CN109977921B - Method for detecting hidden danger of power transmission line - Google Patents
Method for detecting hidden danger of power transmission line Download PDFInfo
- Publication number
- CN109977921B CN109977921B CN201910289546.6A CN201910289546A CN109977921B CN 109977921 B CN109977921 B CN 109977921B CN 201910289546 A CN201910289546 A CN 201910289546A CN 109977921 B CN109977921 B CN 109977921B
- Authority
- CN
- China
- Prior art keywords
- ground
- feature
- remote sensing
- training
- power transmission
- 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.)
- Active
Links
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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/13—Satellite images
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Biophysics (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Astronomy & Astrophysics (AREA)
- Remote Sensing (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The application discloses a method for detecting hidden danger of a power transmission line, which comprises the following steps: acquiring a to-be-detected map shot by satellite inspection; inputting the mapping to be detected into a ground feature classification model to obtain a classification result map output by the ground feature classification model; determining the hidden line danger in each ground object according to the distance between each ground object and the power transmission line in the classification result graph; the ground feature classification model is obtained through the following steps: s1, labeling different ground objects in the original remote sensing image in advance to obtain a labeled remote sensing image; s2, establishing a training set by using the marked remote sensing image and the corresponding original remote sensing image; s3, training the U-Net network by a training set; s4, carrying out accuracy rate inspection on the model obtained after training through the verification set, if the accuracy rate is smaller than a preset value, adjusting the parameters of the current model, and returning to the step S3; if the accuracy is greater than or equal to a preset value, outputting a ground object classification model; the method solves the technical problem that different detection methods are required to be designed according to different hidden danger types of the existing detection method.
Description
Technical Field
The application relates to the technical field of line inspection, in particular to a method for detecting hidden troubles of a power transmission line.
Background
With the continuous expansion of the demand of each business on electric power, the loss caused by power failure in accidents is more and more large. The overhead transmission line is directly built on the ground and is in direct contact with the surrounding environment, so that the overhead transmission line is more easily influenced and frequent in accidents.
How to find, process and prevent natural disasters and accidents damaging the power transmission line in advance and ensure normal power supply all the time becomes the focus of attention of people. The hidden dangers threatening the safety of the power transmission line mainly comprise violation buildings, ultrahigh trees, construction vehicles and the like, and related departments need to continuously patrol the line to search and process the hidden dangers possibly existing.
The conventional line patrol modes include manual line patrol, unmanned plane line patrol, online video monitoring and the like. A large amount of manpower resources are consumed for manual line patrol, and the efficiency is low; the flight distance of the unmanned aerial vehicle is limited, so that the line patrol mode of the unmanned aerial vehicle is difficult to popularize in remote unmanned areas; on-line video monitoring requires continuous shooting of videos and transmission of the videos back to a server, so that a lot of transmission flow is consumed, and the development of the on-line video monitoring is restricted.
There are also detection algorithms for identifying the hidden danger of the line through images, but in these detection algorithms, different detection methods need to be designed for different hidden danger types, a large amount of debugging work is needed during implementation, and after implementation, the difficulty of maintenance and management is also high due to the complex overall structure.
Disclosure of Invention
The application provides a method for detecting hidden dangers of a power transmission line, and solves the technical problems that different detection methods are needed to be designed for different hidden dangers in the existing method for detecting the hidden dangers of the power transmission line, and the implementation is difficult and troublesome.
In view of this, the present application provides a method for detecting hidden troubles of a power transmission line, including:
acquiring a to-be-detected map shot by satellite inspection;
inputting the mapping to be detected into a ground feature classification model to obtain a classification result map output by the ground feature classification model;
determining the hidden line danger in each ground object according to the distance between each ground object and the power transmission line in the classification result graph;
the ground feature classification model is obtained through the following steps:
s1, labeling different ground objects in the original remote sensing image in advance to obtain a labeled remote sensing image;
s2, establishing a training set by using the marked remote sensing image and the corresponding original remote sensing image;
s3, training the U-Net network by the training set;
s4, carrying out accuracy rate inspection on the model obtained after training through the verification set, if the accuracy rate is smaller than a preset value, adjusting the parameters of the current model, and returning to the step S3; and if the accuracy is greater than or equal to a preset value, outputting the ground feature classification model.
