CN116310316A - Electric overhead conductor detection method based on cross checking - Google Patents

Electric overhead conductor detection method based on cross checking Download PDF

Info

Publication number
CN116310316A
CN116310316A CN202310109206.7A CN202310109206A CN116310316A CN 116310316 A CN116310316 A CN 116310316A CN 202310109206 A CN202310109206 A CN 202310109206A CN 116310316 A CN116310316 A CN 116310316A
Authority
CN
China
Prior art keywords
overhead conductor
straight line
image
line segment
detection
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.)
Pending
Application number
CN202310109206.7A
Other languages
Chinese (zh)
Inventor
杨凡
童莹
何睿清
乔家齐
蔡昊
沈伟
胡欣
卫江春
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 Institute of Technology
Original Assignee
Nanjing Institute of Technology
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 Institute of Technology filed Critical Nanjing Institute of Technology
Priority to CN202310109206.7A priority Critical patent/CN116310316A/en
Publication of CN116310316A publication Critical patent/CN116310316A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Energy or water supply
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
    • G06V10/443—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components by matching or filtering
    • G06V10/449—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters
    • G06V10/451—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters with interaction between the filter responses, e.g. cortical complex cells
    • G06V10/454—Integrating the filters into a hierarchical structure, e.g. convolutional neural networks [CNN]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/52—Scale-space analysis, e.g. wavelet analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/80—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level
    • G06V10/806—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level of extracted features
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/10—Terrestrial scenes
    • G06V20/17—Terrestrial scenes taken from planes or by drones
    • G—PHYSICS
    • G07—CHECKING-DEVICES
    • G07C—TIME OR ATTENDANCE REGISTERS; REGISTERING OR INDICATING THE WORKING OF MACHINES; GENERATING RANDOM NUMBERS; VOTING OR LOTTERY APPARATUS; ARRANGEMENTS, SYSTEMS OR APPARATUS FOR CHECKING NOT PROVIDED FOR ELSEWHERE
    • G07C1/00—Registering, indicating or recording the time of events or elapsed time, e.g. time-recorders for work people
    • G07C1/20—Checking timed patrols, e.g. of watchman
    • 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)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Evolutionary Computation (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Artificial Intelligence (AREA)
  • Business, Economics & Management (AREA)
  • Computing Systems (AREA)
  • Economics (AREA)
  • Software Systems (AREA)
  • Medical Informatics (AREA)
  • Databases & Information Systems (AREA)
  • Water Supply & Treatment (AREA)
  • Biodiversity & Conservation Biology (AREA)
  • Remote Sensing (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Molecular Biology (AREA)
  • Public Health (AREA)
  • Biomedical Technology (AREA)
  • Human Resources & Organizations (AREA)
  • Strategic Management (AREA)
  • Primary Health Care (AREA)
  • Marketing (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Image Analysis (AREA)

Abstract

Electric overhead conductor detection method based on cross check, input electric overhead conductor image and use the electric overhead conductor imagePerforming electric power overhead conductor detection on the deep Labv3+ segmentation network, performing binarization operation to obtain a binarized image, performing electric power overhead conductor detection on the electric power overhead conductor by using an MCMLSD straight line detection algorithm, outputting a finally detected straight line segment, and obtaining coordinates of a start point and an end point of the straight line segment; discretizing coordinates of a start point and an end point of the straight line segment, determining the coordinates of the start point and the end point of the straight line segment, and imaging the electric overhead conductor I Drawing corresponding straight line segments and marking; and outputting the marked image to obtain a final complete electric overhead conductor detection image. The invention realizes complete extraction of the electric power overhead conductor by using the cross checking method, avoids the problems of false detection, missing detection, breakage and the like of the power line in the traditional method, and provides technical support for automatically and accurately detecting the electric power overhead conductor by the unmanned aerial vehicle.

