CN112651454B - Infrared data acquisition system and spiral data processing method for power equipment - Google Patents
Infrared data acquisition system and spiral data processing method for power equipment Download PDFInfo
- Publication number
- CN112651454B CN112651454B CN202011616638.XA CN202011616638A CN112651454B CN 112651454 B CN112651454 B CN 112651454B CN 202011616638 A CN202011616638 A CN 202011616638A CN 112651454 B CN112651454 B CN 112651454B
- Authority
- CN
- China
- Prior art keywords
- data
- module
- training
- infrared
- power equipment
- 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
- 238000003672 processing method Methods 0.000 title claims abstract description 17
- 238000012549 training Methods 0.000 claims abstract description 72
- 238000000034 method Methods 0.000 claims abstract description 19
- 238000012216 screening Methods 0.000 claims description 36
- 238000012360 testing method Methods 0.000 claims description 18
- 230000009466 transformation Effects 0.000 claims description 14
- 238000011176 pooling Methods 0.000 claims description 9
- 238000004422 calculation algorithm Methods 0.000 claims description 7
- 238000004891 communication Methods 0.000 claims description 7
- 230000003993 interaction Effects 0.000 claims description 7
- 230000005284 excitation Effects 0.000 claims description 6
- 230000002708 enhancing effect Effects 0.000 claims description 4
- 238000005070 sampling Methods 0.000 claims description 4
- 238000013519 translation Methods 0.000 claims description 4
- 230000004913 activation Effects 0.000 claims description 3
- 238000013528 artificial neural network Methods 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- 230000006835 compression Effects 0.000 claims description 2
- 238000007906 compression Methods 0.000 claims description 2
- 238000012546 transfer Methods 0.000 claims description 2
- 238000011160 research Methods 0.000 abstract description 10
- 238000012545 processing Methods 0.000 abstract description 9
- 238000012423 maintenance Methods 0.000 abstract description 8
- 238000013135 deep learning Methods 0.000 abstract description 6
- 238000013473 artificial intelligence Methods 0.000 abstract description 5
- 238000012986 modification Methods 0.000 abstract description 4
- 230000004048 modification Effects 0.000 abstract description 4
- 230000008569 process Effects 0.000 abstract description 4
- 230000007774 longterm Effects 0.000 abstract description 2
- 238000012544 monitoring process Methods 0.000 abstract description 2
- 238000013461 design Methods 0.000 description 7
- 238000010586 diagram Methods 0.000 description 4
- 230000006870 function Effects 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000010276 construction Methods 0.000 description 2
- 230000005855 radiation Effects 0.000 description 2
- 238000011426 transformation method Methods 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 1
- 238000013434 data augmentation Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 230000002401 inhibitory effect Effects 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 239000013307 optical fiber Substances 0.000 description 1
- 238000012634 optical imaging Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- 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
- G06Q—INFORMATION 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
- G06Q10/00—Administration; Management
- G06Q10/20—Administration of product repair or maintenance
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION 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
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Economics (AREA)
- General Engineering & Computer Science (AREA)
- Human Resources & Organizations (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Strategic Management (AREA)
- General Business, Economics & Management (AREA)
- Tourism & Hospitality (AREA)
- Marketing (AREA)
- General Health & Medical Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Primary Health Care (AREA)
- Computational Linguistics (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computing Systems (AREA)
- Entrepreneurship & Innovation (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
Abstract
The invention provides an infrared data acquisition system and a spiral data processing method for power equipment, which realize a semi-automatic data processing mode by working with experienced workers (such as experts), improve the processing efficiency of a data set, reduce manual operation and save a large amount of manpower and data processing cost. The invention fills the blank of the infrared image data set of the power equipment, provides data for the research and the modification of the infrared image data set of the power equipment on artificial intelligence and deep learning, and lays a foundation for the long-term research of the power equipment and the research of the power equipment in other fields. According to the invention, the infrared images of the power equipment are collected at any time by monitoring for a long time in the operation process of the power equipment, the maintenance, operation and maintenance time of the power equipment is matched, the collection and training do not need to be carried out in a centralized manner in a short time, and the maintenance cost is reduced.