Preferably, the S3 further includes before:
and pre-training the weight of the U-Net network according to the VGG16 network, and taking the weight obtained after pre-training as the weight of the U-Net network.
Preferably, the contracted path of the U-Net network passes information from the front of the network to the back through a multi-line connection operation.
Preferably, the transmitting the information in front of the network to the rear by the multi-line connection operation specifically includes:
performing feature fusion on the feature images generated by the first group of convolutional layers and the input images to generate new feature images;
step X: taking the new characteristic image as an input image of a next group of convolution layers, and performing characteristic fusion with the characteristic image generated by the next group of convolution layers to obtain a new characteristic image;
and (4) circulating the step X until the next group of convolution layers is the last group of convolution layers, and outputting a new characteristic image obtained finally.
Preferably, the U-Net network performs feature extraction on targets with different sizes through various convolution kernels.
Preferably, the U-Net network performs feature extraction specifically by using a 3 × 3 convolution kernel corresponding to a target of a first preset size;
extracting features of the target with the second preset size through a 5 x 5 convolution kernel;
extracting features of the target with the third preset size corresponding to the 7 × 7 convolution kernel;
the first preset size is smaller than the second preset size, and the second preset size is smaller than the third preset size.
Preferably, the S3 further includes before:
and performing data amplification on the training data in the training set through rotation angle and turning, brightness and saturation adjustment, translation, noise addition and distortion to obtain an amplified training set.
Preferably, the S1 includes:
cutting an original remote sensing image in advance, removing irrelevant parts in the original remote sensing image, and labeling different ground objects in the cut original remote sensing image to obtain a labeled remote sensing image.
Preferably, the determining of the line hidden danger in each feature according to the distance between each feature and the power transmission line in the classification result map specifically includes:
calculating the distance between each object and the power transmission line in the classification result graph;
and for each ground object, judging whether the distance between the ground object and the power transmission line falls into a preset hidden danger distance range corresponding to the ground object, and if so, determining that the ground object is a line hidden danger.
Preferably, the acquiring of the to-be-detected map shot by the satellite tour specifically includes:
and acquiring a diagram to be detected shot by the satellite in a regular inspection tour.
According to the technical scheme, the method has the following advantages:
according to the power transmission line hidden danger detection method, a ground object classification model is established in advance. The ground feature classification model is established based on the U-Net network, and the marked remote sensing image and the corresponding original remote sensing image are used for training the U-Net network to obtain the ground feature classification model. The obtained ground object classification model fully utilizes the advantages of the U-Net network in semantic segmentation, can accurately classify the ground objects of the remote sensing image, can detect various ground objects which possibly cause circuit hidden dangers through the model, does not need to design a detection method aiming at a hidden danger type, greatly reduces the difficulty of realizing compared with the existing detection algorithm, and can meet the accuracy of ground object classification.
Drawings
Fig. 1 is a flowchart of a method for detecting hidden troubles of a power transmission line according to a first embodiment of the present application on an application level;
FIG. 2 is a flowchart of a method for building a terrain classification model according to a first embodiment of the present application;
fig. 3 is a flowchart of a method for detecting hidden troubles in a power transmission line according to a second embodiment of the present application on an application level;
FIG. 4 is a flowchart of a method for building a terrain classification model according to a second embodiment of the present application;
fig. 5 is a U-net network structure optimized based on multi-line operation and VGG16 network pre-training weights according to a second embodiment of the present application.
Detailed Description
In order to make the technical solutions of the present application better understood, the technical solutions of the present application will be clearly and completely described below with reference to the drawings in the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all 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 application.
