CN114708267A - Image detection processing method for corrosion defect of tower stay wire on power transmission line - Google Patents

Image detection processing method for corrosion defect of tower stay wire on power transmission line Download PDF

Info

Publication number
CN114708267A
CN114708267A CN202210635427.3A CN202210635427A CN114708267A CN 114708267 A CN114708267 A CN 114708267A CN 202210635427 A CN202210635427 A CN 202210635427A CN 114708267 A CN114708267 A CN 114708267A
Authority
CN
China
Prior art keywords
convolution
image
layer
pooling
defect
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202210635427.3A
Other languages
Chinese (zh)
Other versions
CN114708267B (en
Inventor
陈松波
郭创新
杨强
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.)
Zhejiang University ZJU
Original Assignee
Zhejiang University ZJU
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 Zhejiang University ZJU filed Critical Zhejiang University ZJU
Priority to CN202210635427.3A priority Critical patent/CN114708267B/en
Publication of CN114708267A publication Critical patent/CN114708267A/en
Application granted granted Critical
Publication of CN114708267B publication Critical patent/CN114708267B/en
Priority to GB2219702.4A priority patent/GB2619576A/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

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
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • 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
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • 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
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/098—Distributed learning, e.g. federated learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10004—Still image; Photographic image
    • 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
    • G06T2207/10—Image acquisition modality
    • G06T2207/10032—Satellite or aerial image; Remote sensing
    • 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
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30108—Industrial image inspection
    • G06T2207/30136—Metal
    • 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
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30181—Earth observation
    • G06T2207/30184—Infrastructure
    • 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
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30232—Surveillance
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
    • Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
    • Y04S10/00—Systems supporting electrical power generation, transmission or distribution
    • Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications

Abstract

The embodiment of the invention provides a method for detecting and processing corrosion defect images of tower guy wires on a power transmission line, which comprises the steps of collecting tower guy wire images on the power transmission line through an unmanned aerial vehicle, segmenting the tower guy wire images through a full convolution neural network to obtain a plurality of image blocks, and taking the image blocks containing the defect tower guy wires as target image blocks; extracting a characteristic vector of a tower stay wire based on the target image block; screening out the feature vectors meeting the preset conditions to form a target screening feature set; extracting a typical defect characteristic value of a characteristic vector of each tower stay wire; similarity calculation is carried out on the typical defect characteristic value on each tower stay wire and the target screening characteristic set to obtain defect similarity, whether the current tower stay wire has strand breakage or corrosion defects is judged based on the defect similarity, and the precision of detecting the corrosion defects of the tower stay wire is improved.