Description

Electric overhead conductor detection method based on cross checking
Technical Field
The invention relates to the technical field of power systems, in particular to a cross-checking-based power overhead conductor detection method.
Background
In a power transmission system, an overhead conductor is one of the main components. In order to ensure the normal operation of the overhead conductor, a large amount of manpower and material resources are required for power line inspection. The manual inspection is time-consuming and labor-consuming, and the inspection quality of a long-distance high-voltage transmission line is difficult to ensure. In recent years, unmanned aerial vehicle technology development is rapid, and this makes unmanned aerial vehicle inspection can replace artifical inspection gradually, and especially in long distance high tension transmission system inspection, unmanned aerial vehicle inspection has played important effect. In the field, based on pictures or videos shot by unmanned aerial vehicles, accurate extraction of overhead conductors is a key and precondition for realizing automatic high-precision unmanned aerial vehicle inspection.
At present, the traditional power overhead conductor detection methods are mainly divided into two types: the method is based on an edge detection operator, and mainly uses edge detection algorithms such as Sobel, canny and Gabor to process the characteristics of the electric overhead conductor, and extracts the electric overhead conductor from the background; the other type is to extract power lines or fit power line segments by using a power line to have straight lines or characteristics similar to straight lines based on trend characteristics and shape characteristics of edges of objects in images and adopting straight line detection algorithms such as LSD, cannyLines and Hough transformation. In recent years, the deep learning technology is also widely applied to the power industry, wherein a semantic segmentation network represented by deep labv3+ can realize pixel-level classification of an electric overhead conductor, and compared with a traditional image processing method, the method has less interference by linear morphological characteristics in images and good scene robustness.
The defects of the prior art scheme are as follows:
1. the traditional power wire detection method based on image processing has poor real-time performance and low efficiency, and cannot be suitable for various complex inspection scenes (such as forests, fields, cities and the like). For example, the principle of the edge detection algorithm is to adopt an edge detection operator to extract the shallow layer characteristics of the target for judgment, so that the wire detection is realized. When the target characteristics of the wires in the inspection acquisition image are not obvious or the scene is complex, the target shallow layer characteristics are easily interfered, so that the detection effect of the electric overhead wire is not ideal. Also, the straight line detection algorithm represented by Hough transformation is easy to be interfered by straight line information of a non-overhead conductor in an image, a large number of false detections are generated, and the difficulty of conductor detection is increased.
2. Although the deep semantic segmentation network represented by deep Labv3+ has robustness to complex scenes and is not easily interfered by other factors such as acquired image quality, scenes and the like, the situation that a segmentation area is discontinuous, broken and the like can occur in a wire segmentation result, so that a detection wire does not have integrity, and the positioning of a later unmanned aerial vehicle to an overhead wire is affected.
Disclosure of Invention
The invention is applied to automatic detection of the electric power overhead wire under the complex background in the inspection of the unmanned aerial vehicle, realizes complete extraction of the electric power overhead wire by using a cross inspection method, avoids the problems of false detection, missing detection, breakage and the like of the electric power wire in the traditional method, and provides technical support for automatic and accurate detection of the electric power overhead wire by the unmanned aerial vehicle.
A detection method of an electric overhead conductor based on cross checking comprises the following steps:
step S1: inputting a 512×512 power overhead conductor image, labeled I;
step S2: carrying out electric overhead conductor detection on the electric overhead conductor image I by using a deep Labv3+ segmentation network to obtain a detected overhead conductor image I 1 ；
Step S3: for the detected overhead conductor image I 1 Performing binarization operation to obtain a binarized image I 2 Will I 2 The set of medium image pixels with value 1 is labeled P 0 The set is a straight line identification area;
step S4: detecting the electric overhead conductor by using an MCMLSD straight line detection algorithm on the electric overhead conductor image I, outputting a finally detected straight line segment, and obtaining coordinates of a starting point and an ending point of the straight line segment;
step S5: discretizing coordinates of a starting point and an ending point of the straight line segment, reserving coordinate information of the straight line segment through discretization, and marking the obtained straight line segment set as P i ，P i Discrete coordinate information representing the i-th straight line segment, i=1, 2, 3..n;