The existing detection algorithm for the hidden danger of the power transmission line needs to design different detection methods according to different hidden danger types. The traditional image processing methods such as color and texture analysis, background modeling and the like need manual design features, so that not only are workers required to know the application fields and data of the traditional image processing methods, but also a large amount of debugging work is needed, and after the traditional image processing methods are realized, the detection results are greatly influenced by the surrounding environment; although the detection method based on the statistical model and the convolutional neural network can realize the detection of partial hidden dangers, the system needs to design different detection methods for different hidden danger types (such as buildings, trees, vehicles and the like), the system structure is complex, and the maintenance and management difficulty is high.
Therefore, the method for detecting the hidden danger of the power transmission line is provided, various ground objects which are possibly the hidden danger of the line in the remote sensing image can be obtained through classification and identification of one ground object classification model, and after the possible ground objects are identified, whether the ground objects belong to the hidden danger of the line can be judged through the distance between each ground object and the power transmission line.
Specifically, referring to fig. 1 and fig. 2, fig. 1 is a flowchart of a method for detecting hidden danger of a power transmission line according to a first embodiment of the present application on an application level, and fig. 2 is a flowchart of a method for establishing a ground feature classification model according to the first embodiment of the present application.
The method for detecting hidden danger of power transmission line provided by this embodiment includes, in an application layer:
The current line mode of patrolling commonly used (artifical line, unmanned aerial vehicle patrols line, online video monitoring etc.), no matter which equal efficiency is not high, need consume more manpower and materials. Therefore, the method and the device have the advantages that the hidden danger points under the lines such as trees, construction vehicles and illegal buildings are identified and risk assessment are carried out by means of the satellite remote sensing technology, the satellite patrol replaces the human patrol and the machine patrol, the daily line patrol pressure of teams and groups can be greatly reduced, and the line management and operation and maintenance level is improved.
On the basis of 'satellite patrol', the satellite is controlled, so that the satellite can be regularly patrolled, and hidden danger risks of a line corridor can be regionally checked.
And 102, inputting the mapping to be detected into a ground feature classification model to obtain a classification result map output by the ground feature classification model.
The ground feature classification model that this application was established can carry out ground feature classification with examining the mapping of examining that the satellite was shot, and each ground feature in the mapping of examining is examined in the discernment, like violating regulations building, superelevation trees, construction vehicle etc. these ground features can be through various modes highlight in the classification result picture.
And 103, determining the hidden danger of the line in each ground feature according to the distance between each ground feature and the power transmission line in the classification result diagram.
Specifically, the distance between each feature and the power transmission line can be calculated, and the preset hidden danger distance range corresponding to the feature is determined according to the type of the feature. The preset hidden danger distance range can be set according to the voltage grade of the power transmission line and the attribute of the ground object. For example, if the preset hidden danger distance range of the ground object of the tree class is set to be 5m to 6m, when the ground object is identified as the tree, if the distance between the ground object and the power transmission line is within 5m to 6m, the tree can be determined as the line hidden danger.
The following describes, with reference to fig. 2, a method for building a surface feature classification model provided in this embodiment, where the method includes:
In order to establish a ground feature classification model, materials for training are firstly required to be provided for the training of the model. The original remote sensing image shot by the satellite can be correctly marked by manpower or other modes in advance. Each image corresponds to an annotation file, the annotation file is the image file with the same size as the file to be annotated, and different ground objects are covered by different pixel values.
And 202, establishing a training set by using the labeled remote sensing images and the corresponding original remote sensing images.
And the marked image is the marked remote sensing image. A training set can be established by the marked remote sensing images and the corresponding original remote sensing images, and the U-Net network is trained through the training set.
And step 203, training the U-Net network by using the training set.
In order to carry out accuracy inspection on the trained model, an original remote sensing image can be selected in the verification set as the input of the current model, and a classification result image output by the current model is compared with the remote sensing image which is subjected to correct ground object classification and marking in the verification set. Through multiple application tests on the model, the accuracy of the current model can be calculated, if the accuracy reaches the standard, all parameters of the ground feature classification model are reasonable, and the ground feature classification model can be output. If the accuracy rate does not reach the standard, the initial parameters of the model need to be adjusted again, the training is carried out again, the model obtained by training needs to be verified again, and the steps are repeated until the accuracy rate of the current model reaches the standard.
