CN114120159A - Method and device for detecting pin defects of power transmission line - Google Patents
Method and device for detecting pin defects of power transmission line Download PDFInfo
- Publication number
- CN114120159A CN114120159A CN202111501609.3A CN202111501609A CN114120159A CN 114120159 A CN114120159 A CN 114120159A CN 202111501609 A CN202111501609 A CN 202111501609A CN 114120159 A CN114120159 A CN 114120159A
- Authority
- CN
- China
- Prior art keywords
- image
- sub
- defect detection
- pin defect
- feature map
- 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
Links
- 230000007547 defect Effects 0.000 title claims abstract description 112
- 230000005540 biological transmission Effects 0.000 title claims abstract description 55
- 238000000034 method Methods 0.000 title claims abstract description 52
- 238000001514 detection method Methods 0.000 claims abstract description 168
- 238000013135 deep learning Methods 0.000 claims abstract description 33
- 238000012549 training Methods 0.000 claims description 32
- 230000007246 mechanism Effects 0.000 claims description 19
- 238000011176 pooling Methods 0.000 claims description 17
- 230000008569 process Effects 0.000 claims description 16
- 238000007689 inspection Methods 0.000 claims description 14
- 230000005284 excitation Effects 0.000 claims description 8
- 238000011156 evaluation Methods 0.000 claims description 6
- 230000001629 suppression Effects 0.000 claims description 6
- 230000006870 function Effects 0.000 description 6
- 238000012360 testing method Methods 0.000 description 6
- 238000010586 diagram Methods 0.000 description 4
- 230000009471 action Effects 0.000 description 3
- 238000013527 convolutional neural network Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 230000003044 adaptive effect Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 210000002569 neuron Anatomy 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 238000003860 storage Methods 0.000 description 2
- 238000013526 transfer learning Methods 0.000 description 2
- 230000002776 aggregation Effects 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 238000000889 atomisation Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000009194 climbing Effects 0.000 description 1
- 238000005520 cutting process Methods 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 230000008030 elimination Effects 0.000 description 1
- 238000003379 elimination reaction Methods 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 238000010191 image analysis Methods 0.000 description 1
- 238000011065 in-situ storage Methods 0.000 description 1
- 239000012212 insulator Substances 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a method and a device for detecting pin defects of a power transmission line. The pin defect detection model can be trained by means of a deep learning target detection algorithm, and the pin defect detection model is used for detecting whether the bolt in the target aerial image is missing or not, so that the automatic detection of the bolt missing defect of the power transmission line is realized, and the fault is timely and efficiently removed.
Description
Technical Field
The invention relates to the technical field of electric power equipment detection of a power transmission line, in particular to a pin defect detection method and device of the power transmission line.
Background
The bolt is widely applied to connection among all parts of the power transmission line as a fastener, so that the whole structure is stable. However, the working environment is complex and easy to damage, and part of the pins may be lost, which may cause a large-area transmission line fault and seriously threaten the safety and stability of the power grid. For traditional bolt inspection, climbing the position is the main means of inspecting the bolt, and is both time-consuming and laborious. Due to the distributed distribution of bolts and the diversity of bolt specifications, bolt inspection becomes more difficult. In recent years, unmanned aerial vehicle overhead transmission line inspection is popularized in a power system due to high safety and high efficiency, and intelligent processing can be realized by combining a machine learning target detection technology.
The target detection method based on computer vision deep learning is widely applied in many fields, and solves many different problems. However, the study on the missing of the bolt pin with fine granularity by scholars at home and abroad is less at present, the main direction is in the contents of insulator missing, self-explosion, vibration damper falling detection and the like, and a plurality of problems are needed to be solved in relation to the application of different model detection means in the detection of the missing bolt pin of the power transmission line. How to realize the automatic detection of the missing pin defect of the transmission line bolt and realize the timely and efficient fault elimination is a difficult problem of the power system and is also a long-term research target of the power system.
Disclosure of Invention
In view of the above, in order to solve the above problems, the present invention provides a method and an apparatus for detecting pin defects of a power transmission line, and the technical scheme is as follows:
a method for detecting pin defects of a power transmission line, the method comprising:
acquiring a target aerial image of the power transmission line to be processed;
inputting the target aerial image into a pin defect detection model, wherein the pin defect detection model is obtained by training based on a deep learning target detection algorithm in advance;
and acquiring a pin defect detection result output by the pin defect detection model aiming at the target aerial image.
