CN114418968A - Power transmission line small target defect detection method based on deep learning - Google Patents

Power transmission line small target defect detection method based on deep learning Download PDF

Info

Publication number
CN114418968A
CN114418968A CN202111645927.7A CN202111645927A CN114418968A CN 114418968 A CN114418968 A CN 114418968A CN 202111645927 A CN202111645927 A CN 202111645927A CN 114418968 A CN114418968 A CN 114418968A
Authority
CN
China
Prior art keywords
picture data
defect
small target
transmission line
deep learning
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
CN202111645927.7A
Other languages
Chinese (zh)
Inventor
王江
韦基毅
韦屹健
覃明生
毛云申
韦维
兰建蒙
曾令争
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.)
Hechi Power Supply Bureau of Guangxi Power Grid Co Ltd
Original Assignee
Hechi Power Supply Bureau of Guangxi Power Grid Co Ltd
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 Hechi Power Supply Bureau of Guangxi Power Grid Co Ltd filed Critical Hechi Power Supply Bureau of Guangxi Power Grid Co Ltd
Priority to CN202111645927.7A priority Critical patent/CN114418968A/en
Publication of CN114418968A publication Critical patent/CN114418968A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0004—Industrial image inspection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • 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
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/12—Edge-based segmentation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/13—Edge detection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computing Systems (AREA)
  • Molecular Biology (AREA)
  • Probability & Statistics with Applications (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Quality & Reliability (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a method for detecting defects of small targets of a power transmission line based on deep learning, which comprises the steps of acquiring initial picture data of the power transmission line acquired by a front-end camera; processing the initial picture data to obtain picture data with characteristic directions; carrying out image identification and image feature extraction on the image data with the feature direction, and screening out defective image data; performing corresponding small target defect classification on the screened defect picture data to obtain defect picture data corresponding to different small target defects; establishing a small target defect detection model based on deep learning according to defect picture data corresponding to different small target defects; and detecting the acquired picture data by using a small target defect detection model based on deep learning to obtain a corresponding small target defect result. The method can accurately identify and detect the defects of the small targets of the power transmission line, does not need manual operation at high altitude, improves the detection efficiency, and has higher safety.