Description

Image detection processing method for corrosion defect of tower stay wire on power transmission line
Technical Field
The invention relates to the technical field of computers, in particular to an image detection processing method for corrosion defects of tower stay wires on a power transmission line.
Background
At present, with the application of helicopter routing inspection technology for power transmission lines and the gradual advance of smart grid construction, automatic detection of power line defects is more and more concerned. The tower stay wire is a main material of a tower main body, is influenced by environmental factors such as wind power, icing, temperature and the like, is easy to generate local fatigue damage, even causes local strand breakage, is not enough in stay wire corrosion tension and the like. Therefore, the research on the automatic detection of the defects of the tower stay wire has very important practical significance.
The corrosion phenomenon of the tower stay wire occurs frequently in a long-distance power transmission line system and is extremely harmful, a plurality of methods for identifying the corrosion of the tower stay wire are carried out, the method based on image identification is generally divided into two parts, firstly, the tower stay wire is extracted, and whether the tower stay wire is extracted or not is directly related to the subsequent defect identification of the tower stay wire.
The pole tower stay wire identification method includes the traditional method of extracting features based on Hough transform, extracting HOG (Histogram of Oriented Gradient) features of pole tower stay wires and the like, but the extraction accuracy of the methods is not high due to the fact that the extracted features are single. In recent years, with the rapid development of artificial intelligence technology, neural network technology is widely applied to the identification of objects, a convolutional neural network is proposed to identify pole tower stay wires, a structural neural network structure of an input layer, a plurality of staggered convolutional layers, a pooling layer and a full connecting layer is mainly utilized, the structural mode is single, the convolutional neural network is trained by a large number of pole tower stay wire training sets, the calculated amount is huge, once the shape of the pole tower stay wires changes, the trained neural network model cannot accurately detect the pole tower stay wires, and the mobility of the neural network model in the single mode is poor.
Disclosure of Invention
The invention aims to provide an image detection processing method for corrosion defects of tower stay wires on a power transmission line, which is used for solving the problems in the prior art.
The embodiment of the invention provides a method for detecting and processing corrosion defect images of tower and stay wires on a power transmission line, which comprises the following steps:
collecting tower stay wire images on the power transmission line through an unmanned aerial vehicle;
dividing a tower stay wire image through a self-adaptive tower stay wire defect detection model to obtain a plurality of image blocks, and taking the image blocks containing the defective tower stay wires as target image blocks;
extracting a characteristic vector of a tower stay wire based on a target image block through a pre-trained characteristic extraction network;
screening out the feature vectors meeting the preset conditions to form a target screening feature set;
extracting typical defect characteristics of each tower stay wire in the tower stay wire image by a two-pass scanning method;
obtaining defect similarity between the typical defect feature and each feature vector in the target screening feature set;
and judging whether the current tower stay wire has the defects of strand breakage or corrosion based on the defect similarity.
Optionally, the adaptive tower stay wire defect detection model includes an input layer, an adaptive structure layer and an output layer;
the input layer comprises a convolution node and a pooling node;
the self-adaptive structure layer at least comprises one convolution pooling layer, and each convolution pooling layer comprises a convolution node and a pooling node;
the output layer comprises a convolution layer, a composite layer and a convolution neural network; the convolutional layer includes two convolutional nodes.
Optionally, if the adaptive structural layer includes a convolution pooling layer, the pole tower guy wire image is segmented by the adaptive pole tower guy wire defect detection model to obtain a plurality of image blocks, and the image block including the defective pole tower guy wire is used as a target image block, including:
obtaining a first hog characteristic of a tower stay wire image;
carrying out convolution operation on the tower stay wire image input into the self-adaptive tower stay wire defect detection model through the convolution node of the input layer to obtain a first convolution image; obtaining a first convolution hog feature of a first convolution image;
performing pooling operation on tower stay wire images input into the self-adaptive tower stay wire defect detection model through a pooling node of an input layer to obtain first pooled images; obtaining a first pooling hog feature of the first pooled image;
obtaining a first difference value and a second difference value; the first difference is equal to a difference between the first hog feature and the first convolution hog feature; the second difference is equal to a difference between the first hog feature and the first pooled hog feature;
if the first difference is smaller than or equal to the second difference, the input of the self-adaptive structural layer is a first convolution image;
carrying out convolution operation on the first convolution image by the convolution node of the layer convolution pooling layer to obtain a second convolution image; obtaining a second convolution hog feature of a second convolution image;
pooling the first convolution image by a pooling node of the layer convolution pooling layer to obtain a second pooling image; obtaining a second pooling hog feature of the second pooled image;
obtaining a third difference value and a fourth difference value; the third difference is equal to the difference between the second convolved hog signature and the first convolved hog signature; the fourth difference is equal to the difference between the second pooled hog signature and the first pooled hog signature;
if the first difference is greater than the second difference, inputting the self-adaptive structural layer into a first pooling image;
carrying out convolution operation on the first pooled image by the convolution node of the layer convolution pooling layer to obtain a second convolution image; obtaining a second convolution hog feature of a second convolution image;
pooling nodes of the layer convolution pooling layer perform pooling operation on the first pooled image to obtain a second pooled image; obtaining a second pooling hog feature of the second pooled image;
obtaining a third difference value and a fourth difference value; the third difference is equal to the difference between the second convolved hog feature and the first convolved hog feature; the fourth difference is equal to the difference between the second pooled hog signature and the first pooled hog signature;
the input of the two convolution nodes of the convolution layer is respectively a convolution image and a pooled image output by the last convolution pooling layer of the self-adaptive structural layer, and the two convolution nodes of the convolution layer respectively carry out convolution operation on the convolution image and the pooled image output by the last convolution pooling layer;
the composite layer fuses the convolution image and the pooled image output by the last convolution pooled layer after convolution operation to obtain a fused image;
and the convolution neural network carries out segmentation classification output on the fused image, and takes the image block containing the tower stay wire with the defect as a target image block, wherein the full convolution neural network is trained in advance.
Optionally, if the adaptive structure layer includes multiple convolution pooling layers, the method further includes segmenting the tower guy wire image through the adaptive tower guy wire defect detection model to obtain multiple image blocks, and using the image block containing the defective tower guy wire as a target image block:
if the third difference is smaller than or equal to the fourth difference, the input of the next convolution pooling layer is a second convolution image;
and if the third difference is larger than the fourth difference, the input of the next convolution pooling layer is a second pooling image.
Optionally, the feature extraction network includes a convolution pooling structure and a full convolution structure, and the convolution pooling structure includes three convolution layers and two pooling layers.
Optionally, the training method of the feature extraction network includes:
obtaining a training set, wherein the training set comprises a plurality of training target image blocks; the value of the core of the initialized three-layer convolution layer and the core of the two-layer pooling layer is a random number;