Preferably, the method further comprises:
performing sliding window clipping on the target aerial image to obtain a plurality of first sub-images, wherein two adjacent first sub-images have an overlapping area;
correspondingly, the step of inputting the target aerial image into a pin defect detection model obtained by pre-training comprises the following steps:
inputting each first sub-image into the pin defect detection model in sequence;
correspondingly, the obtaining of the pin defect detection result output by the pin defect detection model for the target aerial image includes:
and acquiring a pin defect detection result output by the pin defect detection model aiming at each first sub-image, and performing non-maximum suppression on the pin defect detection result corresponding to each first sub-image.
Preferably, the process of obtaining the pin defect detection model based on deep learning target detection algorithm training in advance includes:
acquiring a sample aerial image acquired by the unmanned aerial vehicle inspection of the power transmission line, wherein the image in the sample aerial image is marked with a label;
performing sliding window clipping on the sample aerial image based on the label to obtain a plurality of second sub-images, wherein two adjacent second sub-images do not have overlapping areas;
and calling a basic model of a deep learning target detection algorithm, and introducing an attention mechanism to train the basic model by using the plurality of second sub-images.
Preferably, the method further comprises:
and performing data enhancement on the sample aerial image.
Preferably, the base model includes a convolutional layer, a pooling layer, and a full-link layer, and the attentiveness introducing mechanism trains the base model using the sample aerial image, including:
extracting a first feature map of each second sub-image through the convolutional layer, wherein the first feature map records image features of the second sub-image under each channel;
compressing the first feature map through the pooling layer to obtain a second feature map of the second sub-image, wherein the second feature map records image features of the second sub-image under each channel and evaluation scores of each channel;
performing excitation operation on the second feature map through the full-connection layer to obtain a third feature map, wherein the third feature map records image features of the second sub-image under each channel and weight of each channel;
and applying the weight of each channel in the third feature map to the image feature of the second sub-image under each channel.
An apparatus for detecting pin defects in a power transmission line, the apparatus comprising:
the image acquisition module is used for acquiring a target aerial image of the power transmission line to be processed;
the detection module is used for inputting the target aerial image into a pin defect detection model, and the pin defect detection model is obtained by training based on a deep learning target detection algorithm in advance; and acquiring a pin defect detection result output by the pin defect detection model aiming at the target aerial image.
Preferably, the image acquisition module is further configured to:
performing sliding window clipping on the target aerial image to obtain a plurality of first sub-images, wherein two adjacent first sub-images have an overlapping area;
correspondingly, the detection module, which is used for inputting the target aerial image into a pin defect detection model obtained by pre-training, is specifically used for:
inputting each first sub-image into the pin defect detection model in sequence;
correspondingly, the detection module, configured to obtain a pin defect detection result output by the pin defect detection model for the target aerial image, is specifically configured to:
and acquiring a pin defect detection result output by the pin defect detection model aiming at each first sub-image, and performing non-maximum suppression on the pin defect detection result corresponding to each first sub-image.
Preferably, the detection module, configured to obtain the pin defect detection model based on deep learning target detection algorithm training in advance, is specifically configured to:
acquiring a sample aerial image acquired by the unmanned aerial vehicle inspection of the power transmission line, wherein the image in the sample aerial image is marked with a label; performing sliding window clipping on the sample aerial image based on the label to obtain a plurality of second sub-images, wherein two adjacent second sub-images do not have overlapping areas; and calling a basic model of a deep learning target detection algorithm, and introducing an attention mechanism to train the basic model by using the plurality of second sub-images.
Preferably, the detection module is configured to obtain the pin defect detection model based on deep learning target detection algorithm training in advance, and is further configured to:
and performing data enhancement on the sample aerial image.
Preferably, the basic model includes a convolutional layer, a pooling layer, and a full-link layer, and the detection module is configured to introduce an attention mechanism to train the basic model using the sample aerial image, and is specifically configured to:
extracting a first feature map of each second sub-image through the convolutional layer, wherein the first feature map records image features of the second sub-image under each channel; compressing the first feature map through the pooling layer to obtain a second feature map of the second sub-image, wherein the second feature map records image features of the second sub-image under each channel and evaluation scores of each channel; performing excitation operation on the second feature map through the full-connection layer to obtain a third feature map, wherein the third feature map records image features of the second sub-image under each channel and weight of each channel; and applying the weight of each channel in the third feature map to the image feature of the second sub-image under each channel.
Compared with the prior art, the invention has the following beneficial effects:
the invention provides a method and a device for detecting pin defects of a power transmission line. The pin defect detection model can be trained by means of a deep learning target detection algorithm, and the pin defect detection model is used for detecting whether the bolt in the target aerial image is missing or not, so that the automatic detection of the bolt missing defect of the power transmission line is realized, and the fault is timely and efficiently removed.
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 embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the provided drawings without creative efforts.
