CN110826432A - Power transmission line identification method based on aerial picture - Google Patents

Power transmission line identification method based on aerial picture Download PDF

Info

Publication number
CN110826432A
CN110826432A CN201911010923.4A CN201911010923A CN110826432A CN 110826432 A CN110826432 A CN 110826432A CN 201911010923 A CN201911010923 A CN 201911010923A CN 110826432 A CN110826432 A CN 110826432A
Authority
CN
China
Prior art keywords
power transmission
sample
transmission line
image
aerial image
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201911010923.4A
Other languages
Chinese (zh)
Other versions
CN110826432B (en
Inventor
李泊
陈诚
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Nanjing Agricultural University
Original Assignee
Nanjing Agricultural University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Nanjing Agricultural University filed Critical Nanjing Agricultural University
Priority to CN201911010923.4A priority Critical patent/CN110826432B/en
Publication of CN110826432A publication Critical patent/CN110826432A/en
Application granted granted Critical
Publication of CN110826432B publication Critical patent/CN110826432B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/10—Terrestrial scenes
    • G06V20/13—Satellite images
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
    • G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
    • Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
    • Y04S10/00—Systems supporting electrical power generation, transmission or distribution
    • Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Biomedical Technology (AREA)
  • Data Mining & Analysis (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Mathematical Physics (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Multimedia (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Astronomy & Astrophysics (AREA)
  • Remote Sensing (AREA)
  • Image Analysis (AREA)

Abstract

The invention relates to a power transmission line identification method based on aerial pictures, which utilizes a multitask deep convolution neural network to complete a power transmission line identification task while realizing a power transmission line detection foundation, gives navigation points of different power transmission lines, and has the advantages of high accuracy, strong robustness, fewer constraint conditions and the like; and because the characteristics of multi-scale depth information are utilized, the method has good adaptability to the external complex environment, and the target of the power transmission line does not need to occupy an obvious position and a larger proportion in the image, so the method is very suitable for being used in a power line patrol video monitoring system, so that image processing and pattern recognition technologies such as a multi-task depth convolution neural network and the like are used, the automatic detection and recognition of the power transmission line in the field aerial image are realized, and a new data acquisition and environment perception mode is finally provided for an intelligent line patrol system.