randomly selecting a training target image block to train the convolution pooling structure, and specifically comprising the following steps: inputting the training target image block into a first layer of convolutional layer, and performing convolutional operation on the training target image block through the first layer of convolutional layer to obtain a first convolutional image block; performing convolution operation on the first convolution image block through the second convolution layer to obtain a second convolution image block; performing convolution operation on the second convolution image block through the third layer of convolution layer to obtain a third convolution image block; performing pooling operation on the third convolution image block through the first pooling layer to obtain a first pooled image block; performing pooling operation on the first pooled image blocks through a second pooling layer to obtain second pooled image blocks; wherein a relationship between the second convolved image block and the first pooled image block satisfies a formula: fj =2(P +1) -Pp, where Fj denotes a size of a convolution kernel of the third-layer convolution layer performing a convolution operation on the second convolution image block, and Pp denotes a size of a kernel of the first pooling first-layer pooling layer performing a pooling operation on the third convolution image block; p represents a value of a filling manner, and is a preset constant; obtaining a first square difference between the pixel value of the second convolution image block and the pixel value of the first pooling image block; obtaining a second variance between the pixel value of the first convolution image block and the pixel value of the second pooling image block; taking a weighted sum between the first variance and the second variance as a first loss function; adjusting the values of the kernels of the three convolutional layers and the two pooling layers based on the first loss function; when the first loss function is converged, determining that the training of the convolution pooling structure is finished;
inputting all training target image blocks in the training set into a trained convolution pooling structure to obtain a plurality of second convolution images;
training a full convolution structure through a plurality of second convolution images, specifically: and training the full convolution structure based on the second convolution image blocks by taking the second convolution image blocks output by the third layer of convolution layer as the input of the full convolution structure, wherein the loss function of the full convolution structure is equal to the product of the cross entropy function and the first loss function, and when the loss function of the full convolution structure is converged, the feature extraction network training is determined to be finished.
Optionally, the adjusting values of the kernels of the three convolutional layers and the two pooling layers based on the first loss function includes:
adding the value of the loss function to the value of each point in the kernel of the three-layer convolutional layer on the basis of the original value;
and subtracting the value of the loss function from the original value of each point in the kernel of the two pooling layers.
Optionally, the screening out the feature vectors meeting the preset condition to form a target screening feature set, including:
obtaining a feature difference value between the feature vector and the shape feature of the standard defect, if the feature difference value is smaller than a threshold value, indicating that the shape feature accords with the shape feature of the standard defect, and arranging the feature vector in a matched target screening feature set;
wherein the standard defect shape features are extracted features based on images of tower stay wires of typical defects, and the typical defects comprise corrosion and strand breakage.
Optionally, the obtaining the defect similarity between the typical defect feature and each feature vector in the target screening feature set includes:
and calculating the cosine value of an included angle between the typical defect characteristic and the characteristic vector, and taking the cosine value of the included angle as the defect similarity.
Optionally, the determining whether the current tower stay wire has a strand breakage or a corrosion defect based on the defect similarity includes:
if the defect similarity is larger than or equal to a set value, determining that the current tower stay wire has a broken strand or corrosion defect;
and if the defect similarity is larger than a set value, determining that the current tower stay wire is normal.
Compared with the prior art, the embodiment of the invention achieves the following beneficial effects:
the embodiment of the invention also provides a tower and stay wire corrosion defect image detection processing method on the power transmission line, which comprises the following steps: collecting tower stay wire images on the power transmission line through an unmanned aerial vehicle; dividing a tower stay wire image through a self-adaptive tower stay wire defect detection model to obtain a plurality of image blocks, and taking the image blocks containing the defective tower stay wires as target image blocks; extracting a characteristic vector of a tower stay wire based on a target image block through a pre-trained characteristic extraction network; screening out the feature vectors meeting the preset conditions to form a target screening feature set; extracting typical defect characteristics of each tower stay wire in the tower stay wire image by a two-pass scanning method; obtaining defect similarity between the typical defect feature and each feature vector in the target screening feature set; and judging whether the current tower stay wire has the defects of strand breakage or corrosion based on the defect similarity.
By adopting the scheme, the tower stay wire defect detection model is self-adaptive, on one hand, the image area containing the defective tower stay wire can be accurately identified without a large amount of training data, and the tower stay wire defect identification efficiency is improved. In another aspect. Because a large amount of training is not needed, the adaptive tower stay wire defect detection model has strong applicability to defect detection of other defects or other objects, namely, the adaptive tower stay wire defect detection model has strong mobility. Even if the shapes of the tower stay wires and the shapes of the defects are changed, the tower stay wires can be detected and identified accurately.
In order to further judge the type of the defect, the defect similarity between the typical defect feature and each feature vector in the target screening feature set is judged, whether the current tower stay wire has a broken strand or corrosion defect is judged based on the defect similarity, effective reference and information are provided for maintenance personnel to maintain the tower stay wire, the maintenance personnel can conveniently carry appropriate maintenance tools and materials to go to field maintenance, and a large amount of manpower and material resources are saved.
Drawings
Fig. 1 is a flowchart of an image detection processing method for corrosion defects of tower and tower guy wires on a power transmission line according to an embodiment of the present invention.
Fig. 2 is a schematic structural diagram of a self-adaptive tower stay wire defect detection model provided by an embodiment of the invention.
Fig. 3 is a schematic structural diagram of another adaptive tower cable defect detection model provided in the embodiment of the present invention.
Fig. 4 is a schematic diagram of a feature extraction network structure according to an embodiment of the present invention.
Fig. 5 is a schematic block structure diagram of an electronic device according to an embodiment of the present invention.
The labels in the figure are: a bus 500; a receiver 501; a processor 502; a transmitter 503; a memory 504; a bus interface 505.
Detailed Description
The present invention will be described in detail below with reference to the accompanying drawings.
Example 1
The embodiment of the invention provides a method for detecting and processing images of corrosion defects of towers and pull wires on a power transmission line, which is used for carrying out detection video image or photo image processing on the corrosion defects of the pull wires in a computer vision mode and carrying out pull wire corrosion defect identification, and the method for detecting and processing the images of the corrosion defects of the towers and the pull wires on the power transmission line comprises the following steps as shown in figure 1:
s101: and acquiring tower stay wire images on the power transmission line through an unmanned aerial vehicle. Wherein, be provided with camera device among the unmanned aerial vehicle, for example be provided with the camera, can be colored camera or black and white night vision camera.
S102: and segmenting the tower stay wire image through the self-adaptive tower stay wire defect detection model to obtain a plurality of image blocks, and taking the image blocks containing the defective tower stay wires as target image blocks.
S103: and extracting the characteristic vector of the tower stay wire based on the target image block through a pre-trained characteristic extraction network.
S104: and screening the feature vectors meeting the preset conditions to form a target screening feature set.
S105: and extracting typical defect characteristics of each tower stay wire in the tower stay wire image by a two-pass scanning method.
S106: and obtaining the defect similarity between the typical defect feature and each feature vector in the target screening feature set.
S107: and judging whether the current tower stay wire has strand breakage or corrosion defects based on the defect similarity.
By adopting the scheme, the tower stay wire defect detection model is self-adaptive, on one hand, the image area containing the defective tower stay wire can be accurately identified without a large amount of training data, and the tower stay wire defect identification efficiency is improved. In another aspect. Because a large amount of training is not needed, the adaptive tower stay wire defect detection model has strong applicability to defect detection of other defects or other objects, namely, the adaptive tower stay wire defect detection model has strong mobility. Even if the shapes of the tower stay wires and the shapes of the defects are changed, the tower stay wires can be detected and identified accurately.
In order to further judge the type of the defect, the defect similarity between the typical defect feature and each feature vector in the target screening feature set is judged, whether the current tower stay wire has a broken strand or corrosion defect is judged based on the defect similarity, effective reference and information are provided for maintenance personnel to maintain the tower stay wire, the maintenance personnel can conveniently carry appropriate maintenance tools and materials to go to field maintenance, and a large amount of manpower and material resources are saved.