Fig. 1 is a flowchart of a method for detecting a pin defect of a power transmission line according to an embodiment of the present invention;
fig. 2 is a schematic diagram of a detection result of the pin defect of the power transmission line provided by the embodiment of the invention.
Fig. 3 is a partial flowchart of a method for detecting a pin defect of a power transmission line according to an embodiment of the present invention;
fig. 4 is a partial flowchart of a method for detecting a pin defect of a power transmission line according to an embodiment of the present invention;
fig. 5 is a schematic structural diagram of the transmission line pin defect detection apparatus according to 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.
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.
In recent years, the field of object detection has been rapidly developed due to the wide application of deep learning. Unlike traditional methods such as SIFT, SURF, and BRISK, the Deep Convolutional Neural Network (DCNN) inherently has the ability to extract abstract features of data, which makes it widely studied and applied in this field. The DCNN-based main target detection framework can be structurally divided into two categories: the two-stage detection framework based on Region Propusal and the single-stage detection framework based on regression, wherein fast R-CNN is a typical representative of the former, SSD and YOLO are representative of the latter.
Aiming at the defects of the prior art, the invention provides a power transmission line pin defect detection scheme, which detects whether a bolt is missing by using an aerial image by means of a deep learning target detection algorithm YOLOv 5. In addition, aiming at the problems that the aerial image is high in resolution ratio and the missing area is low in occupation ratio, the invention provides a YOLOv5 detection model based on an attention mechanism and a detection scheme for suppressing duplication of non-maximum values by cutting a large image, and automatic classification and detection positioning of whether the pins are missing are realized.
Referring to fig. 1, fig. 1 is a flowchart of a method for detecting a pin defect of a power transmission line according to an embodiment of the present invention, where the method for detecting a pin defect of a power transmission line includes the following steps:
and S10, acquiring a target aerial image of the power transmission line to be processed.
In the embodiment of the invention, the target aerial image, namely the aerial image to be processed of the power transmission line, can comprise at least one bolt.
And S20, inputting the target aerial image into a pin defect detection model, wherein the pin defect detection model is obtained by training in advance based on a deep learning target detection algorithm.
In the embodiment of the invention, a pin defect detection model is trained on the basis of a deep learning target detection algorithm YOLOv5, wherein a network structure of YOLOv5 is composed of a backbone (a backbone network for extracting a feature map) and a head (a detection head assembly for predicting the type and the position of a target). Compared with YOLOv3 and YOLOv4, the backbone is composed of a Focus (i.e., a down-sampling module without information loss, which performs a slicing operation and can improve the computation power without information loss), a CSPNet (i.e., a Cross Stage Partial Network, which solves the problem that the prior work needs a large amount of computation in the inference process and enhances the capability of Network Feature fusion), and the like, and the head is composed of an FPN (i.e., a Feature Pyramid, which can better handle the multi-scale change problem in object detection), a PAN (i.e., a Path Aggregation Network, which adds an information transmission Path from the bottom layer to the top layer and can enhance the Feature Pyramid), and the like.
The input end of the network structure of YOLOv5 is enhanced by adopting Mosaic data, and the detection effect of small targets is improved to a certain extent. The network structure of YOLOv5 uses an adaptive anchor box (auto anchor) to compute the optimal anchor box (regression box) for the input image. Moreover, for the input image of the network structure of YOLOv5, adaptive image scaling is required, the input image needs to be scaled to a certain size, and black-filled edges are not used in the training process, but can be adopted in the testing process to improve the inference speed. In addition, a class balance sampling strategy (such as local LOSS, which is a LOSS function in the deep learning network and belongs to a variation of a cross entropy LOSS function) can be adopted to solve the problem of serious imbalance of positive and negative sample ratios in the one-stage (i.e., one-step method, in which a neural network simultaneously completes regression and classification tasks through one-time gradient calculation) training process, and a prediction branch bounding box (a real box) is added with LOSS functions such as GIOU _ LOSS (which is a LOSS function in the deep learning network and belongs to a variation of an iou cross-over LOSS function).
Moreover, model training can be performed by using transfer learning in the training process, the model is retrained on the own data set, a part of initial weights are frozen in situ, and the rest weights are used for calculating loss and are updated by an optimizer, so that the algorithm model parameters of the optimal solution are finally obtained.
Aiming at the problems that the aerial image has high resolution and the missing area occupies a very low area, the following steps can be further executed after the target aerial image is acquired in the embodiment of the invention:
and performing sliding window clipping on the target aerial image to obtain a plurality of first sub-images, wherein two adjacent first sub-images have an overlapping area.
