to perform multi-scale feature extraction on stereo-rectified binocular images, capture image details and global information at different scales, and use residual connections in combination with the ReLU activation function to enhance the training efficiency of the feature extraction model.

[0089] By introducing the attention mechanism, the deep feature map is processed to generate the attention deep feature map. The attention mechanism improves the expressiveness of the feature map by weighting the importance of different features. Adaptive feature normalization is used, combined with the normalized exponential function and the hyperbolic tangent activation function, to enhance the contrast of the deep feature map.

[0090] This embodiment captures the details and global information of the image at different scales through multi-scale feature extraction, thereby improving the accuracy of feature extraction. The attention mechanism improves the expressiveness of the feature map by weighting the importance of different features. The adaptive feature normalization improves the contrast and detail expression of the feature map, thereby enhancing the recognition ability in complex environments.

[0091] Step S4 includes:

[0092] S41, selecting the best matching pair through a feature matching algorithm, and obtaining the left eye feature points and the right eye feature points corresponding to the best matching pair;

[0093] S42: Calculate disparity data between the left eye feature point and the right eye feature point based on a triangulation method, and acquire depth data based on the disparity data.

[0094] Specifically, this embodiment selects the best matching pair through a feature matching algorithm, identifies the correspondence between the left eye feature points and the right eye feature points, calculates the disparity data between the left eye feature points and the right eye feature points based on the triangulation method, obtains the depth data, and uses the geometric relationship to ensure the accuracy of the depth information.

[0095] Through the optimized feature matching and parallax calculation process, it is possible to maintain high recognition stability and accuracy when dealing with complex scenes and environmental changes, ensuring that under different construction conditions, this embodiment can still effectively identify engineering disturbance areas.

[0096] Accurate feature matching and depth calculation provide high-quality data input for the identification and quantification of engineering disturbance areas, improve processing efficiency, ensure rapid response and real-time monitoring capabilities, and through precise depth data acquisition, can adapt to different types of transmission line construction sites and support engineering monitoring and evaluation in a variety of application scenarios.

[0097] Step S4 also includes:

[0098] The calculation formula of the feature matching algorithm is:

[0099]

[0100] Among them, M(x L ) is the left eye feature point x L The best matching feature point, D L and D R are the feature descriptors of the left and right eye feature points, respectively, x R is the right eye feature point, Solve the function for minimizing the independent variable.

[0101] The calculation formula for disparity data and depth data is:

[0102] d = x L -x R ;

[0103]

[0104] Wherein, d is the disparity data, Z is the depth data, f is the focal length of the binocular imaging device, and B is the baseline length of the binocular imaging device.

[0105] Specifically, this embodiment selects the best matching pair by minimizing the difference between the feature descriptors of the left-eye feature points and the right-eye feature points, utilizes the similarity of the feature descriptors to improve the matching accuracy, and introduces the minimum independent variable solving function to ensure that the best matching point can be found quickly during the feature matching process, thereby improving the matching efficiency. Feature descriptors are used to represent the local information of feature points, and matching is performed by comparing the similarities of these descriptors. This can handle noise and deformation in the image and improve the robustness of the matching.

[0106] Step S5 includes:

[0107] S51, using a depth threshold segmentation algorithm, marking the area in the depth data that exceeds a preset threshold as a potential engineering disturbance area, and optimizing the potential engineering disturbance area in combination with morphological operations to obtain an optimized engineering disturbance area;

[0108] S52, segmenting the optimized engineering disturbance area through a contour extraction algorithm based on image segmentation, and extracting the boundary contour of the engineering disturbance area;

[0109] S53, calculating the area of ​​the optimized engineering disturbance region by an integration method.

[0110] Step S5 also includes:

[0111] The calculation formula of the depth threshold segmentation algorithm is:

[0112]

[0113] in, is the segmentation mask, is the feature point in the depth enhanced feature map The depth value, T d is the reference depth threshold, λ is the adaptive coefficient, σ d is the standard deviation of the depth data;

[0114] The calculation formula of the contour extraction algorithm based on image segmentation is:

[0115] C = argmin C (λ 1 E internal (C)+λ 2 E external (C));

[0116] Among them, C is the contour to be optimized, E internal (C) and E external (C) are internal energy term and external energy term, respectively, λ 1 and λ 2 are the internal energy weight and the external energy weight respectively, and argmin(·) is the function for finding the minimum independent variable.

[0117] Specifically, this embodiment introduces an adaptive coefficient λ and uses the standard deviation σ of the depth data d To dynamically adjust the threshold, a more flexible and accurate depth threshold segmentation can be achieved, so that the algorithm can adapt to the distribution of depth data and enhance the controllability of contour extraction by minimizing the linear combination of internal energy and external energy. 1 and λ 2 This allows users to adjust extraction priorities according to specific needs, thereby optimizing contour accuracy.