According to the method for detecting the hidden danger of the power transmission line, a ground object classification model is established in advance. The ground feature classification model is established based on the U-Net network, and the marked remote sensing image and the corresponding original remote sensing image are used for training the U-Net network to obtain the ground feature classification model. The obtained ground object classification model fully utilizes the advantages of the U-Net network in semantic segmentation, can accurately classify the ground objects of the remote sensing image, can detect various ground objects which possibly cause circuit hidden dangers through the model, does not need to design a detection method aiming at a hidden danger type, greatly reduces the difficulty of realizing compared with the existing detection algorithm, and can meet the accuracy of ground object classification.
Referring to fig. 3 and fig. 4, fig. 3 is a flowchart of a transmission line hidden danger detection method according to a second embodiment of the present application on an application level, and fig. 4 is a flowchart of a ground feature classification model establishment method according to the second embodiment of the present application.
The method for detecting hidden danger of power transmission line provided by this embodiment includes, in an application layer:
and 301, acquiring a diagram to be detected which is shot by the satellite in a regular inspection tour.
The description of this step can be found in step 101 in the first embodiment described above.
And step 302, inputting the to-be-detected map into a ground feature classification model to obtain a classification result map output by the ground feature classification model.
The description of this step can be found in step 102 of the first embodiment described above.
And step 303, calculating the distance between each ground object and the power transmission line in the classification result graph.
And 304, judging whether the distance between the ground object and the power transmission line falls into a preset hidden danger distance range corresponding to the ground object or not for each ground object, and if so, determining that the ground object is a line hidden danger.
The description of the two steps can be referred to step 103 in the first embodiment described above.
The feature classification model provided in this embodiment will be explained below.
In the first embodiment, the feature classification model is obtained based on the training of the U-Net network, so that the method has the advantage of semantic segmentation of the U-Net network. On the basis, the U-Net network can be further optimized.
Considering that the classical U-net network starts training with random initialization weights, which avoid overfitting of the network training, the data set needs to be large enough, but the remote sensing image data of the transmission line is limited and far from the required order of magnitude. Therefore, network initialization training is carried out on the ImageNet data set, the U-Net network weight is pre-trained on the basis of the VGG16 network, the overfitting problem caused by insufficient data volume can be solved, and the detection performance of the U-Net network is improved.
In this embodiment, the U-Net network includes a contracted path to capture context information and a symmetric expanded path for precise positioning. The systolic path follows a typical convolutional network architecture, i.e., alternating convolution and pooling operations, and downsamples the feature maps step by step while increasing the number of feature maps layer by layer.
In particular, in order to multiplex and strengthen the features, ensure denser features and maximize the mobility of information, in this embodiment, the contraction path adopts a multi-line connection operation to transmit information in front of the network to the back. The input image is fused behind the feature image generated by each original group of convolution layers, the new feature image generated after feature fusion can directly use the information of the original input image, and the information of the original input image processed by the convolution layers is also used.
Referring to fig. 5, fig. 5 is a U-net network structure optimized based on multi-line operation and VGG16 network pre-training weights according to a second embodiment of the present application. The lower black curved arrow part in the figure is a multi-line connection operation, mainly realizing the connection of two or more inputs, i.e. multi-input to single output, and aiming at transmitting information in front of the network to the rear.
In particular, the following steps can be performed.
Step W: the feature images generated by the first set of convolutional layers can be feature fused with the input image to generate a new feature image.
Step X: and taking the new characteristic image as an input image of the next group of convolution layers, and performing characteristic fusion with the characteristic image generated by the next group of convolution layers to obtain a new characteristic image.
Step Y: and (5) circulating the step X until the next group of convolution layers is the last group of convolution layers, and outputting a new characteristic image obtained finally.
For the expanding path of the U-Net network, each stage consists of an up-sampling of a feature map followed by a convolution. Expanding the branches may improve the resolution of the output. For localization, the dilation path combines the upsampling feature with the high resolution feature from the contraction path by skipping connections.
Furthermore, the classical U-Net network adopts a single convolution kernel to extract features, and all layers are connected in series, so that the utilization rate of information is low, and the requirements of practical engineering application are difficult to satisfy. Also, using a fixed size convolution kernel, the size of the receptive field will be limited. Therefore, in this embodiment, convolution kernels of different sizes are used to perform feature extraction on targets of different sizes, so that sense fields of different sizes can be provided, and global information and local information can be combined more closely. Specifically, the present embodiment provides three convolution kernels, i.e., 3 × 3, 5 × 5, and 7 × 7, which are respectively used for predicting a small (target of a first preset size), a medium (target of a second preset size), and a large (target of a third preset size). The new network structure does not need to be retrained, and only network models with different scales are used for respectively predicting, and finally all available data are fused.
The network structure of the U-Net network semantic segmentation is an encoder-decoder structure, the encoder gradually reduces the spatial dimension, the decoder gradually restores the details and the spatial dimension of an object, and meanwhile, the encoder and the decoder are connected to help the decoder to better restore the details of a target. Encoders and decoders may be built based on VGG16 networks. Specifically, to construct the encoder, all fully connected layers of VGG16 are removed and replaced with a single convolutional layer, which is the bottleneck middle part of the network, separating the encoder and decoder. To construct a decoder, a transposed convolutional layer of twice the size of the feature map is used, while reducing the number of channels by half. The output of the transposed convolution is then connected to the output of the corresponding part of the decoder. The resulting signature is processed by convolution operations to keep the number of channels the same as the symmetric encoder terms.
Referring again to fig. 5, the entire U-Net network provided in this embodiment has 28 convolutional layers, including a contraction path and an expansion path, where the contraction path includes 3 × 3 convolutional layers, a ReLU activation function, and a max pooling layer for down-sampling, and each down-sampling reduces the channel of the feature map by half. The expansion path contains an upsampled (3 × 3 transposed convolution) halved feature map channel, a corresponding feature of the contraction path, then repeated 3 × 3 convolutions, a ReLU function, which cuts when copying the feature map of the contraction path because each convolution loses image edges, and finally a convolution of 1 × 1, so that the feature map with depth of 64 maps to a class label.
With reference to fig. 4, the method for building a surface feature classification model according to this embodiment includes:
In order to reduce the hardware pressure in the training process, the original remote sensing image can be preprocessed, namely, the part irrelevant to the hidden danger of the power transmission line in the original remote sensing image is proposed in advance. The rest may refer to step 201 in the first embodiment described above.
And 402, establishing a training set by using the marked remote sensing image and the corresponding original remote sensing image.
And 403, performing data amplification on the training data in the training set through rotation angle and inversion, brightness and saturation adjustment, translation, noise addition and distortion to obtain an amplified training set.
When the model is actually trained, the data enhancement can be carried out in the training data considering that the training data is not complete enough, so that the U-Net can be told what is needed to be the invariant required by the task and what is needed to be the attribute to be learned. Specific data amplification methods include, but are not limited to: rotation angle and flip, brightness and saturation adjustment, translation, noise addition, and distortion.
It should be noted that the data may be randomly divided into a training set, a validation set, and a test set during the training process.
And step 404, pre-training the weight of the U-Net network according to the VGG16 network, and taking the weight obtained after pre-training as the weight of the U-Net network.
As described above, after the weights of the U-Net network are pre-trained by the VGG16 network, the overfitting problem caused by insufficient data volume can be solved, and the detection performance of the U-Net network can be improved.
And 405, training the U-Net network by using the amplified training set.
In this embodiment, a feature classification model is established in advance. The ground feature classification model is based on a U-Net network and optimized on a classical U-Net network, and the method comprises the following steps: pre-training the weights of the U-net network based on the VGG16 network; and a network optimization strategy of multi-line connection operation and multi-convolution cores is provided.
The U-net network with the improved pre-training weight based on the VGG16 network solves the overfitting problem caused by insufficient data volume, meanwhile, the classification of various different ground objects such as illegal buildings, trees, vehicles and the like is achieved, the pixel-level classification of various ground objects can be accurately achieved, and the defects that an existing detection system is complex and is difficult to maintain and manage are effectively overcome.
The multi-line connection operation and multi-convolution kernel network optimization strategy further improves the U-net network, enhances the multiplexing and strengthening of characteristic information, maximizes the flowing of information, ensures more dense characteristics, lays a good data foundation for characteristic extraction and meets the practical requirement of hidden danger detection.
Meanwhile, the ground object classification model of the embodiment utilizes the advantages of the U-Net network in semantic segmentation, can accurately classify the ground objects of the remote sensing image, can detect various ground objects which possibly cause circuit hidden dangers through the model, does not need to design a detection method aiming at one hidden danger type, greatly reduces the difficulty of realization compared with the existing detection algorithm, and can meet the accuracy of ground object classification.
The terms "first," "second," "third," "fourth," and the like in the description of the application and the above-described figures, if any, are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used is interchangeable under appropriate circumstances such that the embodiments of the application described herein are, for example, capable of operation in sequences other than those illustrated or otherwise described herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.
The above embodiments are only used for illustrating the technical solutions of the present application, and not for limiting the same; although the present application has been described in detail with reference to the foregoing embodiments, it should be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions in the embodiments of the present application.
Claims (3)
1. A method for detecting hidden troubles of a power transmission line is characterized by comprising the following steps:
acquiring a diagram to be detected shot by a satellite tour at regular intervals;
inputting the mapping to be detected into a ground feature classification model to obtain a classification result map output by the ground feature classification model;
determining the hidden line danger in each ground object according to the distance between each ground object and the power transmission line in the classification result graph; the method specifically comprises the following steps: calculating the distance between each ground feature and the power transmission line, determining a preset hidden danger distance range corresponding to the ground feature according to the type of the ground feature, judging whether the distance between the ground feature and the power transmission line falls into the preset hidden danger distance range corresponding to the ground feature, and if so, determining that the ground feature is a line hidden danger; the preset hidden danger distance range can be set according to the voltage grade of the power transmission line and the type of ground objects;
the ground feature classification model is obtained through the following steps:
s1, labeling different ground objects in the original remote sensing image in advance to obtain a labeled remote sensing image;
s2, establishing a training set by using the marked remote sensing image and the corresponding original remote sensing image;
s3, pre-training the weight of the U-Net network according to the VGG16 network, and taking the weight obtained after pre-training as the weight of the U-Net network;
s4, training the U-Net network by the training set to obtain a terrain classification model based on the U-Net network; the U-Net network is based on a classical U-Net network, is optimized on the classical U-Net network, provides multi-line connection operation, and adopts various convolution checks to perform feature extraction on preset targets with different sizes;
the method specifically comprises the following steps: firstly, a contraction path adopts multi-line connection operation, information in front of a network is transmitted to the back, feature fusion is carried out on a feature image generated by a first group of convolution layers and an input image, a new feature image is generated after feature fusion, the new feature image is used as an input image of a next group of convolution layers and is subjected to feature fusion with the feature image generated by the next group of convolution layers to obtain a new feature image, the steps are circulated until the next group of convolution layers is a last group of convolution layers to obtain a new feature image, secondly, the U-Net network specifically carries out feature extraction on a target with a first preset size through a 3 x 3 convolution kernel, carries out feature extraction on a target with a second preset size through a 5 x 5 convolution kernel, carries out feature extraction on a target with a third preset size through a 7 x 7 convolution kernel, and the first preset size is smaller than the second preset size, the second preset size is smaller than the third preset size;
s5, carrying out accuracy rate inspection on the model obtained after training through the verification set, if the accuracy rate is smaller than a preset value, adjusting the parameters of the current model, and returning to the step S4; and if the accuracy is greater than or equal to a preset value, outputting the ground feature classification model.