In the embodiment of the present invention, sliding window cropping may be performed on both the target aerial image and the input image during the test process, for example, the target aerial image may be cropped to an image with a height h of 640 and a width w of 640. For the target aerial image, sliding window cropping is performed on the target aerial image, so that a plurality of sub-images forming the target aerial image, namely first sub-images, can be obtained, and two adjacent first sub-images in the target aerial image have an overlapping area, for example, an overlapping area of 15%.
Correspondingly, in the process of inputting the target aerial image into the pin defect detection model obtained through pre-training, each first sub-image can be sequentially input into the pin defect detection model, so that the bolt missing pin defect detection of each first sub-image is completed.
And S30, acquiring a pin defect detection result output by the pin defect detection model aiming at the target aerial image.
In the embodiment of the invention, the target aerial image is input into the pin defect detection model, so that the pin detection frame output by the pin detection model aiming at the target aerial image and the coordinate position of the pin detection frame can be obtained.
Of course, if the target aerial image is cropped into a plurality of first sub-images by the sliding window, for each first sub-image, the pin detection frame output by the pin detection model for the first sub-image and the coordinate position of the pin detection frame may be obtained, and further, the absolute coordinate position of the pin detection frame in the target aerial image may be obtained according to the coordinate position of any corner point (for example, the upper left corner point) of the first sub-image. And finally, performing Non-Maximum Suppression (NMS) on all the pin detection frames on the target aerial image to obtain the pin detection frame corresponding to the final target aerial image and the coordinate position of the pin detection frame. Fig. 2 is a schematic diagram of a detection result of the pin defect of the power transmission line provided by the embodiment of the invention.
In the specific implementation process, the process of training to obtain the pin defect detection model based on the deep learning target detection algorithm in advance comprises the following steps, and the flow chart of the method is shown in fig. 3:
s101, acquiring a sample aerial image acquired by the unmanned aerial vehicle inspection of the power transmission line, wherein the image in the sample aerial image is marked with a label.
In the embodiment of the invention, aerial images serving as samples, namely sample aerial images, can be acquired through the unmanned aerial vehicle inspection of the power transmission line. And performing pin marking on the sample aerial image by using image marking software to obtain a pin candidate frame, the coordinate position of the pin candidate frame and a tag of the pin candidate frame in the sample aerial image, wherein the tag comprises two types of pin missing (lack split pin) and normal (normal split pin).
In the training process, the sample aerial images can be further processed according to a certain proportion, such as 8: and 2, dividing a training set and a test set in a mode of 2, wherein the training set is used for training the model, and the test set is used for testing the model training result.
And S102, performing sliding window clipping on the sample aerial image based on the label to obtain a plurality of second sub-images, wherein two adjacent second sub-images do not have overlapping areas.
In the embodiment of the present invention, the sample aerial image may be clipped by a sliding window, for example, the sample aerial image may be clipped as an image with a height h of 640 and a width w of 640. Therefore, after the sliding window cropping, the sample aerial image can be cropped into a plurality of sub-images forming the sample aerial image, namely the second sub-images, and two adjacent sub-images in the sample aerial image do not have overlapping areas. At the same time, an xml annotation file may also be derived relative to the second sub-image, and the second sub-image with no pin candidate box or no tag in the pin candidate box may be deleted.
S103, calling a basic model of the deep learning target detection algorithm, and introducing an attention mechanism to train the basic model by using a plurality of second sub-images.
In the embodiment of the invention, the basic model of the deep learning target detection algorithm can be a network of YOLOv5, and can also be a model obtained by transfer learning. And (3) introducing the advantages of an attention mechanism, combining the attention mechanism with YOLOv5, inputting a plurality of second sub-images into a basic model for training, and performing loop iteration and model parameter optimization to finally obtain model parameters of an optimal solution.
In addition, the base model includes a convolutional layer, a pooling layer, and a fully-connected layer. Therefore, the process of using the sample aerial image to train the base model by introducing the attention mechanism can adopt the following steps, and the flow chart of the method is shown in fig. 4:
s201, extracting a first feature map of each second sub-image through the convolution layer, wherein the first feature map records the image features of the second sub-image under each channel.
In the embodiment of the present invention, for each second sub-image, the convolutional layer may extract image features of the convolutional layer under each channel, so as to obtain a feature map output by the convolutional layer, that is, a first feature map, where the first feature map is a vector of 1 × 1 × C, where C is the number of channels.
And S202, compressing the first feature map through the pooling layer to obtain a second feature map of the second sub-image, wherein the image features of the second sub-image under each channel and the evaluation scores of each channel are recorded in the second feature map.