[0118] See also Figure 2 The present invention also provides a system for quantitatively identifying disturbance areas of power transmission line engineering based on binocular vision, the system comprising:

[0119] An image acquisition module is used to acquire images of a transmission line construction site by using a binocular imaging device carried by a drone, based on the binocular imaging device, to obtain an original binocular image;

[0120] An image processing module is used to preprocess the original binocular image to obtain a stereo-corrected binocular image;

[0121] A feature extraction module, used for performing feature extraction and feature enhancement on the stereo-rectified binocular image using a feature extraction model to obtain a depth enhancement feature map of the stereo-rectified binocular image;

[0122] A depth data module, used for matching left eye feature points and right eye feature points of the depth enhancement feature map, establishing a binocular pixel correspondence relationship of the binocular imaging device, calculating disparity data between the left eye feature points and the right eye feature points, and acquiring depth data based on the disparity data;

[0123] A quantitative identification module is used to identify potential engineering disturbance areas based on a depth threshold segmentation algorithm and the depth data, optimize the potential engineering disturbance areas to obtain engineering disturbance areas, segment the engineering disturbance areas, extract the boundary contours of the engineering disturbance areas, and calculate the area of ​​the engineering disturbance areas.

[0124] Specifically, in this embodiment, a binocular imaging device is carried on an unmanned aerial vehicle, and the system can quickly and comprehensively collect original binocular images of the transmission line construction site over a large range, ensuring the timeliness and comprehensiveness of the data. The image processing module pre-processes the original binocular image to obtain a stereo corrected binocular image, ensuring the geometric consistency of the left and right images and reducing errors caused by image distortion.

[0125] The feature extraction module uses a convolutional neural network to extract and enhance the features of stereo-rectified binocular images, which can extract high-quality depth-enhanced feature maps and improve the recognition accuracy of engineering disturbance areas.

[0126] The depth data module calculates the disparity data and obtains the depth data by matching the left and right eye feature points of the depth enhancement feature map, thus ensuring the accuracy of the depth information.

[0127] The quantitative identification module identifies potential engineering disturbance areas based on the depth threshold segmentation algorithm and depth data, and performs optimization, segmentation, and boundary contour extraction, enabling the system to accurately assess the impact of construction on the environment.

[0128] The present invention also discloses an electronic device, comprising: at least one processor, at least one memory communication interface and a bus: wherein the processor, memory and communication interface communicate with each other through the bus; the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to implement a method for quantitatively identifying disturbance areas of power transmission line projects based on binocular vision.

[0129] The present invention also discloses a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, wherein the computer instructions enable the computer to implement all or part of the steps of a method for quantitatively identifying a disturbance zone of a power transmission line project based on binocular vision according to an embodiment of the present invention. The storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory ROM, a random access memory RAM, a magnetic disk or an optical disk.

[0130] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for quantitatively identifying disturbance areas of power transmission line engineering based on binocular vision, characterized in that: The following steps are involved: S1, using a binocular imaging device mounted on an unmanned aerial vehicle to collect images of a transmission line construction site based on the binocular imaging device to obtain an original binocular image; S2, preprocessing the original binocular image to obtain a stereo corrected binocular image; S3, using a feature extraction model to perform feature extraction and feature enhancement on the stereo-rectified binocular image to obtain a depth enhancement feature map of the stereo-rectified binocular image; S4, matching the left eye feature points and the right eye feature points of the depth enhancement feature map, establishing a binocular pixel correspondence relationship of the binocular imaging device, calculating the disparity data between the left eye feature points and the right eye feature points, and acquiring depth data based on the disparity data; S5. Based on the depth threshold segmentation algorithm and the depth data, a potential engineering disturbance area is identified, the potential engineering disturbance area is optimized to obtain an engineering disturbance area, the engineering disturbance area is segmented, and the boundary contour of the engineering disturbance area is extracted, and the area of ​​the engineering disturbance area is calculated.

2. A method for quantitatively identifying disturbance areas of power transmission line engineering based on binocular vision as claimed in claim 1, characterized in that: Step S2 includes: S21, performing image denoising on the original binocular image by a convolutional neural network to obtain a denoised binocular image, and enhancing the denoised binocular image by an adaptive histogram equalization method to obtain a denoised enhanced binocular image; S22. Based on the intrinsic parameter matrix and the extrinsic parameter matrix of the binocular imaging device, the epipolar equation of the denoised and enhanced binocular image is calculated, and the denoised and enhanced binocular image is reprojected by the epipolar equation so that the same-name points of the left and right images of the denoised and enhanced binocular image are located on the same horizontal line to obtain a reprojected binocular image, and the reprojected binocular image is resampled by a bilinear interpolation method to obtain a stereo corrected binocular image.