step S6: let i=1, solve straight-line segment set P 1 And P 0 Is the intersection of (a), i.e
Figure BDA0004076213180000031
Step S7: p pair P i Each line segment of the set P is operated according to the step S6 to obtain a straight line segment set P i And P 0 Is set as
Figure BDA0004076213180000032
Step S8: computing a set
Figure BDA0004076213180000033
P in each subset 0 The number of intersections, resulting in a set->
Figure BDA0004076213180000034
i＝1,2,3...n；
Step S9: removal of
Figure BDA0004076213180000035
The subsets of zero in (1) are ordered in descending order, the set at this time being marked +.>
Figure BDA0004076213180000036
Step S10: selecting
Figure BDA0004076213180000037
The first 75% of the subsets are indexed by the index line segment number of the superscript i in the subset, by P in step S5 i Determining coordinate values of starting and ending points of the straight line segments, drawing corresponding straight line segments on the electric overhead conductor image I and marking the corresponding straight line segments;
step S11: outputting the marked image to obtain a final complete electric overhead conductor detection image I 0 。
Preferably, in step S2 of the present invention, the power overhead conductor image I is detected by using the deeplbvv3+ split network to obtain a detected overhead conductor image I 1 The method comprises the steps of carrying out a first treatment on the surface of the The specific process is as follows:
the deep Labv3+ partition network is in a coding and decoding structure and comprises a coding area and a decoding area, in the coding area, the deep convolutional neural network is utilized to conduct feature extraction on an original input image to obtain two different feature information, then a spatial pyramid pooling module is utilized to conduct multi-scale cavity convolution sampling on the feature information of a higher level extracted by the deep convolutional neural network, a plurality of feature layers obtained by cavity convolution with different expansion rates are stacked together, and 1X 1 convolution is conducted after combination, so that a high-level semantic feature layer is formed;
in the decoding area part, carrying out 1X 1 convolution on the lower-level characteristic information extracted from the deep convolutional neural network to form a low-level semantic characteristic layer, carrying out characteristic fusion with the high-level semantic characteristic layer after 4 times up sampling, and supplementing image position information; after fusion ofThe complete semantic feature layer is subjected to feature extraction of 3×3 convolution, and then 4 times of up-sampling is used for adjusting an output image to be as large as an input image, so that a detected overhead conductor image I is obtained 1 。
Preferably, in step S3 of the invention, the detected overhead conductor image I is displayed 1 Performing binarization operation to obtain a binarized image I 2 The method comprises the steps of carrying out a first treatment on the surface of the The specific process is as follows:
image binarization is carried out by calculating a global threshold value based on a gray average value of a histogram, an iterative optimal threshold value and an OTSU algorithm, superposing the global threshold value and background distribution to form a local threshold value, carrying out binarization processing according to the obtained global threshold value and the local threshold value, determining a pixel value of each pixel point in an image, wherein values 1 and 0 respectively correspond to pixels which are identified as an overhead conductor and a non-overhead conductor by a network, and finally obtaining a binarized image I 2 。
Preferably, in step S4 of the present invention, an mcml sd linear detection algorithm is used to detect the electric overhead conductor for the electric overhead conductor image I, and the final detected linear segment is output, and coordinates of a start point and an end point of the linear segment are obtained;
firstly, detecting a set of straight line segments in a picture by using global probability Hough transformation on an electric overhead conductor image I, analyzing all detected straight line segments in the image, determining line segment positions corresponding to peaks in Hough domains, thus obtaining a group of line segments to be selected, modeling edge points on each line segment to be selected by using a linear Markov chain model, obtaining a corresponding probability optimal label, carrying out saliency sequencing on all the line segments to be selected according to the probability optimal label, and finally outputting the finally detected straight line segments according to a sequencing result, and obtaining coordinates of a starting point and an ending point of the straight line segments.
Preferably, in step S5 of the present invention, coordinates of a start point and an end point of a straight line segment are discretized; the specific process is as follows:
the starting point coordinates of the marked straight line segment are as follows
Figure BDA0004076213180000054
The coordinates of the ending point of the straight line segment are +.>
Figure BDA0004076213180000051
Calculating the number of pixel point intervals between two points as M; x direction->
Figure BDA0004076213180000052
And->
Figure BDA0004076213180000053
The row index coordinate set with interval 1 is marked as x k Similarly, a column index coordinate set in the y direction is obtained and is marked as y k ，k＝1,2,3...M。