In the embodiment of the present invention, the first feature map output by the convolutional layer may be compressed (Squeeze) by the pooling layer, that is, global average pooling (global average pooling) is performed, so as to obtain the feature map output by the pooling layer, that is, the second feature map, where the second feature map is also a vector of 1 × 1 × C.
And S203, carrying out excitation operation on the second feature map through the full-connection layer to obtain a third feature map, wherein the image features of the second sub-image under each channel and the weight of each channel are recorded in the third feature map.
In the embodiment of the invention, the number of the full connection layers is two. The second feature map output by the pooling layer may be subjected to Excitation (Excitation) operation by the two fully-connected layers, so as to obtain feature maps output by the two fully-connected layers, that is, a third feature map, where the third feature map includes a 1 × 1 × C × SERadio vector output by one fully-connected layer and a 1 × 1 × C output by the other fully-connected layer, where SERadio is a scaling parameter, and the purpose of this parameter is to reduce the number of channels and thus reduce the amount of computation. The first fully connected layer has C × SERatio neurons, with an input of 1 × 1 × C, and an output of 1 × 1 × C × SERadio. The second fully connected layer has C neurons with an input of 1 × 1 × C × SERADio and an output of 1 × 1 × C.
And S204, applying the weight of each channel in the third feature map to the image features of the second sub-image under each channel.
In the embodiment of the present invention, Scale (vector dot product) operation is finally performed, so that the weight of each channel is applied to the image feature under each channel, that is, the weight of each channel is multiplied by the image feature, and a final result is obtained. A general combination mode using CBAM, i.e., a combination mode using a channel attention mechanism and then a spatial attention mechanism. CBAM is a lightweight, universal module that can be integrated into convolutional layer architecture with negligible overhead, and can be trained end-to-end with standard convolutional layers.
In addition, the embodiment of the invention can also use NAS-FPN, AC-FPN or a hole convolution to replace the last layer of the FPN, construct a pin defect detection model based on an attention mechanism to acquire the position information of the missing pin, and realize the pin missing detection.
In other embodiments, after the sample aerial image is acquired, data enhancement may also be performed on the sample aerial image. In the embodiment of the invention, the sample aerial image can be subjected to the preprocessing operation of random inversion and brightness and contrast change, and the sample aerial image is expanded to generate a similar image so as to achieve the purpose of data enhancement.
As the collected data set is limited by various conditions, various scenes in a real environment, such as heavy fog, rain, snow and the like, can be better simulated. And carrying out operations of increasing random turning, brightness, contrast, noise, atomization, translation, rotation and distortion on the collected sample aerial images, and expanding the collected sample aerial images to generate similar images so as to achieve the purpose of data enhancement.
In deep learning, the more complex the convolutional neural network model is, the stronger the ability to express objects, which will result in good training data and poor test data results. Therefore, massive data are needed to avoid the over-fitting condition, and the trained model is ensured to have good detection effect on new data. In order to avoid the influence of too high fitting degree on the detection effect, a large amount of sample data support is needed. The richer the sample data is, the higher the detection accuracy is. Therefore, the pin defect detection model in the embodiment of the invention has good identification precision and model robustness for the pins of the power transmission line.
In addition, in the aspect of expansion of the sample aerial image, an expansion mode for the target can be used, the target is pasted to any position in the image, a new mark is generated, and random transformation (zooming, turning over and rotating) can be performed on the pasted target. It should be noted that the target is a target object to be identified, i.e. a pin in the present invention.
Aiming at the problem of defect detection of the pins in the high-resolution aerial images, the problems of identification difficulty and analysis difficulty of various algorithms are solved, a scheme based on deep learning target detection is provided, and a YOLOv5 model based on an attention mechanism is constructed to obtain the position information of the missing pins. The invention realizes the automatic detection of the missing pin defect of the bolt of the power transmission line, can remove the fault timely and efficiently, needs a large amount of professional technicians compared with the traditional image analysis for detecting the power transmission line, and needs to amplify and drag the image. The invention improves the inspection efficiency and the inspection safety and meets the diagnosis precision of the inspection task of the power transmission line. The invention realizes the intelligent processing of the target detection technology combining the unmanned aerial vehicle overhead transmission line inspection and deep learning.
Based on the method for detecting the pin defect of the power transmission line provided by the embodiment, an embodiment of the present invention further provides a device for executing the method for detecting the pin defect of the power transmission line, and a schematic structural diagram of the device is shown in fig. 5, and the device includes:
the image acquisition module 10 is used for acquiring a target aerial image of the power transmission line to be processed;
the detection module 20 is used for inputting the target aerial image into a pin defect detection model, wherein the pin defect detection model is obtained by training based on a deep learning target detection algorithm in advance; and acquiring a pin defect detection result output by the pin defect detection model aiming at the target aerial image.