In order to improve the accuracy of tower stay wire defect detection and improve the efficiency of tower stay wire defect detection, the embodiment of the invention provides a self-adaptive tower stay wire defect detection model, which can accurately detect an image block of a tower stay wire containing a defect without carrying out a large amount of training data, and improves the accuracy of tower stay wire defect detection and the efficiency of tower stay wire defect detection. As shown in fig. 2, the adaptive tower guy line defect detection model provided in the embodiment of the present invention includes an input layer, an adaptive structure layer, and an output layer. Wherein the input layer comprises a convolution node and a pooling node. The adaptive structure layer at least comprises one convolution pooling layer, and each convolution pooling layer comprises one convolution node and one pooling node. The output layer comprises a convolution layer, a composite layer and a convolution neural network; the convolutional layer includes two convolutional nodes.
Regarding the structure of the adaptive tower stay wire defect detection model, an example in practical application is taken as an explanation: for example, a tower stay wire image is segmented by a self-adaptive tower stay wire defect detection model to obtain a plurality of image blocks, and the image blocks containing the defective tower stay wires are used as target image blocks, specifically:
and obtaining a first hog characteristic of the tower stay wire image.
The input layer includes a convolution node and a pooling node.
And carrying out convolution operation on the tower stay wire image input into the self-adaptive tower stay wire defect detection model through the convolution node of the input layer to obtain a first convolution image. A first convolved hog feature of the first convolved image is then obtained.
And performing pooling operation on the tower stay wire image input into the self-adaptive tower stay wire defect detection model through the pooling node of the input layer to obtain a first pooled image. A first pooled hog feature of the first pooled image is then obtained.
A first difference equal to a difference between the first hog feature and the first convolved hog feature and a second difference equal to a difference between the first hog feature and the first pooled hog feature are obtained.
The adaptive connection between the input layer and the adaptive structure layer is explained as follows:
if the first difference is smaller than or equal to the second difference, the input of the self-adaptive structural layer is a first convolution image. I.e. the connection relationship of the input layer of the adaptation domain is the structure shown by the solid line connection in fig. 2. If the first difference is greater than the second difference, the input of the adaptive structural layer is the first pooled image. I.e. the connection relationship of the input layer of the adaptation domain is the structure shown by the dashed connection in fig. 2. Therefore, the adaptivity of the connection relation between the input layer and the adaptive structure layer is embodied, namely the adaptivity of the adaptive structure layer for processing tower stay wire image information is embodied.
The adaptive connection relation between the input layer and the adaptive structure is described based on the data flow direction and processing angles, and specifically comprises the following steps:
if the self-adaptive structure layer has a convolution pooling layer, the specific connection structure is as follows: if the first difference is less than or equal to the second difference, the input of the adaptive structural layer is the first convolution image. Namely, the convolution node of the layer convolution pooling layer performs convolution operation on the first convolution image to obtain a second convolution image. Obtaining a second convolution hog feature of a second convolution image; pooling the first convolution image by a pooling node of the layer convolution pooling layer to obtain a second pooling image; a second pooled hog feature of the second pooled image is obtained. Obtaining a third difference value and a fourth difference value; the third difference is equal to the difference between the second convolved hog feature and the first convolved hog feature. If the first difference is larger than the second difference, the input of the self-adaptive structural layer is the first pooled image, and the convolution node of the layer convolution pooled layer performs convolution operation on the first pooled image to obtain a second convolution image; obtaining a second convolution hog characteristic of a second convolution image, and performing pooling operation on the first pooled image by a pooling node of a layer convolution pooling layer to obtain a second pooled image; a second pooled hog feature of a second pooled image is obtained.
If the self-adaptive structure layer has a plurality of convolution pooling layers, the first convolution pooling layer of the self-adaptive structure layer is connected with the input layer in a self-adaptive mode, and the method specifically comprises the following steps:
and carrying out convolution operation on the first convolution image by the convolution node of the first convolution pooling layer to obtain a second convolution image and obtain a second convolution hog characteristic of the second convolution image. Pooling operation is carried out on the first convolution image by the pooling node of the first layer of convolution pooling layer to obtain a second pooling image, and a second pooling hog characteristic of the second pooling image is obtained. The connection relationship between the input layer and the adaptive structure layer has been clarified for the case where the first difference is less than or equal to the second difference. In the embodiment of the invention, not only the connection relationship between the input layer and the adaptive structural layer is adaptive, but also the connection relationship between each convolution pooling layer inside the adaptive structural layer is adaptive when the adaptive structural layer comprises a plurality of convolution pooling layers.
Returning to the explanation of the connection relationship between the input layer and the adaptive structure layer. If the first difference is greater than the second difference, the input of the adaptive structural layer is the first pooled image. I.e. the connection relationship of the input layer of the adaptation domain is the structure shown by the dashed connection in fig. 2. Specifically, the convolution node of the layer convolution pooling layer performs convolution operation on the first pooled image to obtain a second convolution image, and a second convolution hog feature of the second convolution image is obtained. And performing pooling operation on the first pooled image by the pooling node of the layer convolution pooling layer to obtain a second pooled image, and obtaining a second pooling hog characteristic of the second pooled image.
The adaptive structure layer at least comprises one convolution pooling layer, each convolution pooling layer comprises one convolution node and one pooling node, and if the adaptive structure layer only comprises one convolution pooling layer, the convolution pooling layer of the adaptive structure layer is adaptively connected with the input layer. If the self-adaptive structure layer has a plurality of convolution pooling layers, the first convolution pooling layer of the self-adaptive structure layer is self-adaptively connected with the input layer, and then the connection relationship among the plurality of convolution pooling layers of the self-adaptive structure layer is also self-adaptive.
In the above, it has been clarified that, when the adaptive structure layer has only one convolution pooling layer, the adaptive connection relationship between the input layer and the adaptive structure layer is in both cases where the first difference is greater than the second difference and the first difference is less than or equal to the second difference.
Then referring to fig. 3, the adaptive connection relationship between convolution pooling layers within the adaptive structural layer when there are multiple convolution pooling layers within the adaptive structural layer will be explained next.
The adaptive connection relationship between convolution pooling layers inside the adaptive structure layer can be determined by referring to the adaptive connection mode, and the connection relationship between the next convolution pooling layer is determined correspondingly, specifically: obtaining a third difference value and a fourth difference value, wherein the third difference value is equal to the difference between the second convolved hog feature and the first convolved hog feature, and the fourth difference value is equal to the difference between the second pooled hog feature and the first pooled hog feature. If the third difference is less than or equal to the fourth difference, the input of the next convolution pooling layer (the second convolution pooling layer) is a second convolution image, as shown by the solid line in fig. 3. If the third difference is greater than the fourth difference, the input to the next convolutional pooling layer is a second pooled image, as shown by the dashed line in FIG. 3. And analogizing in sequence until the last layer of convolution pooling layer processes the output of the previous layer of convolution pooling layer, wherein the output of the last layer of convolution pooling layer is also two, one is the output of the convolution node of the last layer of convolution pooling layer, the other is the output of the pooling node of the last layer of convolution pooling layer, and the outputs of the two are both images. In the embodiment of the present invention, the number of layers of the adaptive structure layer including the convolution pooling layer may be 2, 3, 4, 5.
The output layer comprises a convolutional layer, a composite layer and a full convolutional neural network.
The convolution layer comprises two convolution nodes, the input of the two convolution nodes of the convolution layer is respectively a convolution image and a pooling image output by the last convolution pooling layer of the self-adaptive structure layer, and the two convolution nodes of the convolution layer respectively carry out convolution operation on the convolution image and the pooling image output by the last convolution pooling layer. Namely: and the two convolution nodes of the convolution layer respectively carry out convolution operation on the last layer of convolution image and the last layer of pooling image to respectively and correspondingly obtain the convolution image to be fused and the pooling image to be fused.
And then, the composite layer fuses the convolution image output by the last convolution pooling layer after the convolution operation and the pooling image to obtain a fused image. The specific composite layer performs pixel value superposition on corresponding pixel points of the convolution image to be fused and the pooling image to be fused to obtain a fused image. Therefore, the fused image contains the local features and the global features in the original tower stay wire image, and the accuracy of defect detection based on the fused image is improved.
And finally, the convolutional neural network performs segmentation classification output on the fused image, and the image block containing the tower stay wire with the defect is used as a target image block. In the embodiment of the present invention, the Convolutional neural network is trained in advance, and the Convolutional neural network may be a Full Convolutional Network (FCN).
Based on the self-adaptive tower stay wire defect detection model, the input of the internal nodes of the self-adaptive tower stay wire defect detection model can be self-adaptively selected according to the loss condition of data, so that the image finally used for identifying the defects has enough local characteristics and global characteristics of tower stay wires, and the accuracy of defect identification is improved. In addition, the self-adaptive tower stay wire defect detection model has self-adaptability to data processing, the mobility of the self-adaptive tower stay wire defect detection model is improved, the self-adaptive tower stay wire defect detection model can process different data, and when the scene changes, the defect identification accuracy of the self-adaptive tower stay wire defect detection model is still high.
After identifying a tower pull wire as defective, the specific type of defect is also identified. The method specifically comprises the following steps:
firstly, extracting the characteristics of the defects through a characteristic extraction network. The feature extraction network comprises a convolution pooling structure and a full convolution structure, wherein the convolution pooling structure comprises three convolution layers and two pooling layers. Reference may be made to fig. 4. The structure of the feature extraction network is explained below with reference to fig. 4.
In the embodiment of the present invention, the feature extraction network is a field training method for a feature extraction network, and the specific training method for a feature extraction network includes:
obtaining a training set, wherein the training set comprises a plurality of training target image blocks; the values of the initialized cores of the three convolutional layers and the two pooling layers are random numbers.
Randomly selecting a training target image block to train the convolution pooling structure, and specifically comprising the following steps: and inputting the training target image block into a first layer of convolution layer, and performing convolution operation on the training target image block through the first layer of convolution layer to obtain a first convolution image block. And performing convolution operation on the first convolution image block through the second convolution layer to obtain a second convolution image block. And performing convolution operation on the second convolution image block through the third layer of convolution layer to obtain a third convolution image block. And performing pooling operation on the third convolution image block through the first pooling layer to obtain a first pooled image block. And performing pooling operation on the first pooled image block through the second pooling layer to obtain a second pooled image block. Wherein a relationship between the second convolved image block and the first pooled image block satisfies a formula: fj =2(P +1) -Pp, where Fj denotes a size of a convolution kernel of the third-layer convolution layer performing a convolution operation on the second convolution image block, and Pp denotes a size of a kernel of the first pooling first-layer pooling layer performing a pooling operation on the third convolution image block; p represents a value of the filling method and is a predetermined constant. In the embodiment of the present invention, the step length of the convolution and pooling operations is 1 pixel point, the value of P may be 0 or 1, which indicates whether the image is to be expanded after the convolution or pooling operation, and the expansion mode, which will not be elaborated again. With the above formula, the same size between the second convolved image block and the first pooled image block can be satisfied.
In the embodiment of the present invention, in order to maintain the scale invariance of the convolved-pooled image and further obtain the first loss function with high accuracy, the embodiment of the present invention determines that the sizes of the first convolved image block, the second convolved image block, the third convolved image block, the first pooled image block, and the second pooled image block are the same.
Obtaining a first variance between the pixel value of the second convolved image block and the pixel value of the first pooled image block, obtaining a second variance between the pixel value of the first convolved image block and the pixel value of the second pooled image block, and taking a weighted sum between the first variance and the second variance as a first loss function. The following formula is shown in detail: loss1= a × D1+ (1-a) × D2, Loss1 represents the first Loss function, D1 represents the first variance, D2 represents the second variance, a represents the weighted weight, and a takes on an arbitrary number between 0 and 1. Typically, a = 0.6.
And adjusting the values of the cores of the three convolutional layers and the two pooling layers based on the first loss function. And after the values of the kernels of the three convolutional layers and the two pooling layers are adjusted, carrying out convolution and pooling on the training target image block by the three convolutional layers and the two pooling layers according to the flow again until the obtained first loss function is converged. And when the first loss function converges, determining that the training of the convolution pooling structure is finished.
In this embodiment of the present invention, the adjusting the values of the kernels of the three convolutional layers and the two pooling layers based on the first loss function includes:
adding the value of the loss function to the value of each point in the kernel of the three-layer convolutional layer on the basis of the original value; and subtracting the value of the loss function from the original value of each point in the kernel of the two pooling layers.
Therefore, the global features can be weakened, the local features can be strengthened, the finally output second convolution image contains rich and obvious defect characteristics, the characteristics of the defects can be accurately expressed based on the features extracted from the second convolution image, and a foundation is laid for further improving the accuracy of defect identification.
By adopting the scheme, the three layers of convolution layers of the feature extraction network can be trained by only one training target image block, the training data volume is small, and the feature extraction speed of the feature extraction network is improved.
After the three layers of convolutional layers of the feature extraction network are trained, all training target image blocks in a training set are input into the convolutional pooling structure trained in the process, and a plurality of second convolutional images are correspondingly obtained.
Training a full convolution structure through a plurality of second convolution images, specifically: and taking the second convolution image blocks output by the third layer of convolution layer as the input of the full convolution structure, training the full convolution structure based on the second convolution image blocks, wherein the Loss function of the full convolution structure is equal to the product of the cross entropy function and the first Loss function, namely Loss2= Loss 1L, Loss2 represents the Loss function of the full convolution structure, and L represents the cross entropy function.
And when the loss function of the full convolution structure is converged, determining that the feature extraction network training is finished. In an embodiment of the invention, the full convolution structure may be a full convolution neural network.
By adopting the scheme, the calculated amount in the training process can be reduced, and the feature extraction accuracy of the feature extraction network is improved.
Extracting the characteristic vector of the tower stay wire based on the target image block through a pre-trained characteristic extraction network, which specifically comprises the following steps: inputting the target image block into a convolution pooling structure comprising three convolution layers, inputting the output of the last convolution layer into a full convolution structure, and extracting the characteristic vector of the tower stay wire by the full convolution structure.
The feature vectors meeting the preset conditions are screened out to form a target screening feature set, and the method comprises the following steps:
and obtaining a feature difference value between the feature vector and the shape feature of the standard defect, if the feature difference value is smaller than a threshold value, indicating that the shape feature accords with the shape feature of the standard defect, and setting the feature vector in a screening feature set matched with a target.
Wherein the standard defect shape features are extracted features based on images of tower stay wires of typical defects, and the typical defects comprise corrosion and strand breakage.
The obtaining of the defect similarity between the typical defect feature and each feature vector in the target screening feature set may be: and calculating the cosine value of an included angle between the typical defect characteristic and the characteristic vector, and taking the cosine value of the included angle as the defect similarity.
Whether the current tower stay wire has a broken strand or corrosion defect or not is judged based on the defect similarity, and the method specifically comprises the following steps: and if the defect similarity is larger than or equal to a set value, determining that the current tower stay wire has a broken strand or corrosion defect, and if the defect similarity is larger than the set value, determining that the current tower stay wire is normal.
Through adopting above scheme, can accurately discern the shaft tower and act as go-between to accurately discern the defective shaft tower and act as go-between, still discern the type of defect in the shaft tower is acted as go-between, maintainer acts as go-between to the shaft tower and maintains and embodies and provide effectual reference and information, and maintainer carries suitable repair tools and material to the field maintenance of being convenient for, has saved a large amount of manpower and materials.
Example 2
Based on the method for detecting and processing the corrosion defect image of the tower and the stay wire on the power transmission line, the embodiment of the invention also provides a system for detecting and processing the corrosion defect image of the tower and the stay wire on the power transmission line, which is used for executing the method for detecting and processing the corrosion defect image of the tower and the stay wire on the power transmission line. The image acquisition module is used for acquiring tower stay wire images on the power transmission line through the unmanned aerial vehicle. And the defect identification module is used for segmenting the tower stay wire image through the self-adaptive tower stay wire defect detection model to obtain a plurality of image blocks, and taking the image blocks containing the defective tower stay wires as target image blocks. The defect classification module is used for extracting a feature vector of the tower stay wire based on the target image block through a pre-trained feature extraction network; screening out the feature vectors meeting the preset conditions to form a target screening feature set; extracting typical defect characteristics of each tower stay wire in the tower stay wire image by a two-pass scanning method; obtaining defect similarity between the typical defect feature and each feature vector in the target screening feature set; and judging whether the current tower stay wire has the defects of strand breakage or corrosion based on the defect similarity.
The specific manner in which the respective modules perform operations has been described in detail in the embodiments related to the method, and will not be elaborated upon here.
An embodiment of the present invention further provides an electronic device, as shown in fig. 5, which includes a memory 504, a processor 502, and a computer program stored in the memory 504 and executable on the processor 502, where the processor 502 executes the program to implement the steps of any one of the foregoing methods for detecting and processing the corrosion defect image of the tower and cable on the power transmission line.
Where in fig. 5 a bus architecture (represented by bus 500) is shown, bus 500 may include any number of interconnected buses and bridges, and bus 500 links together various circuits including one or more processors, represented by processor 502, and memory, represented by memory 504. The bus 500 may also link together various other circuits such as peripherals, voltage regulators, power management circuits, and the like, which are well known in the art, and therefore, will not be described any further herein. A bus interface 505 provides an interface between the bus 500 and the receiver 501 and transmitter 503. The receiver 501 and the transmitter 503 may be the same element, i.e. a transceiver, providing a means for communicating with various other apparatus over a transmission medium. The processor 502 is responsible for managing the bus 500 and general processing, and the memory 504 may be used for storing data used by the processor 502 in performing operations.
The embodiment of the invention also provides a computer readable storage medium, on which a computer program is stored, which when executed by a processor, implements the steps of any one of the methods for detecting and processing the corrosion defect image of the tower and the stay wire on the power transmission line, and the above mentioned data.
The algorithms and displays presented herein are not inherently related to any particular computer, virtual machine, or other apparatus. Various general purpose systems may also be used with the teachings herein. The required structure for constructing such a system will be apparent from the description above. Moreover, the present invention is not directed to any particular programming language. It is appreciated that a variety of programming languages may be used to implement the teachings of the present invention as described herein, and any descriptions of specific languages are provided above to disclose the best mode of the invention.
In the description provided herein, numerous specific details are set forth. It is understood, however, that embodiments of the invention may be practiced without these specific details. In some instances, well-known methods, structures and techniques have not been shown in detail in order not to obscure an understanding of this description.
Similarly, it should be appreciated that in the foregoing description of exemplary embodiments of the invention, various features of the invention are sometimes grouped together in a single embodiment, figure, or description thereof for the purpose of streamlining the disclosure and aiding in the understanding of one or more of the various inventive aspects. However, the disclosed method should not be interpreted as reflecting an intention that: that is, the claimed invention requires more features than are expressly recited in a claim. Rather, as the following claims reflect, inventive aspects lie in less than all features of a single foregoing disclosed embodiment. Thus, the claims following the detailed description are hereby expressly incorporated into this detailed description, with the claims themselves being directed to separate embodiments of the present invention.
Those skilled in the art will appreciate that the modules in the device in an embodiment may be adaptively changed and disposed in one or more devices different from the embodiment. The modules or units or components of the embodiments may be combined into one module or unit or component, and furthermore they may be divided into a plurality of sub-modules or sub-units or sub-components. All of the features disclosed in this specification (including any accompanying claims, abstract and drawings), and all of the processes or elements of any method or apparatus so disclosed, may be combined in any combination, except combinations where at least some of such features and/or processes or elements are mutually exclusive. The features disclosed in this specification (including any accompanying claims, abstract and drawings) may be replaced by alternative features serving the same, equivalent or similar purpose, unless expressly stated otherwise.
Furthermore, those skilled in the art will appreciate that while some embodiments herein include some features included in other embodiments, rather than other features, combinations of features of different embodiments are meant to be within the scope of the invention and form different embodiments. For example, in the following claims, any of the claimed embodiments may be used in any combination.
The various component embodiments of the invention may be implemented in hardware, or in software modules running on one or more processors, or in a combination thereof. Those skilled in the art will appreciate that a microprocessor or Digital Signal Processor (DSP) may be used in practice to implement some or all of the functionality of some or all of the components in an apparatus according to an embodiment of the invention. The present invention may also be embodied as apparatus or device programs (e.g., computer programs and computer program products) for performing a portion or all of the methods described herein. Such programs implementing the present invention may be stored on computer-readable media or may be in the form of one or more signals. Such a signal may be downloaded from an internet website, or provided on a carrier signal, or provided in any other form.
It should be noted that the above-mentioned embodiments illustrate rather than limit the invention, and that those skilled in the art will be able to design alternative embodiments without departing from the scope of the appended claims. In the claims, any reference signs placed between parentheses shall not be construed as limiting the claim. The word "comprising" does not exclude the presence of elements or steps not listed in a claim. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. The invention may be implemented by means of hardware comprising several distinct elements, and by means of a suitably programmed computer. In the unit claims enumerating several means, several of these means may be embodied by one and the same item of hardware. The usage of the words first, second and third, etcetera do not indicate any ordering. These words may be interpreted as names.