Description

Power transmission line identification method based on aerial picture
Technical Field
The invention relates to a power transmission line identification method based on aerial pictures, and belongs to the technical field of image processing and pattern identification.
Background
With the development of aircraft technology and power industry, more and more image data are collected by various line patrol aircrafts, so that an intelligent automatic monitoring function capable of replacing human eyes is realized, and the intelligent automatic monitoring function is applied to an actual line patrol system and becomes a common research target in the field of video monitoring and intelligent power line patrol.
The position information of the transmission line is used as basic information of the transmission line and plays an important role in power line patrol. The automatic identification system of the power transmission line can be widely applied to the field line patrol process, and can realize automatic navigation and obstacle avoidance of an aircraft, selection of a power transmission line fault monitoring area and the like. However, in the application of the power transmission line video monitoring system, as the camera is carried on the aircraft, the shooting scene and the visual field of the camera are wide, the power transmission line target is not very obvious, and the difference between the power transmission line target and the image shot by the fixed video monitoring of the power distribution station is large; meanwhile, the flight attitude of the aircraft can also bring uncertainty of a shooting angle, and complex backgrounds and changeable external environments such as weather and illumination which may exist in a monitoring scene need to be considered in application.
Through the search of the prior art documents, almost all the current power transmission line detection methods only distinguish the power transmission lines through artificial rules; a paper "forward automatic power line detection for a UAV reporting and system using pulse coupled neural network and improved hough transform (using pulse coupled neural network and improved hough transform to achieve automatic power line detection of unmanned aerial vehicle monitoring system)" published in 2010 by z.r.li et al removes misdetected line segments and determines the attribution of line segments.
Disclosure of Invention
The invention aims to solve the technical problem of providing a power transmission line identification method based on aerial pictures, which aims at processing aerial images and can efficiently realize the identification of the position of a power transmission line.
The invention adopts the following technical scheme for solving the technical problems: the invention designs a power transmission line identification method based on an aerial picture, which is used for realizing the identification of a power transmission line aiming at a target aerial image containing the power transmission line, and comprises the following steps:
step A, obtaining a preset number of sample aerial images containing power transmission lines, marking the serial numbers of the power transmission lines according to the sample aerial images, marking the power transmission lines according to the power transmission lines, and entering the step B;
step B, respectively aiming at each sample aerial image, obtaining a sample binary image with a power line position pixel value of 1 and other position pixel values of 0 according to the mark of a power line in the sample aerial image, and simultaneously obtaining a sample gray-scale image containing each power line connecting line, namely obtaining the sample binary image and the sample gray-scale image which respectively correspond to each sample aerial image, and then entering the step C;
step C, taking each sample aerial image, and a sample binary image and a sample gray-scale image corresponding to each sample aerial image as input, carrying out sample training on a neural network which is based on a preset convolution neural network and is designed to have two branch outputs of a binary segmentation image and a pixel multi-dimensional feature image, obtaining a target neural network, and then entering step D; wherein, the pixel value of the position of the power line in the two-value segmentation graph is 1, and the pixel values of the other positions are 0;
d, processing the target aerial image by applying a target neural network to obtain a two-value segmentation graph and a pixel multi-dimensional feature graph corresponding to the target aerial image, and then entering the step E;
and E, according to the two-value segmentation graph and the pixel multi-dimensional feature graph corresponding to the target aerial image, clustering all pixels of the positions of the power transmission lines based on the multi-dimensional features of the pixels, namely, all pixels in all clusters respectively form all power transmission lines in the target aerial image, and power transmission line identification is achieved.
As a preferred technical scheme of the invention: step F is also included, and step F is entered after step E is executed;
and F, respectively aiming at each power transmission line in the target aerial image, and aiming at each pixel on the position of the power transmission line, fitting based on the coordinate position of each pixel in the target aerial image to obtain a navigation point corresponding to the power transmission line, namely obtaining the navigation point corresponding to each power transmission line in the target aerial image.