2. The method for detecting hidden danger in power transmission line according to claim 1, wherein the step of S4 further comprises:
and performing data amplification on the training data in the training set through rotation angle and turning, brightness and saturation adjustment, translation, noise addition and distortion to obtain an amplified training set.
3. The method for detecting hidden danger in power transmission line according to claim 1, wherein the step S1 includes:
cutting an original remote sensing image in advance, removing irrelevant parts in the original remote sensing image, and labeling different ground objects in the cut original remote sensing image to obtain a labeled remote sensing image.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910289546.6A CN109977921B (en) | 2019-04-11 | 2019-04-11 | Method for detecting hidden danger of power transmission line |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910289546.6A CN109977921B (en) | 2019-04-11 | 2019-04-11 | Method for detecting hidden danger of power transmission line |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109977921A CN109977921A (en) | 2019-07-05 |
| CN109977921B true CN109977921B (en) | 2022-02-11 |
Family
ID=67084125
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910289546.6A Active CN109977921B (en) | 2019-04-11 | 2019-04-11 | Method for detecting hidden danger of power transmission line |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109977921B (en) |
Families Citing this family (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110705414A (en) * | 2019-09-24 | 2020-01-17 | 智洋创新科技股份有限公司 | Power transmission line construction machinery hidden danger detection method based on deep learning |
| CN112784632B (en) * | 2019-11-06 | 2023-05-30 | 济南信通达电气科技有限公司 | Method and device for detecting potential safety hazards of power transmission line |
| CN111027402B (en) * | 2019-11-15 | 2022-09-02 | 国网江西省电力有限公司电力科学研究院 | Intelligent detection and evaluation method for natural disaster risk |
| CN110929618B (en) * | 2019-11-15 | 2023-06-20 | 国网江西省电力有限公司电力科学研究院 | Potential safety hazard detection and assessment method for power distribution network crossing type building |
| CN111223071A (en) * | 2019-11-20 | 2020-06-02 | 智洋创新科技股份有限公司 | Power transmission line scene semantic understanding method based on deep learning technology |
| CN111325708B (en) * | 2019-11-22 | 2023-06-30 | 济南信通达电气科技有限公司 | Transmission line detection method and server |
| CN111062529A (en) * | 2019-12-10 | 2020-04-24 | 国网北京市电力公司 | Power line processing method and device, storage medium and processor |
| CN111695493B (en) * | 2020-06-10 | 2023-12-08 | 国网山东省电力公司电力科学研究院 | Method and system for detecting hidden danger of power transmission line |
| CN112380945B (en) * | 2020-11-06 | 2021-12-21 | 中国电力科学研究院有限公司 | Method and system for identifying potential hazards of power transmission channel environment based on satellite remote sensing image |
| CN112884026B (en) * | 2021-02-02 | 2023-05-12 | 贵州电网有限责任公司 | Image identification-assisted power transmission line laser LiDAR point cloud classification method |
| CN114067245A (en) * | 2021-11-16 | 2022-02-18 | 中国铁路兰州局集团有限公司 | Method and system for identifying hidden danger of external environment of railway |
| CN115272310B (en) * | 2022-09-26 | 2023-01-03 | 江苏智云天工科技有限公司 | Method and device for detecting defects of workpiece |
| CN116958907B (en) * | 2023-09-18 | 2023-12-26 | 四川泓宝润业工程技术有限公司 | Method and system for inspecting surrounding hidden danger targets of gas pipeline |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108037770A (en) * | 2017-11-22 | 2018-05-15 | 国网山东省电力公司济宁供电公司 | Unmanned plane power transmission line polling system and method based on artificial intelligence |
| CN108470185A (en) * | 2018-02-12 | 2018-08-31 | 北京佳格天地科技有限公司 | The atural object annotation equipment and method of satellite image |
| CN108537192A (en) * | 2018-04-17 | 2018-09-14 | 福州大学 | A kind of remote sensing image ground mulching sorting technique based on full convolutional network |