Description
Technical Field
The invention belongs to the technical field of electric power, and particularly relates to an infrared data acquisition system and a spiral data processing method for electric power equipment.
Background
At present, power enterprises increase the force of automatic construction of power equipment, infrared images are in an important position in detecting faults of power transformation equipment, systems based on thermal infrared imagers or unmanned aerial vehicles are mostly adopted in the industry to collect the infrared images of the power equipment, and target identification and diagnosis are carried out through a traditional image processing method, so that fault occurrence points are measured to guarantee stable and reliable operation of the equipment. However, the prior art has the following defects:
1) Infrared image data sets related to power equipment are few or no, and research on artificial intelligence and deep learning cannot be conducted;
2) In the prior art, the infrared image data set of the power equipment is marked in an artificial mode, so that the consumption of manpower and material resources is high, and intelligent transformation is needed.
Therefore, the acquisition system and the data sorting method for the infrared image data set of the power equipment need to be improved.
Disclosure of Invention
The technical problem to be solved by the invention is as follows: the utility model provides a power equipment infrared data acquisition system and spiral data processing method for improve the processing efficiency to the data set.
The technical scheme adopted by the invention for solving the technical problems is as follows: an infrared data acquisition system of power equipment comprises an infrared sensor, a data acquisition module, a data enhancement module, an expert screening module and a data training module; the signal output end of the infrared sensor is connected with the signal input end of the data acquisition module, and the infrared sensor is used for acquiring an infrared image of the power equipment and sending the infrared image to the data acquisition module; the signal output end of the data acquisition module is connected with the signal input end of the expert screening module, and the data acquisition module is used for storing information including infrared images of the power equipment, the number of the infrared images and the infrared temperature of the power equipment and sending the information to the expert screening module; the signal output end of the expert screening module is connected with the signal input end of the data enhancement module, and the expert screening module is a human-computer interaction module and is used for sorting data and judging a trained result so as to realize semi-automatic sorting of the infrared image data set; the data enhancement module is used for enhancing the received infrared image data and preventing overfitting and data set shortage; the signal input end of the data training module is connected with the signal output end of the data enhancement module, and the signal output end of the data training module is connected with the signal input end of the data acquisition module; the initial model of the data training module is a network training model and is used for training the enhanced and manually screened data sets to obtain and store the trained use model.
According to the scheme, the device further comprises a controller, a communication module, an input module and an output module; the signal control end of the controller is respectively connected with the infrared sensor, the data acquisition module, the data enhancement module, the expert screening module and the data training module and is used for controlling information transfer among the modules; the communication module is connected between the infrared sensor and the data acquisition module in series and is used for ensuring the communication stability of the infrared sensor and the data acquisition module; the signal output end of the input module is connected with the signal input end of the expert screening module and used for converting the operation of an expert into a control signal and sending the control signal to the expert screening module; the signal input end of the output module is connected with the signal output end of the data acquisition module and used for converting the data set into a form visible to an operator so that the operator can observe the data acquisition condition in real time.
According to the scheme, the device further comprises a data query module, wherein the signal transceiving end of the data query module is connected with the signal transceiving end of the data acquisition module and is used for an operator to query the information stored in the data acquisition module.
A spiral data processing method comprises the following steps:
s1: acquiring infrared image data of the field power equipment through an infrared sensor, and transmitting the data to a data acquisition module in real time;
s2: the data acquisition module stores data, the expert screening module, the data enhancement module and the data training module call the data according to data flow, and an infrared image data set of the power equipment is formed by arranging through a human-computer interaction semi-automatic arrangement method.
Further, in the step S2, the specific steps are as follows:
s21: performing primary training on the acquired data to obtain a training data set;
s22: carrying out spiral training on data classification through a network model and expert assistance, and improving the precision of the network model to obtain a final network model;
s23: and automatically classifying the acquired data by applying the final network model to obtain an infrared image data set of the power equipment.
Further, in the step S21, the specific steps are:
s211: when the data capacity stored by the data acquisition module reaches a preset capacity value, sending the data to the expert screening module;
s212: an expert selects first training data from the data to a specified data set through an expert screening module and sends the first training data to a data enhancement module;
s213: the data enhancement module enhances data by using a data enhancement mode of geometric transformation including translation, rotation, random cutting, deformation and scaling to form a training data set and sends the training data set to the data training module;
s214: the data training module trains an SE-ResNet network model by adopting the enhanced data and the manually screened data;
s215: judging the accuracy of the SE-ResNet network model to the data classification, and if the accuracy of the SE-ResNet network model to the data classification reaches 50% of the accuracy of the test set data, finishing the primary training; if the accuracy of the test set data is not 50%, step S214 is executed.
Further, in step S215, the test set is a data set formed by randomly sampling 50% of the existing data set.
Further, in step S22, the specific steps are:
s221: judging whether the classification result reaches 98% of the accuracy of the data in the test set according to the data acquired by sorting the SE-ResNet network model after the initial training, and if so, finishing the algorithm training; if not, the expert performs a second screening;
s222: collecting misjudged data to form a misjudged data set, adding new classified data into the original data set, judging whether the capacity of the data set reaches a preset capacity value, if not, performing data enhancement again to form a new data set and training an SE-ResNet network model; if so, directly training the SE-ResNet network model;
s223: combining a new data set test SE-ResNet network model through the test set and the misjudgment data set, judging the accuracy of the SE-ResNet network model on data classification by an expert, and completing spiral training if the requirements are met to obtain a final network model; if not, step S222 is executed until the final network model is obtained.
Further, the SE-ResNet network is a ResNet network with an SE module; the SE module is used for carrying out feature recalibration, after obtaining output U from input X through convolution, carrying out global average pooling on each channel of U, obtaining a weight value of each channel by utilizing two full-connection layers, and carrying out reweighing on U according to the channels to obtain final output; the ResNet network is a residual network comprising a convolutional layer, a batch normalization layer and a nonlinear activation function, and is used for performing feature compression on a feature map of an infrared image on a spatial dimension by using a global average pooling feature channel at a residual output part of the ResNet network, compressing each feature channel into a real number with a global receptive field, and finally compressing all the feature channels into a real number set; and then, adding an excitation step to perform global average pooling to obtain a feature map with a global receptive field, performing nonlinear transformation by using a fully-connected neural network through excitation operation, and multiplying the result serving as weight to the input features.
Further, the method also comprises the following steps:
s3: the data query module requests the data acquisition module to query the data set, the data acquisition module sends the data set to the output module, and the output module converts the data set into a form visible to an operator for the operator to observe the data acquisition condition in real time.
The beneficial effects of the invention are as follows:
1. according to the infrared data acquisition system and the spiral data processing method for the power equipment, disclosed by the invention, through the mutual work with experienced workers (such as experts), a semi-automatic data processing mode is realized, the processing efficiency of a data set is improved, the manual operation is reduced, and a large amount of manpower and data processing cost are saved.
2. The invention fills the blank of the infrared image data set of the power equipment, provides data for the research and the modification of the infrared image data set of the power equipment on artificial intelligence and deep learning, and lays a foundation for the long-term research of the power equipment and the research of the power equipment in other fields.
3. According to the invention, the infrared images of the power equipment are collected at any time by monitoring for a long time in the operation process of the power equipment, the maintenance, operation and maintenance time of the power equipment is matched, the collection and training do not need to be carried out in a centralized manner in a short time, and the maintenance cost is reduced.
Drawings
FIG. 1 is a functional block diagram of hardware of an embodiment of the present invention.
FIG. 2 is a functional block diagram of software of an embodiment of the present invention.
Fig. 3 is a block diagram of an SE-ResNet according to an embodiment of the present invention.
FIG. 4 is a diagram of a network spiral training mode according to an embodiment of the present invention.
Fig. 5 is a flow chart of an embodiment of the present invention.
Detailed Description
The invention is described in further detail below with reference to the drawings and the detailed description.
Referring to fig. 1, the embodiment of the present invention includes a hardware device and a software application system combined together, the hardware device collects infrared image data of a field power device and transmits a data set to a software platform in real time, and the software platform organizes the image data set to form a parallel flow with the hardware device; the human-computer interaction interface of the software platform is used for observing the data acquisition condition in real time; after the data are collected, a large number of data sets are sorted through a semi-automatic sorting method in a man-machine interaction mode, a foundation is laid for research and modification of an infrared image processing process of power equipment in the fields of artificial intelligence and deep learning, the workload of manual operation for sorting the data is reduced, and the labor cost is saved.
1. Hardware framework of data acquisition system
The infrared thermal imager is used for collecting the infrared image of the electric equipment, a high-performance computer is used for data processing, a hardware framework of a data collecting system is shown as figure 1, the infrared thermal imager consists of a plurality of common elements, but mainly comprises two components, namely a lens and a detector, and an infrared detector and an optical imaging objective lens are used for receiving the infrared radiation energy distribution pattern of a detected target and reflecting the infrared radiation energy distribution pattern on a photosensitive element of the infrared detector so as to obtain an infrared thermal image, and the thermal image corresponds to the thermal distribution field on the surface of an object, so that the infrared image is formed. The system adopts gigabit optical fiber Ethernet based on a TCP/IP protocol as a communication network, thereby realizing the real-time performance of the thermal infrared imager and a computer software platform and better carrying out information sharing.
2. Software platform
In order to sort the collected data set, a set of software for sorting the infrared image data set is specially designed to realize a semi-automatic sorting method, and the modules of the method comprise a data collecting module, a data enhancing module, a data training module, an expert screening module and a data query module, which are shown in figure 2.
A data acquisition module: the infrared thermal imager is associated with the infrared thermal imager and the data enhancement module, the infrared thermal imager inputs infrared images to the data acquisition module, the module records the number of the infrared images and the infrared temperature of the power equipment, the input infrared images of the power equipment are stored, an operator can set related parameters such as the size of the images to be acquired by the infrared thermal imager, the number of the images and the like, and the operator can check the infrared images acquired by the infrared thermal imager so as to operate the infrared thermal imager; the infrared image output by the module is input by the data enhancement module, so that the data source of the data enhancement module is ensured.
The data enhancement module: the module is set for preventing overfitting and few acquired data sets, mainly takes data output by the acquisition module as input, enhances the input infrared image data set and lays a foundation for the data training module.
A data training module: the module is used for realizing one step in the semi-automatic sorting method, comprises two parts of a network training model and a use model, adopts a data set after data enhancement and manual screening for training, and uses the trained model.
An expert screening module: the module is a man-machine interaction module, and aims to realize one step in a semi-automatic sorting method, the expert referred to in the text is an experienced power grid worker and well knows the faults of power equipment and power equipment, and can also be understood as a power equipment maintenance worker, the expert can complete two works through the module, one work is to complete a small amount of screening of infrared images, and the other work is to judge the result after training.
A data query module: when the infrared image data set of the power equipment is in the software platform, a worker can inquire the data set through the module, after the infrared image data set is processed by the module, the worker or a software user can inquire the data set, and the target data set is selected and downloaded from the software platform, so that the target data set can be conveniently used as a data source in other research directions.
3. Core algorithm
In order to complete the complete design of the software platform, the algorithm of related modules needs to be designed, the main modules have data enhancement and data training, and the design scheme is as follows:
(1) Data enhancement algorithm design
Data enhancement, also known as data augmentation, is to prevent overfitting during training of a data set, and in the present invention can be understood colloquially as changing an infrared image or images into multiple infrared images. Common data enhancement modes comprise geometric transformation, color transformation and generation based on GAN, wherein the geometric transformation method is to perform geometric transformation on an image and comprises various operations such as turning, rotating, cutting, deforming, zooming and the like; the color transformation method changes the content of the image itself, which is a part of the selected image or redistributes pixels, and commonly includes noise, blur, color transformation, erasure, filling, and the like; a data enhancement mode based on GAN belongs to unsupervised learning, and data enhancement is carried out by adopting a generation network and a countermeasure network.
The invention mainly aims at infrared images, color transformation is not needed, a generation mode based on GAN needs a large amount of data for training, and the color transformation mode and the generation mode based on GAN are not suitable for the invention, so the invention adopts a geometric transformation mode to enhance the data, and mainly comprises five enhancement modes of translation, rotation, random cutting, deformation and scaling, wherein the translation is to move the images along the directions of an x axis and a y axis or two directions; the rotation is according to clockwise or anticlockwise rotation; the random cutting adopts two modes, one mode is to randomly select a part from the image, then cut the part of the image out and adjust the part of the image into the size of the original image, and the other mode is to enlarge the image, wherein the size of the enlarged image is larger than the original size, and then cut the enlarged image according to the original size. After the infrared image of the power equipment is processed, the purpose of data enhancement can be achieved.
(2) Data training algorithm design
The invention adopts an SE-ResNet network for training, wherein the SE-ResNet network is provided with an SE module in the ResNet network. After an SE module obtains an output U from an input X through convolution, global average pooling is conducted on each channel of the U, then a weight value of each channel is obtained through two layers of full connection layers, the U is re-weighted according to the channels to obtain a final output, and the process is called as characteristic re-calibration. The SE module has the advantages of being easy to integrate with the existing deep learning network, enhancing beneficial features, inhibiting useless features and being small in calculation amount; the ResNet network is a residual network, which mainly comprises a convolution layer, a batch normalization layer and a nonlinear activation function.
The structure of the SE-ResNet network is shown in fig. 3, a feature graph of an infrared image is feature-compressed in a spatial dimension by using a global-average pooling feature channel in a residual output part of the ResNet network, each feature channel is compressed into a real number with a global receptive field, and finally, all the feature channels are compressed into a real number set, and then, an excitation step is added to perform global-average pooling to obtain a feature graph, so that the feature graph has the global receptive field, and then, a result after the previous step is nonlinearly transformed by excitation operation and using a fully-connected neural network, and finally, the result is used as a weight and multiplied to an input feature. The invention is based on SE-ResNet network, works with expert screening module, adopts wave type, spiral type training mode, the training mode is shown in figure 4.
4. Description of the preferred embodiment
The construction and connection mode of hardware (infrared thermal imager and computer) can be operated by referring to relevant product specifications, and detailed description is omitted. The invention mainly designs an electric power equipment infrared image data set acquisition method, realizes the function of acquiring an electric power equipment infrared image data set, provides a data source for artificial intelligence and deep learning research, and provides a semi-automatic arrangement method aiming at the marking of the image data set, thereby reducing the manual operation. The invention focuses on the design of a software platform and a software core algorithm, a software algorithm flow chart is shown in FIG. 5, and the implementation mode can be summarized into the following three steps:
the method comprises the following steps that an infrared thermal imager or an unmanned aerial vehicle is used for obtaining infrared images, when the data capacity reaches a certain capacity, namely 20% of a data set required by one-time training is achieved, an expert (experienced personnel or power equipment maintenance personnel) selects a first batch of training data (a small amount of about 200) from the collected data set through an expert screening module, and then data enhancement (about 1500 pieces of training data can be obtained by using five data enhancement modes of geometric transformation) is carried out to form a training data set;
and secondly, after training is carried out by using the SE-Net network, a network assistance expert sorts data from the collected data into a specified data set. After the SE-Net training is completed, the expert determines whether the network is available, with the determination requirement being that 50% accuracy of the data test of each test set can be achieved, where a test set is a data set formed by randomly sampling 50% of the existing data set. And when the judgment result is usable, classifying and screening the original data through the network of the training network, after the network screening is finished, performing expert-check type secondary screening, collecting misjudged data to form a misjudged data set, and adding new classified data into the original data set.
And thirdly, when the number of the data sets reaches 200, performing data enhancement again to form a new data set (stopping the data enhancement after the number of the original data sets reaches 20000), retraining the network by using the new data set, comparing the accuracy of the network by an expert, and combining the test set and the misjudgment data set into a new data set for testing, wherein the test set is a data set formed by performing 50% random sampling on the existing data set, and the misjudgment data set is a misjudgment data set formed by collecting misjudgment data. After the network is trained, the network and the experts are combined to screen data again, so that the network precision is promoted in a wavy and spiral manner, the workload of the experts is gradually reduced until the experts are not required to examine at last, and the network automatically finishes the classification of the data.
The above embodiments are only used for illustrating the design idea and features of the present invention, and the purpose of the present invention is to enable those skilled in the art to understand the content of the present invention and implement the present invention accordingly, and the protection scope of the present invention is not limited to the above embodiments. Therefore, all equivalent changes and modifications made in accordance with the principles and concepts disclosed herein are intended to be included within the scope of the present invention.
Claims (8)
1. A spiral data processing method for infrared data of electrical equipment is characterized by comprising the following steps: the method comprises the following steps:
s0: building an infrared data acquisition system of the power equipment, wherein the infrared data acquisition system comprises an infrared sensor, a data acquisition module, a data enhancement module, an expert screening module and a data training module;
the signal output end of the infrared sensor is connected with the signal input end of the data acquisition module;
the signal output end of the data acquisition module is connected with the signal input end of the expert screening module;
the signal output end of the expert screening module is connected with the signal input end of the data enhancement module;
the signal input end of the data training module is connected with the signal output end of the data enhancement module, and the signal output end of the data training module is connected with the signal input end of the data acquisition module;
s1: acquiring infrared image data of the field power equipment through an infrared sensor, and transmitting the data to a data acquisition module in real time;
s2: the data acquisition module stores data, the expert screening module, the data enhancement module and the data training module call the data according to data flow, and an infrared image data set of the power equipment is formed through a human-computer interaction semi-automatic sorting method; the method comprises the following specific steps:
s21: performing primary training on the acquired data to obtain a training data set;
s22: performing spiral training on data classification through a network model and expert assistance, and improving the precision of the network model to obtain a final network model; the method comprises the following specific steps:
s221: judging whether the classification result reaches 98% of the accuracy of the data in the test set according to the data acquired by sorting the SE-ResNet network model after the initial training, and if so, finishing the algorithm training; if not, the expert performs a second screening;
s222: collecting misjudged data to form a misjudged data set, adding new classified data into the original data set, judging whether the capacity of the data set reaches a preset capacity value, if not, performing data enhancement again to form a new data set and training an SE-ResNet network model; if so, directly training the SE-ResNet network model;
s223: combining a new data set test SE-ResNet network model through the test set and the misjudgment data set, judging the accuracy of the SE-ResNet network model on data classification by an expert, and completing spiral training if the accuracy meets the requirement to obtain a final network model; if not, executing step S222 until obtaining the final network model;
s23: and automatically classifying the acquired data by using the final network model to obtain an infrared image data set of the power equipment.
2. The spiral data processing method for the infrared data of the power equipment according to claim 1, wherein the spiral data processing method comprises the following steps: in the step S21, the specific steps are:
s211: when the data capacity stored by the data acquisition module reaches a preset capacity value, sending the data to the expert screening module;
s212: an expert selects first training data from the data to a specified data set through an expert screening module and sends the first training data to a data enhancement module;
s213: the data enhancement module enhances data by using a data enhancement mode of geometric transformation including translation, rotation, random cutting, deformation and scaling to form a training data set and sends the training data set to the data training module;
s214: the data training module adopts the enhanced data and the manually screened data to train an SE-ResNet network model;
s215: judging the accuracy of the SE-ResNet network model to the data classification, and if the accuracy of the SE-ResNet network model to the data classification reaches 50% of the accuracy of the test set data, finishing the primary training; if the accuracy of the test set data is less than 50%, step S214 is executed.
3. The spiral data processing method for the infrared data of the power equipment according to claim 2, wherein the spiral data processing method comprises the following steps: in step S215, the test set is a data set formed by randomly sampling 50% of the existing data set.
4. The spiral data processing method for infrared data of electric power equipment according to claim 2 or claim 1, characterized in that: the SE-ResNet network is a ResNet network with an SE module;
the SE module is used for carrying out feature recalibration, after obtaining output U from input X through convolution, carrying out global average pooling on each channel of U, obtaining a weight value of each channel by utilizing two full-connection layers, and carrying out reweighing on U according to the channels to obtain final output;
the ResNet network is a residual network comprising a convolutional layer, a batch normalization layer and a nonlinear activation function, and is used for performing feature compression on a feature map of an infrared image on a spatial dimension by using a global average pooling feature channel at a residual output part of the ResNet network, compressing each feature channel into a real number with a global receptive field, and finally compressing all the feature channels into a real number set; and then, adding an excitation step to perform global average pooling to obtain a feature map with a global receptive field, performing nonlinear transformation by using a fully-connected neural network through excitation operation, and multiplying the result serving as weight to the input features.
5. The spiral data processing method for the infrared data of the power equipment according to claim 1, wherein the spiral data processing method comprises the following steps: further comprising the steps of:
s3: the data query module requests the data acquisition module to query the data set, the data acquisition module sends the data set to the output module, and the output module converts the data set into a form visible to an operator for the operator to observe the data acquisition condition in real time.
6. An electric power equipment infrared data acquisition system used for the spiral data processing method of electric power equipment infrared data in any one of claims 1 to 5, characterized in that: the system comprises an infrared sensor, a data acquisition module, a data enhancement module, an expert screening module and a data training module;
the signal output end of the infrared sensor is connected with the signal input end of the data acquisition module, and the infrared sensor is used for acquiring an infrared image of the power equipment and sending the infrared image to the data acquisition module;
the signal output end of the data acquisition module is connected with the signal input end of the expert screening module, and the data acquisition module is used for storing information including infrared images of the power equipment, the number of the infrared images and the infrared temperature of the power equipment and sending the information to the expert screening module;
the signal output end of the expert screening module is connected with the signal input end of the data enhancement module, and the expert screening module is a man-machine interaction module and is used for sorting data and judging a trained result so as to realize semi-automatic sorting of the infrared image data set;
the data enhancement module is used for enhancing the received infrared image data and preventing overfitting and data set shortage;
the signal input end of the data training module is connected with the signal output end of the data enhancement module, and the signal output end of the data training module is connected with the signal input end of the data acquisition module; the initial model of the data training module is a network training model and is used for training the enhanced and manually screened data sets to obtain and store a trained use model.
7. The infrared data acquisition system of the power equipment as recited in claim 6, wherein: the device also comprises a controller, a communication module, an input module and an output module;
the signal control end of the controller is respectively connected with the infrared sensor, the data acquisition module, the data enhancement module, the expert screening module and the data training module and is used for controlling information transfer among the modules;
the communication module is connected between the infrared sensor and the data acquisition module in series and is used for ensuring the communication stability of the infrared sensor and the data acquisition module;
the signal output end of the input module is connected with the signal input end of the expert screening module and used for converting the operation of an expert into a control signal and sending the control signal to the expert screening module;
the signal input end of the output module is connected with the signal output end of the data acquisition module and used for converting the data set into a form visible to an operator so that the operator can observe the data acquisition condition in real time.
8. The infrared data acquisition system of the power equipment as recited in claim 6, wherein: the data acquisition module is used for acquiring the information stored in the data acquisition module, and the data acquisition module is used for acquiring the information stored in the data acquisition module.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011616638.XA CN112651454B (en) | 2020-12-31 | 2020-12-31 | Infrared data acquisition system and spiral data processing method for power equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011616638.XA CN112651454B (en) | 2020-12-31 | 2020-12-31 | Infrared data acquisition system and spiral data processing method for power equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112651454A CN112651454A (en) | 2021-04-13 |
| CN112651454B true CN112651454B (en) | 2022-11-29 |
Family
ID=75364557
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011616638.XA Active CN112651454B (en) | 2020-12-31 | 2020-12-31 | Infrared data acquisition system and spiral data processing method for power equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112651454B (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111949809A (en) * | 2020-07-16 | 2020-11-17 | 广东电网有限责任公司 | Intelligent processing method for infrared inspection data of power transmission line |
| CN112149634A (en) * | 2020-10-23 | 2020-12-29 | 北京百度网讯科技有限公司 | Training method, device and equipment of image generator and storage medium |
Family Cites Families (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8972310B2 (en) * | 2012-03-12 | 2015-03-03 | The Boeing Company | Method for identifying structural deformation |
| US9489598B2 (en) * | 2014-08-26 | 2016-11-08 | Qualcomm Incorporated | Systems and methods for object classification, object detection and memory management |
| CN105300528A (en) * | 2015-10-12 | 2016-02-03 | 国家电网公司 | Infrared image diagnosis method and infrared image diagnosis system for transformer station equipment |
| CN109448009A (en) * | 2018-11-21 | 2019-03-08 | 国网江苏省电力有限公司扬州供电分公司 | Infrared Image Processing Method and device for transmission line faultlocating |
| CN110136132A (en) * | 2019-05-24 | 2019-08-16 | 国网河北省电力有限公司沧州供电分公司 | Detection method, detection device and the terminal device of equipment heating failure |
| CN110276394A (en) * | 2019-06-21 | 2019-09-24 | 扬州大学 | Power equipment classification method based on deep learning under a kind of small sample |
| CN110378424A (en) * | 2019-07-23 | 2019-10-25 | 国网河北省电力有限公司电力科学研究院 | Bushing shell for transformer failure Infrared image recognition based on convolutional neural networks |
-
2020
- 2020-12-31 CN CN202011616638.XA patent/CN112651454B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111949809A (en) * | 2020-07-16 | 2020-11-17 | 广东电网有限责任公司 | Intelligent processing method for infrared inspection data of power transmission line |
| CN112149634A (en) * | 2020-10-23 | 2020-12-29 | 北京百度网讯科技有限公司 | Training method, device and equipment of image generator and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112651454A (en) | 2021-04-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112734692B (en) | Defect identification method and device for power transformation equipment | |
| CN107688856B (en) | Indoor robot scene active identification method based on deep reinforcement learning | |
| CN114973032B (en) | Deep convolutional neural network-based photovoltaic panel hot spot detection method and device | |
| CN113205039B (en) | Power equipment fault image recognition disaster investigation system and method based on multiple DCNN networks | |
| CN112528979B (en) | Transformer substation inspection robot obstacle distinguishing method and system | |
| CN112367400B (en) | Intelligent inspection method and system for power internet of things with edge cloud coordination | |
| CN112232328A (en) | Remote sensing image building area extraction method and device based on convolutional neural network | |
| CN113408087A (en) | Substation inspection method based on cloud side system and video intelligent analysis | |
| Chen et al. | Surface defect detection of electric power equipment in substation based on improved YOLOv4 algorithm | |
| CN113837994B (en) | Photovoltaic panel defect diagnosis method based on edge detection convolutional neural network | |
| CN112348003A (en) | Airplane refueling scene recognition method and system based on deep convolutional neural network | |
| CN113436184A (en) | Power equipment image defect judging method and system based on improved twin network | |
| CN115100554A (en) | Unmanned aerial vehicle power inspection system based on intelligent vision and detection method thereof | |
| CN113570571A (en) | Industrial edge end power battery defect detection method and system | |
| CN109297978A (en) | The inspection of power circuit unmanned plane and fault intelligence diagnosis system based on binocular imaging | |
| CN113256620B (en) | Vehicle body welding quality information judging method based on difference convolution neural network | |
| CN117056865B (en) | Method and device for diagnosing operation faults of machine pump equipment based on feature fusion | |
| CN112651454B (en) | Infrared data acquisition system and spiral data processing method for power equipment | |
| CN116485802B (en) | Insulator flashover defect detection method, device, equipment and storage medium | |
| CN116168019B (en) | Power grid fault detection method and system based on machine vision technology | |
| CN113284103B (en) | Substation equipment defect online detection method based on space transformation fast R-CNN model | |
| CN111259981B (en) | Automatic classification system after remote sensing image processing | |
| CN115293952A (en) | Hardware crimping defect evaluation system and method based on unmanned aerial vehicle load DR digital imaging | |
| CN115439566B (en) | Compression sensing system and method based on memory and calculation integrated architecture | |
| CN115456989A (en) | High-precision aircraft skin defect detection identification method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |