CN114170144A - Power transmission line pin defect detection method, equipment and medium - Google Patents

Power transmission line pin defect detection method, equipment and medium Download PDF

Info

Publication number
CN114170144A
CN114170144A CN202111334727.XA CN202111334727A CN114170144A CN 114170144 A CN114170144 A CN 114170144A CN 202111334727 A CN202111334727 A CN 202111334727A CN 114170144 A CN114170144 A CN 114170144A
Authority
CN
China
Prior art keywords
pin
image
defect
detection model
feature
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
CN202111334727.XA
Other languages
Chinese (zh)
Inventor
陈杰
朱兴红
孙嫱
沈滨
汤奕琛
赵凌杰
沈如榕
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.)
State Grid Fujian Electric Power Co Ltd
Zhangzhou Power Supply Co of State Grid Fujian Electric Power Co Ltd
Original Assignee
State Grid Fujian Electric Power Co Ltd
Zhangzhou Power Supply Co of State Grid Fujian Electric Power 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 State Grid Fujian Electric Power Co Ltd, Zhangzhou Power Supply Co of State Grid Fujian Electric Power Co Ltd filed Critical State Grid Fujian Electric Power Co Ltd
Priority to CN202111334727.XA priority Critical patent/CN114170144A/en
Publication of CN114170144A publication Critical patent/CN114170144A/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
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • G06F18/253—Fusion techniques of extracted features
    • 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
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/047—Probabilistic or stochastic 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
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/048—Activation functions
    • 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
    • G06T5/00—Image enhancement or restoration
    • G06T5/50—Image enhancement or restoration using two or more images, e.g. averaging or subtraction
    • 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
    • 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
    • G06T2207/20084—Artificial neural networks [ANN]
    • 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
    • G06T2207/20112—Image segmentation details
    • G06T2207/20132—Image cropping
    • 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
    • G06T2207/20212—Image combination
    • G06T2207/20221—Image fusion; Image merging

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • Computing Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Computer Vision & Pattern Recognition (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 relates to a pin defect detection method for a power transmission line, which comprises the following steps: collecting and preprocessing an electric transmission line inspection image containing a pin to obtain a pin image data set; constructing a small target detection model; training the small target detection model by using the pin image data set to obtain a pin detection model; cutting out a target frame subgraph containing the pin from the image in the pin image dataset; preprocessing the target frame image, and adding a defect label for the pin in the image to obtain a pin defect data set; constructing a classification model, and training the classification model by using the pin defect data set to obtain a pin defect detection model; inputting an image to be detected, detecting pins in the image through a pin detection model, and outputting a target frame; and cutting according to the target frame to obtain a target frame subgraph, inputting the target frame subgraph into a pin defect detection model to detect whether defects exist, and outputting a result.

Description

Power transmission line pin defect detection method, equipment and medium
Technical Field
The invention relates to a pin defect detection method, equipment and medium for a power transmission line, and belongs to the technical field of small target detection and power transmission line inspection.
Background
For traditional pin detection, the mode of climbing through the manual work is the main means, and is time-consuming and hard, but because the pin has the characteristics of very extensive distribution and changeable specification and style in transmission line, makes its inspection become more difficult. Along with the development of unmanned aerial vehicle technique, the characteristics of its volume less and high security make it by a large amount of applications in electric power patrols and examines, very big improvement patrol and examine efficiency. However, in the case of a large number of pins and a small size, the inspection personnel can inspect the pins shown in the inspection image of the unmanned aerial vehicle one by one only by means of amplification and pulling, which may cause extremely high workload and missed inspection rate. And the visual difference between the defective pins and the normal pins in the aerial image is extremely small, and the defect feature classification of the defective pins in the aerial image is more challenging due to the irregularity of the defect features.
At present, the pin defect in the unmanned aerial vehicle transmission line inspection image is mainly detected by adopting a deep learning algorithm. The detection of the defects of the pins is carried out by adopting a Faster-RCNN algorithm, which mainly comprises 3 parts: a feature extraction network CNN, a target detection network RCNN and a regional recommendation network (RPN). Firstly, the feature extraction network part uses VGG16 as a network frame to extract shallow features from an input picture, and outputs an obtained feature graph as the input of a regional suggestion network RPN; then, traversing convolution is carried out on the feature map in an RPN by adopting a sliding window, the feature map is mapped into a low-dimensional vector, the low-dimensional vector is respectively sent into a classification layer and a regression positioning layer, and a suggestion frame which possibly comprises a target object is output; the target detection network RCNN combines a characteristic diagram output by the CNN and a region suggestion frame output by the RPN, firstly obtains a region of interest (ROI) by using a non-maximum suppression algorithm, and then downsamples the ROI to a certain fixed size through an ROI pooling layer. And finally, sending the low-dimensional characteristic vector of the ROI into a SoftMax classifier, and further adjusting the confidence coefficient of the target classification and the position of the positioning rectangular frame. The algorithm has a good detection effect on the target object with a large area and obvious characteristics. The detection precision of small targets such as pins is low, and the missing detection rate and the false detection rate are high. Therefore, how to accurately position the pins in the massive images and identify whether the pins are abnormal is an urgent problem to be solved aiming at the characteristics that the missing detection rate is high due to the fact that the pin image background is large and small, and the false detection rate is high due to the fact that the defect characteristics are not obvious.
Disclosure of Invention
In order to solve the problems in the prior art, the invention provides a pin defect detection method for a power transmission line, which is characterized in that a small target detection model and a classification model are established according to the characteristic of small pin size, the pin position is framed by the small target detection model and is cut, and then a sub-image of the cut target frame is input into the classification model for pin defect detection, so that the problem of high pin defect identification false detection rate in a power transmission line inspection image is solved.
The technical scheme of the invention is as follows:
on one hand, the invention provides a method for detecting the pin defect of the power transmission line, which comprises the following steps:
manufacturing a pin image dataset; collecting and preprocessing an electric transmission line inspection image containing a pin to obtain a pin image data set;
constructing a small target detection model based on a neural network; training the small target detection model by using the pin image data set to obtain a pin detection model;
making a defect data set, and cutting out a target frame subgraph containing a pin from an image in the pin image data set; preprocessing the target frame image, and adding a defect label for the pin in the image to obtain a pin defect data set;
constructing a classification model based on a neural network, and training the classification model by using the pin defect data set to obtain a pin defect detection model;
inputting an image to be detected, detecting a pin in the image through a pin detection model, and outputting position information of a target frame of which the pin is detected; and cutting the image according to the position information of the target frame to obtain a target frame subgraph, inputting the target frame subgraph to a pin defect detection model to detect whether defects exist, and outputting a result.
As a preferred embodiment, the small target detection model includes two parts, wherein:
the first part comprises a VGG16 feature extraction network, and an FM focusing module is embedded in the VGG16 feature extraction network; extracting a feature matrix of the input image through a VGG16 feature extraction network; the FM focusing module firstly performs a global average pooling operation on the feature matrix to obtain a feature matrix of 1 multiplied by Channel, then performs a feature interaction process, firstly compresses the number of channels, then reconstructs the original channels, and finally generates an attention weight of 0-1 among the channels through a sigmoid function and then multiplies the attention weight by the original input feature matrix; dividing the image into sub-modules with a plurality of sizes, and outputting the sub-modules to a second part in batches;
the second part comprises a ResNet101 network, an RPN layer and an FC classification layer; the ResNet101 network comprises four residual blocks of RS 1-RS 4 which are sequentially connected, wherein each residual block comprises convolution, pooling and ReLU activation function operation, and a feature matrix of input data is extracted; the feature matrix extracted by the ResNet101 network is output to the RPN layer, and the RPN layer outputs the region of interest RoI to the FC classification layer according to the input feature matrix; and the FC classification layer comprises a full connection layer, a Softmax function and a position regression function, corrects the position of the region of interest RoI and outputs a target frame.
In a preferred embodiment, a feature strengthening operation is embedded in the RS3 module; the feature enhancement operation adopts a bilinear interpolation method to improve the resolution of the features, and specifically comprises the following steps:
let F (w)i,hj) The feature value of any point on the feature map is the F (w) of any positioni+u,hj+ v) is calculated as:
Figure BDA0003350167630000041
and in the ResNet101 network, a multi-scale feature fusion method is used for fusing the feature matrix output by the RS2 module and the enhanced feature matrix output by the RS3 module to obtain a fused feature matrix, and then outputting the fused feature matrix to the RPN layer.
In a preferred embodiment, the RPN layer sets the following loss function metric to measure the deviation between the predicted target frame and the real frame in the learning process:
Figure BDA0003350167630000042
where i is the index number of the suggestion box, diIs the predicted probability of the box i being proposed as the target,
Figure BDA0003350167630000043
is a label of the real frame uiIs a vector of four parameterized coordinates representing the predicted target box,
Figure BDA0003350167630000044
is and uiAssociated real box, NregRepresenting the number of anchor positions, NclsIs a constant corresponding to the training image, LclsAnd LregFrom NclsAnd NregNormalization yields that λ represents a balance weight, L ({ d })i},{ui}) is the final loss obtained after classification regression.
In a preferred embodiment, the pin defect detection model adopts an EfficientNet-B0 network, which comprises 16 MBConv layers, 2 convolutional layers, 1 global average pooling layer and 1 FC classification layer;
the pin defect detection model trains a network by adopting a transfer learning method.
In another aspect, the present invention further provides a device for detecting pin defects of a power transmission line, including a memory, a processor, and a computer program stored in the memory and executable on the processor, where the processor executes the computer program to implement the detection method according to any embodiment of the present invention.
In yet another aspect, the present invention further provides a computer-readable storage medium, on which a computer program is stored, which when executed by a processor, implements the detection method according to any one of the embodiments of the present invention.
The invention has the following beneficial effects:
1. the invention relates to a pin defect detection method for a power transmission line, which is characterized in that a small target detection model and a classification model are established according to the characteristic that the pin size is small, the pin position is framed by the small target detection model and is cut, and then a cut target frame sub-image is input into the classification model for pin defect detection, so that the problem of high pin defect identification false detection rate in a power transmission line inspection image is solved.
2. The invention discloses a pin defect detection method for a power transmission line, and provides a small target detection model based on a mixed structure of VGG16+ ResNet101 as a feature extraction network, and an FM focusing module is added in a first part VGG16 of the feature extraction network aiming at a small target, so that the feature extraction capability of the small target is enhanced.
3. The invention relates to a pin defect detection method for a power transmission line, which embeds characteristic strengthening operation in one module in a ResNet101 network to improve the resolution of characteristics; and a multi-scale feature fusion method is adopted to fuse the enhanced features and other features, so that the abstraction of high-level semantic features is kept, and the noise redundancy of low-level features is inhibited to a certain extent.
Drawings
FIG. 1 is a flow chart of a first embodiment of the present invention;
FIG. 2 is a diagram illustrating an example of training of a pin detection model according to an embodiment of the present invention;
fig. 3 is a schematic training diagram of a pin defect detection model in the embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
It should be understood that the step numbers used herein are for convenience of description only and are not intended as limitations on the order in which the steps are performed.
It is to be understood that the terminology used in the description of the invention herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. As used in the specification of the present invention and the appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise.
The terms "comprises" and "comprising" indicate the presence of the described features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof.
The term "and/or" refers to and includes any and all possible combinations of one or more of the associated listed items.
The first embodiment is as follows:
referring to fig. 1, a method for detecting a pin defect of a power transmission line includes the following steps:
manufacturing a pin image dataset; the method comprises the steps that an unmanned aerial vehicle is used for power transmission line inspection, power transmission line inspection images containing pins are collected from inspection images, preprocessing and data enhancement operations are carried out, wherein the preprocessing comprises the steps of removing trembling, preventing noise, implementing rotation, cutting, stretching and the like, and scene enhancement is carried out on images of different scenes; obtaining a pin image data set after processing; and dividing a training set, a verification set and a test set, manually marking the training set and the verification set by using labelImg, and obtaining an XML file corresponding to the file name, the width and the height information of the image and the coordinate information of the upper left point and the lower right point of the pin target frame after marking.
Constructing a VFM-ResEPNet small target detection model based on a neural network; and training the VFM-ResEPNet small target detection model by utilizing the training set to the pin image data set, continuously verifying the model performance through the verification set in the training process, and solidifying the model to obtain the pin detection model when the expected value is reached according to the change condition of each network parameter.
Making a defect data set, and cutting out a target frame subgraph containing a pin from an image in the pin image data set; and (3) carrying out data preprocessing and image enhancement on the target frame image, dividing a training set, a verification set and a test set, and marking the class labels only when in manual marking, wherein the number of normal pin images is equal to that of pin-lacking images, so as to keep the balance of the sample and obtain a pin-lacking data set.
Constructing a classification model based on a neural network, and training the classification model by using the pin defect data set to obtain a pin defect detection model;
inputting an image to be detected, detecting pins in the image through a pin detection model, and obtaining an output result of a rectangular target frame with target position information after detection is finished; and cutting the image according to the position information of the target frame to obtain a target frame subgraph, inputting the target frame subgraph into a pin defect detection model to detect whether defects exist, outputting a classification result of the image, wherein the result comprises an abnormal type and a normal type, and when the abnormal result is output, outputting the abnormal type, namely the defect type of the pin.
Referring to fig. 2 in detail, as a preferred implementation manner of this embodiment, the VFM-respnet small target detection model includes two parts, where:
the first part comprises a VGG16 feature extraction network, and an FM focusing module is embedded in the VGG16 feature extraction network; extracting a feature matrix of the input image through a VGG16 feature extraction network; the FM focusing module is a focusing module aiming at small target features and can enhance the feature extraction capability of the small target, firstly, a global average pooling operation is carried out on an input feature matrix to obtain a feature matrix of 1 multiplied by Channel, then, a feature interaction process is carried out, the number of channels is firstly compressed and then reconstructed back to the original Channel, and finally, the attention weight of 0-1 among the channels is generated through a sigmoid function and then multiplied back to the original input feature matrix; after the image is subjected to (VGG16+ FM) feature extraction, the resolution is changed into 900 × 700, the image is divided into a plurality of sub-modules with the size of 128 × 128, and after the sub-modules are subjected to convolution and pooling, the sub-modules are output to the second part in batches;