Claims (10)

1. A method for detecting and processing corrosion defect images of tower guy wires on a transmission line is characterized by comprising the following steps:
collecting tower stay wire images on the power transmission line through an unmanned aerial vehicle;
dividing a tower stay wire image through a self-adaptive tower stay wire defect detection model to obtain a plurality of image blocks, and taking the image blocks containing the defective tower stay wires as target image blocks;
extracting a characteristic vector of a tower stay wire based on a target image block through a pre-trained characteristic extraction network;
screening out the feature vectors meeting the preset conditions to form a target screening feature set;
extracting typical defect characteristics of each tower stay wire in the tower stay wire image by a two-pass scanning method;
obtaining defect similarity between the typical defect feature and each feature vector in the target screening feature set;
and judging whether the current tower stay wire has the defects of strand breakage or corrosion based on the defect similarity.
2. The image detection processing method for corrosion defects of tower and tower guy wires on a transmission line according to claim 1,
the self-adaptive tower stay wire defect detection model comprises an input layer, a self-adaptive structural layer and an output layer;
the input layer comprises a convolution node and a pooling node;
the self-adaptive structure layer at least comprises one convolution pooling layer, and each convolution pooling layer comprises a convolution node and a pooling node;
the output layer comprises a convolution layer, a composite layer and a convolution neural network; the convolutional layer includes two convolutional nodes.
3. The method for detecting and processing the corrosion defect image of the tower stay wire on the transmission line according to claim 2, wherein if the adaptive structure layer comprises a convolution pooling layer, the tower stay wire image is segmented by the adaptive tower stay wire defect detection model to obtain a plurality of image blocks, and the image blocks containing the tower stay wire with the defects are used as target image blocks, comprising the following steps:
obtaining a first hog characteristic of a tower stay wire image;
carrying out convolution operation on the tower stay wire image input into the self-adaptive tower stay wire defect detection model through the convolution node of the input layer to obtain a first convolution image; obtaining a first convolution hog feature of a first convolution image;
performing pooling operation on tower stay wire images input into the self-adaptive tower stay wire defect detection model through a pooling node of an input layer to obtain first pooled images; obtaining a first pooling hog feature of the first pooled image;
obtaining a first difference value and a second difference value; the first difference is equal to a difference between the first hog feature and the first convolution hog feature; the second difference is equal to a difference between the first hog feature and the first pooled hog feature;
if the first difference is smaller than or equal to the second difference, the input of the self-adaptive structural layer is a first convolution image;
carrying out convolution operation on the first convolution image by the convolution node of the layer convolution pooling layer to obtain a second convolution image; obtaining a second convolution hog feature of a second convolution image;
pooling the first convolution image by a pooling node of the layer convolution pooling layer to obtain a second pooling image; obtaining a second pooling hog feature of the second pooled image;
obtaining a third difference value and a fourth difference value; the third difference is equal to the difference between the second convolved hog feature and the first convolved hog feature; the fourth difference is equal to the difference between the second pooled hog signature and the first pooled hog signature;
if the first difference is greater than the second difference, inputting the self-adaptive structural layer into a first pooling image;
carrying out convolution operation on the first pooled image by the convolution node of the layer convolution pooling layer to obtain a second convolution image; obtaining a second convolution hog feature of a second convolution image;
pooling nodes of the layer convolution pooling layer perform pooling operation on the first pooled image to obtain a second pooled image; obtaining a second pooling hog feature of the second pooled image;
obtaining a third difference value and a fourth difference value; the third difference is equal to the difference between the second convolved hog signature and the first convolved hog signature; the fourth difference is equal to the difference between the second pooled hog signature and the first pooled hog signature;
the input of the two convolution nodes of the convolution layer is respectively a convolution image and a pooled image output by the last convolution pooling layer of the self-adaptive structural layer, and the two convolution nodes of the convolution layer respectively carry out convolution operation on the convolution image and the pooled image output by the last convolution pooling layer;
the composite layer fuses the convolution image and the pooled image output by the last convolution pooling layer after the convolution operation to obtain a fused image;
and the convolution neural network carries out segmentation classification output on the fused image, and takes the image block containing the tower stay wire with the defect as a target image block, wherein the full convolution neural network is trained in advance.
4. The method for detecting and processing the corrosion defect image of the tower stay wire on the transmission line according to claim 1, wherein if the adaptive structure layer comprises a plurality of convolution pooling layers, the tower stay wire image is segmented by the adaptive tower stay wire defect detection model to obtain a plurality of image blocks, and the image blocks containing the tower stay wire with the defects are used as target image blocks, further comprising:
if the third difference is smaller than or equal to the fourth difference, the input of the next convolution pooling layer is a second convolution image;
and if the third difference is larger than the fourth difference, the input of the next convolution pooling layer is a second pooling image.
5. The image detection processing method for the corrosion defects of the tower guy wires on the power transmission line according to claim 1, wherein the feature extraction network comprises a convolution pooling structure and a full convolution structure, and the convolution pooling structure comprises three convolution layers and two pooling layers.
6. The image detection processing method for the corrosion defect of the tower stay wire on the transmission line according to claim 5, wherein the training method for the feature extraction network comprises the following steps:
obtaining a training set, wherein the training set comprises a plurality of training target image blocks; the initialized core of the three convolution layers and the two pooling layers takes the value of a random number;
randomly selecting a training target image block to train the convolution pooling structure, and specifically comprising the following steps: inputting the training target image block into a first layer of convolutional layer, and performing convolutional operation on the training target image block through the first layer of convolutional layer to obtain a first convolutional image block; performing convolution operation on the first convolution image block through the second convolution layer to obtain a second convolution image block; performing convolution operation on the second convolution image block through the third layer of convolution layer to obtain a third convolution image block; performing pooling operation on the third convolution image block through the first pooling layer to obtain a first pooled image block; performing pooling operation on the first pooled image block through a second pooling layer to obtain a second pooled image block; wherein a relationship between the second convolved image block and the first pooled image block satisfies a formula: fj =2(P +1) -Pp, where Fj denotes a size of a convolution kernel of the third-layer convolution layer performing a convolution operation on the second convolution image block, and Pp denotes a size of a kernel of the first pooling first-layer pooling layer performing a pooling operation on the third convolution image block; p represents a value of a filling manner, and is a preset constant; obtaining a first square difference between the pixel value of the second convolution image block and the pixel value of the first pooling image block; obtaining a second variance between the pixel value of the first convolution image block and the pixel value of the second pooling image block; taking a weighted sum between the first variance and the second variance as a first loss function; adjusting the values of the kernels of the three convolutional layers and the two pooling layers based on the first loss function; when the first loss function is converged, determining that the training of the convolution pooling structure is finished;
inputting all training target image blocks in the training set into a trained convolution pooling structure to obtain a plurality of second convolution images;
training a full convolution structure through a plurality of second convolution images, specifically: and training the full convolution structure based on the second convolution image blocks by taking the second convolution image blocks output by the third layer of convolution layer as the input of the full convolution structure, wherein the loss function of the full convolution structure is equal to the product of the cross entropy function and the first loss function, and when the loss function of the full convolution structure is converged, the feature extraction network training is determined to be finished.
7. The image detection processing method for corrosion defects of tower guy wires on transmission lines according to claim 5, wherein the adjusting of the values of the kernels of the three convolutional layers and the two pooling layers based on the first loss function comprises:
adding the value of the loss function to the value of each point in the kernel of the three-layer convolutional layer on the basis of the original value;
and subtracting the value of the loss function from the original value of each point in the kernel of the two pooling layers.
8. The method for detecting and processing the corrosion defect image of the tower stay wire on the transmission line according to claim 1, wherein the step of screening the feature vectors meeting the preset condition to form a target screening feature set comprises the following steps:
obtaining a feature difference value between the feature vector and the shape feature of the standard defect, if the feature difference value is smaller than a threshold value, indicating that the shape feature accords with the shape feature of the standard defect, and arranging the feature vector in a matched target screening feature set;
wherein the standard defect shape features are extracted features based on images of tower stay wires of typical defects, and the typical defects comprise corrosion and strand breakage.
9. The method for detecting and processing the corrosion defect image of the tower and the stay wire on the power transmission line according to claim 1, wherein the obtaining the defect similarity between the typical defect feature and each feature vector in the target screening feature set comprises:
and calculating the cosine value of an included angle between the typical defect characteristic and the characteristic vector, and taking the cosine value of the included angle as the defect similarity.
10. The method for detecting and processing the corrosion defect images of the tower stay wires on the electric transmission line according to claim 1, wherein the step of judging whether the current tower stay wire has a strand breakage or a corrosion defect based on the defect similarity comprises the following steps:
if the defect similarity is larger than or equal to a set value, determining that the current tower stay wire has a strand breakage or corrosion defect;
and if the defect similarity is larger than a set value, determining that the current tower stay wire is normal.
CN202210635427.3A 2022-06-07 2022-06-07 Image detection processing method for corrosion defect of tower stay wire on power transmission line Active CN114708267B (en)