Optionally, the image obtaining module 10 is further configured to:
the method comprises the steps that sliding window clipping is conducted on a target aerial image to obtain a plurality of first sub-images, and two adjacent first sub-images have overlapping areas;
correspondingly, the detection module 20 for inputting the target aerial image into the pre-trained pin defect detection model is specifically configured to:
inputting each first sub-image into a pin defect detection model in sequence;
correspondingly, the detection module 20, configured to obtain a pin defect detection result output by the pin defect detection model for the target aerial image, is specifically configured to:
and acquiring a pin defect detection result output by the pin defect detection model aiming at each first sub-image, and performing non-maximum suppression on the pin defect detection result corresponding to each first sub-image.
Optionally, the detection module 20 is configured to obtain a pin defect detection model based on deep learning target detection algorithm training in advance, and is specifically configured to:
acquiring a sample aerial image acquired by the unmanned aerial vehicle inspection of the power transmission line, wherein the image in the sample aerial image is marked with a label; performing sliding window clipping on the sample aerial image based on the label to obtain a plurality of second sub-images, wherein two adjacent second sub-images do not have an overlapping area; and calling a basic model of the deep learning target detection algorithm, and introducing an attention mechanism to train the basic model by using a plurality of second sub-images.
Optionally, the detection module 20 is configured to obtain a pin defect detection model based on deep learning target detection algorithm training in advance, and is further configured to:
and performing data enhancement on the sample aerial image.
Optionally, the basic model includes a convolution layer, a pooling layer, and a full connection layer, and the detection module 20 is configured to introduce an attention mechanism and train the basic model using the sample aerial image, and is specifically configured to:
extracting a first feature map of each second sub-image through the convolution layer, wherein the first feature map records image features of the second sub-image under each channel; compressing the first feature map through the pooling layer to obtain a second feature map of the second sub-image, wherein the second feature map records image features of the second sub-image under each channel and evaluation scores of each channel; carrying out excitation operation on the second feature map through the full-connection layer to obtain a third feature map, wherein the third feature map records the image features of the second sub-image under each channel and the weight of each channel; and applying the weight of each channel in the third feature map to the image feature of the second sub-image under each channel.
It should be noted that, for the detailed functions of each functional module in the embodiment of the present invention, reference may be made to the corresponding disclosure of the foregoing power transmission line pin defect detection method embodiment, and details are not described here again.
Based on the method for detecting the pin defect of the power transmission line provided by the embodiment, the embodiment of the invention also provides electronic equipment, wherein the electronic equipment comprises: at least one memory and at least one processor; the memory stores programs, the processor calls the programs stored in the memory, and the programs are used for realizing the power transmission line pin defect detection method.
Based on the method for detecting the pin defect of the power transmission line provided by the embodiment, the embodiment of the invention further provides a storage medium, wherein the storage medium stores computer-executable instructions, and the computer-executable instructions are used for executing the method for detecting the pin defect of the power transmission line.
The method and the device for detecting the pin defect of the power transmission line provided by the invention are described in detail, a specific example is applied in the method to explain the principle and the implementation mode of the invention, and the description of the embodiment is only used for helping to understand the method and the core idea of the invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, there may be variations in the specific embodiments and the application scope, and in summary, the content of the present specification should not be construed as a limitation to the present invention.
It should be noted that, in the present specification, the embodiments are all described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments may be referred to each other. The device disclosed by the embodiment corresponds to the method disclosed by the embodiment, so that the description is simple, and the relevant points can be referred to the method part for description.
It is further noted that, herein, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include or include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in a process, method, article, or apparatus that comprises the element.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims (10)
1. A pin defect detection method for a power transmission line is characterized by comprising the following steps:
acquiring a target aerial image of the power transmission line to be processed;
inputting the target aerial image into a pin defect detection model, wherein the pin defect detection model is obtained by training based on a deep learning target detection algorithm in advance;
and acquiring a pin defect detection result output by the pin defect detection model aiming at the target aerial image.
2. The method of claim 1, further comprising:
performing sliding window clipping on the target aerial image to obtain a plurality of first sub-images, wherein two adjacent first sub-images have an overlapping area;
correspondingly, the step of inputting the target aerial image into a pin defect detection model obtained by pre-training comprises the following steps:
inputting each first sub-image into the pin defect detection model in sequence;
correspondingly, the obtaining of the pin defect detection result output by the pin defect detection model for the target aerial image includes:
and acquiring a pin defect detection result output by the pin defect detection model aiming at each first sub-image, and performing non-maximum suppression on the pin defect detection result corresponding to each first sub-image.