the second part comprises a ResNet101 network, an RPN layer and an FC classification layer; the ResNet101 network comprises four residual blocks of RS 1-RS 4 which are sequentially connected, wherein each residual block comprises convolution, pooling and ReLU activation function operation, and a feature matrix of input data is extracted; the feature matrix extracted by the ResNet101 network is output to the RPN layer, and the RPN layer outputs the region of interest RoI to the FC classification layer according to the input feature matrix; and the FC classification layer comprises a full connection layer, a Softmax function and a position regression function, corrects the position of the region of interest RoI and outputs a target frame.
As a preferred implementation of this embodiment, an Enhance operation, that is, a feature enhancement method, is embedded in the RS3 module; in order to prevent the high-level semantic features from being excessively corroded, bilinear interpolation is added to improve the resolution of the features, and the method specifically comprises the following steps:
let F (w)i,hj) The feature value of any point on the feature map is the F (w) of any positioni+u,hj+ v) is calculated as:
Figure BDA0003350167630000091
then, in this embodiment, a multi-scale feature fusion method is used between the RS2 module and the RS3 module, and a feature matrix output by the RS2 module and an enhanced feature matrix output by the RS3 module are fused, so that abstraction of high-level semantic features is retained, and noise redundancy of low-level features is suppressed to a certain extent; and putting the fused feature matrix into an RPN layer, applying an anchor mechanism, setting the proportion of anchors to be 0.5, 1, 2 and 3, and calculating the confidence degrees of the foreground and the background in the fused feature matrix by using 16 anchor point frames with different sizes for each anchor point.
The embodiment provides an XRoI operation (expansion ROI) to solve the problem of feature loss of a small target in the original ROI layer, and a bilinear interpolation method is added on the basis of the ROI pooling operation, so that deep visual features of a small object are well retained, deep features of a foreground are fully extracted, and for ROIs of different sizes, 7 × 7 fixed-size ROI is adopted for feature extraction: first a 14 x 14 feature area is obtained by XRoI operation, followed by maximum pooling at step 2. The resulting tensor is 7 × 7 × 1536.
In the learning process, the RPN layer sets the following loss function measures to predict the deviation between the target frame and the real frame:
Figure BDA0003350167630000101
where i is the index number of the suggestion box, diIs the predicted probability of the box i being proposed as the target,
Figure BDA0003350167630000102
is a label of the real frame uiIs a vector of four parameterized coordinates representing the predicted target box,
Figure BDA0003350167630000103
is and uiAssociated real box, NregRepresenting the number of anchor positions, NclsIs a constant corresponding to the training image, LclsAnd LregFrom NclsAnd NregNormalization yields that λ represents a balance weight, L ({ d })i},{ui}) is the final loss obtained after classification regression.
The tensor of 7 × 7 × 1536 flows through a full connection layer of 1000 nodes, which is respectively connected to full connection layers with lengths of (k +1) and 4 × (k +1) (k is the number of categories of the object to be detected), and then classification probability and target position correction are obtained through Softmax function calculation and position regression calculation, and a redundant object suggestion box adopts non-maximum suppression operation. The calculation mode of the final classification and regression loss function is consistent with the calculation mode of the RPN layer loss function.
With specific reference to fig. 3, as a preferred embodiment of this embodiment, the pin defect detection model adopts an EfficientNet-B0 network, which includes 16 MBConv layers, 2 convolutional layers, 1 global average pooling layer, and 1 FC classification layer;
in order to accelerate the model training speed, the network adopts a transfer learning method to accelerate the training speed, and uses a deep model pre-trained on an ImageNet data set in a large scale as a universal feature extractor.
Example two:
the embodiment provides a pin defect detection device for a power transmission line, which comprises a memory, a processor and a computer program which is stored on the memory and can run on the processor, wherein the processor executes the program to realize the detection method according to any embodiment of the invention.
Example three:
the present embodiment provides a computer-readable storage medium, on which a computer program is stored, which when executed by a processor implements a detection method according to any of the embodiments of the present invention.
The above description is only an embodiment of the present invention, and not intended to limit the scope of the present invention, and all modifications of equivalent structures and equivalent processes performed by the present specification and drawings, or directly or indirectly applied to other related technical fields, are included in the scope of the present invention.