Overhead conductor detection is an important item in power inspection. The traditional overhead conductor detection method can better identify the morphological characteristics of the overhead conductor, but has limited scene robustness and is difficult to eliminate surrounding environment interference. The overhead conductor detection based on the depth semantic segmentation is an emerging detection method in recent years, the scene robustness is good, but in practice, the detection result of the method has the problem of discontinuous segmentation areas. Aiming at the problems, the invention combines the traditional straight line detection algorithm with the deep learning network, adopts the MSMLSD algorithm (Markov Chain Marginal Line Segment Detector) and the deep Labv3+ segmentation network to detect the overhead conductor respectively, provides an overhead conductor detection method based on cross test, creatively fuses the overhead conductor coordinate information extracted by the traditional straight line detection algorithm with the overhead conductor picture information obtained by the deep learning network, successfully extracts the complete overhead conductor and inhibits the influence of the line segment structure in the surrounding environment on the detection result by unifying and comparing different data information. Experiments prove that the method has good robustness in the scenes such as sky, land, forest and the like, provides a new thought for automatic inspection of unmanned aerial vehicle frame empty wires, and advances the technical development of the field.
Drawings
FIG. 1 is a schematic diagram of an implementation framework of the present invention.
FIG. 2 is a DeepLabv3+ network structure model.
FIG. 3 is a flow chart of the detection method of the present invention.
FIG. 4 is a graph showing the comparison of the effects of the conventional detection methods and the detection method of the present invention.
Fig. 5 is a comparison of images of an electrical overhead conductor obtained using different methods in a complex setting.
Detailed Description
An implementation framework of the electric overhead conductor detection of the present invention is shown in fig. 1. Firstly, inputting an original image of an electric power overhead conductor shot by an unmanned aerial vehicle, performing power line semantic segmentation on the original image by using a deep Labv3+ segmentation network, performing binarization processing on a semantic segmentation effect image to obtain a power line semantic segmentation binary image, and simultaneously performing power line extraction on the original image by using an MCMLSD straight line detection algorithm to obtain a power line straight line detection image. And finally, comparing and fusing the semantic segmentation binary image of the power line with the linear detection image of the power line by adopting a cross checking method, and finally detecting to obtain a complete power overhead conductor area.
The specific implementation flow of the invention is shown in figure 3. The steps are described as follows:
step S1: a 512 x 512 image of the power overhead conductor is input, labeled I.
Step S2: carrying out electric overhead conductor detection on the electric overhead conductor image I by using a deep Labv3+ segmentation network to obtain a detected overhead conductor image I 1 。
The structure of the deep labv3+ split network model is shown in fig. 2. The network model is a coding and decoding structure and mainly consists of two parts, namely an encoding region (Encoder) and a decoding region (Decode). In the encoding region (Encoder), a deep convolutional neural network DCNN (Deep Convolution Neural, DCNN) is utilized to perform Feature extraction on an original input image to obtain two different Feature information, then a spatial pyramid pooling module ASPP (Atrous Spatial Pyramid Pooling, ASPP) is utilized to perform multi-scale cavity convolutional sampling on the Feature information of a higher Level extracted by the DCNN, a plurality of Feature layers obtained by cavity convolution with different expansion rates are stacked together, and 1X 1 convolution is performed after combination to form an advanced semantic Feature layer (High-Level Feature).
Extracting DCNN from a decoding area (Decoder) partAnd (3) carrying out 1X 1 convolution on the obtained lower-Level Feature information to form a Low-Level semantic Feature layer (Low-Level Feature), carrying out Feature fusion with the high-Level semantic Feature layer after 4 times up-sampling, and supplementing image position information. The integrated complete semantic feature layer is subjected to feature extraction of 3×3 convolution, and then 4 times of up-sampling is used for adjusting an output image to be the same as an input image in size, so that an overhead conductor image I after detection is obtained 1 。
Step S3: for the detected overhead conductor image I 1 And performing binarization operation. Image binarization (Image Binarization) calculates a global threshold value through a traditional histogram-based gray average value, an iterative optimal threshold value and an OTSU algorithm, and superimposes the global threshold value and background distribution to form a local threshold value, binarizes the global threshold value and the local threshold value to determine the pixel value of each pixel point in the image, wherein values 1 and 0 respectively correspond to pixels identified as an overhead conductor and a non-overhead conductor by a network, and finally a binarized image I is obtained 2 ，I 2 The set of medium image pixels with value 1 is labeled P 0 The set is a straight line identification area.
Step S4: and carrying out electric overhead conductor detection on the electric overhead conductor image I by using an MCMLSD straight line detection algorithm. Firstly, detecting a set of straight line segments in a picture by using global probability Hough transformation on an electric overhead conductor image I, analyzing all detected straight line segments in the image, determining line segment positions corresponding to peaks in Hough domains, thus obtaining a group of line segments to be selected, modeling edge points on each line segment to be selected by using a linear Markov chain model, obtaining a corresponding probability optimal label, carrying out saliency sequencing on all the line segments to be selected according to the probability optimal label, and finally outputting the finally detected line segments according to a sequencing result, and obtaining coordinates of a starting point and an ending point of the straight line segments.
Step S5: discretizing coordinates of a starting point and an ending point of the straight line segment. The starting point coordinates of the marked straight line segment are as follows
Figure BDA0004076213180000071
The coordinates of the ending point of the straight line segment are +.>
Figure BDA0004076213180000072
Calculating the number of pixel point intervals between two points as M; x direction->
Figure BDA0004076213180000073
And->
Figure BDA0004076213180000074
The row index coordinate set with interval 1 is marked as x k Similarly, a column index coordinate set in the y direction is obtained and is marked as y k K=1, 2,3. Through discretization, the coordinate information of the straight line segments is reserved, and meanwhile, the obtained straight line segment set is marked as P i ，P i Discrete coordinate information representing the i-th straight line segment, i=1, 2,3.
Step S6: let i=1, solve straight-line segment set P 1 And P 0 Is the intersection of (a), i.e
Figure BDA0004076213180000081
Step S7: p pair P i Each line segment of the set P is operated according to the step S6 to obtain a straight line segment set P i And P 0 Is set as
Figure BDA0004076213180000082
Step S8: computing a set
Figure BDA0004076213180000083
P in each subset 0 The number of intersections, resulting in a set->
Figure BDA0004076213180000084
i＝1,2,3...n；
Step S9: removal of
Figure BDA0004076213180000085
The subsets of zero in (1) are ordered in descending order, the set at this time being marked +.>
Figure BDA0004076213180000086
Step S10: selecting
Figure BDA0004076213180000087
The first 75% of the subsets are indexed by the index line segment number of the superscript i in the subset, by P in step S5 i Determining coordinate values of starting and ending points of the straight line segments, drawing corresponding straight line segments on the electric overhead conductor image I and marking the corresponding straight line segments;
step S11: outputting the marked image to obtain a final complete electric overhead conductor detection image I 0 。
1. The invention is compared with the effect of different power line detection algorithms
The results are detected by five different straight line detection algorithms as shown in fig. 4. The traditional classical linear detection algorithm Hough linear detection, cannylines linear detection and MCMLSD linear detection are used for detecting the overhead conductor, and detection results show that the MCMLSD linear detection algorithm can directly detect the whole overhead conductor target, the other two linear detection algorithms can not detect the whole target, the MCMLSD algorithm has more advantages for detecting the overhead conductor and reserving the whole conductor target, and the detected result is convenient for implementation of subsequent algorithms. Although the classical linear detection algorithm can detect the overhead conductor target, a large amount of interference exists in the detection process, the conductor target cannot be accurately detected, and the detection accuracy is low. As shown in fig. 4, although the method of deep labv3+ semantic segmentation is used alone to detect the overhead conductor target and reduce the interference target, the detection result of the method is a discontinuous straight line, and the detection target cannot be completely reflected. The invention combines the MCMLSD straight line detection algorithm with the deep Labv3+ segmentation network and performs cross check, the effect is shown in figure 4, the interference in the background is completely removed, the electric overhead conductor can be completely extracted, and the detection and extraction of the electric overhead conductor can be well completed.
2. Application of the invention in complex inspection scene
In order to further verify the detection effectiveness of the method, the method is added with a plurality of complex environments on the basis of detecting the single background of sky background, land background and forest background, so as to verify the detection capability of the method under the complex background. The MCMLSD straight line detection algorithm, the deep Labv3+ semantic segmentation detection algorithm and the invention are adopted for comparison, and the experimental result is shown in figure 5. Experiments prove that the invention can still eliminate interference and completely detect and extract the electric overhead conductor under the complex background of urban roads, factories, forests, grasslands, parks and the like.