Description

Power transmission line small target defect detection method based on deep learning
Technical Field
The invention relates to the field of power transmission line defect detection, in particular to a power transmission line small target defect detection method based on deep learning.
Background
With the vigorous development of the electric power industry in China, the layout of the transmission line is complex, the transmission line is divided into an overhead transmission line and a cable line, the overhead transmission line is composed of a line tower, a lead, a line hardware fitting, an insulator, a stay wire, a grounding device and the like, the distribution is wide, and the transmission line is distributed in various terrains such as fields, urban areas, deserts, lakes and the like. The natural environment and climate where the transmission line is located are changeable, and the transmission line runs outdoors for a long time, and is subjected to extreme weather impact such as violent storm and solarization, so that the transmission line is inevitably damaged by various artificial or non-artificial factors, and parts such as wires, hardware fittings, insulators and the like are easy to rust, damage, strand breakage and the like. Meanwhile, the irregular installation of the components brings hidden troubles to the safe operation of the power transmission line. Regular inspection and maintenance of the transmission line is required.
The small target samples in the power transmission line comprise line hardware, insulators, stay wires, grounding devices and the like, the traditional small target sample inspection mainly depends on a manual inspection mode or an RCNN algorithm mode, the former inspection mode mainly depends on that a professional goes to an inspection site to perform manual climbing work under a high-voltage environment, the manual inspection not only influences the working efficiency of workers, but also easily causes inaccurate detection results, and has great safety problems; the latter detection mode easily causes image distortion, occupies a large memory, and has low detection speed and inaccurate detection precision.
Disclosure of Invention
The invention aims to provide a method for detecting defects of small targets of a power transmission line based on deep learning, which can solve the problems of low detection speed and inaccurate detection precision caused by image distortion easily caused by a detection mode in the prior art.
The purpose of the invention is realized by the following technical scheme:
the invention provides a power transmission line small target defect detection method based on deep learning, which comprises the following steps of:
step S1, acquiring initial picture data of the power transmission line acquired by the front-end camera;
step S2, processing the initial picture data to obtain picture data with characteristic directions;
step S3, carrying out image recognition and image feature extraction on the picture data with the feature direction, and screening out defective picture data;
step S4, corresponding small target defect classification is carried out on the screened defect picture data to obtain defect picture data corresponding to different small target defects;
step S5, establishing a small target defect detection model based on deep learning according to defect picture data corresponding to different small target defects;
and step S6, detecting the acquired picture data by using a small target defect detection model based on deep learning to obtain a corresponding small target defect result.
Further, the processing the initial picture data to obtain the picture data with the characteristic direction specifically includes:
step S201, obtaining the resolution of each picture in the initial picture data to obtain the picture data with the same resolution;
step S202, extracting the outline of the picture data with the same resolution ratio to obtain an outline curve of the picture data;
step S203, performing derivation on the profile curve to obtain the curvature of the profile curve;
step S204, taking the central point as a characteristic point of the profile curve, and obtaining a direction vector of the characteristic point pointing to the initial end of the profile curve and a direction vector of the characteristic point pointing to the tail end of the profile curve;
and S205, summing the direction vector of the characteristic point pointing to the initial end of the outline curve and the direction vector of the characteristic point pointing to the tail end of the outline curve to obtain the image data with the characteristic direction.
Further, the image recognition and image feature extraction of the image data with the feature direction, and screening out the defective image data includes:
and carrying out image recognition on the image data with the characteristic direction by a gray scale method, carrying out image characteristic extraction on the image data after the image recognition, and screening out the defective image data according to the extracted image characteristics.
Further, the initial picture data includes defective picture data and normal picture data.
The invention has the beneficial effects that:
according to the invention, the picture data with the characteristic direction is processed for the obtained picture data of the power transmission line, and the small target defect detection model is established according to the processed picture data, so that the defect identification detection for the small target defect of the power transmission line is realized, and the detection accuracy is high.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.
FIG. 1 is a schematic step diagram of an embodiment of a method for detecting defects of a small target of a power transmission line based on deep learning;
FIG. 2 is a schematic step diagram of another embodiment of a method for detecting defects of small targets of a power transmission line based on deep learning;
fig. 3 is a schematic diagram of an RCNN model algorithm for defect detection in the prior art.
Detailed Description
The embodiments of the present disclosure are described in detail below with reference to the accompanying drawings.
The embodiments of the present disclosure are described below with specific examples, and other advantages and effects of the present disclosure will be readily apparent to those skilled in the art from the disclosure in the specification. It is to be understood that the described embodiments are merely illustrative of some, and not restrictive, of the embodiments of the disclosure. The disclosure may be embodied or carried out in various other specific embodiments, and various modifications and changes may be made in the details within the description without departing from the spirit of the disclosure. It is to be noted that the features in the following embodiments and examples may be combined with each other without conflict. All other embodiments, which can be derived by a person skilled in the art from the embodiments disclosed herein without making any creative effort, shall fall within the protection scope of the present disclosure.
The invention provides a power transmission line small target defect detection method based on deep learning, which comprises the following steps of:
step S1, acquiring initial picture data of the power transmission line acquired by the front-end camera;
the front-end camera can be a camera installed on the unmanned aerial vehicle, and picture data shooting of automatic inspection of the power transmission line is completed by the aid of the camera on the unmanned aerial vehicle. The front-end camera does not limit the camera installed on the unmanned aerial vehicle, and can be other cameras capable of acquiring pictures of the power transmission line, and is not particularly limited herein.
Wherein the initial picture data includes defective picture data and normal picture data.
Step S2, processing the initial picture data to obtain picture data with characteristic directions;
step S3, carrying out image recognition and image feature extraction on the picture data with the feature direction, and screening out defective picture data;
step S4, corresponding small target defect classification is carried out on the screened defect picture data to obtain defect picture data corresponding to different small target defects;
small target defect categories include hardware, insulators, guy wires, grounding devices, etc.
Step S5, establishing a small target defect detection model based on deep learning according to defect picture data corresponding to different small target defects;
the defect picture data needs more than 20 pieces, and the image and the labeling information of the wires, the fittings, the insulators and the like are stored in the xml file of the defect picture data, and comprise the image name, the image size, the target category and the target rectangular frame size.
And step S6, detecting the acquired picture data by using a small target defect detection model based on deep learning to obtain a corresponding small target defect result.
Further, in a preferred embodiment of the present application, the processing the initial picture data to obtain the picture data with the characteristic direction specifically includes:
step S201, obtaining the resolution of each picture in the initial picture data to obtain the picture data with the same resolution;
step S202, extracting the outline of the picture data with the same resolution ratio to obtain an outline curve of the picture data;
the profile curve (X, Y) is formulated as
Figure BDA0003445133590000051
Wherein the content of the first and second substances,
Figure BDA0003445133590000052
is the point at which the surface changes over time,
Figure BDA0003445133590000053
is an angle that varies with time;
step S203, performing derivation on the profile curve to obtain the curvature of the profile curve;
step S204, taking the central point as a characteristic point of the profile curve, and obtaining a direction vector of the characteristic point pointing to the initial end of the profile curve and a direction vector of the characteristic point pointing to the tail end of the profile curve;
and S205, summing the direction vector of the characteristic point pointing to the initial end of the outline curve and the direction vector of the characteristic point pointing to the tail end of the outline curve to obtain the image data with the characteristic direction.
The image identification and image feature extraction are carried out on the image data with the feature direction, and the defect image data are screened out by the method comprising the following steps:
carrying out image recognition on the picture data with the characteristic direction by a gray scale method, wherein the image recognition c (x, y) formula is as follows:
Figure BDA0003445133590000061
wherein:
(x, y) indicating the position of a pixel of the picture data having the characteristic direction;
p represents the gray scale of the picture data with the characteristic direction;
i is the total amount of pixel points of the picture data with the characteristic direction;
g (x, y) represents the gray coefficient of the pixel point of the picture data with the characteristic direction;
then, after image recognition, image feature extraction is carried out, wherein the formula of the image feature extraction p (x, y) is as follows:
Figure BDA0003445133590000062
wherein:
q (x, y) is grayscale image data;
j is a gray average value corresponding to the gray image, p (x, y) is extracted according to the image characteristics, and the defect picture data are screened out.
It should be noted that, image recognition is performed on the picture data with the characteristic direction by using a grayscale method, the image definition of the picture data with the characteristic direction needs to be ensured, then feature extraction is performed on the image, and feature extraction is performed on the images of the same power transmission line at different angles at the same time.
Further, in a preferred embodiment of the present application, the establishing a small target defect detection model based on deep learning includes:
establishing a loss function of fast RCNN of a small target defect detection model, wherein the loss function of the fast RCNN is as follows:
Figure BDA0003445133590000063
wherein:
l1 is the classification loss function;
l2 is a regression loss function;
aito represent the i-th anchor point area as a target;
ai1classifying probability for the small target defect sample of the power transmission line;
xirepresenting a parameterized feature vector of an ith anchor point region in a small target defect sample of the power transmission line;
xi1a parameterized feature vector representing a true target region;
lambda is a weight parameter;
n1, N2 represent the normalized parameters of the loss function;
loss function of FasterRCNN:
L1(ai,ai1)＝-log[aiai1+(1-ai)(1-ai1)]，
Figure BDA0003445133590000071
smooth is a regression process robustness function.
Performing target probability calculation on the defect picture data through an RPN (regional candidate network) network, and generating a candidate frame;
sharing a convolution layer by the RPN and the fast RCNN detection model, and randomly initializing the parameter weight of a non-shared convolution network in the RPN by utilizing Gaussian distribution with standard deviation of 0.01 and mean value of 0;
and training the RPN by using the training samples, and replacing the candidate frame region calculated by the RPN with the fast RCNN parameter trained by the candidate frame in the fast RCNN detection model.
Initializing an RPN (resilient packet network) by using network parameters of a fast RCNN (radar cross-correlation network) detection model, and training parameters of a non-shared convolutional network in the RPN;
converting the candidate frame region generated by the RPN into a feature vector with fixed dimension through a pooling layer;
and judging the object of the candidate frame area by using a Softmax activation function, identifying the object, and marking the candidate area of the defect picture data.
The RCNN model algorithm in the prior art mainly includes four parts, as follows:
generation of candidate regions: generating a plurality of candidate regions for the input image data, wherein the candidate regions can be mutually overlapped and mutually contained, and fewer windows are selected by utilizing information such as textures, colors, edges and the like in the image;
feature extraction: for each candidate region, extracting a feature vector by using a CNN network;
and (4) classification: performing target classification on the features extracted by the network by using a linear Support Vector Machine (SVM);
boundary regression: the position of the candidate frame is corrected using a regressor to generate a bounding box of the identified object that is more rigid.
Compared with the RCNN model algorithm in the prior art, the fast RCNN model algorithm in the electric transmission line small target defect model based on deep learning is as follows:
the FasterRCNN algorithm can further improve the speed of target detection, and the main improvement is that a region candidate network (RPN) is used for replacing a traditional region search method to select a candidate frame, so that the operation speed is greatly improved on the premise of not sacrificing the detection precision. The RPN network divides the feature map extracted for the convolutional layer into different regions using sliding windows, each generating a feature vector by fully connecting and taking ReLU as an activation function. The feature vector is connected to two fully connected layers, generating a 4 k-dimensional vector for regression of the bounding box and a 2 k-dimensional vector for classification. A reference box, also called an Anchor box (Anchor), of different size and scale is then used in each region. In the Fast RCNN algorithm, the RPN network and the Fast RCNN detection module share a convolutional layer. Through experimental data, the defect detection efficiency of the Faster RCNN algorithm generally reaches 98%, the defect detection efficiency of the RCNN algorithm is about 75%, and the defect detection efficiency of the Faster RCNN algorithm is much higher than that of the RCNN algorithm.
It is clear to those skilled in the art that, for convenience and brevity of description, the specific working processes of the above-described systems, apparatuses and units may refer to the corresponding processes in the foregoing method embodiments, and are not described herein again.
The above description is for the purpose of illustrating embodiments of the invention and is not intended to limit the invention, and it will be apparent to those skilled in the art that any modification, equivalent replacement, or improvement made without departing from the spirit and principle of the invention shall fall within the protection scope of the invention.

Claims (4)

1. A method for detecting defects of small targets of a power transmission line based on deep learning is characterized by comprising the following steps:
step S1, acquiring initial picture data of the power transmission line acquired by the front-end camera;
step S2, processing the initial picture data to obtain picture data with characteristic directions;
step S3, carrying out image recognition and image feature extraction on the picture data with the feature direction, and screening out defective picture data;
step S4, corresponding small target defect classification is carried out on the screened defect picture data to obtain defect picture data corresponding to different small target defects;
step S5, establishing a small target defect detection model based on deep learning according to defect picture data corresponding to different small target defects;
and step S6, detecting the acquired picture data by using a small target defect detection model based on deep learning to obtain a corresponding small target defect result.
2. The method for detecting the defect of the small target of the power transmission line based on the deep learning of claim 1, wherein the processing the initial picture data to obtain the picture data with the characteristic direction specifically comprises:
step S201, obtaining the resolution of each picture in the initial picture data to obtain the picture data with the same resolution;
step S202, extracting the outline of the picture data with the same resolution ratio to obtain an outline curve of the picture data;
step S203, performing derivation on the profile curve to obtain the curvature of the profile curve;
step S204, taking the central point as a characteristic point of the profile curve, and obtaining a direction vector of the characteristic point pointing to the initial end of the profile curve and a direction vector of the characteristic point pointing to the tail end of the profile curve;
and S205, summing the direction vector of the characteristic point pointing to the initial end of the outline curve and the direction vector of the characteristic point pointing to the tail end of the outline curve to obtain the image data with the characteristic direction.
3. The method for detecting the defect of the small target of the power transmission line based on the deep learning of claim 1, wherein the step of carrying out image recognition and image feature extraction on the picture data with the feature direction and screening out the defect picture data comprises the following steps:
and carrying out image recognition on the image data with the characteristic direction by a gray scale method, carrying out image characteristic extraction on the image data after the image recognition, and screening out the defective image data according to the extracted image characteristics.
4. The method for detecting the defect of the small target of the power transmission line based on the deep learning of claim 1 or 2, wherein the initial picture data comprises defect picture data and normal picture data.
CN202111645927.7A 2021-12-30 2021-12-30 Power transmission line small target defect detection method based on deep learning Pending CN114418968A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111645927.7A CN114418968A (en) 2021-12-30 2021-12-30 Power transmission line small target defect detection method based on deep learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111645927.7A CN114418968A (en) 2021-12-30 2021-12-30 Power transmission line small target defect detection method based on deep learning

Publications (1)

Publication Number Publication Date
CN114418968A true CN114418968A (en) 2022-04-29

Family

ID=81269644

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111645927.7A Pending CN114418968A (en) 2021-12-30 2021-12-30 Power transmission line small target defect detection method based on deep learning

Country Status (1)

Country Link
CN (1) CN114418968A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116452513A (en) * 2023-03-20 2023-07-18 山东未来智能技术有限公司 Automatic identification method for corrugated aluminum sheath defects of submarine cable

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116452513A (en) * 2023-03-20 2023-07-18 山东未来智能技术有限公司 Automatic identification method for corrugated aluminum sheath defects of submarine cable
CN116452513B (en) * 2023-03-20 2023-11-21 山东未来智能技术有限公司 Automatic identification method for corrugated aluminum sheath defects of submarine cable

Similar Documents

Publication Publication Date Title
CN109118479B (en) Capsule network-based insulator defect identification and positioning device and method
CN111460995B (en) Unmanned aerial vehicle-based power line inspection method and inspection system
CN108734689B (en) Method for detecting scattered strands of conducting wires based on region growth
CN111046950B (en) Image processing method and device, storage medium and electronic device
CN110610483B (en) Crack image acquisition and detection method, computer equipment and readable storage medium
CN113592822B (en) Insulator defect positioning method for electric power inspection image
CN113436169A (en) Industrial equipment surface crack detection method and system based on semi-supervised semantic segmentation
CN111914720B (en) Method and device for identifying insulator burst of power transmission line
CN112101138A (en) Bridge inhaul cable surface defect real-time identification system and method based on deep learning
CN110705397A (en) YOLOv3 pruning identification insulator defect method suitable for small field sample amount
CN113297915A (en) Insulator recognition target detection method based on unmanned aerial vehicle inspection
CN115588021A (en) Ceramic insulator and hardware unmanned aerial vehicle infrared inspection image fusion and segmentation method
CN115147383A (en) Insulator state rapid detection method based on lightweight YOLOv5 model
CN116612098A (en) Insulator RTV spraying quality evaluation method and device based on image processing
CN111667473A (en) Insulator hydrophobicity grade judging method based on improved Canny algorithm
CN114418968A (en) Power transmission line small target defect detection method based on deep learning
CN114266963A (en) Method and device for identifying winding gap of wire and storage medium
CN113326734A (en) Rotary target detection method based on YOLOv5
CN112991432A (en) Icing shape identification method based on image processing
CN111931577A (en) Intelligent inspection method for specific foreign matters of power grid line
CN109389152B (en) Refined identification method for power transmission line falling object
CN116309407A (en) Method for detecting abnormal state of railway contact net bolt
CN116385950A (en) Electric power line hidden danger target detection method under small sample condition
CN114037840A (en) Power transmission line visual object extraction method and system based on multi-feature fusion
CN113205487B (en) Cable state detection method based on residual error network fusion heterogeneous data

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