3. A method for quantitatively identifying disturbance areas of power transmission line engineering based on binocular vision as claimed in claim 1, characterized in that: Step S3 includes: S31, constructing an original feature extraction model based on a convolutional neural network, obtaining historical image data of a transmission line construction site, performing transfer learning on the original feature extraction model through pre-trained model parameters and the historical image data, obtaining a feature extraction model, and using the feature extraction model to perform multi-scale feature extraction on the stereo-rectified binocular image to obtain depth feature maps of the left eye and the right eye respectively; S32. Use the attention mechanism to process the depth feature map to obtain an attention depth feature map, and perform adaptive feature normalization on the attention depth feature map to obtain a depth enhancement feature map.

4. A method for quantitatively identifying disturbance areas of power transmission line engineering based on binocular vision as claimed in claim 3, characterized in that: Step S3 also includes: The calculation formula for multi-scale feature extraction is: Among them, F is the deep feature map, L is the number of network layers of the feature extraction model, and W (l) is the weight of the ,th layer network of the feature extraction model, b (l) is the bias parameter of the lth layer network of the feature extraction model, I (l) is the representation of the stereo-rectified binocular image in the l-th layer network of the feature extraction model, * is the convolution operator, ReLU(·) is the ReLU activation function, and W r (l) is the weight of the residual connection of the ,th layer network of the feature extraction model, b r (l) is the bias of the residual connection of the ,th layer network of the feature extraction model, I (l-1) It is the representation of the stereo-rectified binocular image in the l-1th layer network of the feature extraction model; The calculation formula of the depth enhancement feature map is: F′=α·f Attention (F)+β·f AdaNorm (F)+(1-α-β)·F; f Attention (F)=softmax(W a ·tanh(W h ·F+W x ·X)+b a )·F; Among them, F′ is the depth enhancement feature map, α and β are the weight of the attention mechanism and the weight of adaptive normalization, respectively, and f Attention (F) is the result of the attention mechanism processing of the deep feature map F, f AdaNorm (F) is the adaptive normalization result of the deep feature map, softmax(·) is the normalized exponential function, tanh(·) is the hyperbolic tangent activation function, W a , W h and W x They are the attention weight matrix, feature map mapping weight matrix and context mapping weight matrix, respectively, a is the attention mechanism bias parameter, X is the context information input vector, γ is the learnable scaling parameter, δ is the learnable offset parameter, τ is the anti-zero constant, μ is the mean of the feature map, σ 2 is the variance of the feature map.

5. The method for quantitatively identifying disturbance areas of power transmission line engineering based on binocular vision according to claim 1, characterized in that: Step S4 includes: S41, selecting the best matching pair through a feature matching algorithm, and obtaining the left eye feature points and the right eye feature points corresponding to the best matching pair; S42: Calculate disparity data between the left eye feature point and the right eye feature point based on a triangulation method, and acquire depth data based on the disparity data.

6. A method for quantitatively identifying disturbance areas of power transmission line engineering based on binocular vision as claimed in claim 5, characterized in that: Step S4 also includes: The calculation formula of the feature matching algorithm is: Among them, M(x L ) is the left eye feature point x L The best matching feature point, D L and D R are the feature descriptors of the left and right eye feature points, respectively, R is the right eye feature point, Solve the function for minimizing the independent variable.

7. The method for quantitatively identifying disturbance areas of power transmission line engineering based on binocular vision according to claim 1, characterized in that: Step S5 includes: S51, using a depth threshold segmentation algorithm, marking the area in the depth data that exceeds a preset threshold as a potential engineering disturbance area, and optimizing the potential engineering disturbance area in combination with morphological operations to obtain an engineering disturbance area; S52, segmenting the engineering disturbance area by a contour extraction algorithm based on image segmentation, and extracting the boundary contour of the engineering disturbance area; S53. Calculate the area of ​​the engineering disturbance region by an integration method.

8. A quantitative identification system for disturbance areas of power transmission line engineering based on binocular vision, characterized in that: The system comprises: An image acquisition module is used to acquire images of a transmission line construction site by using a binocular imaging device carried by a drone, based on the binocular imaging device, to obtain an original binocular image; An image processing module, used for preprocessing the original binocular image to obtain a stereo-corrected binocular image; A feature extraction module, used to extract and enhance the features of the stereo-rectified binocular image using a feature extraction model to obtain a depth enhancement feature map of the stereo-rectified binocular image; A depth data module is used to match the left eye feature points and the right eye feature points of the depth enhancement feature map, establish a binocular pixel point correspondence relationship of the binocular imaging device, calculate the disparity data between the left eye feature points and the right eye feature points, and obtain depth data based on the disparity data; A quantitative identification module is used to identify potential engineering disturbance areas based on a depth threshold segmentation algorithm and the depth data, optimize the potential engineering disturbance areas to obtain engineering disturbance areas, segment the engineering disturbance areas, extract the boundary contours of the engineering disturbance areas, and calculate the area of ​​the engineering disturbance areas.

9. An electronic device, characterized in that: include: at least one processor, at least one memory, a communication interface, and a bus; The processor, memory, and communication interface communicate with each other via the bus, the memory stores program instructions executable by the processor, and the processor calls the program instructions to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions enable a computer to implement the method according to any one of claims 1 to 7.

Images