Claims (5)

1. The electric overhead conductor detection method based on cross checking is characterized by comprising the following steps:
step S1: inputting a 512×512 power overhead conductor image, labeled I;
step S2: carrying out electric overhead conductor detection on the electric overhead conductor image I by using a deep Labv3+ segmentation network to obtain a detected overhead conductor image I 1 ；
Step S3: for the detected overhead conductor image I 1 Performing binarization operation to obtain a binarized image I 2 Will I 2 The set of medium image pixels with value 1 is labeled P 0 The set is a straight line identification area;
step S4: detecting the electric overhead conductor by using an MCMLSD straight line detection algorithm on the electric overhead conductor image I, outputting a finally detected straight line segment, and obtaining coordinates of a starting point and an ending point of the straight line segment;
step S5: discretizing coordinates of a starting point and an ending point of the straight line segment, reserving coordinate information of the straight line segment through discretization, and marking the obtained straight line segment set as P i ，P i Discrete coordinate information representing the i-th straight line segment, i=1, 2, 3..n;
step S6: let i=1, solve straight-line segment set P 1 And P 0 Is the intersection of (a), i.e
Figure FDA0004076213170000011
i＝1；
Step S7: p pair P i Each line segment is operated according to the step S6 to obtain a straight line segment setP i And P 0 Is set as
Figure FDA0004076213170000012
Step S8: computing a set
Figure FDA0004076213170000013
P in each subset 0 The number of intersections, resulting in a set->
Figure FDA0004076213170000014
Figure FDA0004076213170000015
Step S9: removal of
Figure FDA0004076213170000016
The subsets of zero in (1) are ordered in descending order, the set at this time being marked +.>
Figure FDA0004076213170000017
Step S10: selecting
Figure FDA0004076213170000018
The first 75% of the subsets are indexed by the index line segment number of the superscript i in the subset, by P in step S5 i Determining coordinate values of starting and ending points of the straight line segments, drawing corresponding straight line segments on the electric overhead conductor image I and marking the corresponding straight line segments;
step S11: outputting the marked image to obtain a final complete electric overhead conductor detection image I 0 。
2. The method for detecting an electric overhead conductor based on cross-checking as claimed in claim 1, wherein the step S2 is performed on the electric overhead conductor image I by using deep labv3+ split network to obtain a detected overhead conductor image I 1 The method comprises the steps of carrying out a first treatment on the surface of the The specific process is as follows:
the deep Labv3+ partition network is in a coding and decoding structure and comprises a coding area and a decoding area, in the coding area, the deep convolutional neural network is utilized to conduct feature extraction on an original input image to obtain two different feature information, then a spatial pyramid pooling module is utilized to conduct multi-scale cavity convolution sampling on the feature information of a higher level extracted by the deep convolutional neural network, a plurality of feature layers obtained by cavity convolution with different expansion rates are stacked together, and 1X 1 convolution is conducted after combination, so that a high-level semantic feature layer is formed;
in the decoding area part, carrying out 1X 1 convolution on the lower-level characteristic information extracted from the deep convolutional neural network to form a low-level semantic characteristic layer, carrying out characteristic fusion with the high-level semantic characteristic layer after 4 times up sampling, and supplementing image position information; the integrated complete semantic feature layer is subjected to feature extraction of 3×3 convolution, and then 4 times of up-sampling is used for adjusting an output image to be the same as an input image in size, so that an overhead conductor image I after detection is obtained 1 。
3. The method for detecting an electric overhead conductor based on cross-checking as claimed in claim 2, wherein the detected overhead conductor image I is subjected to step S3 1 Performing binarization operation to obtain a binarized image I 2 The method comprises the steps of carrying out a first treatment on the surface of the The specific process is as follows:
image binarization is carried out by calculating a global threshold value based on a gray average value of a histogram, an iterative optimal threshold value and an OTSU algorithm, superposing the global threshold value and background distribution to form a local threshold value, carrying out binarization processing according to the obtained global threshold value and the local threshold value, determining a pixel value of each pixel point in an image, wherein values 1 and 0 respectively correspond to pixels which are identified as an overhead conductor and a non-overhead conductor by a network, and finally obtaining a binarized image I 2 。
4. The method for detecting the electric overhead conductor based on the cross check as claimed in claim 3, wherein in the step S4, the electric overhead conductor image I is detected by using an MCMLSD straight line detection algorithm, a finally detected straight line segment is output, and coordinates of a starting point and an ending point of the straight line segment are obtained;
firstly, detecting a set of straight line segments in a picture by using global probability Hough transformation on an electric overhead conductor image I, analyzing all detected straight line segments in the image, determining line segment positions corresponding to peaks in Hough domains, thus obtaining a group of line segments to be selected, modeling edge points on each line segment to be selected by using a linear Markov chain model, obtaining a corresponding probability optimal label, carrying out saliency sequencing on all the line segments to be selected according to the probability optimal label, and finally outputting the finally detected straight line segments according to a sequencing result, and obtaining coordinates of a starting point and an ending point of the straight line segments.
5. The cross-check-based power overhead conductor detection method according to claim 4, wherein in step S5, coordinates of a start point and an end point of the straight line segment are discretized; the specific process is as follows:
the starting point coordinates of the marked straight line segment are (x) i 1 ,y i 1 ) The straight line segment end point coordinates are (x i 2 ,y i 2 ) Calculating the number of pixel point intervals between two points as M; in the x direction
Figure FDA0004076213170000031
And->
Figure FDA0004076213170000032
The row index coordinate set with interval 1 is marked as x k Similarly, a column index coordinate set in the y direction is obtained and is marked as y k ，k＝1,2,3…M。
CN202310109206.7A 2023-02-14 2023-02-14 Electric overhead conductor detection method based on cross checking Pending CN116310316A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310109206.7A CN116310316A (en) 2023-02-14 2023-02-14 Electric overhead conductor detection method based on cross checking