| CN108875561A (en) * | 2018-04-27 | 2018-11-23 | 山东信通电子股份有限公司 | A kind of hidden danger repetition method of discrimination of transmission line of electricity monitoring hidden danger early warning image |
-
2019
- 2019-04-11 CN CN201910289546.6A patent/CN109977921B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108037770A (en) * | 2017-11-22 | 2018-05-15 | 国网山东省电力公司济宁供电公司 | Unmanned plane power transmission line polling system and method based on artificial intelligence |
| CN108470185A (en) * | 2018-02-12 | 2018-08-31 | 北京佳格天地科技有限公司 | The atural object annotation equipment and method of satellite image |
| CN108537192A (en) * | 2018-04-17 | 2018-09-14 | 福州大学 | A kind of remote sensing image ground mulching sorting technique based on full convolutional network |
| CN108875561A (en) * | 2018-04-27 | 2018-11-23 | 山东信通电子股份有限公司 | A kind of hidden danger repetition method of discrimination of transmission line of electricity monitoring hidden danger early warning image |
Non-Patent Citations (1)
| Title |
|---|
| 基于无人机激光扫描的输电通道安全距离诊断技术;彭向阳 等;《电网技术》;20141130;第38卷(第11期);第3254-3259页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109977921A (en) | 2019-07-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109977921B (en) | Method for detecting hidden danger of power transmission line | |
| CN109118479B (en) | Capsule network-based insulator defect identification and positioning device and method | |
| Chakma et al. | Image-based air quality analysis using deep convolutional neural network | |
| CN108596277B (en) | Vehicle identity recognition method and device and storage medium | |
| CN111179345B (en) | Front vehicle line-crossing violation behavior automatic detection method and system based on vehicle-mounted machine vision | |
| CN111339882B (en) | Power transmission line hidden danger detection method based on example segmentation | |
| CN110084165B (en) | Intelligent identification and early warning method for abnormal events in open scene of power field based on edge calculation | |
| CN112837315B (en) | Deep learning-based transmission line insulator defect detection method | |
| CN110070530A (en) | A kind of powerline ice-covering detection method based on deep neural network | |
| CN110942456B (en) | Tamper image detection method, device, equipment and storage medium | |
| CN106339657A (en) | Straw incineration monitoring method and device based on monitoring video | |
| CN113642474A (en) | Hazardous area personnel monitoring method based on YOLOV5 | |
| CN113284144B (en) | Tunnel detection method and device based on unmanned aerial vehicle | |
| CN115861210B (en) | Transformer substation equipment abnormality detection method and system based on twin network | |
| CN112037142A (en) | Image denoising method and device, computer and readable storage medium | |
| Li et al. | Image manipulation localization using attentional cross-domain CNN features | |
| CN115082798A (en) | Power transmission line pin defect detection method based on dynamic receptive field | |
| CN109241893B (en) | Road selection method and device based on artificial intelligence technology and readable storage medium | |
| Manninen et al. | Multi-stage deep learning networks for automated assessment of electricity transmission infrastructure using fly-by images | |
| Katsamenis et al. | A Few-Shot Attention Recurrent Residual U-Net for Crack Segmentation | |
| CN116994161A (en) | Insulator defect detection method based on improved YOLOv5 | |
| CN116543333A (en) | Target recognition method, training method, device, equipment and medium of power system | |
| CN115690505A (en) | Photovoltaic module fault detection method and device, computer equipment and storage medium | |
| CN111898564B (en) | Time sequence convolution network model, model training method and device and target recognition method and device | |
| CN116434100A (en) | Liquid leakage detection method, device, system and machine-readable storage medium |
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 |