3. The method according to claim 2, wherein the process of training the pin defect detection model based on the deep learning target detection algorithm in advance comprises:
acquiring a sample aerial image acquired by the unmanned aerial vehicle inspection of the power transmission line, wherein the image in the sample aerial image is marked with a label;
performing sliding window clipping on the sample aerial image based on the label to obtain a plurality of second sub-images, wherein two adjacent second sub-images do not have overlapping areas;
and calling a basic model of a deep learning target detection algorithm, and introducing an attention mechanism to train the basic model by using the plurality of second sub-images.
4. The method of claim 3, further comprising:
and performing data enhancement on the sample aerial image.
5. The method of claim 3, wherein the base model comprises a convolutional layer, a pooling layer, and a fully-connected layer, and wherein the attentive mechanism trains the base model using the sample aerial image, comprising:
extracting a first feature map of each second sub-image through the convolutional layer, wherein the first feature map records image features of the second sub-image under each channel;
compressing the first feature map through the pooling layer to obtain a second feature map of the second sub-image, wherein the second feature map records image features of the second sub-image under each channel and evaluation scores of each channel;
performing excitation operation on the second feature map through the full-connection layer to obtain a third feature map, wherein the third feature map records image features of the second sub-image under each channel and weight of each channel;
and applying the weight of each channel in the third feature map to the image feature of the second sub-image under each channel.
6. The utility model provides a transmission line pin defect detecting device which characterized in that, the device includes:
the image acquisition module is used for acquiring a target aerial image of the power transmission line to be processed;
the detection module is used for inputting the target aerial image into a pin defect detection model, and the pin defect detection model is obtained by training based on a deep learning target detection algorithm in advance; and acquiring a pin defect detection result output by the pin defect detection model aiming at the target aerial image.
7. The apparatus of claim 6, wherein the image acquisition module is further configured to:
performing sliding window clipping on the target aerial image to obtain a plurality of first sub-images, wherein two adjacent first sub-images have an overlapping area;
correspondingly, the detection module, which is used for inputting the target aerial image into a pin defect detection model obtained by pre-training, is specifically used for:
inputting each first sub-image into the pin defect detection model in sequence;
correspondingly, the detection module, configured to obtain a pin defect detection result output by the pin defect detection model for the target aerial image, is specifically configured to:
and acquiring a pin defect detection result output by the pin defect detection model aiming at each first sub-image, and performing non-maximum suppression on the pin defect detection result corresponding to each first sub-image.
8. The apparatus according to claim 7, wherein the detection module for obtaining the pin defect detection model based on deep learning target detection algorithm training in advance is specifically configured to:
acquiring a sample aerial image acquired by the unmanned aerial vehicle inspection of the power transmission line, wherein the image in the sample aerial image is marked with a label; performing sliding window clipping on the sample aerial image based on the label to obtain a plurality of second sub-images, wherein two adjacent second sub-images do not have overlapping areas; and calling a basic model of a deep learning target detection algorithm, and introducing an attention mechanism to train the basic model by using the plurality of second sub-images.
9. The apparatus according to claim 8, wherein the detection module for training the pin defect detection model based on a deep learning target detection algorithm in advance is further configured to:
and performing data enhancement on the sample aerial image.
10. The apparatus according to claim 8, wherein the base model comprises a convolutional layer, a pooling layer and a full-link layer, the detection module for introducing an attention mechanism for training the base model using the sample aerial image is specifically configured to:
extracting a first feature map of each second sub-image through the convolutional layer, wherein the first feature map records image features of the second sub-image under each channel; compressing the first feature map through the pooling layer to obtain a second feature map of the second sub-image, wherein the second feature map records image features of the second sub-image under each channel and evaluation scores of each channel; performing excitation operation on the second feature map through the full-connection layer to obtain a third feature map, wherein the third feature map records image features of the second sub-image under each channel and weight of each channel; and applying the weight of each channel in the third feature map to the image feature of the second sub-image under each channel.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111501609.3A CN114120159A (en) | 2021-12-09 | 2021-12-09 | Method and device for detecting pin defects of power transmission line |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111501609.3A CN114120159A (en) | 2021-12-09 | 2021-12-09 | Method and device for detecting pin defects of power transmission line |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114120159A true CN114120159A (en) | 2022-03-01 |
Family
ID=80363854
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111501609.3A Pending CN114120159A (en) | 2021-12-09 | 2021-12-09 | Method and device for detecting pin defects of power transmission line |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114120159A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114998576A (en) * | 2022-08-08 | 2022-09-02 | 广东电网有限责任公司佛山供电局 | Method, device, equipment and medium for detecting loss of cotter pin of power transmission line |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110688925A (en) * | 2019-09-19 | 2020-01-14 | 国网山东省电力公司电力科学研究院 | Cascade target identification method and system based on deep learning |
| CN110827251A (en) * | 2019-10-30 | 2020-02-21 | 江苏方天电力技术有限公司 | Power transmission line locking pin defect detection method based on aerial image |
| US20200104993A1 (en) * | 2018-10-01 | 2020-04-02 | Skc Co., Ltd. | Film defect detection method and system |
| CN112990392A (en) * | 2021-05-20 | 2021-06-18 | 四川大学 | New material floor defect target detection system based on improved YOLOv5 algorithm |
| CN113673326A (en) * | 2021-07-14 | 2021-11-19 | 南京邮电大学 | Unmanned aerial vehicle platform crowd counting method and system based on image deep learning |
-
2021
- 2021-12-09 CN CN202111501609.3A patent/CN114120159A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20200104993A1 (en) * | 2018-10-01 | 2020-04-02 | Skc Co., Ltd. | Film defect detection method and system |
| CN110688925A (en) * | 2019-09-19 | 2020-01-14 | 国网山东省电力公司电力科学研究院 | Cascade target identification method and system based on deep learning |
| CN110827251A (en) * | 2019-10-30 | 2020-02-21 | 江苏方天电力技术有限公司 | Power transmission line locking pin defect detection method based on aerial image |
| CN112990392A (en) * | 2021-05-20 | 2021-06-18 | 四川大学 | New material floor defect target detection system based on improved YOLOv5 algorithm |
| CN113673326A (en) * | 2021-07-14 | 2021-11-19 | 南京邮电大学 | Unmanned aerial vehicle platform crowd counting method and system based on image deep learning |
Non-Patent Citations (1)
| Title |
|---|
| 麦俊佳 等: "基于深度学习的输电线路航拍照片目标检测应用", 《广东电力》, vol. 33, no. 9, 10 October 2020 (2020-10-10), pages 174 - 182 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114998576A (en) * | 2022-08-08 | 2022-09-02 | 广东电网有限责任公司佛山供电局 | Method, device, equipment and medium for detecting loss of cotter pin of power transmission line |
| CN114998576B (en) * | 2022-08-08 | 2022-12-30 | 广东电网有限责任公司佛山供电局 | Method, device, equipment and medium for detecting loss of cotter pin of power transmission line |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109165623B (en) | Rice disease spot detection method and system based on deep learning | |
| CN113177560A (en) | Universal lightweight deep learning vehicle detection method | |
| CN113947590A (en) | Surface defect detection method based on multi-scale attention guidance and knowledge distillation | |
| CN113255589B (en) | Target detection method and system based on multi-convolution fusion network | |
| CN114973002A (en) | Improved YOLOv 5-based ear detection method | |
| CN114973032B (en) | Deep convolutional neural network-based photovoltaic panel hot spot detection method and device | |
| CN117496384B (en) | Unmanned aerial vehicle image object detection method | |
| CN114596278A (en) | Method and device for detecting hot spot defects of photovoltaic panel of photovoltaic power station | |
| CN111507398A (en) | Transformer substation metal instrument corrosion identification method based on target detection | |
| CN115908354A (en) | Photovoltaic panel defect detection method based on double-scale strategy and improved YOLOV5 network | |
| CN115410087A (en) | Transmission line foreign matter detection method based on improved YOLOv4 | |
| CN115830535A (en) | Method, system, equipment and medium for detecting accumulated water in peripheral area of transformer substation | |
| CN115937736A (en) | Small target detection method based on attention and context awareness | |
| CN109829421B (en) | Method and device for vehicle detection and computer readable storage medium | |
| CN116824543A (en) | Automatic driving target detection method based on OD-YOLO | |
| CN118038379A (en) | Vehicle small target detection method and device based on lightweight network design | |
| CN116503354A (en) | Method and device for detecting and evaluating hot spots of photovoltaic cells based on multi-mode fusion | |
| CN116579992A (en) | Small target bolt defect detection method for unmanned aerial vehicle inspection | |
| CN114120159A (en) | Method and device for detecting pin defects of power transmission line | |
| CN116485802B (en) | Insulator flashover defect detection method, device, equipment and storage medium | |
| CN113537013A (en) | Multi-scale self-attention feature fusion pedestrian detection method | |
| CN116596895A (en) | Substation equipment image defect identification method and system | |
| CN113343749A (en) | Fruit identification method and system based on D2Det model | |
| CN116342531B (en) | Device and method for detecting quality of welding seam of high-altitude steel structure of lightweight large-scale building | |
| CN114612803B (en) | Improved CENTERNET transmission line insulator defect detection method |
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 |