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310109206.7A CN116310316A (en) 2023-02-14 2023-02-14 Electric overhead conductor detection method based on cross checking

Publications (1)

Publication Number Publication Date
CN116310316A true CN116310316A (en) 2023-06-23

Family

ID=86786004

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310109206.7A Pending CN116310316A (en) 2023-02-14 2023-02-14 Electric overhead conductor detection method based on cross checking

Country Status (1)

Country Link
CN (1) CN116310316A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117690063A (en) * 2024-02-04 2024-03-12 广东电网有限责任公司广州供电局 Cable line detection method, device, electronic equipment and computer readable medium

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117690063A (en) * 2024-02-04 2024-03-12 广东电网有限责任公司广州供电局 Cable line detection method, device, electronic equipment and computer readable medium
CN117690063B (en) * 2024-02-04 2024-04-12 广东电网有限责任公司广州供电局 Cable line detection method, device, electronic equipment and computer readable medium

Similar Documents

Publication Publication Date Title
CN111626146B (en) Merging cell table segmentation recognition method based on template matching
CN108830171B (en) Intelligent logistics warehouse guide line visual detection method based on deep learning
Piccioli et al. Robust road sign detection and recognition from image sequences
CN110751154B (en) Complex environment multi-shape text detection method based on pixel-level segmentation
CN113436157B (en) Vehicle-mounted image identification method for pantograph fault
CN111460927B (en) Method for extracting structured information of house property evidence image
CN108509950B (en) Railway contact net support number plate detection and identification method based on probability feature weighted fusion
CN111951284B (en) Optical remote sensing satellite image refined cloud detection method based on deep learning
CN112926556B (en) Semantic segmentation-based aerial photography transmission line broken strand identification method and system
CN109815802A (en) A kind of monitor video vehicle detection and recognition method based on convolutional neural networks
CN111046950A (en) Image processing method and device, storage medium and electronic device
CN116310316A (en) Electric overhead conductor detection method based on cross checking
CN113435319A (en) Classification method combining multi-target tracking and pedestrian angle identification
CN113240623A (en) Pavement disease detection method and device
Hu Intelligent road sign inventory (IRSI) with image recognition and attribute computation from video log
CN115457017A (en) Wire defect detection method and device, computer equipment and storage medium
CN115841669A (en) Pointer instrument detection and reading identification method based on deep learning technology
Fang et al. Towards real-time crack detection using a deep neural network with a Bayesian fusion algorithm
CN113989604A (en) Tire DOT information identification method based on end-to-end deep learning
CN113326846A (en) Rapid bridge apparent disease detection method based on machine vision
CN116843829A (en) Concrete structure crack three-dimensional reconstruction and length quantization method based on binocular video
CN109325487B (en) Full-category license plate recognition method based on target detection
Zhu et al. A minimum cover approach for extracting the road network from airborne LIDAR data
CN111931689B (en) Method for extracting video satellite data identification features on line
Hossen et al. License plate detection and recognition system based on morphological approach and feed-forward neural network

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