CN112115986A - Power transmission line scene classification method based on lightweight neural network - Google Patents

Power transmission line scene classification method based on lightweight neural network Download PDF

Info

Publication number
CN112115986A
CN112115986A CN202010893898.5A CN202010893898A CN112115986A CN 112115986 A CN112115986 A CN 112115986A CN 202010893898 A CN202010893898 A CN 202010893898A CN 112115986 A CN112115986 A CN 112115986A
Authority
CN
China
Prior art keywords
neural network
transmission line
power transmission
lightweight neural
network
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
CN202010893898.5A
Other languages
Chinese (zh)
Other versions
CN112115986B (en
Inventor
李弘宸
杨忠
姜遇红
韩家明
张驰
赖尚祥
方千慧
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.)
Nanjing University of Aeronautics and Astronautics
Original Assignee
Nanjing University of Aeronautics and Astronautics
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 Nanjing University of Aeronautics and Astronautics filed Critical Nanjing University of Aeronautics and Astronautics
Priority to CN202010893898.5A priority Critical patent/CN112115986B/en
Publication of CN112115986A publication Critical patent/CN112115986A/en
Application granted granted Critical
Publication of CN112115986B publication Critical patent/CN112115986B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • G06F18/20Analysing
    • G06F18/24Classification techniques
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/04Architecture, e.g. interconnection topology
    • G06N3/045Combinations of networks
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/04Architecture, e.g. interconnection topology
    • G06N3/048Activation functions
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/08Learning methods
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06QINFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/06Energy or water supply
    • YGENERAL 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
    • Y04INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
    • Y04SSYSTEMS 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
    • Y04S10/00Systems supporting electrical power generation, transmission or distribution
    • Y04S10/50Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Health & Medical Sciences (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Computational Linguistics (AREA)
  • Mathematical Physics (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Business, Economics & Management (AREA)
  • Software Systems (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Economics (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Public Health (AREA)
  • Water Supply & Treatment (AREA)
  • Human Resources & Organizations (AREA)
  • Marketing (AREA)
  • Primary Health Care (AREA)
  • Strategic Management (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a power transmission line scene classification method based on a lightweight neural network, which comprises the following steps of: (1) configuring the operating environment of the power transmission line scene classification system; (2) designing an optimization prediction module containing a convolution layer to build a lightweight neural network; (3) capturing aerial images of the power transmission line by using a camera carried by an unmanned aerial vehicle, constructing a scene classification data set of the power transmission line, and dividing the aerial images of the power transmission line into a training set and a test set; (4) inputting the images of the training set into the lightweight neural network built in the step (2), training the lightweight neural network, and storing the obtained network weight; (5) and (4) inputting the test set image into the network model obtained in the step (4) for testing, and verifying the performance of the network model. The method has strong adaptability to the complex scene of the power transmission line, compared with VGG-16, the FAST-VGG-16 has low memory occupation ratio and high classification precision, invalid information is effectively removed, and the unmanned aerial vehicle inspection efficiency is improved.

Description

Power transmission line scene classification method based on lightweight neural network
Technical Field
The invention belongs to the field of pattern recognition, and particularly relates to a power transmission line scene classification method based on a lightweight neural network.
Background
With the continuous development of modern society, the daily life and industry of human beings have great dependence on electric power. To meet the demand for electric power in modern society, the length of transmission lines is increasing year by year. In order to guarantee the normal operation of transmission line, the circuit is patrolled and examined absolutely. Because the security of artifical patrolling and examining is poor, and is inefficient, so in recent years artifical patrolling and examining is patrolled and examined gradually by unmanned aerial vehicle and is replaced.
However, in the unmanned aerial vehicle inspection process, a camera can capture a large number of images which do not contain key components of the power transmission line. Removing these invalid data can improve the inspection efficiency, so it is necessary to classify the transmission line scene. The traditional classification mainly depends on mechanical learning algorithms such as a support vector machine and a neural network, the working modes of the traditional mechanical learning algorithms such as the support vector machine depend on an image feature extraction algorithm, and the features obtained by the feature extraction algorithm directly influence the classification precision of the traditional machine learning. However, the image features extracted by the feature extraction algorithm have a large difference from the high-level semantics of the image, and the generalization is poor, which results in that the traditional machine learning algorithm has a poor effect in classifying complex scenes. The power transmission line scene comprises various targets such as towers, trees, grasslands and buildings, and the power transmission line scene characteristics change greatly along with the change of external conditions such as illumination, so that the traditional machine learning is difficult to meet the performance requirements of the power transmission line scene classification problem.
With the development of the convolutional neural network, the deep learning obtains remarkable results in various fields such as semantic segmentation, target detection, image classification and the like. Common classification networks include: AlexNet, GoogleNet, VGG, ResNet, and the like. The convolutional neural network does not depend on an artificial image feature extraction algorithm, image features are directly extracted through a convolutional layer and a pooling layer, finally, the image categories are predicted by utilizing a full-connection layer, network weights are modified through a gradient descent method, and the neural network can deal with the classification problem of complex scenes through the working mode. However, the full connection layer for prediction contains a large number of parameters, and the excessive parameters can cause the network operation speed to be reduced and the occupied memory to be increased, so that the full connection layer is not beneficial to being carried on the unmanned aerial vehicle platform.
Disclosure of Invention
The purpose of the invention is as follows: the invention aims to provide a power transmission line scene classification method based on a lightweight neural network, which can remove invalid data and improve the inspection efficiency of an unmanned aerial vehicle.
The technical scheme is as follows: the invention discloses a power transmission line scene classification method based on a lightweight neural network, which comprises the following steps of:
(1) configuring the operating environment of the power transmission line scene classification system;
(2) designing an optimization prediction module containing a convolutional layer, and building a lightweight neural network by using the optimization prediction module;
(3) constructing a scene classification data set of the power transmission line, and dividing aerial images of the power transmission line into a training set and a test set;
(4) inputting the images of the training set into the lightweight neural network built in the step (2), training the lightweight neural network, and storing the obtained network weight;
(5) and (4) inputting the test set image into the network model obtained in the step (4) for testing, and verifying the performance of the network model.
In step (2), the optimized prediction module includes two 3 × 3 convolutional layers, a 1 × 1 convolutional layer, two BN layers, and two maximum pooling layers.
In the optimization prediction module, sigmoid is adopted by the last layer of the convolution layer as an activation function, and Leaky ReLU is adopted by the rest of the convolution layers as the activation function.
The expression of the activation function leak ReLU is as follows:
Figure BDA0002657807170000021
in the formula, a represents a fixed parameter in the interval (1, + ∞).
And (4) training the lightweight neural network by using a random gradient descent method.
The training of the lightweight neural network by using the stochastic gradient descent method specifically comprises the following steps:
(4.1) activating a neural network function library Tensorflow, constructing a Fast-VGG-16 network, and activating a neural computation acceleration library cuDNN; .
(4.2) inputting the training set picture into a neural network in a tensor form, wherein the positive samples and the negative samples are respectively stored in different files;
(4.3) calculating the gradient descending direction theta' by the following formula:
Figure BDA0002657807170000022
in the formula, Θ represents the network weight before updating; thetajRepresenting updated network weights; α represents an initial learning rate; i-1 denotes the initial number of iterations, i-imaxThe maximum number of iterations is indicated.
(4.4) inputting the training set picture into a Fast-VGG-16 network by using the SGD training strategy, and training;
(4.5) Loop step S44 until a maximum number of iterations i is reachedmaxAnd ending the loop and saving the network weight.
Has the advantages that: compared with the prior art, the invention has the beneficial effects that: the method has strong adaptability to complex scenes of the power transmission line, compared with VGG-16, FAST-VGG-16 has lower memory occupation ratio and higher classification precision, invalid information can be effectively removed, and the unmanned aerial vehicle inspection efficiency is improved.
Drawings
FIG. 1 is a schematic flow diagram of the present invention;
FIG. 2 is an aerial photography scenario of the power transmission line of the present invention;
FIG. 3 is a network structure diagram of the lightweight neural network Fast-VGG-16 constructed by the invention.
Detailed Description
The invention is described in further detail below with reference to specific embodiments and the attached drawings.
As shown in fig. 1, the power transmission line scene classification method based on the lightweight neural network of the present invention includes the following steps:
(1) configuring the operating environment of the power transmission line scene classification system; the operating environment of the power transmission line scene classification system comprises a software system and a hardware system, wherein the software system comprises a windows10 operating system, a Tensorflow open source software library, a CUDA and a convolutional neural network acceleration library cuDNN; the hardware system adopts NVIDIA Jetson AGX Xavier.
(2) Designing an optimization prediction module containing a convolution layer, and building a lightweight neural network by using the optimization prediction module; the optimized prediction module does not comprise a full connection layer, and consists of two 3 x 3 convolution layers, a 1 x 1 convolution layer, two Batch Normalization (BN) layers and two maximum pooling layers, in the optimized prediction module, except that the activation function adopted by the last convolution layer is sigmoid, the rest layers are all adopted as activation functions, and the module is used for replacing the original prediction to build FAST-VGG-16.
The expression of the activation function Leaky ReLU in the optimization prediction module is as follows:
Figure BDA0002657807170000031
in the formula, a represents a fixed parameter in the interval (1, + ∞).
The light weight neural network Fast-VGG-16 takes VGG-16 as a reference network, a traditional prediction module is replaced by an optimized prediction module, network parameters and network operation loss are reduced, and the network configuration of the Fast-VGG-16 is shown in Table 1:
TABLE 1 network configuration Table for Fast-VGG-16
Figure BDA0002657807170000032
Figure BDA0002657807170000041
(3) Capturing aerial images of the power transmission line by using a camera carried by an unmanned aerial vehicle, establishing a scene classification data set of the power transmission line, and establishing a training set and a testing set according to a ratio of 3: 1.
(4) Inputting the images of the training set into the lightweight neural network built in the step (2), training the lightweight neural network by using a random gradient descent method, and storing the obtained network weight; the method comprises the following specific steps:
(4.1) activating a neural network function library Tensorflow, constructing a lightweight neural network, and activating a neural computation acceleration library cuDNN;
(4.2) inputting the training set images into the lightweight neural network in a tensor form; wherein, the positive sample and the negative sample are respectively stored in different files; as shown in fig. 2, graph (a) shows a positive sample, and graph (b) shows a negative sample;
(4.3) calculating the gradient descending direction theta' by the following formula:
Figure BDA0002657807170000051
in the formula, Θ represents the network weight before updating; thetajRepresenting updated network weights; α represents an initial learning rate; i-1 denotes the initial number of iterations, i-imaxThe maximum number of iterations is indicated.
(4.4) inputting the training set picture into a Fast-VGG-16 network by using the SGD training strategy, and training;
(4.5) Loop step S44 until a maximum number of iterations i is reachedmaxEnding the cycle and savingNetwork weight.
(5) And (4) inputting the test set image into the Fast-VGG-16 network model obtained in the step (4) for testing, and verifying the performance of the Fast-VGG-16 network model. As shown in FIG. 3, in order to verify the performance of the network, the test set pictures containing the positive and negative samples are input into the Fast-VGG-16 and VGG-16 models which are trained, so as to obtain the classification accuracy of the two networks.
TABLE 2 Fast-VGG-16 vs VGG-16 network Performance
Figure BDA0002657807170000052

Claims (6)

1. A power transmission line scene classification method based on a lightweight neural network is characterized by comprising the following steps:
(1) configuring the operating environment of the power transmission line scene classification system;
(2) designing an optimization prediction module containing a convolutional layer, and building a lightweight neural network by using the optimization prediction module;
(3) capturing aerial images of the power transmission line by using a camera carried by an unmanned aerial vehicle, constructing a scene classification data set of the power transmission line, and dividing the aerial images of the power transmission line into a training set and a test set;
(4) inputting the images of the training set into the lightweight neural network built in the step (2), training the lightweight neural network, and storing the obtained network weight;
(5) and (4) inputting the test set image into the network model obtained in the step (4) for testing, and verifying the performance of the network model.
2. The power transmission line scene classification method based on the lightweight neural network as claimed in claim 1, wherein: in step (2), the optimized prediction module includes two 3 × 3 convolutional layers, a 1 × 1 convolutional layer, two BN layers, and two maximum pooling layers.
3. The power transmission line scene classification method based on the lightweight neural network as claimed in claim 2, wherein: in the optimization prediction module, sigmoid is adopted by the last layer of the convolution layer as an activation function, and Leaky ReLU is adopted by the rest of the convolution layers as the activation function.
4. The power transmission line scene classification method based on the lightweight neural network as claimed in claim 2, wherein the expression of the activation function Leaky ReLU is as follows:
Figure FDA0002657807160000011
in the formula, a represents a fixed parameter in the interval (1, + ∞).
5. The power transmission line scene classification method based on the lightweight neural network as claimed in claim 1, wherein: and (4) training the lightweight neural network by using a random gradient descent method.
6. The power transmission line scene classification method based on the lightweight neural network as claimed in claim 5, wherein the training of the lightweight neural network by using the stochastic gradient descent method specifically comprises the following steps:
(4.1) activating a neural network function library Tensorflow, constructing a lightweight neural network, and activating a neural computation acceleration library cuDNN;
(4.2) inputting the training set images into the lightweight neural network in a tensor form; wherein, the positive sample and the negative sample are respectively stored in different files;
(4.3) calculating the gradient descending direction theta' by the following formula:
Figure FDA0002657807160000021
in the formula, Θ represents the network weight before updating; thetajRepresenting updated network weights; alpha meterInitializing an initial learning rate; i-1 denotes the initial number of iterations, i-imaxThe maximum number of iterations is indicated.
(4.4) inputting the training set picture into a Fast-VGG-16 network by using the SGD training strategy, and training;
(4.5) Loop step S44 until a maximum number of iterations i is reachedmaxAnd ending the loop and saving the network weight.
CN202010893898.5A 2020-08-31 2020-08-31 Lightweight neural network-based power transmission line scene classification method Active CN112115986B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010893898.5A CN112115986B (en) 2020-08-31 2020-08-31 Lightweight neural network-based power transmission line scene classification method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010893898.5A CN112115986B (en) 2020-08-31 2020-08-31 Lightweight neural network-based power transmission line scene classification method

Publications (2)

Publication Number Publication Date
CN112115986A true CN112115986A (en) 2020-12-22
CN112115986B CN112115986B (en) 2024-04-16

Family

ID=73804216

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010893898.5A Active CN112115986B (en) 2020-08-31 2020-08-31 Lightweight neural network-based power transmission line scene classification method

Country Status (1)

Country Link
CN (1) CN112115986B (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113486865A (en) * 2021-09-03 2021-10-08 国网江西省电力有限公司电力科学研究院 Power transmission line suspended foreign object target detection method based on deep learning
CN113569672A (en) * 2021-07-16 2021-10-29 国网电力科学研究院有限公司 Lightweight target detection and fault identification method, device and system

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107274451A (en) * 2017-05-17 2017-10-20 北京工业大学 Isolator detecting method and device based on shared convolutional neural networks
CN110349146A (en) * 2019-07-11 2019-10-18 中原工学院 The building method of fabric defect identifying system based on lightweight convolutional neural networks
CN110827251A (en) * 2019-10-30 2020-02-21 江苏方天电力技术有限公司 Power transmission line locking pin defect detection method based on aerial image

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107274451A (en) * 2017-05-17 2017-10-20 北京工业大学 Isolator detecting method and device based on shared convolutional neural networks
CN110349146A (en) * 2019-07-11 2019-10-18 中原工学院 The building method of fabric defect identifying system based on lightweight convolutional neural networks
CN110827251A (en) * 2019-10-30 2020-02-21 江苏方天电力技术有限公司 Power transmission line locking pin defect detection method based on aerial image

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113569672A (en) * 2021-07-16 2021-10-29 国网电力科学研究院有限公司 Lightweight target detection and fault identification method, device and system
CN113486865A (en) * 2021-09-03 2021-10-08 国网江西省电力有限公司电力科学研究院 Power transmission line suspended foreign object target detection method based on deep learning

Also Published As

Publication number Publication date
CN112115986B (en) 2024-04-16

Similar Documents

Publication Publication Date Title
CN110084281B (en) Image generation method, neural network compression method, related device and equipment
CN112115986B (en) Lightweight neural network-based power transmission line scene classification method
WO2023280113A1 (en) Data processing method, training method for neural network model, and apparatus
CN112861602A (en) Face living body recognition model compression and transplantation method based on depth separable convolution
CN112597324A (en) Image hash index construction method, system and equipment based on correlation filtering
Zhang et al. A real-time recognition method of static gesture based on DSSD
CN110599459A (en) Underground pipe network risk assessment cloud system based on deep learning
CN115222950A (en) Lightweight target detection method for embedded platform
CN116630608A (en) Multi-mode target detection method for complex scene
CN111008979A (en) Robust night image semantic segmentation method
WO2023282569A1 (en) Method and electronic device for generating optimal neural network (nn) model
CN111401405B (en) Image classification method and system integrated by multiple neural networks
CN116824270A (en) Small sample vehicle type recognition method based on attention mechanism improvement
CN117576540A (en) Low-illumination target detection method based on convolutional neural network
CN111126303B (en) Multi-parking-place detection method for intelligent parking
CN117132910A (en) Vehicle detection method and device for unmanned aerial vehicle and storage medium
CN116824330A (en) Small sample cross-domain target detection method based on deep learning
CN116310902A (en) Unmanned aerial vehicle target detection method and system based on lightweight neural network
CN116246305A (en) Pedestrian retrieval method based on hybrid component transformation network
CN114612758A (en) Target detection method based on deep grouping separable convolution
Tan et al. An application of an improved FCOS algorithm in detection and recognition of industrial instruments
Zhu et al. Research on the innovative application of Shen Embroidery cultural heritage based on convolutional neural network
CN114120046B (en) Lightweight engineering structure crack identification method and system based on phantom convolution
Zhao et al. Rapid and accurate object detection on drone based embedded devices with dilated, deformable and pyramid convolution
WO2022131399A1 (en) Method and device for training bsdcn designed by optimizing deformable convolutional network using static calculation scheme, and test method and apparatus using same

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