As a preferred technical scheme of the invention: in the step F, the following steps F1 to F3 are executed for each power line in the target aerial image, respectively, to obtain a navigation point corresponding to each power line in the target aerial image;
f1, fitting all pixel coordinates on the position of the power transmission line through a preset 3-order polynomial to obtain a fitting function corresponding to the power transmission line;
step F2., obtaining the minimum value and the maximum value on the x axis according to the coordinates of each pixel on the power transmission line, and calculating the difference value between the minimum value and the maximum value as the x axis difference value; obtaining the minimum value and the maximum value on the y axis, and calculating the difference between the minimum value and the maximum value as the y axis difference; then selecting the axis corresponding to the maximum difference value as an input axis from the x-axis difference value and the y-axis difference value;
step F3., selecting coordinates of each pixel on the input shaft corresponding to the position of the power transmission line according to a preset step interval, using the coordinates as the input of the fitting function corresponding to the power transmission line, and calculating to obtain the coordinates on the other shaft corresponding to the input shaft, namely, using the coordinates as the navigation points corresponding to the power transmission line.
As a preferred technical scheme of the invention: in the step B, in the sample gray level image which includes the connection lines of the power transmission lines and corresponds to the sample aerial image, the gray level value of each pixel point at each power transmission line position is determined according to the following formula:
Figure BDA0002244174850000021
wherein I is more than or equal to 1 and less than or equal to I, I represents the number of transmission lines in the sample gray-scale image, ViRepresenting the grey value of a pixel point on the ith power line in the sample aerial image, ImaxAnd the maximum value of the number of the power transmission lines in all the sample aerial images and in a single sample aerial image is represented.
As a preferred technical scheme of the invention: in the step C, in the process of training a sample for the designed neural network with two branch outputs of the binary segmentation graph and the pixel multidimensional feature graph, single processing is implemented for the sample aerial image in the following manner to obtain a loss function result of the neural network corresponding to the single processing, so that each sample aerial image, the sample binary graph and the sample gray-scale graph corresponding to the sample aerial image are used as input to complete sample training for designing the neural network;
firstly, processing a sample aerial image by using the neural network to obtain a two-value segmentation graph and a pixel multi-dimensional feature graph corresponding to the sample aerial image; then, calculating a branch loss function between the binary segmentation image and a sample binary image corresponding to the sample aerial image, and calculating a branch loss function between the pixel multi-dimensional feature image and a sample gray-scale image corresponding to the sample aerial image; and finally, according to the loss function result of each branch, combining with the preset loss function weight of each branch, and calculating in a weighting mode to obtain the loss function result of the neural network.
As a preferred technical scheme of the invention: processing a sample aerial image based on the application of the neural network to obtain a two-value segmentation graph and a pixel multi-dimensional feature graph corresponding to the sample aerial image, and then calculating a branch loss function between the two-value segmentation graph and the sample two-value graph corresponding to the sample aerial image by adopting a cross entropy loss function; and calculating a branch loss function between the pixel multi-dimensional feature map and a sample gray scale map corresponding to the sample aerial image by adopting a discriminant loss function.
As a preferred technical scheme of the invention: the preset convolutional neural network is a VGG16 convolutional neural network.
As a preferred technical scheme of the invention: and E, performing clustering processing on all pixels of the position of the power transmission line by applying a Mean-Shift clustering algorithm based on the multi-dimensional characteristics of the pixels according to the two-value segmentation graph and the pixel multi-dimensional characteristic graph corresponding to the target aerial image.
Compared with the prior art, the power transmission line identification method based on the aerial picture has the following technical effects:
the power transmission line identification method based on the aerial picture is designed, aiming at the requirement of acquiring line information by using a video monitoring system in intelligent power line patrol, determining an aerial picture marking strategy and an automatic sample generation method by using related technologies of computer vision, image processing and mode identification, and completing a power transmission line identification task by using a multi-task deep convolutional neural network while realizing a power transmission line detection basis, so as to provide navigation points of different power transmission lines, and has the advantages of high accuracy, strong robustness, fewer constraint conditions and the like; and because the characteristics of multi-scale depth information are utilized, the method has good adaptability to the external complex environment, and the target of the power transmission line does not need to occupy an obvious position and a larger proportion in the image, so the method is very suitable for being used in a power line patrol video monitoring system, so that image processing and pattern recognition technologies such as a multi-task depth convolution neural network and the like are used, the automatic detection and recognition of the power transmission line in the field aerial image are realized, and a new data acquisition and environment perception mode is finally provided for an intelligent line patrol system.
Drawings
FIG. 1 is a schematic flow chart of the method for identifying an aerial picture-based power transmission line according to the present invention;
fig. 2 is a schematic flow chart of generation of navigation points in the aviation picture-based power transmission line identification method.
Detailed Description
The following description will explain embodiments of the present invention in further detail with reference to the accompanying drawings.
The invention designs a power transmission line identification method based on an aerial picture, which is used for realizing the identification of a power transmission line aiming at a target aerial image containing the power transmission line, and in practical application, as shown in figure 1, the method specifically comprises the following steps.
And A, obtaining a preset number of sample aerial images containing power transmission lines, marking the serial numbers of the power transmission lines according to the sample aerial images, marking the power transmission lines according to the power transmission lines, and entering the step B.
In the practical application of step a, 525 images of the power transmission line shot by the fixed-wing inspection unmanned aerial vehicle in the field environment are used as aerial images of each sample, and the source and resolution of the images are not limited, such as 4288 × 2848.
And B, respectively aiming at each sample aerial image, obtaining a sample binary image with the pixel value of the position of the power line being 1 and the pixel values of the rest positions being 0 according to the mark of the power line, simultaneously obtaining a sample gray-scale image containing the connecting line of each power line, namely obtaining the sample binary image and the sample gray-scale image which respectively correspond to each sample aerial image, and then entering the step C.
In practical application, in a sample gray-scale image corresponding to the sample aerial image and containing the connection lines of the power transmission lines, the gray-scale value of each pixel point at each power transmission line position is determined according to the following formula:
Figure BDA0002244174850000041
wherein I is more than or equal to 1 and less than or equal to I, I represents the number of transmission lines in the sample gray-scale image, ViRepresenting the grey value of a pixel point on the ith power line in the sample aerial image, ImaxAnd the maximum value of the number of the power transmission lines in all the sample aerial images and in a single sample aerial image is represented.
Based on the practical embodiment of the step a, in the step B, the marking tool is used to mark the transmission line point by point, and the position information recorded by the mark is used to generate a binary image and a grayscale image, which is as follows.
And B1, reading in the sample aerial image by using an open source labeling tool Labelme, and manually searching for power transmission lines in the image according to a certain sequence, wherein the power transmission lines are numbered from left to right in the example, and I is greater than or equal to 1 and less than or equal to I. And the power transmission line is marked point by manpower, the number of the marks is more than 3, the connection line between every two marks is ensured to fall on the power transmission line, and finally the head mark and the tail mark are connected to finish a power transmission line marking task. The number of the power line marking tasks in the single sample aerial image is determined by the number I of the power lines seen by human eyes in the image.
Step B2, aiming at each sample aerial image, storing the marked position coordinates marked by the marking tasks, and generating connecting lines with the pixel value of 1 in the binary image according to the marked position coordinates of each power transmission line, wherein the number of the connecting lines is the same as that of the power transmission lines, the width of the connecting lines is determined according to the maximum width of the power transmission lines, and the width value in the embodiment is 10 pixels; and simultaneously, generating connecting lines with the same number as the transmission lines in the gray-scale image according to the marked position coordinates of each transmission line, wherein the width of each connecting line is consistent with that of the binary image.