Priority Applications (2)

Application Number Priority Date Filing Date Title
CN202210635427.3A CN114708267B (en) 2022-06-07 2022-06-07 Image detection processing method for corrosion defect of tower stay wire on power transmission line
GB2219702.4A GB2619576A (en) 2022-06-07 2022-12-23 Method for detecting and processing corrosion defect image of tower guy wire on power transmission line

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210635427.3A CN114708267B (en) 2022-06-07 2022-06-07 Image detection processing method for corrosion defect of tower stay wire on power transmission line

Publications (2)

Publication Number Publication Date
CN114708267A true CN114708267A (en) 2022-07-05
CN114708267B CN114708267B (en) 2022-09-13

Family

ID=82178149

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210635427.3A Active CN114708267B (en) 2022-06-07 2022-06-07 Image detection processing method for corrosion defect of tower stay wire on power transmission line

Country Status (2)

Country Link
CN (1) CN114708267B (en)
GB (1) GB2619576A (en)

Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180253836A1 (en) * 2015-06-16 2018-09-06 South China University Of Technology Method for automated detection of defects in cast wheel products
CN108596470A (en) * 2018-04-19 2018-09-28 浙江大学 A kind of power equipments defect text handling method based on TensorFlow frames
CN109064462A (en) * 2018-08-06 2018-12-21 长沙理工大学 A kind of detection method of surface flaw of steel rail based on deep learning
CN110148117A (en) * 2019-04-22 2019-08-20 南方电网科学研究院有限责任公司 Power equipments defect recognition methods, device and storage medium based on electric power image
CN111311569A (en) * 2020-02-12 2020-06-19 江苏方天电力技术有限公司 Pole tower defect identification method based on unmanned aerial vehicle inspection
CN111402203A (en) * 2020-02-24 2020-07-10 杭州电子科技大学 Fabric surface defect detection method based on convolutional neural network
CN111784633A (en) * 2020-05-26 2020-10-16 西安理工大学 Insulator defect automatic detection algorithm for power inspection video
US20210334587A1 (en) * 2018-09-04 2021-10-28 Boe Technology Group Co., Ltd. Method and apparatus for training a convolutional neural network to detect defects
CN114581388A (en) * 2022-02-24 2022-06-03 国能包神铁路集团有限责任公司 Contact net part defect detection method and device

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104483326B (en) * 2014-12-19 2017-12-22 长春工程学院 High-voltage line defects of insulator detection method and system based on depth belief network
CN114596375A (en) * 2020-11-19 2022-06-07 英业达科技有限公司 Method for generating reconstructed image
CN112819784A (en) * 2021-02-01 2021-05-18 广东电网有限责任公司广州供电局 Method and system for detecting broken strands and scattered strands of wires of distribution line
CN112884741B (en) * 2021-02-22 2023-01-24 西安理工大学 Printing apparent defect detection method based on image similarity comparison
CN114049316A (en) * 2021-11-03 2022-02-15 青岛明思为科技有限公司 Steel wire rope defect detection method based on metallic luster area

Patent Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180253836A1 (en) * 2015-06-16 2018-09-06 South China University Of Technology Method for automated detection of defects in cast wheel products
CN108596470A (en) * 2018-04-19 2018-09-28 浙江大学 A kind of power equipments defect text handling method based on TensorFlow frames
CN109064462A (en) * 2018-08-06 2018-12-21 长沙理工大学 A kind of detection method of surface flaw of steel rail based on deep learning
US20210334587A1 (en) * 2018-09-04 2021-10-28 Boe Technology Group Co., Ltd. Method and apparatus for training a convolutional neural network to detect defects
CN110148117A (en) * 2019-04-22 2019-08-20 南方电网科学研究院有限责任公司 Power equipments defect recognition methods, device and storage medium based on electric power image
CN111311569A (en) * 2020-02-12 2020-06-19 江苏方天电力技术有限公司 Pole tower defect identification method based on unmanned aerial vehicle inspection
CN111402203A (en) * 2020-02-24 2020-07-10 杭州电子科技大学 Fabric surface defect detection method based on convolutional neural network
CN111784633A (en) * 2020-05-26 2020-10-16 西安理工大学 Insulator defect automatic detection algorithm for power inspection video
CN114581388A (en) * 2022-02-24 2022-06-03 国能包神铁路集团有限责任公司 Contact net part defect detection method and device

Non-Patent Citations (6)

* Cited by examiner, † Cited by third party
Title
HUAGANG LIANG, ET AL: "Detection and Evaluation Method of Transmission Line Defects Based on Deep Learning", 《IEEE ACCESS》 *
SONGBO CHEN, ET AL: "Insulator Fault Detection in Aerial Images based on the Mixed-grouped Fire Single-shotMultibox Detector", 《JOURNAL OF IMAGING SCIENCE AND TECHNOLOGY》 *
冯斌等: "基于BiLSTM-Attention神经网络的电力设备缺陷文本挖掘", 《中国电机工程学报》 *
徐守坤等: "基于HOG-CSLBP及YOLOv2的行人检测", 《计算机工程与设计》 *
董召杰等: "基于边缘计算的机巡图像缺陷识别算法研究", 《自动化与仪器仪表》 *
金昊等: "基于Faster Ｒ-CNN的高压电线缺陷检测方法", 《计算机应用》 *

Also Published As

Publication number Publication date
GB2619576A (en) 2023-12-13
GB202219702D0 (en) 2023-02-08
CN114708267B (en) 2022-09-13

Similar Documents

Publication Publication Date Title
CN112183788B (en) Domain adaptive equipment operation detection system and method
CN109977921B (en) Method for detecting hidden danger of power transmission line
CN111179249A (en) Power equipment detection method and device based on deep convolutional neural network
CN110533086B (en) Semi-automatic image data labeling method
CN107179479B (en) Visible light image-based transmission conductor strand scattering defect detection method
CN115294476B (en) Edge computing intelligent detection method and device for unmanned aerial vehicle power inspection
CN110910440B (en) Power transmission line length determination method and system based on power image data
CN112101138A (en) Bridge inhaul cable surface defect real-time identification system and method based on deep learning
CN109389322A (en) The disconnected broken lot recognition methods of grounded-line based on target detection and long memory models in short-term
CN115147383A (en) Insulator state rapid detection method based on lightweight YOLOv5 model
CN112327930A (en) Routing inspection path determining method and device
CN111524121A (en) Road and bridge fault automatic detection method based on machine vision technology
CN109902730B (en) Power transmission line broken strand detection method based on deep learning
CN114965328A (en) Ultraviolet multispectral high-voltage line fault detection and positioning method and device
CN113962973A (en) Power transmission line unmanned aerial vehicle intelligent inspection system and method based on satellite technology
Manninen et al. Multi-stage deep learning networks for automated assessment of electricity transmission infrastructure using fly-by images
CN114119454A (en) Device and method for smoke detection of power transmission line
CN114708267B (en) Image detection processing method for corrosion defect of tower stay wire on power transmission line
CN113393442A (en) Method and system for detecting abnormality of train parts, electronic device and storage medium
CN116071299A (en) Insulator RTV spraying defect detection method and system
CN116309303A (en) Electrical equipment defect detection method and device based on infrared image and related equipment
CN114418968A (en) Power transmission line small target defect detection method based on deep learning
CN110782431B (en) High-voltage wire icing area detection method based on deep learning
CN110705695A (en) Method, device, equipment and storage medium for searching model structure
CN116188998B (en) Method, device, equipment and storage medium for identifying defects of overhead transmission line

Legal Events

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