Claims (7)

1. A pin defect detection method for a power transmission line is characterized by comprising the following steps:
manufacturing a pin image dataset; collecting and preprocessing an electric transmission line inspection image containing a pin to obtain a pin image data set;
constructing a small target detection model based on a neural network; training the small target detection model by using the pin image data set to obtain a pin detection model;
making a defect data set, and cutting out a target frame subgraph containing a pin from an image in the pin image data set; preprocessing the target frame image, and adding a defect label for the pin in the image to obtain a pin defect data set;
constructing a classification model based on a neural network, and training the classification model by using the pin defect data set to obtain a pin defect detection model;
inputting an image to be detected, detecting a pin in the image through a pin detection model, and outputting position information of a target frame of which the pin is detected; and cutting the image according to the position information of the target frame to obtain a target frame subgraph, inputting the target frame subgraph to a pin defect detection model to detect whether defects exist, and outputting a result.
2. The method for detecting the defect of the pin of the power transmission line according to claim 1, wherein the small target detection model comprises two parts, wherein:
the first part comprises a VGG16 feature extraction network, and an FM focusing module is embedded in the VGG16 feature extraction network; extracting a feature matrix of the input image through a VGG16 feature extraction network; the FM focusing module firstly performs a global average pooling operation on the feature matrix to obtain a feature matrix of 1 multiplied by Channel, then performs a feature interaction process, firstly compresses the number of channels, then reconstructs the original channels, and finally generates an attention weight of 0-1 among the channels through a sigmoid function and then multiplies the attention weight by the original input feature matrix; dividing the image into sub-modules with a plurality of sizes, and outputting the sub-modules to a second part in batches;
the second part comprises a ResNet101 network, an RPN layer and an FC classification layer; the ResNet101 network comprises four residual blocks of RS 1-RS 4 which are sequentially connected, wherein each residual block comprises convolution, pooling and ReLU activation function operation, and a feature matrix of input data is extracted; the feature matrix extracted by the ResNet101 network is output to the RPN layer, and the RPN layer outputs the region of interest RoI to the FC classification layer according to the input feature matrix; and the FC classification layer comprises a full connection layer, a Softmax function and a position regression function, corrects the position of the region of interest RoI and outputs a target frame.
3. The method for detecting the pin defect of the power transmission line according to claim 2, wherein a characteristic strengthening operation is embedded in the RS3 module; the feature enhancement operation adopts a bilinear interpolation method to improve the resolution of the features, and specifically comprises the following steps:
let F (w)i,hj) The feature value of any point on the feature map is the F (w) of any positioni+u,hj+ v) is calculated as:
Figure FDA0003350167620000021
and in the ResNet101 network, a multi-scale feature fusion method is used for fusing the feature matrix output by the RS2 module and the enhanced feature matrix output by the RS3 module to obtain a fused feature matrix, and then outputting the fused feature matrix to the RPN layer.
4. The method for detecting the pin defect of the power transmission line according to claim 2, wherein the RPN layer sets the following loss function metric to predict the deviation between the target frame and the real frame in the learning process:
Figure FDA0003350167620000031
where i is the index number of the suggestion box, diIs the predicted probability of the box i being proposed as the target,
Figure FDA0003350167620000032
is a label of the real frame uiIs a vector of four parameterized coordinates representing the predicted target box,
Figure FDA0003350167620000033
is and uiAssociated real box, NregRepresenting the number of anchor positions, NclsIs a constant corresponding to the training image, LclsAnd LregFrom NclsAnd NregNormalization yields that λ represents a balance weight, L ({ d })i},{ui}) is the final loss obtained after classification regression.
5. The method for detecting the defect of the pin of the power transmission line according to claim 1, wherein the method comprises the following steps: the pin defect detection model adopts an EfficientNet-B0 network, and comprises 16 MBConv layers, 2 convolutional layers, 1 global average pooling layer and 1 FC classification layer;
the pin defect detection model trains a network by adopting a transfer learning method.
6. An electric transmission line pin defect detection device, comprising a memory, a processor and a computer program stored on the memory and operable on the processor, wherein the processor implements the detection method according to any one of claims 1 to 5 when executing the program.
7. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the detection method according to any one of claims 1 to 5.
CN202111334727.XA 2021-11-11 2021-11-11 Power transmission line pin defect detection method, equipment and medium Pending CN114170144A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111334727.XA CN114170144A (en) 2021-11-11 2021-11-11 Power transmission line pin defect detection method, equipment and medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111334727.XA CN114170144A (en) 2021-11-11 2021-11-11 Power transmission line pin defect detection method, equipment and medium

Publications (1)

Publication Number Publication Date
CN114170144A true CN114170144A (en) 2022-03-11

Family

ID=80478873

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111334727.XA Pending CN114170144A (en) 2021-11-11 2021-11-11 Power transmission line pin defect detection method, equipment and medium

Country Status (1)

Country Link
CN (1) CN114170144A (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115601363A (en) * 2022-12-14 2023-01-13 中建科技集团有限公司（Cn） Small-target detection algorithm-based assembly type building product defect detection method
CN115731478A (en) * 2022-11-24 2023-03-03 国网湖北省电力有限公司超高压公司 Power transmission line cotter pin target detection method based on multistage target detection
CN116012375A (en) * 2023-03-22 2023-04-25 成都唐源电气股份有限公司 Method and system for detecting cotter pin defects of overhead contact system soft crossing suspension pulley
CN116385952A (en) * 2023-06-01 2023-07-04 华雁智能科技(集团)股份有限公司 Distribution network line small target defect detection method, device, equipment and storage medium

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115731478A (en) * 2022-11-24 2023-03-03 国网湖北省电力有限公司超高压公司 Power transmission line cotter pin target detection method based on multistage target detection
CN115731478B (en) * 2022-11-24 2023-12-22 国网湖北省电力有限公司超高压公司 Power transmission line cotter pin target detection method based on multistage target detection
CN115601363A (en) * 2022-12-14 2023-01-13 中建科技集团有限公司（Cn） Small-target detection algorithm-based assembly type building product defect detection method
CN116012375A (en) * 2023-03-22 2023-04-25 成都唐源电气股份有限公司 Method and system for detecting cotter pin defects of overhead contact system soft crossing suspension pulley
CN116385952A (en) * 2023-06-01 2023-07-04 华雁智能科技(集团)股份有限公司 Distribution network line small target defect detection method, device, equipment and storage medium
CN116385952B (en) * 2023-06-01 2023-09-01 华雁智能科技(集团)股份有限公司 Distribution network line small target defect detection method, device, equipment and storage medium

Similar Documents

Publication Publication Date Title
CN114170144A (en) Power transmission line pin defect detection method, equipment and medium
CN111723654B (en) High-altitude parabolic detection method and device based on background modeling, YOLOv3 and self-optimization
CN111612008B (en) Image segmentation method based on convolution network
CN111598860B (en) Lithium battery defect detection method based on yolov3 network embedded into self-attention door module
CN114663346A (en) Strip steel surface defect detection method based on improved YOLOv5 network
CN115131283A (en) Defect detection and model training method, device, equipment and medium for target object
CN110942456B (en) Tamper image detection method, device, equipment and storage medium
CN108710893A (en) A kind of digital image cameras source model sorting technique of feature based fusion
CN115439694A (en) High-precision point cloud completion method and device based on deep learning
CN114332473A (en) Object detection method, object detection device, computer equipment, storage medium and program product
CN116385958A (en) Edge intelligent detection method for power grid inspection and monitoring
CN113516146A (en) Data classification method, computer and readable storage medium
CN115661932A (en) Fishing behavior detection method
CN113469950A (en) Method for diagnosing abnormal heating defect of composite insulator based on deep learning
CN115546742A (en) Rail foreign matter identification method and system based on monocular thermal infrared camera
CN116205876A (en) Unsupervised notebook appearance defect detection method based on multi-scale standardized flow
CN115861210A (en) Transformer substation equipment abnormity detection method and system based on twin network
Shit et al. An encoder‐decoder based CNN architecture using end to end dehaze and detection network for proper image visualization and detection
CN112802048B (en) Method and device for generating layer generation countermeasure network with asymmetric structure
CN114067268A (en) Method and device for detecting safety helmet and identifying identity of electric power operation site
CN112149528A (en) Panorama target detection method, system, medium and equipment
CN115620083B (en) Model training method, face image quality evaluation method, equipment and medium
CN116030050A (en) On-line detection and segmentation method for surface defects of fan based on unmanned aerial vehicle and deep learning
CN115564031A (en) Detection network for glass defect detection
CN115601675A (en) Surgical instrument counting method based on Swin-transducer and yolov5 model integration

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