Step C, taking each sample aerial image, and a sample binary image and a sample gray-scale image corresponding to each sample aerial image as input, carrying out sample training on a neural network which is based on a preset convolution neural network and is designed to have two branch outputs of a binary segmentation image and a pixel multi-dimensional feature image, obtaining a target neural network, and then entering step D; wherein, the pixel value of the power line position in the two-value segmentation graph is 1, and the pixel values of the other positions are 0.
In the practical application of the step C, the preset convolutional neural network is specifically selected as the VGG16 convolutional neural network, and in the process of training the sample for the designed neural network with two branch outputs of the two-value segmentation graph and the pixel multi-dimensional feature graph, single processing is implemented for the sample aerial image in the following manner to obtain the loss function result of the neural network corresponding to the single processing, so that each sample aerial image, and the sample binary graph and the sample gray scale graph corresponding to the sample aerial image are used as input to complete the sample training for designing the neural network.
Firstly, processing a sample aerial image by using the neural network to obtain a two-value segmentation graph and a pixel multi-dimensional feature graph corresponding to the sample aerial image; then, calculating a branch loss function between the binary segmentation image and a sample binary image corresponding to the sample aerial image by adopting a cross entropy loss function, and calculating a branch loss function between the pixel multidimensional feature image and a sample gray-scale image corresponding to the sample aerial image by adopting a discriminant loss function; and finally, according to the loss function result of each branch, combining with the preset loss function weight of each branch, and calculating in a weighting mode to obtain the loss function result of the neural network.
In practical application, the training operation is specifically as follows:
and C1, aiming at the detection and identification tasks, determining a multi-task convolutional neural network framework, adopting a coding-decoding (Encoder-Decoder) network structure, wherein a decoding end is of a double-branch structure and is respectively responsible for binary segmentation and feature extraction tasks.
The coding network designed in step C2. is responsible for extracting image features of the input image under different scales, in this embodiment, the front 13 layers of the VGG-16 network are used as the coding network, and the weight of the VGG-16 under the ImageNet image public data set is used as the initial assignment of the weight of the coding network.
Step C3: the designed decoding network is responsible for pixel-by-pixel attribute prediction. In this embodiment, the category attribute (wire or background) and the multi-dimensional feature attribute of each pixel in the image need to be predicted. The FCN is used as a decoder network, a double-branch structure is added at the tail end of the decoder network, and a binary segmentation prediction image and a multi-dimensional pixel characteristic image are respectively output.
Step C4: loss functions are designed for the two branches respectively, and in the embodiment, a Cross Entropy Loss function (Cross Entropy Loss) is adopted for the binary segmentation branch. A discriminant Loss Function (discriminant Loss Function) is adopted for the pixel multi-dimensional feature extraction branch.
Step C5: and designing a total loss function based on the loss functions of the two branches, and iteratively optimizing a network weight based on the total loss function to obtain a model. In this embodiment, the total loss function is equal to a weighted sum of two-branch loss functions, each branch having a weight of 0.5. And training the network by adopting a random gradient descent (SGD) method in the iterative optimization process, and optimizing the weight of the network.
And D, applying a target neural network, processing the target aerial image to obtain a two-value segmentation graph and a pixel multi-dimensional feature graph corresponding to the target aerial image, and then entering the step E.
And E, according to the binary segmentation graph and the pixel multi-dimensional feature graph corresponding to the target aerial image, based on the multi-dimensional features of the pixels, applying a Mean-Shift clustering algorithm, clustering all pixels at the positions of the power transmission lines, namely, all pixels in all clusters respectively form all power transmission lines in the target aerial image, realizing power transmission line identification, and then entering the step F.
The clustering specifically includes the following:
step e1, extracting the characteristics of the power line pixels from the pixel characteristic map according to the non-zero pixel position coordinates in the binary map, and forming M × (N +2) matrices by using all the power line pixel characteristics and the pixel coordinates thereof in this embodiment, where M is the non-zero pixel number, N is the pixel characteristic dimension, and the last two columns of each row are the pixel coordinates.
And E2, clustering the pixels of the power transmission line according to the characteristics of the pixels, wherein M N-dimensional characteristic vectors in the matrix are clustered by using a Mean-Shift clustering algorithm in the embodiment. The maximum clustering category number of the Mean-Shift algorithm is set as the maximum transmission line number seen by human eyes in a target aerial image in the image library. And adding a column to record the cluster type id of each vector after the M (N +2) matrix.
And F, respectively aiming at each power transmission line in the target aerial image, aiming at each pixel on the position of the power transmission line, and performing fitting according to the coordinate position of each pixel in the target aerial image as shown in FIG. 2 in the following steps F1 to F3 to obtain the navigation point corresponding to the power transmission line, namely obtaining the navigation point corresponding to each power transmission line in the target aerial image.
And F1, fitting all pixel coordinates on the position of the power transmission line through a preset 3-order polynomial to obtain a fitting function corresponding to the power transmission line.
Step F2., obtaining the minimum value and the maximum value on the x axis according to the coordinates of each pixel on the power transmission line, and calculating the difference value between the minimum value and the maximum value as the x axis difference value; obtaining the minimum value and the maximum value on the y axis, and calculating the difference between the minimum value and the maximum value as the y axis difference; and then selecting the axis corresponding to the maximum difference value as the input axis from the x-axis difference value and the y-axis difference value.
Step F3., selecting coordinates of each pixel on the input shaft corresponding to the position of the power transmission line according to a preset step interval, using the coordinates as the input of the fitting function corresponding to the power transmission line, and calculating to obtain the coordinates on the other shaft corresponding to the input shaft, namely, using the coordinates as the navigation points corresponding to the power transmission line.
The power transmission line identification method based on the aerial picture is designed by the technical scheme, the linear characteristics of the power transmission line are utilized, a reasonable labeling mode is designed, and a binary segmentation graph and gray level graphs facing different power transmission line examples are automatically generated; then, according to the detection and identification of two types of tasks, a multi-task convolutional neural network is designed, and a model is iteratively trained on the basis of a sample library; obtaining a binary segmentation graph and a pixel feature graph in real time through a model; by utilizing the binary segmentation graph, the non-zero pixel position is the position of the power transmission line, and the task of detecting the power transmission line is completed; and then extracting the multidimensional characteristics of the pixels of the power transmission line from the pixel characteristic graph through a binary segmentation graph, finishing pixel clustering through a clustering algorithm, and marking the pixels of the power transmission line with corresponding clustering categories id. Powerline pixels with the same id belong to the same powerline. Because the depth features of different scales of the depth convolution network are utilized, the detection result of the method is more robust in a complex environment, the constraints on the external environment and the shooting angle are less, and the detection and identification accuracy is improved; meanwhile, due to the highly parallel design of the deep convolutional network, the real-time performance of the algorithm can be improved on a GPU platform.
The power transmission line identification method based on the aerial pictures is designed by the technical scheme, aiming at the requirement of acquiring line information by using a video monitoring system in intelligent power line patrol, determining an aerial picture marking strategy and an automatic sample generation method by using related technologies of computer vision, image processing and mode identification, completing a power transmission line identification task by using a multi-task deep convolutional neural network while realizing a power transmission line detection basis, providing navigation points of different power transmission lines, and having the advantages of high accuracy, strong robustness, fewer constraint conditions and the like; and because the characteristics of multi-scale depth information are utilized, the method has good adaptability to the external complex environment, and the target of the power transmission line does not need to occupy an obvious position and a larger proportion in the image, so the method is very suitable for being used in a power line patrol video monitoring system, so that image processing and pattern recognition technologies such as a multi-task depth convolution neural network and the like are used, the automatic detection and recognition of the power transmission line in the field aerial image are realized, and a new data acquisition and environment perception mode is finally provided for an intelligent line patrol system.
The embodiments of the present invention have been described in detail with reference to the drawings, but the present invention is not limited to the above embodiments, and various changes can be made within the knowledge of those skilled in the art without departing from the gist of the present invention.

Claims (8)

1. A power transmission line identification method based on aerial pictures is used for realizing identification of power transmission lines of target aerial images containing the power transmission lines, and is characterized by comprising the following steps:
step A, obtaining a preset number of sample aerial images containing power transmission lines, marking the serial numbers of the power transmission lines according to the sample aerial images, marking the power transmission lines according to the power transmission lines, and entering the step B;
step B, respectively aiming at each sample aerial image, obtaining a sample binary image with a power line position pixel value of 1 and other position pixel values of 0 according to the mark of a power line in the sample aerial image, and simultaneously obtaining a sample gray-scale image containing each power line connecting line, namely obtaining the sample binary image and the sample gray-scale image which respectively correspond to each sample aerial image, and then entering the step C;
step C, taking each sample aerial image, and a sample binary image and a sample gray-scale image corresponding to each sample aerial image as input, carrying out sample training on a neural network which is based on a preset convolution neural network and is designed to have two branch outputs of a binary segmentation image and a pixel multi-dimensional feature image, obtaining a target neural network, and then entering step D; wherein, the pixel value of the position of the power line in the two-value segmentation graph is 1, and the pixel values of the other positions are 0;
d, processing the target aerial image by applying a target neural network to obtain a two-value segmentation graph and a pixel multi-dimensional feature graph corresponding to the target aerial image, and then entering the step E;
and E, according to the two-value segmentation graph and the pixel multi-dimensional feature graph corresponding to the target aerial image, clustering all pixels of the positions of the power transmission lines based on the multi-dimensional features of the pixels, namely, all pixels in all clusters respectively form all power transmission lines in the target aerial image, and power transmission line identification is achieved.
2. The aerial-picture-based power transmission line identification method according to claim 1, characterized in that: step F is also included, and step F is entered after step E is executed;
and F, respectively aiming at each power transmission line in the target aerial image, and aiming at each pixel on the position of the power transmission line, fitting based on the coordinate position of each pixel in the target aerial image to obtain a navigation point corresponding to the power transmission line, namely obtaining the navigation point corresponding to each power transmission line in the target aerial image.
3. The aerial-picture-based power transmission line identification method according to claim 2, characterized in that: in the step F, the following steps F1 to F3 are executed for each power line in the target aerial image, respectively, to obtain a navigation point corresponding to each power line in the target aerial image;
f1, fitting all pixel coordinates on the position of the power transmission line through a preset 3-order polynomial to obtain a fitting function corresponding to the power transmission line;
step F2., obtaining the minimum value and the maximum value on the x axis according to the coordinates of each pixel on the power transmission line, and calculating the difference value between the minimum value and the maximum value as the x axis difference value; obtaining the minimum value and the maximum value on the y axis, and calculating the difference between the minimum value and the maximum value as the y axis difference; then selecting the axis corresponding to the maximum difference value as an input axis from the x-axis difference value and the y-axis difference value;
step F3., selecting coordinates of each pixel on the input shaft corresponding to the position of the power transmission line according to a preset step interval, using the coordinates as the input of the fitting function corresponding to the power transmission line, and calculating to obtain the coordinates on the other shaft corresponding to the input shaft, namely, using the coordinates as the navigation points corresponding to the power transmission line.
4. The aerial-picture-based power transmission line identification method according to any one of claims 1 to 3, characterized in that: in the step B, in the sample gray level image which includes the connection lines of the power transmission lines and corresponds to the sample aerial image, the gray level value of each pixel point at each power transmission line position is determined according to the following formula:
Figure FDA0002244174840000021
wherein I is more than or equal to 1 and less than or equal to I, I represents the number of transmission lines in the sample gray-scale image, ViRepresenting the grey value of a pixel point on the ith power line in the sample aerial image, ImaxAnd the maximum value of the number of the power transmission lines in all the sample aerial images and in a single sample aerial image is represented.
5. The aerial-picture-based power transmission line identification method according to any one of claims 1 to 3, characterized in that: in the step C, in the process of training a sample for the designed neural network with two branch outputs of the binary segmentation graph and the pixel multidimensional feature graph, single processing is implemented for the sample aerial image in the following manner to obtain a loss function result of the neural network corresponding to the single processing, so that each sample aerial image, the sample binary graph and the sample gray-scale graph corresponding to the sample aerial image are used as input to complete sample training for designing the neural network;
firstly, processing a sample aerial image by using the neural network to obtain a two-value segmentation graph and a pixel multi-dimensional feature graph corresponding to the sample aerial image; then, calculating a branch loss function between the binary segmentation image and a sample binary image corresponding to the sample aerial image, and calculating a branch loss function between the pixel multi-dimensional feature image and a sample gray-scale image corresponding to the sample aerial image; and finally, according to the loss function result of each branch, combining with the preset loss function weight of each branch, and calculating in a weighting mode to obtain the loss function result of the neural network.
6. The aerial-picture-based power transmission line identification method according to claim 5, wherein: processing a sample aerial image based on the application of the neural network to obtain a two-value segmentation graph and a pixel multi-dimensional feature graph corresponding to the sample aerial image, and then calculating a branch loss function between the two-value segmentation graph and the sample two-value graph corresponding to the sample aerial image by adopting a cross entropy loss function; and calculating a branch loss function between the pixel multi-dimensional feature map and a sample gray scale map corresponding to the sample aerial image by adopting a discriminant loss function.
7. The aerial-picture-based power transmission line identification method according to claim 5, wherein: the preset convolutional neural network is a VGG16 convolutional neural network.
8. The aerial-picture-based power transmission line identification method according to any one of claims 1 to 3, characterized in that: and E, performing clustering processing on all pixels of the position of the power transmission line by applying a Mean-Shift clustering algorithm based on the multi-dimensional characteristics of the pixels according to the two-value segmentation graph and the pixel multi-dimensional characteristic graph corresponding to the target aerial image.
CN201911010923.4A 2019-10-23 2019-10-23 Power transmission line identification method based on aviation picture Active CN110826432B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201911010923.4A CN110826432B (en) 2019-10-23 2019-10-23 Power transmission line identification method based on aviation picture

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201911010923.4A CN110826432B (en) 2019-10-23 2019-10-23 Power transmission line identification method based on aviation picture

Publications (2)

Publication Number Publication Date
CN110826432A true CN110826432A (en) 2020-02-21
CN110826432B CN110826432B (en) 2023-09-22

Family

ID=69550175

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201911010923.4A Active CN110826432B (en) 2019-10-23 2019-10-23 Power transmission line identification method based on aviation picture

Country Status (1)

Country Link
CN (1) CN110826432B (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112990293A (en) * 2021-03-10 2021-06-18 深圳一清创新科技有限公司 Point cloud marking method and device and electronic equipment

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2018028255A1 (en) * 2016-08-11 2018-02-15 深圳市未来媒体技术研究院 Image saliency detection method based on adversarial network
CN108009515A (en) * 2017-12-14 2018-05-08 杭州远鉴信息科技有限公司 A kind of power transmission line positioning identifying method of the unmanned plane image based on FCN
CN109543595A (en) * 2018-11-19 2019-03-29 上海交通大学 The training method and detection method of the electric wire of convolutional neural networks are separated based on depth

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2018028255A1 (en) * 2016-08-11 2018-02-15 深圳市未来媒体技术研究院 Image saliency detection method based on adversarial network
CN108009515A (en) * 2017-12-14 2018-05-08 杭州远鉴信息科技有限公司 A kind of power transmission line positioning identifying method of the unmanned plane image based on FCN
CN109543595A (en) * 2018-11-19 2019-03-29 上海交通大学 The training method and detection method of the electric wire of convolutional neural networks are separated based on depth

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112990293A (en) * 2021-03-10 2021-06-18 深圳一清创新科技有限公司 Point cloud marking method and device and electronic equipment
CN112990293B (en) * 2021-03-10 2024-03-29 深圳一清创新科技有限公司 Point cloud labeling method and device and electronic equipment

Also Published As

Publication number Publication date
CN110826432B (en) 2023-09-22

Similar Documents

Publication Publication Date Title
CN110472627B (en) End-to-end SAR image recognition method, device and storage medium
CN109584248B (en) Infrared target instance segmentation method based on feature fusion and dense connection network
US20210390329A1 (en) Image processing method, device, movable platform, unmanned aerial vehicle, and storage medium
CN108694386B (en) Lane line detection method based on parallel convolution neural network
CN106897681B (en) Remote sensing image contrast analysis method and system
Biasutti et al. Lu-net: An efficient network for 3d lidar point cloud semantic segmentation based on end-to-end-learned 3d features and u-net
Girisha et al. Uvid-net: Enhanced semantic segmentation of uav aerial videos by embedding temporal information
CN110175615B (en) Model training method, domain-adaptive visual position identification method and device
CN111967313B (en) Unmanned aerial vehicle image annotation method assisted by deep learning target detection algorithm
CN110135227B (en) Laser point cloud outdoor scene automatic segmentation method based on machine learning
CN113449594A (en) Multilayer network combined remote sensing image ground semantic segmentation and area calculation method
CN111652273B (en) Deep learning-based RGB-D image classification method
CN114067118B (en) Processing method of aerial photogrammetry data
CN111241970A (en) SAR image sea surface ship detection method based on yolov3 algorithm and sliding window strategy
US20220237908A1 (en) Flight mission learning using synthetic three-dimensional (3d) modeling and simulation
CN113838064B (en) Cloud removal method based on branch GAN using multi-temporal remote sensing data
CN112712052A (en) Method for detecting and identifying weak target in airport panoramic video
CN113936214A (en) Karst wetland vegetation community classification method based on fusion of aerospace remote sensing images
CN111046756A (en) Convolutional neural network detection method for high-resolution remote sensing image target scale features
CN110472632B (en) Character segmentation method and device based on character features and computer storage medium
CN113033386B (en) High-resolution remote sensing image-based transmission line channel hidden danger identification method and system
CN112528058B (en) Fine-grained image classification method based on image attribute active learning
CN110826432B (en) Power transmission line identification method based on aviation picture
CN117437382A (en) Updating method and system for data center component
CN116630828B (en) Unmanned aerial vehicle remote sensing information acquisition system and method based on terrain environment adaptation

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant